diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -10,7 +10,7 @@
   "log_history": [
     {
       "epoch": 0.0010471204188481676,
-      "grad_norm": 4.721594539542488,
+      "grad_norm": 4.721671007179121,
       "learning_rate": 1.7421602787456448e-07,
       "logits/chosen": -12.0,
       "logits/rejected": -9.25,
@@ -25,52 +25,52 @@
     },
     {
       "epoch": 0.010471204188481676,
-      "grad_norm": 4.7892951137734485,
+      "grad_norm": 4.790387210407091,
       "learning_rate": 1.7421602787456445e-06,
       "logits/chosen": -11.0,
       "logits/rejected": -8.5,
       "logps/chosen": -404.0,
       "logps/rejected": -1784.0,
       "loss": 0.6903,
-      "rewards/accuracies": 0.1736111044883728,
-      "rewards/chosen": 0.000179290771484375,
-      "rewards/margins": 0.002532958984375,
-      "rewards/rejected": -0.00238037109375,
+      "rewards/accuracies": 0.2083333283662796,
+      "rewards/chosen": 0.00098419189453125,
+      "rewards/margins": 0.00102996826171875,
+      "rewards/rejected": -5.435943603515625e-05,
       "step": 10
     },
     {
       "epoch": 0.020942408376963352,
-      "grad_norm": 4.259957666358221,
+      "grad_norm": 4.264809061935147,
       "learning_rate": 3.484320557491289e-06,
       "logits/chosen": -10.8125,
       "logits/rejected": -8.6875,
-      "logps/chosen": -388.0,
+      "logps/chosen": -386.0,
       "logps/rejected": -1632.0,
-      "loss": 0.6431,
-      "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -0.000972747802734375,
+      "loss": 0.6433,
+      "rewards/accuracies": 0.831250011920929,
+      "rewards/chosen": -0.000232696533203125,
       "rewards/margins": 0.111328125,
-      "rewards/rejected": -0.1123046875,
+      "rewards/rejected": -0.11181640625,
       "step": 20
     },
     {
       "epoch": 0.031413612565445025,
-      "grad_norm": 1.9901924501746173,
+      "grad_norm": 1.976605333324745,
       "learning_rate": 5.226480836236934e-06,
       "logits/chosen": -11.5625,
       "logits/rejected": -9.5625,
       "logps/chosen": -416.0,
       "logps/rejected": -1576.0,
-      "loss": 0.386,
+      "loss": 0.3854,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -0.099609375,
-      "rewards/margins": 0.796875,
+      "rewards/chosen": -0.10009765625,
+      "rewards/margins": 0.80078125,
       "rewards/rejected": -0.8984375,
       "step": 30
     },
     {
       "epoch": 0.041884816753926704,
-      "grad_norm": 1.3595644043240016,
+      "grad_norm": 1.3587404881008218,
       "learning_rate": 6.968641114982578e-06,
       "logits/chosen": -12.25,
       "logits/rejected": -10.625,
@@ -85,10 +85,10 @@
     },
     {
       "epoch": 0.05235602094240838,
-      "grad_norm": 0.6267546378111436,
+      "grad_norm": 0.6279240439086688,
       "learning_rate": 8.710801393728225e-06,
       "logits/chosen": -14.75,
-      "logits/rejected": -12.9375,
+      "logits/rejected": -12.875,
       "logps/chosen": -528.0,
       "logps/rejected": -2544.0,
       "loss": 0.0648,
@@ -100,12 +100,12 @@
     },
     {
       "epoch": 0.06282722513089005,
-      "grad_norm": 0.7708331975276697,
+      "grad_norm": 0.7722392402397799,
       "learning_rate": 1.0452961672473868e-05,
-      "logits/chosen": -15.5,
+      "logits/chosen": -15.5625,
       "logits/rejected": -13.8125,
       "logps/chosen": -436.0,
-      "logps/rejected": -2688.0,
+      "logps/rejected": -2672.0,
       "loss": 0.0809,
       "rewards/accuracies": 0.981249988079071,
       "rewards/chosen": -1.0625,
@@ -115,4059 +115,4059 @@
     },
     {
       "epoch": 0.07329842931937172,
-      "grad_norm": 0.5156432978583906,
+      "grad_norm": 0.5347511232996419,
       "learning_rate": 1.2195121951219513e-05,
       "logits/chosen": -15.1875,
       "logits/rejected": -13.9375,
       "logps/chosen": -556.0,
       "logps/rejected": -2928.0,
-      "loss": 0.0399,
+      "loss": 0.0395,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.3671875,
-      "rewards/margins": 12.0625,
-      "rewards/rejected": -13.4375,
+      "rewards/chosen": -1.375,
+      "rewards/margins": 12.125,
+      "rewards/rejected": -13.5,
       "step": 70
     },
     {
       "epoch": 0.08376963350785341,
-      "grad_norm": 0.6947872868390979,
+      "grad_norm": 0.6751099879284532,
       "learning_rate": 1.3937282229965156e-05,
-      "logits/chosen": -14.25,
+      "logits/chosen": -14.1875,
       "logits/rejected": -13.25,
       "logps/chosen": -532.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.043,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -1.46875,
-      "rewards/margins": 11.625,
-      "rewards/rejected": -13.125,
+      "logps/rejected": -2736.0,
+      "loss": 0.0436,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.484375,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -12.875,
       "step": 80
     },
     {
       "epoch": 0.09424083769633508,
-      "grad_norm": 3.820673845719557,
+      "grad_norm": 7.789217326708327,
       "learning_rate": 1.56794425087108e-05,
-      "logits/chosen": -13.125,
-      "logits/rejected": -12.375,
-      "logps/chosen": -588.0,
-      "logps/rejected": -3552.0,
-      "loss": 0.0296,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.328125,
-      "rewards/margins": 18.0,
-      "rewards/rejected": -20.375,
+      "logits/chosen": -13.25,
+      "logits/rejected": -12.5625,
+      "logps/chosen": -592.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0307,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.375,
+      "rewards/margins": 18.625,
+      "rewards/rejected": -21.0,
       "step": 90
     },
     {
       "epoch": 0.10471204188481675,
-      "grad_norm": 0.7059660250016567,
+      "grad_norm": 0.7454476987429359,
       "learning_rate": 1.742160278745645e-05,
-      "logits/chosen": -10.6875,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -616.0,
-      "logps/rejected": -3168.0,
-      "loss": 0.0575,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -1.609375,
-      "rewards/margins": 14.8125,
-      "rewards/rejected": -16.375,
+      "logits/chosen": -10.9375,
+      "logits/rejected": -9.875,
+      "logps/chosen": -600.0,
+      "logps/rejected": -3328.0,
+      "loss": 0.0531,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -1.4609375,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -17.875,
       "step": 100
     },
     {
       "epoch": 0.10471204188481675,
-      "eval_logits/chosen": -10.0,
-      "eval_logits/rejected": -9.75,
-      "eval_logps/chosen": -442.0,
-      "eval_logps/rejected": -444.0,
-      "eval_loss": 0.6934062242507935,
-      "eval_rewards/accuracies": 0.583984375,
-      "eval_rewards/chosen": -1.2421875,
-      "eval_rewards/margins": 0.302734375,
-      "eval_rewards/rejected": -1.546875,
-      "eval_runtime": 49.4073,
-      "eval_samples_per_second": 40.48,
-      "eval_steps_per_second": 0.648,
+      "eval_logits/chosen": -9.9375,
+      "eval_logits/rejected": -9.6875,
+      "eval_logps/chosen": -426.0,
+      "eval_logps/rejected": -432.0,
+      "eval_loss": 0.7023749947547913,
+      "eval_rewards/accuracies": 0.580078125,
+      "eval_rewards/chosen": -1.078125,
+      "eval_rewards/margins": 0.34765625,
+      "eval_rewards/rejected": -1.4296875,
+      "eval_runtime": 48.7538,
+      "eval_samples_per_second": 41.022,
+      "eval_steps_per_second": 0.656,
       "step": 100
     },
     {
       "epoch": 0.11518324607329843,
-      "grad_norm": 0.24235289134114754,
+      "grad_norm": 0.3941882059756442,
       "learning_rate": 1.9163763066202093e-05,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -486.0,
-      "logps/rejected": -3408.0,
-      "loss": 0.0301,
+      "logits/chosen": -11.125,
+      "logits/rejected": -10.25,
+      "logps/chosen": -484.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0305,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -0.984375,
-      "rewards/margins": 15.5,
-      "rewards/rejected": -16.5,
+      "rewards/chosen": -0.96484375,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -19.875,
       "step": 110
     },
     {
       "epoch": 0.1256544502617801,
-      "grad_norm": 0.4499987385913587,
+      "grad_norm": 0.4675274017224052,
       "learning_rate": 2.0905923344947736e-05,
-      "logits/chosen": -13.375,
-      "logits/rejected": -12.5625,
-      "logps/chosen": -520.0,
-      "logps/rejected": -3408.0,
-      "loss": 0.0248,
+      "logits/chosen": -12.625,
+      "logits/rejected": -11.75,
+      "logps/chosen": -446.0,
+      "logps/rejected": -3168.0,
+      "loss": 0.0258,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.6875,
-      "rewards/margins": 17.25,
-      "rewards/rejected": -19.0,
+      "rewards/chosen": -0.95703125,
+      "rewards/margins": 15.6875,
+      "rewards/rejected": -16.625,
       "step": 120
     },
     {
       "epoch": 0.13612565445026178,
-      "grad_norm": 0.5271972587321407,
+      "grad_norm": 1.0490131755991037,
       "learning_rate": 2.264808362369338e-05,
-      "logits/chosen": -14.125,
-      "logits/rejected": -13.25,
-      "logps/chosen": -596.0,
-      "logps/rejected": -3856.0,
-      "loss": 0.0599,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -13.5,
+      "logps/chosen": -628.0,
+      "logps/rejected": -4224.0,
+      "loss": 0.0673,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.0625,
-      "rewards/margins": 21.125,
-      "rewards/rejected": -23.25,
+      "rewards/chosen": -2.359375,
+      "rewards/margins": 24.5,
+      "rewards/rejected": -26.875,
       "step": 130
     },
     {
       "epoch": 0.14659685863874344,
-      "grad_norm": 1.0084732817461486,
+      "grad_norm": 0.2824517609810153,
       "learning_rate": 2.4390243902439026e-05,
-      "logits/chosen": -15.125,
-      "logits/rejected": -13.875,
-      "logps/chosen": -592.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0865,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -1.578125,
-      "rewards/margins": 17.0,
-      "rewards/rejected": -18.625,
+      "logits/chosen": -15.5625,
+      "logits/rejected": -14.4375,
+      "logps/chosen": -556.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.0588,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -1.1953125,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -13.5,
       "step": 140
     },
     {
       "epoch": 0.15706806282722513,
-      "grad_norm": 0.8945331654422062,
+      "grad_norm": 1.1407298977563938,
       "learning_rate": 2.6132404181184672e-05,
-      "logits/chosen": -13.75,
-      "logits/rejected": -11.9375,
-      "logps/chosen": -536.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.0289,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.2421875,
-      "rewards/margins": 14.0625,
-      "rewards/rejected": -15.3125,
+      "logits/chosen": -15.0625,
+      "logits/rejected": -13.4375,
+      "logps/chosen": -540.0,
+      "logps/rejected": -3264.0,
+      "loss": 0.026,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -1.296875,
+      "rewards/margins": 15.125,
+      "rewards/rejected": -16.375,
       "step": 150
     },
     {
       "epoch": 0.16753926701570682,
-      "grad_norm": 4.258702140374296,
+      "grad_norm": 3.798082365553247,
       "learning_rate": 2.7874564459930312e-05,
-      "logits/chosen": -13.125,
-      "logits/rejected": -11.125,
+      "logits/chosen": -15.125,
+      "logits/rejected": -13.0,
       "logps/chosen": -494.0,
-      "logps/rejected": -3088.0,
-      "loss": 0.0765,
+      "logps/rejected": -3136.0,
+      "loss": 0.0763,
       "rewards/accuracies": 0.9937499761581421,
       "rewards/chosen": -1.1171875,
-      "rewards/margins": 13.625,
-      "rewards/rejected": -14.75,
+      "rewards/margins": 14.0,
+      "rewards/rejected": -15.125,
       "step": 160
     },
     {
       "epoch": 0.17801047120418848,
-      "grad_norm": 0.5978788723629429,
+      "grad_norm": 0.6406066650684676,
       "learning_rate": 2.9616724738675962e-05,
-      "logits/chosen": -12.625,
-      "logits/rejected": -11.4375,
-      "logps/chosen": -512.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.0249,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -0.99609375,
-      "rewards/margins": 11.25,
-      "rewards/rejected": -12.25,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -13.1875,
+      "logps/chosen": -524.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.0247,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -1.1171875,
+      "rewards/margins": 13.25,
+      "rewards/rejected": -14.375,
       "step": 170
     },
     {
       "epoch": 0.18848167539267016,
-      "grad_norm": 0.6885389951751066,
+      "grad_norm": 1.639128374248885,
       "learning_rate": 3.13588850174216e-05,
-      "logits/chosen": -12.6875,
-      "logits/rejected": -11.75,
-      "logps/chosen": -580.0,
-      "logps/rejected": -3360.0,
-      "loss": 0.0499,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -1.640625,
-      "rewards/margins": 15.625,
-      "rewards/rejected": -17.25,
+      "logits/chosen": -15.125,
+      "logits/rejected": -14.0,
+      "logps/chosen": -676.0,
+      "logps/rejected": -4096.0,
+      "loss": 0.0519,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.609375,
+      "rewards/margins": 22.125,
+      "rewards/rejected": -24.75,
       "step": 180
     },
     {
       "epoch": 0.19895287958115182,
-      "grad_norm": 0.2175573324841809,
+      "grad_norm": 0.43125493799011,
       "learning_rate": 3.310104529616725e-05,
-      "logits/chosen": -13.375,
-      "logits/rejected": -11.9375,
-      "logps/chosen": -536.0,
-      "logps/rejected": -3392.0,
-      "loss": 0.0729,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -1.6328125,
-      "rewards/margins": 15.0625,
-      "rewards/rejected": -16.75,
+      "logits/chosen": -14.25,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -524.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0674,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -1.515625,
+      "rewards/margins": 15.375,
+      "rewards/rejected": -16.875,
       "step": 190
     },
     {
       "epoch": 0.2094240837696335,
-      "grad_norm": 0.5318831863482857,
+      "grad_norm": 0.6709842514271963,
       "learning_rate": 3.48432055749129e-05,
-      "logits/chosen": -14.9375,
-      "logits/rejected": -13.25,
-      "logps/chosen": -628.0,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -11.5625,
+      "logps/chosen": -612.0,
       "logps/rejected": -3200.0,
-      "loss": 0.0427,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.96875,
+      "loss": 0.0523,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -1.8203125,
       "rewards/margins": 14.1875,
-      "rewards/rejected": -16.125,
+      "rewards/rejected": -16.0,
       "step": 200
     },
     {
       "epoch": 0.2094240837696335,
-      "eval_logits/chosen": -15.0,
-      "eval_logits/rejected": -14.8125,
-      "eval_logps/chosen": -430.0,
-      "eval_logps/rejected": -418.0,
-      "eval_loss": 0.7579062581062317,
-      "eval_rewards/accuracies": 0.50390625,
-      "eval_rewards/chosen": -1.125,
-      "eval_rewards/margins": 0.16015625,
-      "eval_rewards/rejected": -1.28125,
-      "eval_runtime": 46.638,
-      "eval_samples_per_second": 42.883,
+      "eval_logits/chosen": -14.3125,
+      "eval_logits/rejected": -13.9375,
+      "eval_logps/chosen": -428.0,
+      "eval_logps/rejected": -420.0,
+      "eval_loss": 0.7656875252723694,
+      "eval_rewards/accuracies": 0.51953125,
+      "eval_rewards/chosen": -1.1015625,
+      "eval_rewards/margins": 0.2060546875,
+      "eval_rewards/rejected": -1.3125,
+      "eval_runtime": 46.6431,
+      "eval_samples_per_second": 42.879,
       "eval_steps_per_second": 0.686,
       "step": 200
     },
     {
       "epoch": 0.2198952879581152,
-      "grad_norm": 0.7138722197750738,
+      "grad_norm": 0.6282094371827225,
       "learning_rate": 3.6585365853658535e-05,
       "logits/chosen": -14.875,
-      "logits/rejected": -12.6875,
-      "logps/chosen": -486.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.0283,
+      "logits/rejected": -12.0,
+      "logps/chosen": -470.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.0299,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -1.2890625,
-      "rewards/margins": 10.125,
-      "rewards/rejected": -11.375,
+      "rewards/chosen": -1.140625,
+      "rewards/margins": 10.75,
+      "rewards/rejected": -11.875,
       "step": 210
     },
     {
       "epoch": 0.23036649214659685,
-      "grad_norm": 2.0571071973574657,
+      "grad_norm": 3.248721113127671,
       "learning_rate": 3.8327526132404185e-05,
-      "logits/chosen": -14.9375,
-      "logits/rejected": -12.75,
-      "logps/chosen": -576.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.0627,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.6015625,
-      "rewards/margins": 11.3125,
-      "rewards/rejected": -12.9375,
+      "logits/chosen": -15.3125,
+      "logits/rejected": -12.0625,
+      "logps/chosen": -592.0,
+      "logps/rejected": -2992.0,
+      "loss": 0.0659,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -1.75,
+      "rewards/margins": 13.5625,
+      "rewards/rejected": -15.3125,
       "step": 220
     },
     {
       "epoch": 0.24083769633507854,
-      "grad_norm": 0.27271331885571687,
+      "grad_norm": 0.24785510094838845,
       "learning_rate": 4.006968641114983e-05,
-      "logits/chosen": -14.5,
-      "logits/rejected": -12.1875,
-      "logps/chosen": -576.0,
-      "logps/rejected": -2800.0,
-      "loss": 0.0289,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -516.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.0248,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.078125,
-      "rewards/margins": 10.8125,
-      "rewards/rejected": -12.875,
+      "rewards/chosen": -1.4765625,
+      "rewards/margins": 12.4375,
+      "rewards/rejected": -13.875,
       "step": 230
     },
     {
       "epoch": 0.2513089005235602,
-      "grad_norm": 2.7749664833400343,
+      "grad_norm": 0.5986274184940548,
       "learning_rate": 4.181184668989547e-05,
-      "logits/chosen": -13.0,
-      "logits/rejected": -11.25,
-      "logps/chosen": -596.0,
-      "logps/rejected": -3072.0,
-      "loss": 0.0573,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -2.328125,
-      "rewards/margins": 13.5625,
-      "rewards/rejected": -15.875,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -488.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.0827,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.265625,
+      "rewards/margins": 13.3125,
+      "rewards/rejected": -14.5625,
       "step": 240
     },
     {
       "epoch": 0.2617801047120419,
-      "grad_norm": 0.255871273675974,
+      "grad_norm": 4.01493318190849,
       "learning_rate": 4.3554006968641115e-05,
-      "logits/chosen": -12.4375,
-      "logits/rejected": -10.5,
-      "logps/chosen": -656.0,
-      "logps/rejected": -2880.0,
-      "loss": 0.0305,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.3125,
-      "rewards/margins": 12.0,
-      "rewards/rejected": -14.375,
+      "logits/chosen": -12.875,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -612.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.025,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -1.90625,
+      "rewards/margins": 18.125,
+      "rewards/rejected": -20.0,
       "step": 250
     },
     {
       "epoch": 0.27225130890052357,
-      "grad_norm": 0.6496613184540602,
+      "grad_norm": 0.7625488935737561,
       "learning_rate": 4.529616724738676e-05,
-      "logits/chosen": -12.875,
-      "logits/rejected": -11.25,
-      "logps/chosen": -592.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.0318,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.234375,
-      "rewards/margins": 11.625,
-      "rewards/rejected": -13.875,
+      "logits/chosen": -12.25,
+      "logits/rejected": -9.5,
+      "logps/chosen": -552.0,
+      "logps/rejected": -3024.0,
+      "loss": 0.0243,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -1.84375,
+      "rewards/margins": 13.625,
+      "rewards/rejected": -15.5,
       "step": 260
     },
     {
       "epoch": 0.28272251308900526,
-      "grad_norm": 0.6461653166296297,
+      "grad_norm": 0.37869273626745503,
       "learning_rate": 4.703832752613241e-05,
-      "logits/chosen": -14.3125,
-      "logits/rejected": -12.0,
-      "logps/chosen": -544.0,
-      "logps/rejected": -3024.0,
-      "loss": 0.0591,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -1.8125,
-      "rewards/margins": 13.5,
-      "rewards/rejected": -15.3125,
+      "logits/chosen": -11.5,
+      "logits/rejected": -8.5,
+      "logps/chosen": -552.0,
+      "logps/rejected": -2976.0,
+      "loss": 0.0537,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -1.8984375,
+      "rewards/margins": 13.0,
+      "rewards/rejected": -14.875,
       "step": 270
     },
     {
       "epoch": 0.2931937172774869,
-      "grad_norm": 0.267868977624901,
+      "grad_norm": 0.27494687899494014,
       "learning_rate": 4.878048780487805e-05,
-      "logits/chosen": -14.6875,
-      "logits/rejected": -12.5625,
-      "logps/chosen": -500.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0382,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -1.546875,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -10.75,
+      "logits/chosen": -10.5,
+      "logits/rejected": -8.5,
+      "logps/chosen": -496.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0359,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -1.5,
+      "rewards/margins": 9.875,
+      "rewards/rejected": -11.375,
       "step": 280
     },
     {
       "epoch": 0.3036649214659686,
-      "grad_norm": 0.16183546690506861,
+      "grad_norm": 0.14416301961979705,
       "learning_rate": 4.999983293481418e-05,
-      "logits/chosen": -14.75,
-      "logits/rejected": -12.125,
-      "logps/chosen": -620.0,
-      "logps/rejected": -3168.0,
-      "loss": 0.0174,
+      "logits/chosen": -14.375,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -592.0,
+      "logps/rejected": -3120.0,
+      "loss": 0.0153,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.421875,
-      "rewards/margins": 14.75,
-      "rewards/rejected": -17.25,
+      "rewards/chosen": -2.15625,
+      "rewards/margins": 14.5625,
+      "rewards/rejected": -16.75,
       "step": 290
     },
     {
       "epoch": 0.31413612565445026,
-      "grad_norm": 4.011996959148844,
+      "grad_norm": 8.576987120427964,
       "learning_rate": 4.9996862949181595e-05,
-      "logits/chosen": -14.1875,
-      "logits/rejected": -11.9375,
-      "logps/chosen": -812.0,
-      "logps/rejected": -4032.0,
-      "loss": 0.0532,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -4.25,
-      "rewards/margins": 21.5,
-      "rewards/rejected": -25.625,
+      "logits/chosen": -15.125,
+      "logits/rejected": -10.875,
+      "logps/chosen": -652.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0716,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -2.640625,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -22.875,
       "step": 300
     },
     {
       "epoch": 0.31413612565445026,
       "eval_logits/chosen": -13.9375,
-      "eval_logits/rejected": -13.75,
-      "eval_logps/chosen": -948.0,
-      "eval_logps/rejected": -976.0,
-      "eval_loss": 1.6151249408721924,
-      "eval_rewards/accuracies": 0.484375,
-      "eval_rewards/chosen": -6.3125,
-      "eval_rewards/margins": 0.58203125,
-      "eval_rewards/rejected": -6.875,
-      "eval_runtime": 46.6419,
-      "eval_samples_per_second": 42.88,
+      "eval_logits/rejected": -13.5,
+      "eval_logps/chosen": -616.0,
+      "eval_logps/rejected": -640.0,
+      "eval_loss": 1.249453067779541,
+      "eval_rewards/accuracies": 0.513671875,
+      "eval_rewards/chosen": -2.984375,
+      "eval_rewards/margins": 0.5390625,
+      "eval_rewards/rejected": -3.515625,
+      "eval_runtime": 46.6358,
+      "eval_samples_per_second": 42.886,
       "eval_steps_per_second": 0.686,
       "step": 300
     },
     {
       "epoch": 0.32460732984293195,
-      "grad_norm": 0.3521715424281661,
+      "grad_norm": 0.7285709321382741,
       "learning_rate": 4.9990180911526896e-05,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -11.5,
-      "logps/chosen": -724.0,
-      "logps/rejected": -3488.0,
-      "loss": 0.1106,
+      "logits/chosen": -12.8125,
+      "logits/rejected": -9.25,
+      "logps/chosen": -532.0,
+      "logps/rejected": -3056.0,
+      "loss": 0.0304,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.703125,
-      "rewards/margins": 14.0,
-      "rewards/rejected": -17.75,
+      "rewards/chosen": -1.765625,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -13.3125,
       "step": 310
     },
     {
       "epoch": 0.33507853403141363,
-      "grad_norm": 0.5824614401989563,
+      "grad_norm": 0.8482563135709271,
       "learning_rate": 4.997978781413745e-05,
-      "logits/chosen": -13.125,
-      "logits/rejected": -11.6875,
+      "logits/chosen": -10.5625,
+      "logits/rejected": -7.78125,
       "logps/chosen": -784.0,
-      "logps/rejected": -3376.0,
-      "loss": 0.0536,
-      "rewards/accuracies": 0.9937499761581421,
+      "logps/rejected": -3296.0,
+      "loss": 0.0487,
+      "rewards/accuracies": 0.987500011920929,
       "rewards/chosen": -3.578125,
-      "rewards/margins": 15.6875,
-      "rewards/rejected": -19.25,
+      "rewards/margins": 14.9375,
+      "rewards/rejected": -18.5,
       "step": 320
     },
     {
       "epoch": 0.34554973821989526,
-      "grad_norm": 1.804359993687507,
+      "grad_norm": 4.580853167186891,
       "learning_rate": 4.996568520039565e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -12.0625,
-      "logps/chosen": -612.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0905,
+      "logits/chosen": -12.1875,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -648.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.0778,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -1.9765625,
-      "rewards/margins": 11.0625,
-      "rewards/rejected": -13.0625,
+      "rewards/chosen": -2.359375,
+      "rewards/margins": 13.125,
+      "rewards/rejected": -15.5,
       "step": 330
     },
     {
       "epoch": 0.35602094240837695,
-      "grad_norm": 0.24261599886564092,
+      "grad_norm": 0.2204611356467403,
       "learning_rate": 4.994787516454973e-05,
-      "logits/chosen": -15.25,
-      "logits/rejected": -12.4375,
-      "logps/chosen": -604.0,
-      "logps/rejected": -2944.0,
-      "loss": 0.0213,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.125,
-      "rewards/margins": 10.875,
-      "rewards/rejected": -13.0,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -11.0,
+      "logps/chosen": -528.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0192,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -1.3671875,
+      "rewards/margins": 10.375,
+      "rewards/rejected": -11.75,
       "step": 340
     },
     {
       "epoch": 0.36649214659685864,
-      "grad_norm": 1.6704252299175926,
+      "grad_norm": 2.388385711239015,
       "learning_rate": 4.992636035140273e-05,
-      "logits/chosen": -17.0,
-      "logits/rejected": -11.6875,
-      "logps/chosen": -612.0,
-      "logps/rejected": -3376.0,
-      "loss": 0.0594,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -2.625,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -11.125,
+      "logps/chosen": -524.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0414,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -1.7578125,
       "rewards/margins": 15.1875,
-      "rewards/rejected": -17.75,
+      "rewards/rejected": -17.0,
       "step": 350
     },
     {
       "epoch": 0.3769633507853403,
-      "grad_norm": 0.69873441240531,
+      "grad_norm": 0.9729685500352959,
       "learning_rate": 4.99011439559198e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -11.625,
-      "logps/chosen": -648.0,
-      "logps/rejected": -3184.0,
-      "loss": 0.0407,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.625,
-      "rewards/margins": 13.8125,
-      "rewards/rejected": -16.5,
+      "logits/chosen": -14.0,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -592.0,
+      "logps/rejected": -3328.0,
+      "loss": 0.0495,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.046875,
+      "rewards/margins": 15.8125,
+      "rewards/rejected": -17.875,
       "step": 360
     },
     {
       "epoch": 0.387434554973822,
-      "grad_norm": 0.278423213888884,
+      "grad_norm": 0.3605015545260038,
       "learning_rate": 4.987222972275369e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -11.8125,
-      "logps/chosen": -624.0,
-      "logps/rejected": -3040.0,
-      "loss": 0.0338,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -2.359375,
-      "rewards/margins": 13.1875,
-      "rewards/rejected": -15.5,
+      "logits/chosen": -13.375,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -608.0,
+      "logps/rejected": -3392.0,
+      "loss": 0.0257,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.203125,
+      "rewards/margins": 16.75,
+      "rewards/rejected": -19.0,
       "step": 370
     },
     {
       "epoch": 0.39790575916230364,
-      "grad_norm": 0.15722742475848675,
+      "grad_norm": 0.15612090376245558,
       "learning_rate": 4.983962194568872e-05,
-      "logits/chosen": -16.5,
-      "logits/rejected": -12.4375,
-      "logps/chosen": -604.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.0247,
+      "logits/chosen": -13.5,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -596.0,
+      "logps/rejected": -3360.0,
+      "loss": 0.0232,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.046875,
-      "rewards/margins": 15.5,
-      "rewards/rejected": -17.5,
+      "rewards/chosen": -1.96875,
+      "rewards/margins": 16.625,
+      "rewards/rejected": -18.5,
       "step": 380
     },
     {
       "epoch": 0.4083769633507853,
-      "grad_norm": 0.5715376688170342,
+      "grad_norm": 0.4681683980008556,
       "learning_rate": 4.9803325467003084e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -11.9375,
-      "logps/chosen": -588.0,
-      "logps/rejected": -3472.0,
-      "loss": 0.0249,
+      "logits/chosen": -13.625,
+      "logits/rejected": -9.25,
+      "logps/chosen": -616.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.036,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.8828125,
-      "rewards/margins": 16.125,
-      "rewards/rejected": -18.0,
+      "rewards/chosen": -2.1875,
+      "rewards/margins": 18.125,
+      "rewards/rejected": -20.25,
       "step": 390
     },
     {
       "epoch": 0.418848167539267,
-      "grad_norm": 0.8993338333225877,
+      "grad_norm": 0.23591733068055193,
       "learning_rate": 4.9763345676749816e-05,
-      "logits/chosen": -15.375,
-      "logits/rejected": -11.4375,
-      "logps/chosen": -540.0,
-      "logps/rejected": -3088.0,
-      "loss": 0.0436,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -1.6796875,
-      "rewards/margins": 13.5,
-      "rewards/rejected": -15.125,
+      "logits/chosen": -12.875,
+      "logits/rejected": -8.625,
+      "logps/chosen": -524.0,
+      "logps/rejected": -3024.0,
+      "loss": 0.0472,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -1.5078125,
+      "rewards/margins": 13.0,
+      "rewards/rejected": -14.5,
       "step": 400
     },
     {
       "epoch": 0.418848167539267,
-      "eval_logits/chosen": -15.5625,
-      "eval_logits/rejected": -15.5,
-      "eval_logps/chosen": -474.0,
-      "eval_logps/rejected": -456.0,
-      "eval_loss": 0.8565000295639038,
-      "eval_rewards/accuracies": 0.478515625,
-      "eval_rewards/chosen": -1.5625,
-      "eval_rewards/margins": 0.10693359375,
-      "eval_rewards/rejected": -1.671875,
-      "eval_runtime": 46.631,
-      "eval_samples_per_second": 42.89,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -12.3125,
+      "eval_logits/rejected": -11.9375,
+      "eval_logps/chosen": -492.0,
+      "eval_logps/rejected": -486.0,
+      "eval_loss": 0.8801093697547913,
+      "eval_rewards/accuracies": 0.49609375,
+      "eval_rewards/chosen": -1.734375,
+      "eval_rewards/margins": 0.2392578125,
+      "eval_rewards/rejected": -1.96875,
+      "eval_runtime": 53.3967,
+      "eval_samples_per_second": 37.455,
+      "eval_steps_per_second": 0.599,
       "step": 400
     },
     {
       "epoch": 0.4293193717277487,
-      "grad_norm": 0.4031261713265817,
+      "grad_norm": 0.23665357471354573,
       "learning_rate": 4.9719688511956374e-05,
-      "logits/chosen": -15.625,
-      "logits/rejected": -12.875,
-      "logps/chosen": -576.0,
-      "logps/rejected": -3008.0,
-      "loss": 0.0607,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -1.703125,
-      "rewards/margins": 11.5625,
-      "rewards/rejected": -13.25,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -8.25,
+      "logps/chosen": -608.0,
+      "logps/rejected": -3088.0,
+      "loss": 0.0545,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -2.0,
+      "rewards/margins": 11.9375,
+      "rewards/rejected": -14.0,
       "step": 410
     },
     {
       "epoch": 0.4397905759162304,
-      "grad_norm": 0.5770645629908464,
+      "grad_norm": 1.2777211184954786,
       "learning_rate": 4.967236045574294e-05,
-      "logits/chosen": -15.4375,
-      "logits/rejected": -11.375,
+      "logits/chosen": -12.125,
+      "logits/rejected": -7.15625,
       "logps/chosen": -608.0,
-      "logps/rejected": -2960.0,
-      "loss": 0.0245,
-      "rewards/accuracies": 0.987500011920929,
+      "logps/rejected": -2768.0,
+      "loss": 0.0339,
+      "rewards/accuracies": 0.9937499761581421,
       "rewards/chosen": -2.078125,
-      "rewards/margins": 12.5,
-      "rewards/rejected": -14.625,
+      "rewards/margins": 10.625,
+      "rewards/rejected": -12.75,
       "step": 420
     },
     {
       "epoch": 0.450261780104712,
-      "grad_norm": 3.1161465027576907,
+      "grad_norm": 1.5033971815348899,
       "learning_rate": 4.96213685363597e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -11.0,
-      "logps/chosen": -612.0,
-      "logps/rejected": -3312.0,
-      "loss": 0.0836,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.34375,
-      "rewards/margins": 16.0,
-      "rewards/rejected": -18.25,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -6.59375,
+      "logps/chosen": -620.0,
+      "logps/rejected": -3104.0,
+      "loss": 0.0552,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -2.40625,
+      "rewards/margins": 13.875,
+      "rewards/rejected": -16.25,
       "step": 430
     },
     {
       "epoch": 0.4607329842931937,
-      "grad_norm": 2.378496827700141,
+      "grad_norm": 4.35635241938968,
       "learning_rate": 4.956672032614315e-05,
-      "logits/chosen": -14.625,
-      "logits/rejected": -10.5,
-      "logps/chosen": -592.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0524,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.796875,
-      "rewards/margins": 10.875,
-      "rewards/rejected": -12.6875,
+      "logits/chosen": -10.0,
+      "logits/rejected": -6.53125,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.0455,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 13.625,
+      "rewards/rejected": -16.375,
       "step": 440
     },
     {
       "epoch": 0.4712041884816754,
-      "grad_norm": 0.39713967023063335,
+      "grad_norm": 1.0110974570997668,
       "learning_rate": 4.950842394039157e-05,
-      "logits/chosen": -15.3125,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -604.0,
-      "logps/rejected": -3184.0,
-      "loss": 0.0276,
+      "logits/chosen": -13.375,
+      "logits/rejected": -7.8125,
+      "logps/chosen": -632.0,
+      "logps/rejected": -3072.0,
+      "loss": 0.0264,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.34375,
-      "rewards/margins": 12.875,
-      "rewards/rejected": -15.25,
+      "rewards/chosen": -2.609375,
+      "rewards/margins": 11.5625,
+      "rewards/rejected": -14.1875,
       "step": 450
     },
     {
       "epoch": 0.4816753926701571,
-      "grad_norm": 0.6241862729153295,
+      "grad_norm": 0.24607976610378823,
       "learning_rate": 4.944648803615994e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -632.0,
+      "logits/chosen": -15.125,
+      "logits/rejected": -7.84375,
+      "logps/chosen": -692.0,
       "logps/rejected": -3040.0,
-      "loss": 0.0277,
+      "loss": 0.0306,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -2.328125,
-      "rewards/margins": 12.5,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 11.875,
       "rewards/rejected": -14.8125,
       "step": 460
     },
     {
       "epoch": 0.49214659685863876,
-      "grad_norm": 0.12564659581850465,
+      "grad_norm": 0.23899389583878303,
       "learning_rate": 4.938092181097433e-05,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -680.0,
-      "logps/rejected": -3376.0,
-      "loss": 0.0398,
+      "logits/chosen": -14.8125,
+      "logits/rejected": -7.125,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0745,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.828125,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -19.375,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 15.375,
+      "rewards/rejected": -18.5,
       "step": 470
     },
     {
       "epoch": 0.5026178010471204,
-      "grad_norm": 2.283919346361066,
+      "grad_norm": 1.559933644858924,
       "learning_rate": 4.931173500146607e-05,
-      "logits/chosen": -15.75,
-      "logits/rejected": -8.875,
-      "logps/chosen": -668.0,
-      "logps/rejected": -3584.0,
-      "loss": 0.0483,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.96875,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -19.375,
+      "logits/chosen": -15.625,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -572.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.0329,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.03125,
+      "rewards/margins": 11.9375,
+      "rewards/rejected": -14.0,
       "step": 480
     },
     {
       "epoch": 0.5130890052356021,
-      "grad_norm": 0.6163973170410908,
+      "grad_norm": 0.8732602787510814,
       "learning_rate": 4.923893788192582e-05,
-      "logits/chosen": -14.125,
-      "logits/rejected": -7.3125,
-      "logps/chosen": -668.0,
-      "logps/rejected": -3136.0,
-      "loss": 0.02,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.75,
-      "rewards/margins": 13.3125,
-      "rewards/rejected": -16.125,
+      "logits/chosen": -14.5,
+      "logits/rejected": -7.5,
+      "logps/chosen": -632.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.0198,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.375,
+      "rewards/margins": 12.0,
+      "rewards/rejected": -14.375,
       "step": 490
     },
     {
       "epoch": 0.5235602094240838,
-      "grad_norm": 0.5145404605081624,
+      "grad_norm": 1.1187002799562606,
       "learning_rate": 4.9162541262777906e-05,
-      "logits/chosen": -13.0625,
-      "logits/rejected": -6.9375,
-      "logps/chosen": -692.0,
-      "logps/rejected": -3056.0,
-      "loss": 0.0331,
+      "logits/chosen": -15.3125,
+      "logits/rejected": -8.75,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0394,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.953125,
-      "rewards/margins": 12.0,
-      "rewards/rejected": -14.9375,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 14.1875,
+      "rewards/rejected": -17.25,
       "step": 500
     },
     {
       "epoch": 0.5235602094240838,
-      "eval_logits/chosen": -11.4375,
-      "eval_logits/rejected": -10.9375,
-      "eval_logps/chosen": -608.0,
-      "eval_logps/rejected": -600.0,
-      "eval_loss": 0.9543906450271606,
-      "eval_rewards/accuracies": 0.486328125,
-      "eval_rewards/chosen": -2.875,
-      "eval_rewards/margins": 0.25390625,
-      "eval_rewards/rejected": -3.125,
-      "eval_runtime": 46.6208,
-      "eval_samples_per_second": 42.899,
+      "eval_logits/chosen": -16.25,
+      "eval_logits/rejected": -16.125,
+      "eval_logps/chosen": -596.0,
+      "eval_logps/rejected": -584.0,
+      "eval_loss": 0.9541093707084656,
+      "eval_rewards/accuracies": 0.4921875,
+      "eval_rewards/chosen": -2.78125,
+      "eval_rewards/margins": 0.1904296875,
+      "eval_rewards/rejected": -2.96875,
+      "eval_runtime": 46.6384,
+      "eval_samples_per_second": 42.883,
       "eval_steps_per_second": 0.686,
       "step": 500
     },
     {
       "epoch": 0.5340314136125655,
-      "grad_norm": 0.8321526963998709,
+      "grad_norm": 0.8857338866579929,
       "learning_rate": 4.90825564889749e-05,
-      "logits/chosen": -12.25,
-      "logits/rejected": -7.0625,
-      "logps/chosen": -716.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0287,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.921875,
-      "rewards/margins": 14.75,
-      "rewards/rejected": -17.75,
+      "logits/chosen": -16.125,
+      "logits/rejected": -10.625,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0254,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.1875,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -22.125,
       "step": 510
     },
     {
       "epoch": 0.5445026178010471,
-      "grad_norm": 1.8650468461297594,
+      "grad_norm": 2.4497953668569146,
       "learning_rate": 4.899899543831288e-05,
-      "logits/chosen": -13.0625,
-      "logits/rejected": -7.6875,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0315,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -2.875,
-      "rewards/margins": 18.625,
-      "rewards/rejected": -21.5,
+      "logits/chosen": -16.0,
+      "logits/rejected": -10.0,
+      "logps/chosen": -760.0,
+      "logps/rejected": -4048.0,
+      "loss": 0.0331,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.359375,
+      "rewards/margins": 21.25,
+      "rewards/rejected": -24.625,
       "step": 520
     },
     {
       "epoch": 0.5549738219895288,
-      "grad_norm": 1.9887085484294216,
+      "grad_norm": 2.289155576391381,
       "learning_rate": 4.891187051966765e-05,
-      "logits/chosen": -14.0,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -712.0,
+      "logits/chosen": -15.6875,
+      "logits/rejected": -10.25,
+      "logps/chosen": -732.0,
       "logps/rejected": -3472.0,
-      "loss": 0.0561,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -3.03125,
-      "rewards/margins": 16.875,
+      "loss": 0.0488,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 16.625,
       "rewards/rejected": -19.875,
       "step": 530
     },
     {
       "epoch": 0.5654450261780105,
-      "grad_norm": 0.44777227889244137,
+      "grad_norm": 0.46391523143413893,
       "learning_rate": 4.882119467115191e-05,
-      "logits/chosen": -14.5625,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -632.0,
-      "logps/rejected": -3072.0,
-      "loss": 0.0386,
+      "logits/chosen": -15.5,
+      "logits/rejected": -10.5,
+      "logps/chosen": -572.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0223,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.40625,
-      "rewards/margins": 12.75,
-      "rewards/rejected": -15.1875,
+      "rewards/chosen": -1.8125,
+      "rewards/margins": 10.5625,
+      "rewards/rejected": -12.375,
       "step": 540
     },
     {
       "epoch": 0.5759162303664922,
-      "grad_norm": 0.45740116052736923,
+      "grad_norm": 0.5441827646779858,
       "learning_rate": 4.872698135819402e-05,
-      "logits/chosen": -14.3125,
-      "logits/rejected": -9.25,
-      "logps/chosen": -588.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0293,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -2.046875,
-      "rewards/margins": 9.625,
-      "rewards/rejected": -11.6875,
+      "logits/chosen": -16.5,
+      "logits/rejected": -10.0,
+      "logps/chosen": -604.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.0338,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.234375,
+      "rewards/margins": 12.0,
+      "rewards/rejected": -14.25,
       "step": 550
     },
     {
       "epoch": 0.5863874345549738,
-      "grad_norm": 0.7755330928981287,
+      "grad_norm": 2.2688012496463354,
       "learning_rate": 4.862924457153831e-05,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -8.625,
-      "logps/chosen": -568.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0228,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.34375,
-      "rewards/margins": 13.6875,
-      "rewards/rejected": -16.0,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -668.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0356,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 18.75,
+      "rewards/rejected": -22.125,
       "step": 560
     },
     {
       "epoch": 0.5968586387434555,
-      "grad_norm": 0.34459164563098016,
+      "grad_norm": 0.3855277416887269,
       "learning_rate": 4.8527998825167484e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -8.125,
-      "logps/chosen": -648.0,
-      "logps/rejected": -3616.0,
-      "loss": 0.0386,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.828125,
-      "rewards/margins": 17.125,
-      "rewards/rejected": -20.0,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -6.9375,
+      "logps/chosen": -840.0,
+      "logps/rejected": -4480.0,
+      "loss": 0.0384,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 23.875,
+      "rewards/rejected": -28.625,
       "step": 570
     },
     {
       "epoch": 0.6073298429319371,
-      "grad_norm": 0.21961815265940254,
+      "grad_norm": 0.1993755557709174,
       "learning_rate": 4.842325915414728e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.0,
-      "logps/chosen": -644.0,
-      "logps/rejected": -3376.0,
-      "loss": 0.0355,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -2.796875,
-      "rewards/margins": 16.375,
-      "rewards/rejected": -19.25,
+      "logits/chosen": -14.125,
+      "logits/rejected": -8.125,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0419,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 20.375,
+      "rewards/rejected": -23.625,
       "step": 580
     },
     {
       "epoch": 0.6178010471204188,
-      "grad_norm": 0.234344096391646,
+      "grad_norm": 0.31359328843809586,
       "learning_rate": 4.831504111239373e-05,
-      "logits/chosen": -16.375,
-      "logits/rejected": -9.25,
-      "logps/chosen": -748.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.0336,
+      "logits/chosen": -15.125,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.025,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 19.0,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -2.703125,
+      "rewards/margins": 18.125,
+      "rewards/rejected": -20.75,
       "step": 590
     },
     {
       "epoch": 0.6282722513089005,
-      "grad_norm": 5.161477154470557,
+      "grad_norm": 4.000829113141296,
       "learning_rate": 4.8203360770363425e-05,
-      "logits/chosen": -17.0,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -704.0,
-      "logps/rejected": -3744.0,
-      "loss": 0.043,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -3.0,
-      "rewards/margins": 20.0,
-      "rewards/rejected": -23.0,
+      "logits/chosen": -15.3125,
+      "logits/rejected": -10.125,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3552.0,
+      "loss": 0.0707,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -2.671875,
+      "rewards/margins": 18.25,
+      "rewards/rejected": -21.0,
       "step": 600
     },
     {
       "epoch": 0.6282722513089005,
-      "eval_logits/chosen": -16.75,
-      "eval_logits/rejected": -16.625,
-      "eval_logps/chosen": -596.0,
-      "eval_logps/rejected": -580.0,
-      "eval_loss": 0.9331250190734863,
-      "eval_rewards/accuracies": 0.455078125,
-      "eval_rewards/chosen": -2.78125,
-      "eval_rewards/margins": 0.1396484375,
-      "eval_rewards/rejected": -2.921875,
-      "eval_runtime": 46.6258,
-      "eval_samples_per_second": 42.895,
+      "eval_logits/chosen": -15.125,
+      "eval_logits/rejected": -14.9375,
+      "eval_logps/chosen": -544.0,
+      "eval_logps/rejected": -532.0,
+      "eval_loss": 0.8604375123977661,
+      "eval_rewards/accuracies": 0.4921875,
+      "eval_rewards/chosen": -2.25,
+      "eval_rewards/margins": 0.193359375,
+      "eval_rewards/rejected": -2.4375,
+      "eval_runtime": 46.6348,
+      "eval_samples_per_second": 42.886,
       "eval_steps_per_second": 0.686,
       "step": 600
     },
     {
       "epoch": 0.6387434554973822,
-      "grad_norm": 0.5228909692371784,
+      "grad_norm": 0.24229371175464115,
       "learning_rate": 4.808823471266702e-05,
-      "logits/chosen": -17.25,
-      "logits/rejected": -11.625,
-      "logps/chosen": -620.0,
-      "logps/rejected": -3024.0,
-      "loss": 0.0311,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.375,
-      "rewards/margins": 13.0,
-      "rewards/rejected": -15.375,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -10.375,
+      "logps/chosen": -616.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.0321,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.359375,
+      "rewards/margins": 13.1875,
+      "rewards/rejected": -15.5625,
       "step": 610
     },
     {
       "epoch": 0.6492146596858639,
-      "grad_norm": 0.9842904923972754,
+      "grad_norm": 1.2491045649235486,
       "learning_rate": 4.796968003560639e-05,
-      "logits/chosen": -17.0,
-      "logits/rejected": -12.25,
-      "logps/chosen": -676.0,
-      "logps/rejected": -3648.0,
-      "loss": 0.0557,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -2.96875,
-      "rewards/margins": 16.75,
-      "rewards/rejected": -19.75,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -9.5,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0629,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -3.25,
+      "rewards/margins": 15.8125,
+      "rewards/rejected": -19.125,
       "step": 620
     },
     {
       "epoch": 0.6596858638743456,
-      "grad_norm": 0.046708990987790404,
+      "grad_norm": 0.11636428651229724,
       "learning_rate": 4.7847714344635864e-05,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -760.0,
-      "logps/rejected": -4192.0,
-      "loss": 0.0418,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.875,
-      "rewards/margins": 22.0,
-      "rewards/rejected": -25.875,
+      "logits/chosen": -11.0625,
+      "logits/rejected": -8.0,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0379,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 15.4375,
+      "rewards/rejected": -18.5,
       "step": 630
     },
     {
       "epoch": 0.6701570680628273,
-      "grad_norm": 0.5671438483477018,
+      "grad_norm": 0.541839659263218,
       "learning_rate": 4.772235575174776e-05,
-      "logits/chosen": -15.0625,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -824.0,
-      "logps/rejected": -3616.0,
-      "loss": 0.0309,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -4.0625,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -20.625,
+      "logits/chosen": -12.0,
+      "logits/rejected": -7.4375,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3360.0,
+      "loss": 0.0473,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 15.25,
+      "rewards/rejected": -18.25,
       "step": 640
     },
     {
       "epoch": 0.680628272251309,
-      "grad_norm": 0.5393892043321871,
+      "grad_norm": 0.6219786383905689,
       "learning_rate": 4.7593622872782754e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -10.375,
-      "logps/chosen": -680.0,
-      "logps/rejected": -3216.0,
-      "loss": 0.0477,
+      "logits/chosen": -11.625,
+      "logits/rejected": -8.0,
+      "logps/chosen": -652.0,
+      "logps/rejected": -3504.0,
+      "loss": 0.0396,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 12.8125,
-      "rewards/rejected": -16.0,
+      "rewards/chosen": -2.859375,
+      "rewards/margins": 16.0,
+      "rewards/rejected": -18.875,
       "step": 650
     },
     {
       "epoch": 0.6910994764397905,
-      "grad_norm": 0.2088555714731049,
+      "grad_norm": 0.4477057700245674,
       "learning_rate": 4.746153482466542e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -11.625,
-      "logps/chosen": -620.0,
-      "logps/rejected": -2976.0,
-      "loss": 0.0333,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.484375,
-      "rewards/margins": 12.9375,
-      "rewards/rejected": -15.375,
+      "logits/chosen": -13.375,
+      "logits/rejected": -8.75,
+      "logps/chosen": -660.0,
+      "logps/rejected": -3392.0,
+      "loss": 0.0371,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.859375,
+      "rewards/margins": 16.75,
+      "rewards/rejected": -19.625,
       "step": 660
     },
     {
       "epoch": 0.7015706806282722,
-      "grad_norm": 1.6990165259882208,
+      "grad_norm": 2.13255276471122,
       "learning_rate": 4.732611122256537e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -10.875,
-      "logps/chosen": -688.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0331,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -2.703125,
-      "rewards/margins": 14.0625,
-      "rewards/rejected": -16.75,
+      "logits/chosen": -14.4375,
+      "logits/rejected": -8.75,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3936.0,
+      "loss": 0.0396,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -23.25,
       "step": 670
     },
     {
       "epoch": 0.7120418848167539,
-      "grad_norm": 0.9148434816260349,
+      "grad_norm": 1.2946125287386432,
       "learning_rate": 4.718737217698435e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -11.125,
-      "logps/chosen": -664.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.0761,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -9.875,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3568.0,
+      "loss": 0.1288,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.875,
-      "rewards/margins": 13.3125,
-      "rewards/rejected": -16.25,
+      "rewards/chosen": -4.03125,
+      "rewards/margins": 15.375,
+      "rewards/rejected": -19.375,
       "step": 680
     },
     {
       "epoch": 0.7225130890052356,
-      "grad_norm": 0.45245253995016227,
+      "grad_norm": 0.30405034978785483,
       "learning_rate": 4.704533829076982e-05,
-      "logits/chosen": -15.625,
-      "logits/rejected": -10.5,
-      "logps/chosen": -664.0,
-      "logps/rejected": -2912.0,
-      "loss": 0.057,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.703125,
-      "rewards/margins": 10.8125,
-      "rewards/rejected": -13.5,
+      "logits/chosen": -14.8125,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -660.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.0528,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.65625,
+      "rewards/margins": 9.25,
+      "rewards/rejected": -11.9375,
       "step": 690
     },
     {
       "epoch": 0.7329842931937173,
-      "grad_norm": 0.1761121684476406,
+      "grad_norm": 0.16763350421149487,
       "learning_rate": 4.6900030656055455e-05,
-      "logits/chosen": -14.5625,
-      "logits/rejected": -10.625,
-      "logps/chosen": -696.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.037,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 9.75,
-      "rewards/rejected": -12.625,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -9.25,
+      "logps/chosen": -676.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.045,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.703125,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -12.5,
       "step": 700
     },
     {
       "epoch": 0.7329842931937173,
-      "eval_logits/chosen": -13.0625,
-      "eval_logits/rejected": -12.6875,
-      "eval_logps/chosen": -608.0,
-      "eval_logps/rejected": -596.0,
-      "eval_loss": 0.8353281021118164,
-      "eval_rewards/accuracies": 0.515625,
-      "eval_rewards/chosen": -2.90625,
-      "eval_rewards/margins": 0.177734375,
-      "eval_rewards/rejected": -3.078125,
-      "eval_runtime": 46.6196,
-      "eval_samples_per_second": 42.9,
+      "eval_logits/chosen": -15.5,
+      "eval_logits/rejected": -15.25,
+      "eval_logps/chosen": -620.0,
+      "eval_logps/rejected": -612.0,
+      "eval_loss": 0.9377812743186951,
+      "eval_rewards/accuracies": 0.486328125,
+      "eval_rewards/chosen": -3.03125,
+      "eval_rewards/margins": 0.216796875,
+      "eval_rewards/rejected": -3.25,
+      "eval_runtime": 46.6331,
+      "eval_samples_per_second": 42.888,
       "eval_steps_per_second": 0.686,
       "step": 700
     },
     {
       "epoch": 0.743455497382199,
-      "grad_norm": 0.3830201535850538,
+      "grad_norm": 0.6988145836661018,
       "learning_rate": 4.6751470851128906e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.0,
-      "logps/chosen": -712.0,
-      "logps/rejected": -2944.0,
-      "loss": 0.0707,
+      "logits/chosen": -15.625,
+      "logits/rejected": -9.625,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0583,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.9375,
-      "rewards/margins": 11.8125,
-      "rewards/rejected": -14.75,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 14.5,
+      "rewards/rejected": -17.5,
       "step": 710
     },
     {
       "epoch": 0.7539267015706806,
-      "grad_norm": 0.4810244225705128,
+      "grad_norm": 0.47751725341865003,
       "learning_rate": 4.6599680937227454e-05,
-      "logits/chosen": -14.5625,
-      "logits/rejected": -10.375,
-      "logps/chosen": -676.0,
-      "logps/rejected": -3136.0,
-      "loss": 0.0423,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -2.78125,
-      "rewards/margins": 13.0,
-      "rewards/rejected": -15.8125,
+      "logits/chosen": -16.375,
+      "logits/rejected": -11.125,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3600.0,
+      "loss": 0.0321,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.140625,
+      "rewards/margins": 17.25,
+      "rewards/rejected": -20.375,
       "step": 720
     },
     {
       "epoch": 0.7643979057591623,
-      "grad_norm": 0.5340457168242676,
+      "grad_norm": 0.5608195326396617,
       "learning_rate": 4.644468345526183e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -10.5,
-      "logps/chosen": -708.0,
-      "logps/rejected": -3312.0,
-      "loss": 0.0271,
+      "logits/chosen": -16.0,
+      "logits/rejected": -12.375,
+      "logps/chosen": -748.0,
+      "logps/rejected": -3872.0,
+      "loss": 0.0499,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.1875,
-      "rewards/margins": 14.6875,
-      "rewards/rejected": -17.875,
+      "rewards/chosen": -3.625,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -23.375,
       "step": 730
     },
     {
       "epoch": 0.774869109947644,
-      "grad_norm": 0.22586065006737482,
+      "grad_norm": 0.42858168901761934,
       "learning_rate": 4.628650142246894e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3120.0,
-      "loss": 0.0437,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.171875,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -15.4375,
+      "logits/chosen": -15.6875,
+      "logits/rejected": -10.25,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0762,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.484375,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -21.125,
       "step": 740
     },
     {
       "epoch": 0.7853403141361257,
-      "grad_norm": 1.0714552455544113,
+      "grad_norm": 0.6245045108603484,
       "learning_rate": 4.6125158328993765e-05,
-      "logits/chosen": -16.5,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -688.0,
-      "logps/rejected": -2992.0,
-      "loss": 0.0315,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -3.109375,
-      "rewards/margins": 11.125,
-      "rewards/rejected": -14.25,
+      "logits/chosen": -14.625,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3136.0,
+      "loss": 0.0266,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -3.0625,
+      "rewards/margins": 12.5625,
+      "rewards/rejected": -15.625,
       "step": 750
     },
     {
       "epoch": 0.7958115183246073,
-      "grad_norm": 2.3025531204789385,
+      "grad_norm": 1.2284101859003445,
       "learning_rate": 4.5960678134401064e-05,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.875,
-      "logps/chosen": -736.0,
-      "logps/rejected": -2992.0,
-      "loss": 0.0586,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 12.4375,
-      "rewards/rejected": -15.75,
+      "logits/chosen": -15.125,
+      "logits/rejected": -7.0,
+      "logps/chosen": -680.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.0485,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.796875,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -15.4375,
       "step": 760
     },
     {
       "epoch": 0.806282722513089,
-      "grad_norm": 0.029438547575492153,
+      "grad_norm": 0.05241378141096161,
       "learning_rate": 4.579308526411734e-05,
-      "logits/chosen": -17.5,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.0253,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.25,
-      "rewards/margins": 13.5625,
-      "rewards/rejected": -16.75,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -5.09375,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0177,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 13.25,
+      "rewards/rejected": -16.5,
       "step": 770
     },
     {
       "epoch": 0.8167539267015707,
-      "grad_norm": 0.5377522842562789,
+      "grad_norm": 0.46171207252872265,
       "learning_rate": 4.562240460580367e-05,
-      "logits/chosen": -17.625,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -700.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0227,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -3.03125,
-      "rewards/margins": 15.75,
-      "rewards/rejected": -18.75,
+      "logits/chosen": -15.875,
+      "logits/rejected": -5.21875,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3488.0,
+      "loss": 0.0211,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 17.125,
+      "rewards/rejected": -20.25,
       "step": 780
     },
     {
       "epoch": 0.8272251308900523,
-      "grad_norm": 0.07169042095939708,
+      "grad_norm": 0.11486195338872664,
       "learning_rate": 4.544866150565985e-05,
-      "logits/chosen": -17.75,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -708.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.0545,
+      "logits/chosen": -16.875,
+      "logits/rejected": -6.1875,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0463,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 19.5,
-      "rewards/rejected": -22.625,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -21.0,
       "step": 790
     },
     {
       "epoch": 0.837696335078534,
-      "grad_norm": 0.3572409460292667,
+      "grad_norm": 0.438006178585215,
       "learning_rate": 4.527188176466051e-05,
-      "logits/chosen": -17.5,
-      "logits/rejected": -10.25,
-      "logps/chosen": -720.0,
-      "logps/rejected": -3184.0,
-      "loss": 0.0295,
+      "logits/chosen": -16.875,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -676.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.0278,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -3.09375,
-      "rewards/margins": 14.0625,
-      "rewards/rejected": -17.125,
+      "rewards/chosen": -2.625,
+      "rewards/margins": 11.0,
+      "rewards/rejected": -13.625,
       "step": 800
     },
     {
       "epoch": 0.837696335078534,
-      "eval_logits/chosen": -18.375,
-      "eval_logits/rejected": -18.25,
-      "eval_logps/chosen": -604.0,
-      "eval_logps/rejected": -588.0,
-      "eval_loss": 0.9348593950271606,
+      "eval_logits/chosen": -17.75,
+      "eval_logits/rejected": -18.0,
+      "eval_logps/chosen": -568.0,
+      "eval_logps/rejected": -556.0,
+      "eval_loss": 0.9399999976158142,
       "eval_rewards/accuracies": 0.486328125,
-      "eval_rewards/chosen": -2.84375,
-      "eval_rewards/margins": 0.1611328125,
-      "eval_rewards/rejected": -3.015625,
-      "eval_runtime": 46.6194,
-      "eval_samples_per_second": 42.901,
+      "eval_rewards/chosen": -2.515625,
+      "eval_rewards/margins": 0.1640625,
+      "eval_rewards/rejected": -2.671875,
+      "eval_runtime": 46.6355,
+      "eval_samples_per_second": 42.886,
       "eval_steps_per_second": 0.686,
       "step": 800
     },
     {
       "epoch": 0.8481675392670157,
-      "grad_norm": 0.9622070224035102,
+      "grad_norm": 0.8051790334517187,
       "learning_rate": 4.509209163472357e-05,
-      "logits/chosen": -17.75,
-      "logits/rejected": -11.125,
-      "logps/chosen": -720.0,
-      "logps/rejected": -3056.0,
-      "loss": 0.0344,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.140625,
-      "rewards/margins": 12.1875,
-      "rewards/rejected": -15.375,
+      "logits/chosen": -17.0,
+      "logits/rejected": -12.0625,
+      "logps/chosen": -652.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0304,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.46875,
+      "rewards/margins": 10.5625,
+      "rewards/rejected": -13.0,
       "step": 810
     },
     {
       "epoch": 0.8586387434554974,
-      "grad_norm": 1.1976183202812771,
+      "grad_norm": 1.2092358113762114,
       "learning_rate": 4.4909317814811896e-05,
-      "logits/chosen": -17.375,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -736.0,
-      "logps/rejected": -3408.0,
-      "loss": 0.0365,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.3125,
-      "rewards/margins": 14.6875,
-      "rewards/rejected": -18.0,
+      "logits/chosen": -17.875,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3488.0,
+      "loss": 0.0363,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 15.9375,
+      "rewards/rejected": -18.875,
       "step": 820
     },
     {
       "epoch": 0.8691099476439791,
-      "grad_norm": 0.6185431829266129,
+      "grad_norm": 0.8381935675790307,
       "learning_rate": 4.472358744696845e-05,
-      "logits/chosen": -17.5,
-      "logits/rejected": -10.75,
-      "logps/chosen": -744.0,
-      "logps/rejected": -3456.0,
-      "loss": 0.039,
+      "logits/chosen": -18.75,
+      "logits/rejected": -13.1875,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0657,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 16.625,
-      "rewards/rejected": -20.125,
+      "rewards/chosen": -3.46875,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.75,
       "step": 830
     },
     {
       "epoch": 0.8795811518324608,
-      "grad_norm": 0.256274935575436,
+      "grad_norm": 0.4012384088559265,
       "learning_rate": 4.4534928112285645e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -10.375,
-      "logps/chosen": -756.0,
-      "logps/rejected": -3632.0,
-      "loss": 0.0314,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -3.5625,
-      "rewards/margins": 17.625,
-      "rewards/rejected": -21.125,
+      "logits/chosen": -17.125,
+      "logits/rejected": -12.125,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0463,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 18.375,
+      "rewards/rejected": -21.625,
       "step": 840
     },
     {
       "epoch": 0.8900523560209425,
-      "grad_norm": 0.4016852140262369,
+      "grad_norm": 1.6871275676298252,
       "learning_rate": 4.43433678268096e-05,
-      "logits/chosen": -15.875,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -768.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0316,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -3.90625,
-      "rewards/margins": 20.75,
-      "rewards/rejected": -24.625,
+      "logits/chosen": -16.0,
+      "logits/rejected": -10.5,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0257,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -19.25,
       "step": 850
     },
     {
       "epoch": 0.900523560209424,
-      "grad_norm": 0.34849804735893963,
+      "grad_norm": 0.270638893401861,
       "learning_rate": 4.414893503737968e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -7.875,
-      "logps/chosen": -744.0,
-      "logps/rejected": -4096.0,
-      "loss": 0.0353,
+      "logits/chosen": -17.0,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3504.0,
+      "loss": 0.0352,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -3.78125,
-      "rewards/margins": 23.0,
-      "rewards/rejected": -26.875,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 17.875,
+      "rewards/rejected": -20.875,
       "step": 860
     },
     {
       "epoch": 0.9109947643979057,
-      "grad_norm": 3.5384286746592504,
+      "grad_norm": 0.30498384198884454,
       "learning_rate": 4.395165861740414e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -736.0,
-      "logps/rejected": -4032.0,
-      "loss": 0.028,
+      "logits/chosen": -17.0,
+      "logits/rejected": -9.625,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0226,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.734375,
-      "rewards/margins": 21.25,
-      "rewards/rejected": -25.0,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -22.125,
       "step": 870
     },
     {
       "epoch": 0.9214659685863874,
-      "grad_norm": 1.9462295523172257,
+      "grad_norm": 2.434992683965432,
       "learning_rate": 4.375156786257241e-05,
-      "logits/chosen": -16.375,
-      "logits/rejected": -9.375,
-      "logps/chosen": -840.0,
-      "logps/rejected": -4512.0,
-      "loss": 0.0771,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -4.21875,
-      "rewards/margins": 25.5,
-      "rewards/rejected": -29.75,
+      "logits/chosen": -16.25,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -832.0,
+      "logps/rejected": -4224.0,
+      "loss": 0.0548,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -4.125,
+      "rewards/margins": 22.625,
+      "rewards/rejected": -26.75,
       "step": 880
     },
     {
       "epoch": 0.9319371727748691,
-      "grad_norm": 0.6115988075792568,
+      "grad_norm": 0.6590122618738954,
       "learning_rate": 4.354869248650466e-05,
-      "logits/chosen": -16.5,
-      "logits/rejected": -11.5,
-      "logps/chosen": -760.0,
-      "logps/rejected": -3280.0,
-      "loss": 0.0448,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 12.75,
-      "rewards/rejected": -16.25,
+      "logits/chosen": -16.0,
+      "logits/rejected": -8.5,
+      "logps/chosen": -824.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0604,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -4.15625,
+      "rewards/margins": 16.875,
+      "rewards/rejected": -21.125,
       "step": 890
     },
     {
       "epoch": 0.9424083769633508,
-      "grad_norm": 1.412853210014212,
+      "grad_norm": 0.8188599262674754,
       "learning_rate": 4.334306261633931e-05,
-      "logits/chosen": -15.75,
-      "logits/rejected": -11.0,
-      "logps/chosen": -728.0,
-      "logps/rejected": -2928.0,
-      "loss": 0.0344,
+      "logits/chosen": -14.8125,
+      "logits/rejected": -8.0,
+      "logps/chosen": -760.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.018,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.125,
-      "rewards/margins": 10.125,
-      "rewards/rejected": -13.25,
+      "rewards/chosen": -3.484375,
+      "rewards/margins": 13.4375,
+      "rewards/rejected": -16.875,
       "step": 900
     },
     {
       "epoch": 0.9424083769633508,
-      "eval_logits/chosen": -15.25,
-      "eval_logits/rejected": -14.8125,
-      "eval_logps/chosen": -640.0,
-      "eval_logps/rejected": -624.0,
-      "eval_loss": 0.9633281230926514,
-      "eval_rewards/accuracies": 0.482421875,
-      "eval_rewards/chosen": -3.21875,
-      "eval_rewards/margins": 0.11572265625,
-      "eval_rewards/rejected": -3.328125,
-      "eval_runtime": 46.6113,
-      "eval_samples_per_second": 42.908,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -15.625,
+      "eval_logits/rejected": -15.625,
+      "eval_logps/chosen": -704.0,
+      "eval_logps/rejected": -704.0,
+      "eval_loss": 1.1675000190734863,
+      "eval_rewards/accuracies": 0.490234375,
+      "eval_rewards/chosen": -3.84375,
+      "eval_rewards/margins": 0.296875,
+      "eval_rewards/rejected": -4.125,
+      "eval_runtime": 46.6383,
+      "eval_samples_per_second": 42.883,
+      "eval_steps_per_second": 0.686,
       "step": 900
     },
     {
       "epoch": 0.9528795811518325,
-      "grad_norm": 1.9407880419110068,
+      "grad_norm": 0.35754074116484946,
       "learning_rate": 4.313470878825911e-05,
-      "logits/chosen": -15.6875,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0299,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -7.125,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3632.0,
+      "loss": 0.0448,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.234375,
-      "rewards/margins": 13.6875,
-      "rewards/rejected": -16.875,
+      "rewards/chosen": -3.5625,
+      "rewards/margins": 16.625,
+      "rewards/rejected": -20.25,
       "step": 910
     },
     {
       "epoch": 0.9633507853403142,
-      "grad_norm": 5.477367652742189,
+      "grad_norm": 2.707305489151552,
       "learning_rate": 4.292366194295652e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -10.4375,
-      "logps/chosen": -676.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0463,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.875,
-      "rewards/margins": 11.125,
-      "rewards/rejected": -14.0,
+      "logits/chosen": -15.0,
+      "logits/rejected": -7.40625,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3088.0,
+      "loss": 0.0517,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 13.8125,
+      "rewards/rejected": -17.0,
       "step": 920
     },
     {
       "epoch": 0.9738219895287958,
-      "grad_norm": 0.2727532054837141,
+      "grad_norm": 0.7705879467304723,
       "learning_rate": 4.2709953421038964e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -12.375,
-      "logps/chosen": -660.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.0459,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.84375,
-      "rewards/margins": 11.75,
-      "rewards/rejected": -14.5625,
+      "logits/chosen": -14.375,
+      "logits/rejected": -8.125,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3168.0,
+      "loss": 0.0413,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 11.6875,
+      "rewards/rejected": -14.6875,
       "step": 930
     },
     {
       "epoch": 0.9842931937172775,
-      "grad_norm": 0.5629565362593076,
+      "grad_norm": 0.6146071320644448,
       "learning_rate": 4.249361495837478e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -692.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0634,
+      "logits/chosen": -14.25,
+      "logits/rejected": -7.53125,
+      "logps/chosen": -680.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.043,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.03125,
-      "rewards/margins": 13.125,
-      "rewards/rejected": -16.125,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -12.0625,
       "step": 940
     },
     {
       "epoch": 0.9947643979057592,
-      "grad_norm": 0.29555778161173835,
+      "grad_norm": 0.3761939267477344,
       "learning_rate": 4.227467868138035e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -11.625,
-      "logps/chosen": -664.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.0363,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.734375,
-      "rewards/margins": 10.6875,
-      "rewards/rejected": -13.4375,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -7.625,
+      "logps/chosen": -680.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.0352,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -12.6875,
       "step": 950
     },
     {
       "epoch": 1.0052356020942408,
-      "grad_norm": 0.00910754421374954,
+      "grad_norm": 0.013786615060030349,
       "learning_rate": 4.205317710224938e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -11.5625,
-      "logps/chosen": -660.0,
-      "logps/rejected": -2880.0,
-      "loss": 0.0201,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -7.78125,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3056.0,
+      "loss": 0.0182,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.625,
-      "rewards/margins": 11.6875,
-      "rewards/rejected": -14.3125,
+      "rewards/chosen": -2.828125,
+      "rewards/margins": 13.25,
+      "rewards/rejected": -16.125,
       "step": 960
     },
     {
       "epoch": 1.0157068062827226,
-      "grad_norm": 0.10579369726306186,
+      "grad_norm": 0.2178134787226099,
       "learning_rate": 4.1829143114124733e-05,
-      "logits/chosen": -17.5,
-      "logits/rejected": -12.0,
-      "logps/chosen": -656.0,
-      "logps/rejected": -3184.0,
-      "loss": 0.002,
+      "logits/chosen": -14.0,
+      "logits/rejected": -7.78125,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3504.0,
+      "loss": 0.0026,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.671875,
-      "rewards/margins": 13.8125,
-      "rewards/rejected": -16.5,
+      "rewards/chosen": -2.8125,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -19.75,
       "step": 970
     },
     {
       "epoch": 1.0261780104712042,
-      "grad_norm": 0.021382836768259528,
+      "grad_norm": 0.033659585543537276,
       "learning_rate": 4.160260998621387e-05,
-      "logits/chosen": -18.0,
-      "logits/rejected": -12.0625,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0133,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -8.5,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.96875,
-      "rewards/margins": 15.6875,
-      "rewards/rejected": -18.625,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -22.75,
       "step": 980
     },
     {
       "epoch": 1.036649214659686,
-      "grad_norm": 0.03803686721379146,
+      "grad_norm": 0.1618219850870755,
       "learning_rate": 4.137361135884824e-05,
-      "logits/chosen": -17.375,
-      "logits/rejected": -11.5,
+      "logits/chosen": -14.0,
+      "logits/rejected": -8.375,
       "logps/chosen": -708.0,
-      "logps/rejected": -3584.0,
-      "loss": 0.0016,
+      "logps/rejected": -3904.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.046875,
-      "rewards/margins": 16.875,
-      "rewards/rejected": -19.875,
+      "rewards/chosen": -3.0625,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -23.0,
       "step": 990
     },
     {
       "epoch": 1.0471204188481675,
-      "grad_norm": 0.15791541938503503,
+      "grad_norm": 0.07654117607974566,
       "learning_rate": 4.114218123848777e-05,
-      "logits/chosen": -17.5,
-      "logits/rejected": -11.6875,
-      "logps/chosen": -680.0,
-      "logps/rejected": -3744.0,
-      "loss": 0.0067,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.078125,
-      "rewards/margins": 18.375,
-      "rewards/rejected": -21.5,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -684.0,
+      "logps/rejected": -4000.0,
+      "loss": 0.0043,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.140625,
+      "rewards/margins": 20.875,
+      "rewards/rejected": -24.0,
       "step": 1000
     },
     {
       "epoch": 1.0471204188481675,
-      "eval_logits/chosen": -17.75,
-      "eval_logits/rejected": -17.5,
-      "eval_logps/chosen": -652.0,
-      "eval_logps/rejected": -656.0,
-      "eval_loss": 1.0684062242507935,
-      "eval_rewards/accuracies": 0.486328125,
-      "eval_rewards/chosen": -3.34375,
-      "eval_rewards/margins": 0.328125,
-      "eval_rewards/rejected": -3.671875,
-      "eval_runtime": 46.6109,
-      "eval_samples_per_second": 42.908,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -14.4375,
+      "eval_logits/rejected": -14.25,
+      "eval_logps/chosen": -672.0,
+      "eval_logps/rejected": -684.0,
+      "eval_loss": 1.0828437805175781,
+      "eval_rewards/accuracies": 0.484375,
+      "eval_rewards/chosen": -3.515625,
+      "eval_rewards/margins": 0.43359375,
+      "eval_rewards/rejected": -3.953125,
+      "eval_runtime": 46.8503,
+      "eval_samples_per_second": 42.689,
+      "eval_steps_per_second": 0.683,
       "step": 1000
     },
     {
       "epoch": 1.057591623036649,
-      "grad_norm": 0.003677347894430227,
+      "grad_norm": 0.003420502570656448,
       "learning_rate": 4.0908353992670824e-05,
-      "logits/chosen": -17.875,
-      "logits/rejected": -11.8125,
-      "logps/chosen": -668.0,
-      "logps/rejected": -3920.0,
-      "loss": 0.004,
+      "logits/chosen": -14.125,
+      "logits/rejected": -9.625,
+      "logps/chosen": -652.0,
+      "logps/rejected": -4256.0,
+      "loss": 0.0021,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.015625,
-      "rewards/margins": 20.75,
-      "rewards/rejected": -23.75,
+      "rewards/chosen": -2.859375,
+      "rewards/margins": 24.125,
+      "rewards/rejected": -27.0,
       "step": 1010
     },
     {
       "epoch": 1.068062827225131,
-      "grad_norm": 0.00270778571821124,
+      "grad_norm": 0.00129162521018807,
       "learning_rate": 4.067216434491059e-05,
-      "logits/chosen": -17.5,
-      "logits/rejected": -10.8125,
-      "logps/chosen": -740.0,
-      "logps/rejected": -4096.0,
-      "loss": 0.0058,
+      "logits/chosen": -14.375,
+      "logits/rejected": -10.0,
+      "logps/chosen": -696.0,
+      "logps/rejected": -4512.0,
+      "loss": 0.0031,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.609375,
-      "rewards/margins": 22.25,
-      "rewards/rejected": -25.875,
+      "rewards/chosen": -3.1875,
+      "rewards/margins": 26.75,
+      "rewards/rejected": -29.875,
       "step": 1020
     },
     {
       "epoch": 1.0785340314136125,
-      "grad_norm": 0.004209628075165994,
+      "grad_norm": 1.2919247637284166,
       "learning_rate": 4.043364736953866e-05,
-      "logits/chosen": -17.0,
-      "logits/rejected": -11.0,
-      "logps/chosen": -828.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0047,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -880.0,
+      "logps/rejected": -5088.0,
+      "loss": 0.0215,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.796875,
-      "rewards/margins": 22.0,
-      "rewards/rejected": -25.75,
+      "rewards/chosen": -4.3125,
+      "rewards/margins": 31.0,
+      "rewards/rejected": -35.25,
       "step": 1030
     },
     {
       "epoch": 1.0890052356020943,
-      "grad_norm": 0.007913080091272649,
+      "grad_norm": 0.00841434533680797,
       "learning_rate": 4.019283848649643e-05,
-      "logits/chosen": -17.0,
-      "logits/rejected": -10.6875,
-      "logps/chosen": -752.0,
-      "logps/rejected": -4064.0,
-      "loss": 0.0009,
+      "logits/chosen": -14.25,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3952.0,
+      "loss": 0.0189,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.578125,
-      "rewards/margins": 21.25,
-      "rewards/rejected": -24.75,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -23.625,
       "step": 1040
     },
     {
       "epoch": 1.0994764397905759,
-      "grad_norm": 0.07599450502576392,
+      "grad_norm": 0.08157229092728348,
       "learning_rate": 3.9949773456075215e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -10.4375,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0106,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -660.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0094,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.578125,
-      "rewards/margins": 22.375,
-      "rewards/rejected": -26.0,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 15.5,
+      "rewards/rejected": -18.375,
       "step": 1050
     },
     {
       "epoch": 1.1099476439790577,
-      "grad_norm": 0.032165241772376826,
+      "grad_norm": 0.02290890717281907,
       "learning_rate": 3.970448837360585e-05,
-      "logits/chosen": -16.5,
-      "logits/rejected": -10.75,
-      "logps/chosen": -816.0,
-      "logps/rejected": -3872.0,
-      "loss": 0.0017,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.78125,
-      "rewards/margins": 20.125,
-      "rewards/rejected": -24.0,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -8.625,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3136.0,
+      "loss": 0.008,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 13.4375,
+      "rewards/rejected": -16.625,
       "step": 1060
     },
     {
       "epoch": 1.1204188481675392,
-      "grad_norm": 0.009946849247570928,
+      "grad_norm": 0.027412666178410636,
       "learning_rate": 3.945701966409846e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -11.5,
-      "logps/chosen": -728.0,
-      "logps/rejected": -4096.0,
-      "loss": 0.0045,
+      "logits/chosen": -14.625,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3392.0,
+      "loss": 0.0115,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 21.875,
-      "rewards/rejected": -25.375,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 15.1875,
+      "rewards/rejected": -18.25,
       "step": 1070
     },
     {
       "epoch": 1.130890052356021,
-      "grad_norm": 0.01787227822567642,
+      "grad_norm": 0.02135877995551201,
       "learning_rate": 3.9207404076833374e-05,
-      "logits/chosen": -15.9375,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -820.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0032,
+      "logits/chosen": -14.4375,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -812.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.796875,
-      "rewards/margins": 21.75,
-      "rewards/rejected": -25.625,
+      "rewards/chosen": -3.71875,
+      "rewards/margins": 17.375,
+      "rewards/rejected": -21.125,
       "step": 1080
     },
     {
       "epoch": 1.1413612565445026,
-      "grad_norm": 0.02050033524977119,
+      "grad_norm": 0.01715122987150644,
       "learning_rate": 3.895567867990379e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -10.5,
-      "logps/chosen": -804.0,
-      "logps/rejected": -3920.0,
-      "loss": 0.0088,
+      "logits/chosen": -14.4375,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -836.0,
+      "logps/rejected": -4016.0,
+      "loss": 0.0067,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.640625,
-      "rewards/margins": 19.75,
-      "rewards/rejected": -23.5,
+      "rewards/chosen": -3.953125,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -24.5,
       "step": 1090
     },
     {
       "epoch": 1.1518324607329844,
-      "grad_norm": 0.019240491332168475,
+      "grad_norm": 0.006346095718303495,
       "learning_rate": 3.870188085471114e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -760.0,
-      "logps/rejected": -3888.0,
-      "loss": 0.0165,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -7.875,
+      "logps/chosen": -840.0,
+      "logps/rejected": -4288.0,
+      "loss": 0.0034,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.609375,
-      "rewards/margins": 20.125,
-      "rewards/rejected": -23.75,
+      "rewards/chosen": -4.40625,
+      "rewards/margins": 23.5,
+      "rewards/rejected": -27.875,
       "step": 1100
     },
     {
       "epoch": 1.1518324607329844,
-      "eval_logits/chosen": -16.375,
-      "eval_logits/rejected": -15.9375,
-      "eval_logps/chosen": -708.0,
-      "eval_logps/rejected": -720.0,
-      "eval_loss": 1.1375312805175781,
-      "eval_rewards/accuracies": 0.47265625,
-      "eval_rewards/chosen": -3.890625,
-      "eval_rewards/margins": 0.408203125,
-      "eval_rewards/rejected": -4.3125,
-      "eval_runtime": 46.6235,
-      "eval_samples_per_second": 42.897,
+      "eval_logits/chosen": -14.9375,
+      "eval_logits/rejected": -14.8125,
+      "eval_logps/chosen": -808.0,
+      "eval_logps/rejected": -828.0,
+      "eval_loss": 1.2924062013626099,
+      "eval_rewards/accuracies": 0.501953125,
+      "eval_rewards/chosen": -4.90625,
+      "eval_rewards/margins": 0.462890625,
+      "eval_rewards/rejected": -5.375,
+      "eval_runtime": 46.6306,
+      "eval_samples_per_second": 42.89,
       "eval_steps_per_second": 0.686,
       "step": 1100
     },
     {
       "epoch": 1.162303664921466,
-      "grad_norm": 0.02775507935637888,
+      "grad_norm": 0.007829083816249926,
       "learning_rate": 3.844604829041395e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -9.125,
-      "logps/chosen": -724.0,
-      "logps/rejected": -3616.0,
-      "loss": 0.0031,
+      "logits/chosen": -15.0,
+      "logits/rejected": -7.5,
+      "logps/chosen": -812.0,
+      "logps/rejected": -4352.0,
+      "loss": 0.0129,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 17.375,
-      "rewards/rejected": -20.875,
+      "rewards/chosen": -4.375,
+      "rewards/margins": 23.75,
+      "rewards/rejected": -28.125,
       "step": 1110
     },
     {
       "epoch": 1.1727748691099475,
-      "grad_norm": 0.14334234081518102,
+      "grad_norm": 0.7986623454595745,
       "learning_rate": 3.818821897833095e-05,
-      "logits/chosen": -16.5,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -724.0,
-      "logps/rejected": -3328.0,
-      "loss": 0.006,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 15.0625,
-      "rewards/rejected": -18.25,
-      "step": 1120
-    },
-    {
-      "epoch": 1.1832460732984293,
-      "grad_norm": 0.13883148099014533,
-      "learning_rate": 3.792843120629935e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -9.625,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3136.0,
-      "loss": 0.0066,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -7.3125,
+      "logps/chosen": -816.0,
+      "logps/rejected": -4160.0,
+      "loss": 0.0018,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.125,
+      "rewards/margins": 22.25,
+      "rewards/rejected": -26.375,
+      "step": 1120
+    },
+    {
+      "epoch": 1.1832460732984293,
+      "grad_norm": 0.1774577779982222,
+      "learning_rate": 3.792843120629935e-05,
+      "logits/chosen": -14.375,
+      "logits/rejected": -7.84375,
+      "logps/chosen": -792.0,
+      "logps/rejected": -4032.0,
+      "loss": 0.0102,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.046875,
-      "rewards/margins": 12.8125,
-      "rewards/rejected": -15.875,
+      "rewards/chosen": -4.125,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -24.75,
       "step": 1130
     },
     {
       "epoch": 1.193717277486911,
-      "grad_norm": 0.027681289072349582,
+      "grad_norm": 0.007517932464713113,
       "learning_rate": 3.766672355298907e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -9.5,
-      "logps/chosen": -640.0,
-      "logps/rejected": -3136.0,
-      "loss": 0.0071,
+      "logits/chosen": -14.75,
+      "logits/rejected": -8.5625,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3952.0,
+      "loss": 0.0044,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.75,
-      "rewards/margins": 13.5,
-      "rewards/rejected": -16.25,
+      "rewards/chosen": -3.46875,
+      "rewards/margins": 21.0,
+      "rewards/rejected": -24.5,
       "step": 1140
     },
     {
       "epoch": 1.2041884816753927,
-      "grad_norm": 0.006221153127032893,
+      "grad_norm": 0.02755056825794098,
       "learning_rate": 3.740313488217373e-05,
-      "logits/chosen": -17.25,
+      "logits/chosen": -15.625,
       "logits/rejected": -9.25,
-      "logps/chosen": -632.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0015,
+      "logps/chosen": -696.0,
+      "logps/rejected": -3552.0,
+      "loss": 0.0094,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.671875,
-      "rewards/margins": 16.25,
-      "rewards/rejected": -19.0,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 18.125,
+      "rewards/rejected": -21.5,
       "step": 1150
     },
     {
       "epoch": 1.2146596858638743,
-      "grad_norm": 0.01525414037672662,
+      "grad_norm": 0.04192639642320736,
       "learning_rate": 3.7137704336959464e-05,
-      "logits/chosen": -16.5,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -652.0,
-      "logps/rejected": -3760.0,
-      "loss": 0.001,
+      "logits/chosen": -15.8125,
+      "logits/rejected": -11.0,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0016,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 18.75,
-      "rewards/rejected": -21.625,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -19.75,
       "step": 1160
     },
     {
       "epoch": 1.225130890052356,
-      "grad_norm": 0.010197101946367053,
+      "grad_norm": 0.04475028063994474,
       "learning_rate": 3.687047133397201e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -7.59375,
-      "logps/chosen": -668.0,
-      "logps/rejected": -3776.0,
-      "loss": 0.0011,
+      "logits/chosen": -16.0,
+      "logits/rejected": -12.375,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0015,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.96875,
-      "rewards/margins": 19.625,
-      "rewards/rejected": -22.625,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -19.5,
       "step": 1170
     },
     {
       "epoch": 1.2356020942408377,
-      "grad_norm": 0.025036964614841038,
+      "grad_norm": 0.030137849187904627,
       "learning_rate": 3.6601475557503415e-05,
       "logits/chosen": -16.625,
-      "logits/rejected": -7.1875,
-      "logps/chosen": -684.0,
-      "logps/rejected": -4016.0,
-      "loss": 0.001,
+      "logits/rejected": -12.6875,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0021,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 21.625,
-      "rewards/rejected": -24.5,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 17.5,
+      "rewards/rejected": -20.625,
       "step": 1180
     },
     {
       "epoch": 1.2460732984293195,
-      "grad_norm": 0.15406262525919465,
+      "grad_norm": 0.09075149209761485,
       "learning_rate": 3.633075695361882e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.6875,
-      "logps/chosen": -648.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0038,
+      "logits/chosen": -16.5,
+      "logits/rejected": -12.5,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0433,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.890625,
-      "rewards/margins": 22.5,
-      "rewards/rejected": -25.375,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 18.125,
+      "rewards/rejected": -21.375,
       "step": 1190
     },
     {
       "epoch": 1.256544502617801,
-      "grad_norm": 0.16359899117491533,
+      "grad_norm": 1.793661939581947,
       "learning_rate": 3.6058355724224477e-05,
-      "logits/chosen": -17.375,
-      "logits/rejected": -9.25,
-      "logps/chosen": -664.0,
-      "logps/rejected": -4000.0,
-      "loss": 0.0019,
+      "logits/chosen": -16.875,
+      "logits/rejected": -12.9375,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0056,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.859375,
-      "rewards/margins": 21.125,
-      "rewards/rejected": -24.0,
+      "rewards/chosen": -3.609375,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -21.25,
       "step": 1200
     },
     {
       "epoch": 1.256544502617801,
-      "eval_logits/chosen": -17.25,
+      "eval_logits/chosen": -17.125,
       "eval_logits/rejected": -17.25,
-      "eval_logps/chosen": -604.0,
-      "eval_logps/rejected": -596.0,
-      "eval_loss": 0.9504687786102295,
-      "eval_rewards/accuracies": 0.490234375,
-      "eval_rewards/chosen": -2.875,
-      "eval_rewards/margins": 0.197265625,
-      "eval_rewards/rejected": -3.0625,
-      "eval_runtime": 46.6202,
-      "eval_samples_per_second": 42.9,
+      "eval_logps/chosen": -684.0,
+      "eval_logps/rejected": -680.0,
+      "eval_loss": 1.0445468425750732,
+      "eval_rewards/accuracies": 0.494140625,
+      "eval_rewards/chosen": -3.65625,
+      "eval_rewards/margins": 0.2412109375,
+      "eval_rewards/rejected": -3.90625,
+      "eval_runtime": 46.6313,
+      "eval_samples_per_second": 42.89,
       "eval_steps_per_second": 0.686,
       "step": 1200
     },
     {
       "epoch": 1.2670157068062826,
-      "grad_norm": 0.2501885677040669,
+      "grad_norm": 0.3618488236936621,
       "learning_rate": 3.5784312321097744e-05,
-      "logits/chosen": -17.0,
-      "logits/rejected": -8.6875,
-      "logps/chosen": -680.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0031,
+      "logits/chosen": -16.625,
+      "logits/rejected": -12.5625,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3392.0,
+      "loss": 0.0449,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.734375,
-      "rewards/margins": 22.0,
-      "rewards/rejected": -24.75,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -19.5,
       "step": 1210
     },
     {
       "epoch": 1.2774869109947644,
-      "grad_norm": 0.006606685774812871,
+      "grad_norm": 0.2505053989180939,
       "learning_rate": 3.550866743987994e-05,
-      "logits/chosen": -16.375,
-      "logits/rejected": -7.125,
-      "logps/chosen": -728.0,
-      "logps/rejected": -4064.0,
-      "loss": 0.0021,
+      "logits/chosen": -16.25,
+      "logits/rejected": -12.8125,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.0068,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 22.625,
-      "rewards/rejected": -25.75,
+      "rewards/chosen": -3.140625,
+      "rewards/margins": 15.1875,
+      "rewards/rejected": -18.25,
       "step": 1220
     },
     {
       "epoch": 1.2879581151832462,
-      "grad_norm": 0.006482167203350088,
+      "grad_norm": 0.017917408120364006,
       "learning_rate": 3.5231462014033024e-05,
       "logits/chosen": -16.625,
-      "logits/rejected": -7.90625,
-      "logps/chosen": -656.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0151,
+      "logits/rejected": -13.6875,
+      "logps/chosen": -632.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0011,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.875,
-      "rewards/margins": 21.75,
-      "rewards/rejected": -24.625,
+      "rewards/chosen": -2.640625,
+      "rewards/margins": 15.8125,
+      "rewards/rejected": -18.375,
       "step": 1230
     },
     {
       "epoch": 1.2984293193717278,
-      "grad_norm": 2.4701887729086143,
+      "grad_norm": 0.00596531294924745,
       "learning_rate": 3.495273720876095e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -7.84375,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3968.0,
-      "loss": 0.0036,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.125,
-      "rewards/margins": 21.375,
-      "rewards/rejected": -24.5,
+      "logits/chosen": -16.5,
+      "logits/rejected": -13.0625,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3440.0,
+      "loss": 0.0024,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 16.25,
+      "rewards/rejected": -19.25,
       "step": 1240
     },
     {
       "epoch": 1.3089005235602094,
-      "grad_norm": 0.018151824444800773,
+      "grad_norm": 0.04526162228762231,
       "learning_rate": 3.4672534414896616e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -7.375,
-      "logps/chosen": -756.0,
-      "logps/rejected": -3984.0,
-      "loss": 0.002,
+      "logits/chosen": -16.875,
+      "logits/rejected": -12.5625,
+      "logps/chosen": -764.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0032,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.3125,
-      "rewards/margins": 21.375,
-      "rewards/rejected": -24.75,
+      "rewards/chosen": -3.421875,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -21.5,
       "step": 1250
     },
     {
       "epoch": 1.3193717277486912,
-      "grad_norm": 0.02153888139019444,
+      "grad_norm": 0.2508856827199363,
       "learning_rate": 3.439089524275526e-05,
-      "logits/chosen": -15.875,
-      "logits/rejected": -7.0,
-      "logps/chosen": -704.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0017,
+      "logits/chosen": -16.5,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0096,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 22.5,
-      "rewards/rejected": -25.75,
+      "rewards/chosen": -3.5625,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -22.5,
       "step": 1260
     },
     {
       "epoch": 1.3298429319371727,
-      "grad_norm": 0.020095496590698026,
+      "grad_norm": 0.029177640900151583,
       "learning_rate": 3.410786151595533e-05,
-      "logits/chosen": -15.75,
-      "logits/rejected": -8.25,
-      "logps/chosen": -668.0,
-      "logps/rejected": -3984.0,
-      "loss": 0.0017,
+      "logits/chosen": -16.875,
+      "logits/rejected": -12.5,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3632.0,
+      "loss": 0.0023,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 22.125,
-      "rewards/rejected": -25.125,
+      "rewards/chosen": -3.328125,
+      "rewards/margins": 18.25,
+      "rewards/rejected": -21.5,
       "step": 1270
     },
     {
       "epoch": 1.3403141361256545,
-      "grad_norm": 0.04031243768881032,
+      "grad_norm": 0.04041826941755009,
       "learning_rate": 3.382347526520765e-05,
-      "logits/chosen": -15.4375,
-      "logits/rejected": -6.03125,
-      "logps/chosen": -712.0,
-      "logps/rejected": -4000.0,
-      "loss": 0.0187,
+      "logits/chosen": -17.0,
+      "logits/rejected": -11.5,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3552.0,
+      "loss": 0.0158,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.125,
-      "rewards/margins": 21.5,
-      "rewards/rejected": -24.75,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -20.25,
       "step": 1280
     },
     {
       "epoch": 1.350785340314136,
-      "grad_norm": 0.006502584658035976,
+      "grad_norm": 0.006360189818775591,
       "learning_rate": 3.353777872207381e-05,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -6.0625,
-      "logps/chosen": -716.0,
-      "logps/rejected": -3872.0,
-      "loss": 0.0013,
+      "logits/chosen": -16.75,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0025,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.171875,
-      "rewards/margins": 20.75,
-      "rewards/rejected": -23.875,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 16.25,
+      "rewards/rejected": -19.25,
       "step": 1290
     },
     {
       "epoch": 1.3612565445026177,
-      "grad_norm": 0.04039484425165907,
+      "grad_norm": 0.046480605965547046,
       "learning_rate": 3.325081431269475e-05,
-      "logits/chosen": -12.625,
-      "logits/rejected": -6.4375,
-      "logps/chosen": -692.0,
-      "logps/rejected": -4032.0,
-      "loss": 0.0016,
+      "logits/chosen": -16.375,
+      "logits/rejected": -10.5,
+      "logps/chosen": -664.0,
+      "logps/rejected": -3552.0,
+      "loss": 0.0049,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.390625,
-      "rewards/margins": 21.0,
-      "rewards/rejected": -24.375,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -19.5,
       "step": 1300
     },
     {
       "epoch": 1.3612565445026177,
-      "eval_logits/chosen": -13.625,
-      "eval_logits/rejected": -13.125,
-      "eval_logps/chosen": -684.0,
-      "eval_logps/rejected": -688.0,
-      "eval_loss": 1.0801249742507935,
-      "eval_rewards/accuracies": 0.482421875,
-      "eval_rewards/chosen": -3.65625,
-      "eval_rewards/margins": 0.345703125,
-      "eval_rewards/rejected": -4.0,
-      "eval_runtime": 46.6393,
-      "eval_samples_per_second": 42.882,
+      "eval_logits/chosen": -16.75,
+      "eval_logits/rejected": -16.875,
+      "eval_logps/chosen": -660.0,
+      "eval_logps/rejected": -656.0,
+      "eval_loss": 1.0295780897140503,
+      "eval_rewards/accuracies": 0.48828125,
+      "eval_rewards/chosen": -3.40625,
+      "eval_rewards/margins": 0.24609375,
+      "eval_rewards/rejected": -3.65625,
+      "eval_runtime": 46.6192,
+      "eval_samples_per_second": 42.901,
       "eval_steps_per_second": 0.686,
       "step": 1300
     },
     {
       "epoch": 1.3717277486910995,
-      "grad_norm": 0.01963561323245404,
+      "grad_norm": 0.03352564306328381,
       "learning_rate": 3.296262465149046e-05,
-      "logits/chosen": -14.25,
-      "logits/rejected": -7.5625,
-      "logps/chosen": -748.0,
-      "logps/rejected": -3984.0,
-      "loss": 0.003,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 20.75,
-      "rewards/rejected": -24.125,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -11.5,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0117,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 17.125,
+      "rewards/rejected": -20.375,
       "step": 1310
     },
     {
       "epoch": 1.3821989528795813,
-      "grad_norm": 7.9808329122691e-05,
+      "grad_norm": 0.01524638681019545,
       "learning_rate": 3.2673252534831687e-05,
-      "logits/chosen": -14.875,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -688.0,
-      "logps/rejected": -3792.0,
-      "loss": 0.0024,
+      "logits/chosen": -16.25,
+      "logits/rejected": -11.0,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0044,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.015625,
-      "rewards/margins": 20.125,
-      "rewards/rejected": -23.25,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -22.375,
       "step": 1320
     },
     {
       "epoch": 1.3926701570680629,
-      "grad_norm": 0.0009559392153135045,
+      "grad_norm": 0.005943143315185005,
       "learning_rate": 3.2382740934684696e-05,
-      "logits/chosen": -14.75,
-      "logits/rejected": -8.875,
-      "logps/chosen": -696.0,
-      "logps/rejected": -4016.0,
-      "loss": 0.0114,
+      "logits/chosen": -16.125,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -712.0,
+      "logps/rejected": -4080.0,
+      "loss": 0.0095,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 20.875,
-      "rewards/rejected": -24.125,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 21.375,
+      "rewards/rejected": -24.75,
       "step": 1330
     },
     {
       "epoch": 1.4031413612565444,
-      "grad_norm": 1.1508737816097288,
+      "grad_norm": 0.10456849378151957,
       "learning_rate": 3.209113299222982e-05,
-      "logits/chosen": -15.0,
-      "logits/rejected": -9.0,
-      "logps/chosen": -624.0,
-      "logps/rejected": -3824.0,
-      "loss": 0.0118,
+      "logits/chosen": -16.125,
+      "logits/rejected": -8.875,
+      "logps/chosen": -640.0,
+      "logps/rejected": -3920.0,
+      "loss": 0.0079,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.890625,
-      "rewards/margins": 19.375,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.25,
       "step": 1340
     },
     {
       "epoch": 1.4136125654450262,
-      "grad_norm": 0.026595969943652712,
+      "grad_norm": 0.011074338141496845,
       "learning_rate": 3.179847201145505e-05,
-      "logits/chosen": -14.375,
-      "logits/rejected": -7.09375,
-      "logps/chosen": -716.0,
-      "logps/rejected": -3920.0,
-      "loss": 0.0047,
+      "logits/chosen": -15.875,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0036,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.125,
-      "rewards/margins": 19.125,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -3.1875,
+      "rewards/margins": 18.375,
+      "rewards/rejected": -21.5,
       "step": 1350
     },
     {
       "epoch": 1.4240837696335078,
-      "grad_norm": 0.07707864161719753,
+      "grad_norm": 0.03215748436111734,
       "learning_rate": 3.150480145272528e-05,
-      "logits/chosen": -13.875,
-      "logits/rejected": -6.0,
-      "logps/chosen": -752.0,
-      "logps/rejected": -4000.0,
-      "loss": 0.001,
+      "logits/chosen": -15.75,
+      "logits/rejected": -8.5,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0018,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.640625,
-      "rewards/margins": 20.75,
-      "rewards/rejected": -24.375,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 18.625,
+      "rewards/rejected": -21.5,
       "step": 1360
     },
     {
       "epoch": 1.4345549738219896,
-      "grad_norm": 0.01908869194349684,
+      "grad_norm": 0.0034642834908974176,
       "learning_rate": 3.121016492632848e-05,
-      "logits/chosen": -13.875,
-      "logits/rejected": -5.8125,
-      "logps/chosen": -772.0,
-      "logps/rejected": -3936.0,
+      "logits/chosen": -15.6875,
+      "logits/rejected": -8.5625,
+      "logps/chosen": -696.0,
+      "logps/rejected": -3696.0,
       "loss": 0.0035,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.75,
-      "rewards/margins": 20.625,
-      "rewards/rejected": -24.375,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 19.0,
+      "rewards/rejected": -22.0,
       "step": 1370
     },
     {
       "epoch": 1.4450261780104712,
-      "grad_norm": 0.023050835282196147,
+      "grad_norm": 0.018780622759260988,
       "learning_rate": 3.091460618599951e-05,
-      "logits/chosen": -14.75,
-      "logits/rejected": -5.59375,
-      "logps/chosen": -704.0,
-      "logps/rejected": -3584.0,
-      "loss": 0.0059,
+      "logits/chosen": -16.25,
+      "logits/rejected": -8.0,
+      "logps/chosen": -720.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0062,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.140625,
-      "rewards/margins": 19.0,
-      "rewards/rejected": -22.125,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 19.25,
+      "rewards/rejected": -22.5,
       "step": 1380
     },
     {
       "epoch": 1.4554973821989527,
-      "grad_norm": 0.004455469185940304,
+      "grad_norm": 0.009796326790758971,
       "learning_rate": 3.061816912242265e-05,
-      "logits/chosen": -15.125,
-      "logits/rejected": -5.8125,
-      "logps/chosen": -724.0,
-      "logps/rejected": -3712.0,
-      "loss": 0.0015,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -748.0,
+      "logps/rejected": -3952.0,
+      "loss": 0.0011,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.171875,
-      "rewards/margins": 18.25,
-      "rewards/rejected": -21.5,
+      "rewards/chosen": -3.421875,
+      "rewards/margins": 20.375,
+      "rewards/rejected": -23.75,
       "step": 1390
     },
     {
       "epoch": 1.4659685863874345,
-      "grad_norm": 0.0005802799230889754,
+      "grad_norm": 0.002955424596736435,
       "learning_rate": 3.0320897756713785e-05,
-      "logits/chosen": -14.75,
-      "logits/rejected": -7.09375,
-      "logps/chosen": -688.0,
-      "logps/rejected": -3808.0,
-      "loss": 0.0015,
+      "logits/chosen": -16.0,
+      "logits/rejected": -8.875,
+      "logps/chosen": -732.0,
+      "logps/rejected": -4032.0,
+      "loss": 0.0009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.890625,
-      "rewards/margins": 18.75,
-      "rewards/rejected": -21.625,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -23.75,
       "step": 1400
     },
     {
       "epoch": 1.4659685863874345,
-      "eval_logits/chosen": -14.75,
-      "eval_logits/rejected": -14.375,
-      "eval_logps/chosen": -660.0,
-      "eval_logps/rejected": -668.0,
-      "eval_loss": 1.0415312051773071,
-      "eval_rewards/accuracies": 0.501953125,
-      "eval_rewards/chosen": -3.4375,
-      "eval_rewards/margins": 0.36328125,
-      "eval_rewards/rejected": -3.796875,
-      "eval_runtime": 46.6258,
-      "eval_samples_per_second": 42.895,
+      "eval_logits/chosen": -16.625,
+      "eval_logits/rejected": -16.625,
+      "eval_logps/chosen": -700.0,
+      "eval_logps/rejected": -708.0,
+      "eval_loss": 1.090343713760376,
+      "eval_rewards/accuracies": 0.490234375,
+      "eval_rewards/chosen": -3.8125,
+      "eval_rewards/margins": 0.3828125,
+      "eval_rewards/rejected": -4.1875,
+      "eval_runtime": 46.6309,
+      "eval_samples_per_second": 42.89,
       "eval_steps_per_second": 0.686,
       "step": 1400
     },
     {
       "epoch": 1.4764397905759163,
-      "grad_norm": 1.1791292215881397,
+      "grad_norm": 0.021914687658456458,
       "learning_rate": 3.002283623388332e-05,
-      "logits/chosen": -15.125,
-      "logits/rejected": -6.75,
-      "logps/chosen": -664.0,
-      "logps/rejected": -3696.0,
-      "loss": 0.0043,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.125,
-      "rewards/margins": 19.25,
-      "rewards/rejected": -22.375,
+      "logits/chosen": -16.375,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0164,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.328125,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -23.125,
       "step": 1410
     },
     {
       "epoch": 1.486910994764398,
-      "grad_norm": 2.066865645485102,
+      "grad_norm": 2.8893141839466616,
       "learning_rate": 2.9724028816280507e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -7.25,
-      "logps/chosen": -704.0,
-      "logps/rejected": -3936.0,
-      "loss": 0.0113,
+      "logits/chosen": -16.625,
+      "logits/rejected": -11.375,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.028,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.296875,
-      "rewards/margins": 20.75,
-      "rewards/rejected": -24.0,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -23.75,
       "step": 1420
     },
     {
       "epoch": 1.4973821989528795,
-      "grad_norm": 0.02469927504991995,
+      "grad_norm": 0.0049634031713516,
       "learning_rate": 2.9424519877020524e-05,
-      "logits/chosen": -17.375,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -732.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.0019,
+      "logits/chosen": -16.5,
+      "logits/rejected": -12.0625,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0038,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.171875,
-      "rewards/margins": 20.0,
-      "rewards/rejected": -23.125,
+      "rewards/chosen": -3.25,
+      "rewards/margins": 18.125,
+      "rewards/rejected": -21.375,
       "step": 1430
     },
-    {
-      "epoch": 1.5078534031413613,
-      "grad_norm": 0.011538619153821013,
-      "learning_rate": 2.912435389339504e-05,
-      "logits/chosen": -17.25,
-      "logits/rejected": -9.125,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3776.0,
-      "loss": 0.0053,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.96875,
-      "rewards/margins": 19.5,
-      "rewards/rejected": -22.5,
+    {
+      "epoch": 1.5078534031413613,
+      "grad_norm": 7.299615840940962e-05,
+      "learning_rate": 2.912435389339504e-05,
+      "logits/chosen": -16.375,
+      "logits/rejected": -11.875,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0095,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 17.875,
+      "rewards/rejected": -21.125,
       "step": 1440
     },
     {
       "epoch": 1.518324607329843,
-      "grad_norm": 0.024617939984094436,
+      "grad_norm": 0.00963523348072373,
       "learning_rate": 2.8823575440267226e-05,
-      "logits/chosen": -15.1875,
-      "logits/rejected": -6.40625,
-      "logps/chosen": -736.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0046,
+      "logits/chosen": -16.125,
+      "logits/rejected": -11.125,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3984.0,
+      "loss": 0.008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.578125,
-      "rewards/margins": 22.375,
-      "rewards/rejected": -26.0,
+      "rewards/chosen": -3.546875,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.625,
       "step": 1450
     },
     {
       "epoch": 1.5287958115183247,
-      "grad_norm": 0.12372189236877726,
+      "grad_norm": 0.9779842569379582,
       "learning_rate": 2.8522229183452465e-05,
-      "logits/chosen": -14.25,
-      "logits/rejected": -6.46875,
-      "logps/chosen": -740.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0382,
+      "logits/chosen": -16.0,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -768.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0414,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 19.75,
-      "rewards/rejected": -23.125,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 18.625,
+      "rewards/rejected": -22.375,
       "step": 1460
     },
     {
       "epoch": 1.5392670157068062,
-      "grad_norm": 0.027667000244700098,
+      "grad_norm": 0.6532398372793786,
       "learning_rate": 2.8220359873085373e-05,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -7.28125,
-      "logps/chosen": -744.0,
-      "logps/rejected": -3392.0,
-      "loss": 0.0052,
+      "logits/chosen": -15.875,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.0053,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 14.5625,
-      "rewards/rejected": -18.0,
+      "rewards/chosen": -3.578125,
+      "rewards/margins": 13.6875,
+      "rewards/rejected": -17.25,
       "step": 1470
     },
     {
       "epoch": 1.5497382198952878,
-      "grad_norm": 1.613936268247198,
+      "grad_norm": 0.25030425569480075,
       "learning_rate": 2.7918012336974388e-05,
-      "logits/chosen": -12.8125,
-      "logits/rejected": -6.34375,
-      "logps/chosen": -724.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0095,
+      "logits/chosen": -16.375,
+      "logits/rejected": -10.375,
+      "logps/chosen": -692.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.0038,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.3125,
-      "rewards/margins": 14.75,
-      "rewards/rejected": -18.0,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 12.125,
+      "rewards/rejected": -15.125,
       "step": 1480
     },
     {
       "epoch": 1.5602094240837696,
-      "grad_norm": 0.03596904162336614,
+      "grad_norm": 0.036634036125090565,
       "learning_rate": 2.7615231473944815e-05,
-      "logits/chosen": -14.0,
-      "logits/rejected": -7.15625,
-      "logps/chosen": -656.0,
-      "logps/rejected": -3104.0,
-      "loss": 0.0073,
+      "logits/chosen": -16.25,
+      "logits/rejected": -11.4375,
+      "logps/chosen": -652.0,
+      "logps/rejected": -3024.0,
+      "loss": 0.0038,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.875,
-      "rewards/margins": 13.5,
-      "rewards/rejected": -16.375,
+      "rewards/chosen": -2.828125,
+      "rewards/margins": 12.75,
+      "rewards/rejected": -15.5625,
       "step": 1490
     },
     {
       "epoch": 1.5706806282722514,
-      "grad_norm": 0.3170858707393714,
+      "grad_norm": 0.04172386843023788,
       "learning_rate": 2.7312062247171327e-05,
-      "logits/chosen": -14.9375,
-      "logits/rejected": -9.125,
-      "logps/chosen": -648.0,
-      "logps/rejected": -3072.0,
-      "loss": 0.0037,
+      "logits/chosen": -16.375,
+      "logits/rejected": -12.0,
+      "logps/chosen": -644.0,
+      "logps/rejected": -3088.0,
+      "loss": 0.0026,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.859375,
-      "rewards/margins": 13.25,
-      "rewards/rejected": -16.125,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 13.4375,
+      "rewards/rejected": -16.25,
       "step": 1500
     },
     {
       "epoch": 1.5706806282722514,
-      "eval_logits/chosen": -14.375,
-      "eval_logits/rejected": -14.0,
-      "eval_logps/chosen": -632.0,
-      "eval_logps/rejected": -632.0,
-      "eval_loss": 1.0120937824249268,
-      "eval_rewards/accuracies": 0.5,
-      "eval_rewards/chosen": -3.140625,
-      "eval_rewards/margins": 0.29296875,
-      "eval_rewards/rejected": -3.4375,
-      "eval_runtime": 46.6203,
-      "eval_samples_per_second": 42.9,
+      "eval_logits/chosen": -16.75,
+      "eval_logits/rejected": -16.75,
+      "eval_logps/chosen": -624.0,
+      "eval_logps/rejected": -624.0,
+      "eval_loss": 0.9406718611717224,
+      "eval_rewards/accuracies": 0.498046875,
+      "eval_rewards/chosen": -3.078125,
+      "eval_rewards/margins": 0.291015625,
+      "eval_rewards/rejected": -3.359375,
+      "eval_runtime": 46.6282,
+      "eval_samples_per_second": 42.893,
       "eval_steps_per_second": 0.686,
       "step": 1500
     },
     {
       "epoch": 1.581151832460733,
-      "grad_norm": 0.4322792920461349,
+      "grad_norm": 0.1633304929966964,
       "learning_rate": 2.7008549677500876e-05,
-      "logits/chosen": -15.5625,
-      "logits/rejected": -8.4375,
-      "logps/chosen": -720.0,
-      "logps/rejected": -3360.0,
-      "loss": 0.0038,
+      "logits/chosen": -16.625,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.0029,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.09375,
-      "rewards/margins": 14.3125,
-      "rewards/rejected": -17.375,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 14.1875,
+      "rewards/rejected": -17.125,
       "step": 1510
     },
     {
       "epoch": 1.5916230366492146,
-      "grad_norm": 0.009001812007699686,
+      "grad_norm": 0.008836110812112791,
       "learning_rate": 2.670473883676709e-05,
-      "logits/chosen": -15.25,
-      "logits/rejected": -8.5,
-      "logps/chosen": -708.0,
-      "logps/rejected": -3440.0,
-      "loss": 0.0014,
+      "logits/chosen": -16.5,
+      "logits/rejected": -10.9375,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3184.0,
+      "loss": 0.0016,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.125,
-      "rewards/margins": 16.375,
-      "rewards/rejected": -19.5,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 14.0,
+      "rewards/rejected": -17.0,
       "step": 1520
     },
     {
       "epoch": 1.6020942408376964,
-      "grad_norm": 0.04858996739957046,
+      "grad_norm": 0.023388236457014894,
       "learning_rate": 2.6400674841097074e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -7.84375,
-      "logps/chosen": -748.0,
-      "logps/rejected": -4000.0,
-      "loss": 0.0011,
+      "logits/chosen": -15.6875,
+      "logits/rejected": -10.25,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.46875,
-      "rewards/margins": 19.5,
-      "rewards/rejected": -23.0,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 16.625,
+      "rewards/rejected": -20.0,
       "step": 1530
     },
     {
       "epoch": 1.6125654450261782,
-      "grad_norm": 0.373490748784985,
+      "grad_norm": 1.0901343226377085,
       "learning_rate": 2.609640284421158e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -7.90625,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.003,
+      "logits/chosen": -16.25,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3600.0,
+      "loss": 0.0055,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.984375,
-      "rewards/margins": 20.125,
-      "rewards/rejected": -23.125,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -20.125,
       "step": 1540
     },
     {
       "epoch": 1.6230366492146597,
-      "grad_norm": 0.000800147296026108,
+      "grad_norm": 0.02273280873522998,
       "learning_rate": 2.5791968030719726e-05,
       "logits/chosen": -16.25,
-      "logits/rejected": -7.25,
-      "logps/chosen": -744.0,
-      "logps/rejected": -3920.0,
-      "loss": 0.0015,
+      "logits/rejected": -9.125,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3520.0,
+      "loss": 0.0027,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.171875,
-      "rewards/margins": 21.5,
-      "rewards/rejected": -24.625,
+      "rewards/chosen": -3.078125,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -20.625,
       "step": 1550
     },
     {
       "epoch": 1.6335078534031413,
-      "grad_norm": 0.00374900210577065,
+      "grad_norm": 0.03673216390872207,
       "learning_rate": 2.548741560940902e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -7.53125,
-      "logps/chosen": -688.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0009,
+      "logits/chosen": -16.5,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -668.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0029,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.125,
-      "rewards/margins": 21.875,
-      "rewards/rejected": -25.0,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 17.125,
+      "rewards/rejected": -20.0,
       "step": 1560
     },
     {
       "epoch": 1.6439790575916229,
-      "grad_norm": 0.028118741371708094,
+      "grad_norm": 0.23942743097957062,
       "learning_rate": 2.5182790806531787e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -7.90625,
-      "logps/chosen": -760.0,
-      "logps/rejected": -3984.0,
-      "loss": 0.0013,
+      "logits/chosen": -16.375,
+      "logits/rejected": -9.375,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0026,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.453125,
-      "rewards/margins": 22.0,
-      "rewards/rejected": -25.5,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -20.25,
       "step": 1570
     },
     {
       "epoch": 1.6544502617801047,
-      "grad_norm": 0.11483350905035293,
+      "grad_norm": 0.1905973785604278,
       "learning_rate": 2.4878138859089074e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -7.8125,
-      "logps/chosen": -684.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0027,
+      "logits/chosen": -16.875,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0022,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.96875,
-      "rewards/margins": 22.75,
-      "rewards/rejected": -25.75,
+      "rewards/chosen": -2.8125,
+      "rewards/margins": 18.125,
+      "rewards/rejected": -20.875,
       "step": 1580
     },
     {
       "epoch": 1.6649214659685865,
-      "grad_norm": 0.05175941037134179,
+      "grad_norm": 0.10292883528128294,
       "learning_rate": 2.4573505008112924e-05,
       "logits/chosen": -16.5,
-      "logits/rejected": -8.125,
-      "logps/chosen": -704.0,
-      "logps/rejected": -4048.0,
-      "loss": 0.0005,
+      "logits/rejected": -8.625,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 21.5,
-      "rewards/rejected": -24.75,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -21.0,
       "step": 1590
     },
     {
       "epoch": 1.675392670157068,
-      "grad_norm": 0.002117649958991347,
+      "grad_norm": 0.008929125203784105,
       "learning_rate": 2.4268934491948032e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -7.78125,
-      "logps/chosen": -744.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0025,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 22.5,
-      "rewards/rejected": -25.875,
+      "logits/chosen": -16.5,
+      "logits/rejected": -8.875,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0009,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 18.375,
+      "rewards/rejected": -21.5,
       "step": 1600
     },
     {
       "epoch": 1.675392670157068,
-      "eval_logits/chosen": -15.4375,
-      "eval_logits/rejected": -15.125,
-      "eval_logps/chosen": -700.0,
-      "eval_logps/rejected": -708.0,
-      "eval_loss": 1.1921093463897705,
-      "eval_rewards/accuracies": 0.490234375,
-      "eval_rewards/chosen": -3.8125,
-      "eval_rewards/margins": 0.3984375,
-      "eval_rewards/rejected": -4.21875,
-      "eval_runtime": 46.638,
-      "eval_samples_per_second": 42.883,
+      "eval_logits/chosen": -17.0,
+      "eval_logits/rejected": -16.75,
+      "eval_logps/chosen": -652.0,
+      "eval_logps/rejected": -656.0,
+      "eval_loss": 0.9907343983650208,
+      "eval_rewards/accuracies": 0.498046875,
+      "eval_rewards/chosen": -3.34375,
+      "eval_rewards/margins": 0.341796875,
+      "eval_rewards/rejected": -3.671875,
+      "eval_runtime": 46.6293,
+      "eval_samples_per_second": 42.892,
       "eval_steps_per_second": 0.686,
       "step": 1600
     },
     {
       "epoch": 1.6858638743455496,
-      "grad_norm": 0.002932412337814824,
+      "grad_norm": 0.002689921366713306,
       "learning_rate": 2.396447253953385e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -7.59375,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4448.0,
-      "loss": 0.0019,
+      "logits/chosen": -16.375,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 25.0,
-      "rewards/rejected": -28.5,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 19.5,
+      "rewards/rejected": -22.5,
       "step": 1610
     },
     {
       "epoch": 1.6963350785340314,
-      "grad_norm": 0.021672891655779257,
+      "grad_norm": 0.004695245795980241,
       "learning_rate": 2.3660164363687998e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -7.1875,
-      "logps/chosen": -748.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0014,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.9375,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.1875,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 18.375,
+      "rewards/rejected": -21.375,
       "step": 1620
     },
     {
       "epoch": 1.7068062827225132,
-      "grad_norm": 0.010723243328225105,
+      "grad_norm": 0.027243903766862452,
       "learning_rate": 2.33560551543922e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -7.28125,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0042,
+      "logits/chosen": -16.375,
+      "logits/rejected": -9.125,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3392.0,
+      "loss": 0.0062,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 25.125,
-      "rewards/rejected": -28.375,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 16.625,
+      "rewards/rejected": -19.5,
       "step": 1630
     },
     {
       "epoch": 1.7172774869109948,
-      "grad_norm": 0.6939503586114789,
+      "grad_norm": 1.141533437694049,
       "learning_rate": 2.3052190072081493e-05,
       "logits/chosen": -16.25,
-      "logits/rejected": -8.0,
-      "logps/chosen": -720.0,
-      "logps/rejected": -4448.0,
-      "loss": 0.0016,
+      "logits/rejected": -9.25,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3472.0,
+      "loss": 0.003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.203125,
-      "rewards/margins": 25.875,
-      "rewards/rejected": -29.125,
+      "rewards/chosen": -2.75,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -19.25,
       "step": 1640
     },
     {
       "epoch": 1.7277486910994764,
-      "grad_norm": 0.003990055324601182,
+      "grad_norm": 0.02557206809389542,
       "learning_rate": 2.2748614240937864e-05,
-      "logits/chosen": -15.0625,
-      "logits/rejected": -7.5,
-      "logps/chosen": -704.0,
-      "logps/rejected": -4640.0,
-      "loss": 0.0039,
+      "logits/chosen": -14.625,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -660.0,
+      "logps/rejected": -3504.0,
+      "loss": 0.002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 27.875,
-      "rewards/rejected": -31.25,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 16.875,
+      "rewards/rejected": -19.875,
       "step": 1650
     },
     {
       "epoch": 1.738219895287958,
-      "grad_norm": 0.04287829863524718,
+      "grad_norm": 0.03556740008636162,
       "learning_rate": 2.2445372742189334e-05,
-      "logits/chosen": -15.125,
-      "logits/rejected": -6.875,
-      "logps/chosen": -796.0,
-      "logps/rejected": -5024.0,
-      "loss": 0.004,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.953125,
-      "rewards/margins": 29.625,
-      "rewards/rejected": -33.5,
+      "logits/chosen": -13.0,
+      "logits/rejected": -6.25,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0009,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.375,
+      "rewards/margins": 19.0,
+      "rewards/rejected": -22.375,
       "step": 1660
     },
     {
       "epoch": 1.7486910994764397,
-      "grad_norm": 0.0005149145355830644,
+      "grad_norm": 0.00044968681834409293,
       "learning_rate": 2.214251060741528e-05,
-      "logits/chosen": -15.375,
-      "logits/rejected": -6.625,
-      "logps/chosen": -848.0,
-      "logps/rejected": -4736.0,
-      "loss": 0.0017,
+      "logits/chosen": -13.875,
+      "logits/rejected": -6.34375,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0044,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -4.3125,
-      "rewards/margins": 28.75,
-      "rewards/rejected": -33.0,
+      "rewards/chosen": -3.640625,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -23.0,
       "step": 1670
     },
     {
       "epoch": 1.7591623036649215,
-      "grad_norm": 0.0051900117814029185,
+      "grad_norm": 0.1816829023550082,
       "learning_rate": 2.1840072811859302e-05,
-      "logits/chosen": -15.625,
-      "logits/rejected": -6.75,
-      "logps/chosen": -872.0,
-      "logps/rejected": -4960.0,
-      "loss": 0.0098,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -4.59375,
-      "rewards/margins": 30.625,
-      "rewards/rejected": -35.25,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -6.5,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.004,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -23.0,
       "step": 1680
     },
     {
       "epoch": 1.7696335078534031,
-      "grad_norm": 0.5862348093698138,
+      "grad_norm": 0.29437622637916083,
       "learning_rate": 2.1538104267750287e-05,
-      "logits/chosen": -14.375,
-      "logits/rejected": -6.96875,
-      "logps/chosen": -800.0,
-      "logps/rejected": -4896.0,
-      "loss": 0.0017,
+      "logits/chosen": -12.75,
+      "logits/rejected": -7.0,
+      "logps/chosen": -684.0,
+      "logps/rejected": -4000.0,
+      "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -4.34375,
-      "rewards/margins": 28.375,
-      "rewards/rejected": -32.75,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -23.75,
       "step": 1690
     },
     {
       "epoch": 1.7801047120418847,
-      "grad_norm": 0.0015988060852680864,
+      "grad_norm": 0.006594000424464487,
       "learning_rate": 2.1236649817632952e-05,
-      "logits/chosen": -13.875,
-      "logits/rejected": -6.25,
-      "logps/chosen": -792.0,
-      "logps/rejected": -4352.0,
-      "loss": 0.0029,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -4.40625,
-      "rewards/margins": 23.5,
-      "rewards/rejected": -27.875,
+      "logits/chosen": -12.875,
+      "logits/rejected": -6.4375,
+      "logps/chosen": -668.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0078,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.1875,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -22.375,
       "step": 1700
     },
     {
       "epoch": 1.7801047120418847,
-      "eval_logits/chosen": -13.1875,
-      "eval_logits/rejected": -12.8125,
-      "eval_logps/chosen": -812.0,
-      "eval_logps/rejected": -832.0,
-      "eval_loss": 1.4102656841278076,
-      "eval_rewards/accuracies": 0.490234375,
-      "eval_rewards/chosen": -4.9375,
-      "eval_rewards/margins": 0.5078125,
-      "eval_rewards/rejected": -5.4375,
-      "eval_runtime": 46.63,
-      "eval_samples_per_second": 42.891,
+      "eval_logits/chosen": -12.75,
+      "eval_logits/rejected": -12.5,
+      "eval_logps/chosen": -652.0,
+      "eval_logps/rejected": -652.0,
+      "eval_loss": 0.9547187685966492,
+      "eval_rewards/accuracies": 0.494140625,
+      "eval_rewards/chosen": -3.328125,
+      "eval_rewards/margins": 0.291015625,
+      "eval_rewards/rejected": -3.625,
+      "eval_runtime": 46.6314,
+      "eval_samples_per_second": 42.89,
       "eval_steps_per_second": 0.686,
       "step": 1700
     },
     {
       "epoch": 1.7905759162303665,
-      "grad_norm": 0.5545560789982877,
+      "grad_norm": 0.11960565821214279,
       "learning_rate": 2.093575422770872e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -5.875,
-      "logps/chosen": -784.0,
-      "logps/rejected": -4192.0,
-      "loss": 0.0046,
+      "logits/chosen": -13.375,
+      "logits/rejected": -6.53125,
+      "logps/chosen": -636.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0027,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -4.28125,
-      "rewards/margins": 22.625,
-      "rewards/rejected": -26.875,
+      "rewards/chosen": -2.765625,
+      "rewards/margins": 18.625,
+      "rewards/rejected": -21.5,
       "step": 1710
     },
     {
       "epoch": 1.8010471204188483,
-      "grad_norm": 0.07149319918760624,
+      "grad_norm": 0.01706018096948714,
       "learning_rate": 2.0635462181187827e-05,
-      "logits/chosen": -15.0625,
-      "logits/rejected": -6.78125,
-      "logps/chosen": -756.0,
-      "logps/rejected": -3968.0,
-      "loss": 0.0025,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.625,
-      "rewards/margins": 20.875,
-      "rewards/rejected": -24.5,
+      "logits/chosen": -14.0,
+      "logits/rejected": -6.84375,
+      "logps/chosen": -668.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0053,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.765625,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.125,
       "step": 1720
     },
     {
       "epoch": 1.8115183246073299,
-      "grad_norm": 0.034796619694835426,
+      "grad_norm": 0.03608793871091643,
       "learning_rate": 2.0335818271653917e-05,
-      "logits/chosen": -15.75,
-      "logits/rejected": -7.34375,
-      "logps/chosen": -772.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0042,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.71875,
-      "rewards/margins": 21.875,
-      "rewards/rejected": -25.625,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -7.21875,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0085,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.703125,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -22.375,
       "step": 1730
     },
     {
       "epoch": 1.8219895287958114,
-      "grad_norm": 0.38715528710030167,
+      "grad_norm": 0.3795652897483298,
       "learning_rate": 2.0036866996441815e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -8.0,
-      "logps/chosen": -740.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0034,
+      "logits/chosen": -15.0,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -672.0,
+      "logps/rejected": -4096.0,
+      "loss": 0.0147,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.578125,
-      "rewards/margins": 21.125,
-      "rewards/rejected": -24.625,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 21.25,
+      "rewards/rejected": -24.125,
       "step": 1740
     },
     {
       "epoch": 1.8324607329842932,
-      "grad_norm": 0.12316314590589363,
+      "grad_norm": 0.9007251251178866,
       "learning_rate": 1.9738652750029677e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -7.71875,
-      "logps/chosen": -768.0,
+      "logits/chosen": -14.625,
+      "logits/rejected": -6.75,
+      "logps/chosen": -724.0,
       "logps/rejected": -3712.0,
-      "loss": 0.0047,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 19.125,
+      "loss": 0.0015,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.0625,
+      "rewards/margins": 19.5,
       "rewards/rejected": -22.625,
       "step": 1750
     },
     {
       "epoch": 1.8429319371727748,
-      "grad_norm": 0.08384356230386839,
+      "grad_norm": 0.49061629331944495,
       "learning_rate": 1.944121981744631e-05,
-      "logits/chosen": -15.875,
-      "logits/rejected": -7.625,
-      "logps/chosen": -784.0,
-      "logps/rejected": -3648.0,
-      "loss": 0.0053,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 18.125,
-      "rewards/rejected": -21.625,
+      "logits/chosen": -15.0,
+      "logits/rejected": -7.09375,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0035,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -23.125,
       "step": 1760
     },
     {
       "epoch": 1.8534031413612566,
-      "grad_norm": 0.36032440285942113,
+      "grad_norm": 0.6823656643916527,
       "learning_rate": 1.9144612367694887e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -7.6875,
-      "logps/chosen": -708.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0016,
+      "logits/chosen": -15.3125,
+      "logits/rejected": -7.1875,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0029,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.1875,
-      "rewards/margins": 19.25,
-      "rewards/rejected": -22.375,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 20.375,
+      "rewards/rejected": -23.375,
       "step": 1770
     },
     {
       "epoch": 1.8638743455497382,
-      "grad_norm": 0.029839033707389273,
+      "grad_norm": 0.8691943462822609,
       "learning_rate": 1.8848874447193803e-05,
-      "logits/chosen": -15.5625,
-      "logits/rejected": -7.3125,
-      "logps/chosen": -700.0,
-      "logps/rejected": -4016.0,
-      "loss": 0.0011,
+      "logits/chosen": -14.875,
+      "logits/rejected": -6.78125,
+      "logps/chosen": -664.0,
+      "logps/rejected": -4032.0,
+      "loss": 0.0026,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.25,
-      "rewards/margins": 20.75,
-      "rewards/rejected": -24.0,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 21.25,
+      "rewards/rejected": -24.25,
       "step": 1780
     },
     {
       "epoch": 1.8743455497382198,
-      "grad_norm": 0.007948075205731065,
+      "grad_norm": 0.0058496003887458705,
       "learning_rate": 1.855404997323571e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4016.0,
-      "loss": 0.0011,
+      "logits/chosen": -15.6875,
+      "logits/rejected": -7.78125,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3936.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.453125,
-      "rewards/margins": 20.25,
-      "rewards/rejected": -23.625,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -22.75,
       "step": 1790
     },
     {
       "epoch": 1.8848167539267016,
-      "grad_norm": 0.022996487838440136,
+      "grad_norm": 0.019215544127824183,
       "learning_rate": 1.82601827274658e-05,
-      "logits/chosen": -16.5,
-      "logits/rejected": -7.84375,
-      "logps/chosen": -688.0,
-      "logps/rejected": -3792.0,
-      "loss": 0.0009,
+      "logits/chosen": -16.125,
+      "logits/rejected": -7.5,
+      "logps/chosen": -652.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0022,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.265625,
-      "rewards/margins": 20.125,
-      "rewards/rejected": -23.375,
+      "rewards/chosen": -2.875,
+      "rewards/margins": 19.625,
+      "rewards/rejected": -22.625,
       "step": 1800
     },
     {
       "epoch": 1.8848167539267016,
       "eval_logits/chosen": -15.875,
-      "eval_logits/rejected": -15.6875,
-      "eval_logps/chosen": -692.0,
-      "eval_logps/rejected": -688.0,
-      "eval_loss": 1.1240781545639038,
-      "eval_rewards/accuracies": 0.4765625,
-      "eval_rewards/chosen": -3.734375,
-      "eval_rewards/margins": 0.28515625,
-      "eval_rewards/rejected": -4.0,
-      "eval_runtime": 46.6061,
-      "eval_samples_per_second": 42.913,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/rejected": -15.5625,
+      "eval_logps/chosen": -616.0,
+      "eval_logps/rejected": -620.0,
+      "eval_loss": 0.9915937781333923,
+      "eval_rewards/accuracies": 0.4921875,
+      "eval_rewards/chosen": -3.0,
+      "eval_rewards/margins": 0.328125,
+      "eval_rewards/rejected": -3.328125,
+      "eval_runtime": 46.6325,
+      "eval_samples_per_second": 42.888,
+      "eval_steps_per_second": 0.686,
       "step": 1800
     },
     {
       "epoch": 1.8952879581151834,
-      "grad_norm": 0.04406236061401373,
+      "grad_norm": 0.02445182769342151,
       "learning_rate": 1.7967316349380213e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -7.875,
-      "logps/chosen": -672.0,
-      "logps/rejected": -3712.0,
-      "loss": 0.0018,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.9375,
+      "logps/chosen": -616.0,
+      "logps/rejected": -3536.0,
+      "loss": 0.0015,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.046875,
-      "rewards/margins": 20.0,
-      "rewards/rejected": -23.0,
+      "rewards/chosen": -2.46875,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -21.25,
       "step": 1810
     },
     {
       "epoch": 1.905759162303665,
-      "grad_norm": 0.01558601314137932,
+      "grad_norm": 0.01431897451285721,
       "learning_rate": 1.7675494329845516e-05,
       "logits/chosen": -16.5,
-      "logits/rejected": -8.875,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3984.0,
-      "loss": 0.0019,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -664.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0014,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 20.375,
-      "rewards/rejected": -23.75,
+      "rewards/chosen": -2.8125,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -21.75,
       "step": 1820
     },
     {
       "epoch": 1.9162303664921465,
-      "grad_norm": 0.4451340569669243,
+      "grad_norm": 0.573990532180258,
       "learning_rate": 1.738476000464026e-05,
       "logits/chosen": -16.5,
-      "logits/rejected": -8.5625,
-      "logps/chosen": -700.0,
-      "logps/rejected": -3872.0,
-      "loss": 0.0033,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -656.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0028,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 19.75,
-      "rewards/rejected": -23.0,
+      "rewards/chosen": -2.75,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -21.875,
       "step": 1830
     },
     {
       "epoch": 1.9267015706806283,
-      "grad_norm": 0.41022128293485555,
+      "grad_norm": 0.39867458352403784,
       "learning_rate": 1.7095156548019646e-05,
       "logits/chosen": -16.875,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -696.0,
-      "logps/rejected": -4032.0,
-      "loss": 0.0076,
+      "logits/rejected": -9.375,
+      "logps/chosen": -640.0,
+      "logps/rejected": -3872.0,
+      "loss": 0.0031,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 20.75,
-      "rewards/rejected": -24.0,
+      "rewards/chosen": -2.6875,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -22.375,
       "step": 1840
     },
     {
       "epoch": 1.93717277486911,
-      "grad_norm": 0.4071104715365148,
+      "grad_norm": 0.6400579799800042,
       "learning_rate": 1.6806726966304064e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.0,
-      "logps/chosen": -716.0,
-      "logps/rejected": -3872.0,
-      "loss": 0.0028,
+      "logits/chosen": -16.5,
+      "logits/rejected": -8.75,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0054,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.140625,
-      "rewards/margins": 20.125,
-      "rewards/rejected": -23.25,
+      "rewards/chosen": -2.859375,
+      "rewards/margins": 19.625,
+      "rewards/rejected": -22.5,
       "step": 1850
     },
     {
       "epoch": 1.9476439790575917,
-      "grad_norm": 0.0009387174292754044,
+      "grad_norm": 0.0011911661935824552,
       "learning_rate": 1.6519514091492625e-05,
       "logits/chosen": -16.375,
-      "logits/rejected": -8.0,
-      "logps/chosen": -668.0,
-      "logps/rejected": -3936.0,
-      "loss": 0.0021,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -616.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0041,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 21.0,
-      "rewards/rejected": -24.25,
+      "rewards/chosen": -2.640625,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -22.375,
       "step": 1860
     },
     {
       "epoch": 1.9581151832460733,
-      "grad_norm": 0.0118689470576417,
+      "grad_norm": 0.013220794006462451,
       "learning_rate": 1.6233560574902627e-05,
-      "logits/chosen": -16.5,
-      "logits/rejected": -8.75,
-      "logps/chosen": -636.0,
-      "logps/rejected": -4096.0,
-      "loss": 0.0197,
+      "logits/chosen": -16.75,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -576.0,
+      "logps/rejected": -3936.0,
+      "loss": 0.0075,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.9375,
-      "rewards/margins": 20.875,
-      "rewards/rejected": -23.875,
+      "rewards/chosen": -2.328125,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -22.0,
       "step": 1870
     },
     {
       "epoch": 1.9685863874345548,
-      "grad_norm": 0.48482684552079053,
+      "grad_norm": 1.1983946148687459,
       "learning_rate": 1.5948908880835753e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -7.125,
-      "logps/chosen": -704.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0021,
+      "logits/chosen": -16.5,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -652.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0051,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.140625,
-      "rewards/margins": 20.5,
-      "rewards/rejected": -23.75,
+      "rewards/chosen": -2.65625,
+      "rewards/margins": 19.25,
+      "rewards/rejected": -22.0,
       "step": 1880
     },
     {
       "epoch": 1.9790575916230366,
-      "grad_norm": 4.685776649048767,
+      "grad_norm": 1.8984581101916458,
       "learning_rate": 1.5665601280272124e-05,
-      "logits/chosen": -16.375,
-      "logits/rejected": -7.90625,
-      "logps/chosen": -668.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0046,
+      "logits/chosen": -16.125,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -620.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0041,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.109375,
-      "rewards/margins": 20.125,
-      "rewards/rejected": -23.25,
+      "rewards/chosen": -2.640625,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -22.5,
       "step": 1890
     },
     {
       "epoch": 1.9895287958115184,
-      "grad_norm": 0.06775541950192056,
+      "grad_norm": 0.04508497390394497,
       "learning_rate": 1.538367984459298e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -8.5,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0052,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 18.5,
-      "rewards/rejected": -21.75,
+      "logits/chosen": -15.75,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -656.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0041,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.875,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -20.875,
       "step": 1900
     },
     {
       "epoch": 1.9895287958115184,
-      "eval_logits/chosen": -16.375,
-      "eval_logits/rejected": -16.25,
-      "eval_logps/chosen": -684.0,
-      "eval_logps/rejected": -680.0,
-      "eval_loss": 1.1657500267028809,
-      "eval_rewards/accuracies": 0.46875,
-      "eval_rewards/chosen": -3.640625,
-      "eval_rewards/margins": 0.267578125,
-      "eval_rewards/rejected": -3.90625,
-      "eval_runtime": 46.6185,
-      "eval_samples_per_second": 42.901,
+      "eval_logits/chosen": -16.25,
+      "eval_logits/rejected": -16.125,
+      "eval_logps/chosen": -604.0,
+      "eval_logps/rejected": -604.0,
+      "eval_loss": 0.9827812314033508,
+      "eval_rewards/accuracies": 0.505859375,
+      "eval_rewards/chosen": -2.859375,
+      "eval_rewards/margins": 0.296875,
+      "eval_rewards/rejected": -3.15625,
+      "eval_runtime": 46.6341,
+      "eval_samples_per_second": 42.887,
       "eval_steps_per_second": 0.686,
       "step": 1900
     },
     {
       "epoch": 2.0,
-      "grad_norm": 0.0030706632632777875,
+      "grad_norm": 0.01233741892253411,
       "learning_rate": 1.5103186439333133e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -648.0,
-      "logps/rejected": -3648.0,
-      "loss": 0.0041,
+      "logits/chosen": -16.125,
+      "logits/rejected": -9.0,
+      "logps/chosen": -612.0,
+      "logps/rejected": -3536.0,
+      "loss": 0.0084,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.828125,
-      "rewards/margins": 18.625,
-      "rewards/rejected": -21.375,
+      "rewards/chosen": -2.5,
+      "rewards/margins": 17.875,
+      "rewards/rejected": -20.5,
       "step": 1910
     },
     {
       "epoch": 2.0104712041884816,
-      "grad_norm": 0.010848722220843123,
+      "grad_norm": 0.03008228740388698,
       "learning_rate": 1.4824162717963829e-05,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.75,
-      "logps/chosen": -672.0,
-      "logps/rejected": -3872.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.375,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -624.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.03125,
-      "rewards/margins": 20.0,
-      "rewards/rejected": -23.0,
+      "rewards/chosen": -2.515625,
+      "rewards/margins": 18.75,
+      "rewards/rejected": -21.25,
       "step": 1920
     },
     {
       "epoch": 2.020942408376963,
-      "grad_norm": 0.007946842587368216,
+      "grad_norm": 0.05466499699499411,
       "learning_rate": 1.4546650115707192e-05,
-      "logits/chosen": -17.125,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -648.0,
-      "logps/rejected": -3696.0,
-      "loss": 0.0027,
+      "logits/chosen": -16.125,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -608.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.859375,
-      "rewards/margins": 19.5,
-      "rewards/rejected": -22.375,
+      "rewards/chosen": -2.484375,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -21.625,
       "step": 1930
     },
     {
       "epoch": 2.031413612565445,
-      "grad_norm": 0.0074145123281544505,
+      "grad_norm": 0.0032844492076225438,
       "learning_rate": 1.4270689843383111e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.625,
-      "logps/chosen": -720.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.125,
+      "logits/rejected": -8.125,
+      "logps/chosen": -668.0,
+      "logps/rejected": -3872.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.265625,
-      "rewards/margins": 21.25,
-      "rewards/rejected": -24.5,
+      "rewards/chosen": -2.765625,
+      "rewards/margins": 19.25,
+      "rewards/rejected": -22.125,
       "step": 1940
     },
     {
       "epoch": 2.0418848167539267,
-      "grad_norm": 0.008964629053622658,
+      "grad_norm": 0.00674493763075121,
       "learning_rate": 1.3996322881289347e-05,
-      "logits/chosen": -17.125,
-      "logits/rejected": -8.6875,
-      "logps/chosen": -664.0,
-      "logps/rejected": -4000.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.25,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -616.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.078125,
-      "rewards/margins": 21.0,
-      "rewards/rejected": -24.125,
+      "rewards/chosen": -2.59375,
+      "rewards/margins": 18.75,
+      "rewards/rejected": -21.375,
       "step": 1950
     },
     {
       "epoch": 2.0523560209424083,
-      "grad_norm": 0.006222026307179967,
+      "grad_norm": 0.008302600308011493,
       "learning_rate": 1.3723589973115963e-05,
-      "logits/chosen": -17.125,
-      "logits/rejected": -9.0,
-      "logps/chosen": -688.0,
-      "logps/rejected": -4016.0,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -8.25,
+      "logps/chosen": -624.0,
+      "logps/rejected": -3792.0,
       "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 21.0,
-      "rewards/rejected": -24.25,
+      "rewards/chosen": -2.625,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.0,
       "step": 1960
     },
     {
       "epoch": 2.06282722513089,
-      "grad_norm": 0.005521543921954491,
+      "grad_norm": 0.0042110717849586365,
       "learning_rate": 1.3452531619894837e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.625,
-      "logps/chosen": -692.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0002,
+      "logits/chosen": -15.8125,
+      "logits/rejected": -7.8125,
+      "logps/chosen": -648.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.234375,
-      "rewards/margins": 20.75,
-      "rewards/rejected": -24.0,
+      "rewards/chosen": -2.796875,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.125,
       "step": 1970
     },
     {
       "epoch": 2.073298429319372,
-      "grad_norm": 0.006384590170511054,
+      "grad_norm": 0.009290591261412313,
       "learning_rate": 1.3183188073985245e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -688.0,
-      "logps/rejected": -3920.0,
-      "loss": 0.0002,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -7.9375,
+      "logps/chosen": -640.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.203125,
-      "rewards/margins": 21.25,
-      "rewards/rejected": -24.375,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.125,
       "step": 1980
     },
     {
       "epoch": 2.0837696335078535,
-      "grad_norm": 0.002477380755866227,
+      "grad_norm": 0.0005445868655585701,
       "learning_rate": 1.2915599333096351e-05,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.4375,
-      "logps/chosen": -716.0,
-      "logps/rejected": -3952.0,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -7.96875,
+      "logps/chosen": -668.0,
+      "logps/rejected": -3792.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.1875,
-      "rewards/margins": 21.125,
-      "rewards/rejected": -24.25,
+      "rewards/chosen": -2.703125,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -22.5,
       "step": 1990
     },
     {
       "epoch": 2.094240837696335,
-      "grad_norm": 0.0022273867474921225,
+      "grad_norm": 0.001980384511543767,
       "learning_rate": 1.264980513434752e-05,
-      "logits/chosen": -16.875,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -760.0,
-      "logps/rejected": -3936.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.125,
+      "logits/rejected": -7.90625,
+      "logps/chosen": -720.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 20.5,
-      "rewards/rejected": -24.0,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -22.25,
       "step": 2000
     },
     {
       "epoch": 2.094240837696335,
-      "eval_logits/chosen": -16.625,
-      "eval_logits/rejected": -16.5,
-      "eval_logps/chosen": -692.0,
-      "eval_logps/rejected": -692.0,
-      "eval_loss": 1.1422499418258667,
-      "eval_rewards/accuracies": 0.478515625,
-      "eval_rewards/chosen": -3.71875,
-      "eval_rewards/margins": 0.310546875,
-      "eval_rewards/rejected": -4.03125,
-      "eval_runtime": 46.617,
-      "eval_samples_per_second": 42.903,
+      "eval_logits/chosen": -16.25,
+      "eval_logits/rejected": -16.0,
+      "eval_logps/chosen": -636.0,
+      "eval_logps/rejected": -636.0,
+      "eval_loss": 1.0311249494552612,
+      "eval_rewards/accuracies": 0.49609375,
+      "eval_rewards/chosen": -3.15625,
+      "eval_rewards/margins": 0.318359375,
+      "eval_rewards/rejected": -3.46875,
+      "eval_runtime": 46.6444,
+      "eval_samples_per_second": 42.878,
       "eval_steps_per_second": 0.686,
       "step": 2000
     },
     {
       "epoch": 2.1047120418848166,
-      "grad_norm": 0.0661634618945412,
+      "grad_norm": 0.05219794833690374,
       "learning_rate": 1.2385844948367324e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -9.125,
-      "logps/chosen": -672.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0004,
+      "logits/chosen": -16.125,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -636.0,
+      "logps/rejected": -4032.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.296875,
-      "rewards/margins": 22.875,
-      "rewards/rejected": -26.125,
+      "rewards/chosen": -2.9375,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -23.625,
       "step": 2010
     },
     {
       "epoch": 2.115183246073298,
-      "grad_norm": 0.0015786896611314654,
+      "grad_norm": 0.0019893029938409736,
       "learning_rate": 1.2123757973432114e-05,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.625,
-      "logps/chosen": -748.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0008,
+      "logits/chosen": -16.0,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3984.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 21.625,
-      "rewards/rejected": -25.0,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.125,
       "step": 2020
     },
     {
       "epoch": 2.1256544502617802,
-      "grad_norm": 0.7489977143151078,
+      "grad_norm": 0.008011752305716997,
       "learning_rate": 1.1863583129645106e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.5625,
-      "logps/chosen": -708.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.0,
+      "logits/rejected": -7.875,
+      "logps/chosen": -660.0,
+      "logps/rejected": -3920.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.3125,
-      "rewards/margins": 22.375,
-      "rewards/rejected": -25.625,
+      "rewards/chosen": -2.828125,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -23.375,
       "step": 2030
     },
     {
       "epoch": 2.136125654450262,
-      "grad_norm": 0.004693963642544854,
+      "grad_norm": 0.006129003520427554,
       "learning_rate": 1.1605359053156607e-05,
-      "logits/chosen": -17.25,
-      "logits/rejected": -8.375,
-      "logps/chosen": -716.0,
-      "logps/rejected": -4048.0,
-      "loss": 0.0012,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.375,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.125,
-      "rewards/margins": 21.875,
-      "rewards/rejected": -25.0,
+      "rewards/chosen": -2.703125,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -23.375,
       "step": 2040
     },
     {
       "epoch": 2.1465968586387434,
-      "grad_norm": 0.2696077770408498,
+      "grad_norm": 0.000534369620166865,
       "learning_rate": 1.1349124090426644e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.6875,
-      "logps/chosen": -708.0,
-      "logps/rejected": -3952.0,
-      "loss": 0.0008,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -7.59375,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.25,
-      "rewards/margins": 21.375,
-      "rewards/rejected": -24.625,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.0,
       "step": 2050
     },
     {
       "epoch": 2.157068062827225,
-      "grad_norm": 0.009448898476836283,
+      "grad_norm": 0.009494619620820379,
       "learning_rate": 1.1094916292530406e-05,
-      "logits/chosen": -16.5,
-      "logits/rejected": -9.0,
-      "logps/chosen": -728.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0005,
+      "logits/chosen": -15.75,
+      "logits/rejected": -8.25,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.3125,
-      "rewards/margins": 22.75,
-      "rewards/rejected": -26.125,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -23.5,
       "step": 2060
     },
     {
       "epoch": 2.167539267015707,
-      "grad_norm": 0.02865953917383672,
+      "grad_norm": 0.029416024032075667,
       "learning_rate": 1.0842773409507623e-05,
-      "logits/chosen": -17.125,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -688.0,
-      "logps/rejected": -4416.0,
-      "loss": 0.0006,
+      "logits/chosen": -16.375,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -644.0,
+      "logps/rejected": -4096.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.390625,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.375,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 21.5,
+      "rewards/rejected": -24.375,
       "step": 2070
     },
     {
       "epoch": 2.1780104712041886,
-      "grad_norm": 0.0006172221224906784,
+      "grad_norm": 0.0025106656389008012,
       "learning_rate": 1.0592732884756754e-05,
-      "logits/chosen": -17.0,
-      "logits/rejected": -7.78125,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0008,
+      "logits/chosen": -16.0,
+      "logits/rejected": -6.65625,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 23.5,
-      "rewards/rejected": -27.0,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 21.0,
+      "rewards/rejected": -24.25,
       "step": 2080
     },
     {
       "epoch": 2.18848167539267,
-      "grad_norm": 0.0031908155334336667,
+      "grad_norm": 0.010519097344393259,
       "learning_rate": 1.0344831849474507e-05,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.625,
-      "logps/chosen": -744.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.53125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -4000.0,
+      "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 21.625,
+      "rewards/rejected": -24.75,
       "step": 2090
     },
     {
       "epoch": 2.1989528795811517,
-      "grad_norm": 0.0165355516158079,
+      "grad_norm": 0.06009683353561999,
       "learning_rate": 1.0099107117141879e-05,
-      "logits/chosen": -16.875,
-      "logits/rejected": -7.90625,
-      "logps/chosen": -704.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0008,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.59375,
+      "logps/chosen": -660.0,
+      "logps/rejected": -3968.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 23.5,
-      "rewards/rejected": -27.0,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 21.5,
+      "rewards/rejected": -24.5,
       "step": 2100
     },
     {
       "epoch": 2.1989528795811517,
       "eval_logits/chosen": -16.375,
       "eval_logits/rejected": -16.125,
-      "eval_logps/chosen": -728.0,
-      "eval_logps/rejected": -736.0,
-      "eval_loss": 1.2501250505447388,
-      "eval_rewards/accuracies": 0.486328125,
-      "eval_rewards/chosen": -4.09375,
-      "eval_rewards/margins": 0.404296875,
-      "eval_rewards/rejected": -4.5,
-      "eval_runtime": 46.6234,
-      "eval_samples_per_second": 42.897,
+      "eval_logps/chosen": -664.0,
+      "eval_logps/rejected": -672.0,
+      "eval_loss": 1.0823124647140503,
+      "eval_rewards/accuracies": 0.5,
+      "eval_rewards/chosen": -3.46875,
+      "eval_rewards/margins": 0.369140625,
+      "eval_rewards/rejected": -3.84375,
+      "eval_runtime": 46.6275,
+      "eval_samples_per_second": 42.893,
       "eval_steps_per_second": 0.686,
       "step": 2100
     },
     {
       "epoch": 2.2094240837696333,
-      "grad_norm": 0.007941675782305474,
+      "grad_norm": 0.008261521929699606,
       "learning_rate": 9.855595178057334e-06,
-      "logits/chosen": -16.25,
-      "logits/rejected": -7.75,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0003,
+      "logits/chosen": -15.6875,
+      "logits/rejected": -6.9375,
+      "logps/chosen": -724.0,
+      "logps/rejected": -4016.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 21.75,
+      "rewards/rejected": -24.875,
       "step": 2110
     },
     {
       "epoch": 2.2198952879581153,
-      "grad_norm": 0.0017261481963212902,
+      "grad_norm": 0.005096974198110823,
       "learning_rate": 9.614332193917949e-06,
-      "logits/chosen": -16.875,
-      "logits/rejected": -7.96875,
-      "logps/chosen": -796.0,
-      "logps/rejected": -4480.0,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.21875,
+      "logps/chosen": -748.0,
+      "logps/rejected": -4192.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.78125,
-      "rewards/margins": 24.625,
-      "rewards/rejected": -28.375,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 22.375,
+      "rewards/rejected": -25.75,
       "step": 2120
     },
     {
       "epoch": 2.230366492146597,
-      "grad_norm": 0.0013794945744065494,
+      "grad_norm": 0.025749354383095575,
       "learning_rate": 9.375353992449385e-06,
-      "logits/chosen": -17.0,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -740.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -696.0,
+      "logps/rejected": -4016.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 24.125,
-      "rewards/rejected": -27.75,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 22.25,
+      "rewards/rejected": -25.375,
       "step": 2130
     },
     {
       "epoch": 2.2408376963350785,
-      "grad_norm": 0.013113667770158554,
+      "grad_norm": 0.004314312372392489,
       "learning_rate": 9.138696062085442e-06,
-      "logits/chosen": -17.125,
-      "logits/rejected": -9.0,
-      "logps/chosen": -696.0,
-      "logps/rejected": -4320.0,
-      "loss": 0.0006,
+      "logits/chosen": -16.25,
+      "logits/rejected": -8.375,
+      "logps/chosen": -660.0,
+      "logps/rejected": -4096.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.234375,
-      "rewards/margins": 24.25,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -2.859375,
+      "rewards/margins": 22.5,
+      "rewards/rejected": -25.375,
       "step": 2140
     },
     {
       "epoch": 2.25130890052356,
-      "grad_norm": 0.0008050243263897401,
+      "grad_norm": 0.0021792393758785847,
       "learning_rate": 8.904393546698007e-06,
-      "logits/chosen": -17.25,
-      "logits/rejected": -8.25,
-      "logps/chosen": -740.0,
-      "logps/rejected": -4352.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.09375,
+      "logps/chosen": -716.0,
+      "logps/rejected": -4224.0,
+      "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 24.125,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 23.0,
+      "rewards/rejected": -26.125,
       "step": 2150
     },
     {
       "epoch": 2.261780104712042,
-      "grad_norm": 0.003718463988386438,
+      "grad_norm": 0.009922535155424054,
       "learning_rate": 8.672481240378142e-06,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.5625,
-      "logps/chosen": -668.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.25,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -636.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.265625,
-      "rewards/margins": 23.375,
-      "rewards/rejected": -26.625,
+      "rewards/chosen": -2.9375,
+      "rewards/margins": 21.375,
+      "rewards/rejected": -24.25,
       "step": 2160
     },
     {
       "epoch": 2.2722513089005236,
-      "grad_norm": 0.005104869450515169,
+      "grad_norm": 0.016193694994840576,
       "learning_rate": 8.44299358226919e-06,
-      "logits/chosen": -16.375,
-      "logits/rejected": -8.4375,
-      "logps/chosen": -696.0,
-      "logps/rejected": -4224.0,
+      "logits/chosen": -16.125,
+      "logits/rejected": -7.75,
+      "logps/chosen": -668.0,
+      "logps/rejected": -4096.0,
       "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 23.5,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 22.375,
+      "rewards/rejected": -25.375,
       "step": 2170
     },
     {
       "epoch": 2.282722513089005,
-      "grad_norm": 0.0060298484831497546,
+      "grad_norm": 0.009058499497863104,
       "learning_rate": 8.215964651452457e-06,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.125,
-      "logps/chosen": -728.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0003,
+      "logits/chosen": -15.875,
+      "logits/rejected": -7.34375,
+      "logps/chosen": -700.0,
+      "logps/rejected": -4096.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 23.875,
-      "rewards/rejected": -27.375,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 22.25,
+      "rewards/rejected": -25.5,
       "step": 2180
     },
     {
       "epoch": 2.2931937172774868,
-      "grad_norm": 0.009107944652835225,
+      "grad_norm": 0.001962578570237978,
       "learning_rate": 7.991428161886502e-06,
-      "logits/chosen": -16.5,
-      "logits/rejected": -8.25,
-      "logps/chosen": -736.0,
-      "logps/rejected": -4096.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.125,
+      "logits/rejected": -7.8125,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3968.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 22.875,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 21.625,
+      "rewards/rejected": -24.75,
       "step": 2190
     },
     {
       "epoch": 2.303664921465969,
-      "grad_norm": 0.015671060901887333,
+      "grad_norm": 0.006393455701850208,
       "learning_rate": 7.76941745740061e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -7.3125,
-      "logps/chosen": -784.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0002,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -6.46875,
+      "logps/chosen": -768.0,
+      "logps/rejected": -4048.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.734375,
-      "rewards/margins": 23.5,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -3.5625,
+      "rewards/margins": 22.0,
+      "rewards/rejected": -25.5,
       "step": 2200
     },
     {
       "epoch": 2.303664921465969,
       "eval_logits/chosen": -16.375,
       "eval_logits/rejected": -16.25,
-      "eval_logps/chosen": -724.0,
-      "eval_logps/rejected": -736.0,
-      "eval_loss": 1.249750018119812,
-      "eval_rewards/accuracies": 0.490234375,
-      "eval_rewards/chosen": -4.0625,
-      "eval_rewards/margins": 0.41796875,
-      "eval_rewards/rejected": -4.46875,
-      "eval_runtime": 46.6183,
-      "eval_samples_per_second": 42.902,
+      "eval_logps/chosen": -680.0,
+      "eval_logps/rejected": -692.0,
+      "eval_loss": 1.115625023841858,
+      "eval_rewards/accuracies": 0.494140625,
+      "eval_rewards/chosen": -3.625,
+      "eval_rewards/margins": 0.400390625,
+      "eval_rewards/rejected": -4.03125,
+      "eval_runtime": 46.6264,
+      "eval_samples_per_second": 42.894,
       "eval_steps_per_second": 0.686,
       "step": 2200
     },
     {
       "epoch": 2.3141361256544504,
-      "grad_norm": 0.013622997918332847,
+      "grad_norm": 0.004799777313939887,
       "learning_rate": 7.549965506743123e-06,
-      "logits/chosen": -16.75,
-      "logits/rejected": -7.75,
-      "logps/chosen": -720.0,
-      "logps/rejected": -4320.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.875,
+      "logits/rejected": -7.40625,
+      "logps/chosen": -684.0,
+      "logps/rejected": -4128.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 24.25,
-      "rewards/rejected": -27.75,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 22.75,
+      "rewards/rejected": -25.875,
       "step": 2210
     },
     {
       "epoch": 2.324607329842932,
-      "grad_norm": 0.006952519620580333,
+      "grad_norm": 0.003770350060699025,
       "learning_rate": 7.333104898685631e-06,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.375,
-      "logps/chosen": -720.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0006,
+      "logits/chosen": -15.875,
+      "logits/rejected": -8.0,
+      "logps/chosen": -700.0,
+      "logps/rejected": -4016.0,
+      "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.328125,
-      "rewards/margins": 23.375,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 22.5,
+      "rewards/rejected": -25.5,
       "step": 2220
     },
     {
       "epoch": 2.3350785340314135,
-      "grad_norm": 0.0024224926153747673,
+      "grad_norm": 0.002484949330617898,
       "learning_rate": 7.118867837183518e-06,
-      "logits/chosen": -16.875,
-      "logits/rejected": -7.875,
-      "logps/chosen": -832.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.125,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -792.0,
+      "logps/rejected": -4192.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.8125,
-      "rewards/margins": 23.0,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -3.421875,
+      "rewards/margins": 22.375,
+      "rewards/rejected": -25.875,
       "step": 2230
     },
     {
       "epoch": 2.345549738219895,
-      "grad_norm": 0.0037243058610017493,
+      "grad_norm": 0.002656904413044796,
       "learning_rate": 6.9072861365936045e-06,
-      "logits/chosen": -17.25,
-      "logits/rejected": -8.0,
-      "logps/chosen": -812.0,
-      "logps/rejected": -4224.0,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.4375,
+      "logps/chosen": -788.0,
+      "logps/rejected": -4128.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.6875,
-      "rewards/margins": 23.5,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -3.46875,
+      "rewards/margins": 22.75,
+      "rewards/rejected": -26.25,
       "step": 2240
     },
     {
       "epoch": 2.356020942408377,
-      "grad_norm": 0.008846447475605607,
+      "grad_norm": 0.008594946596000632,
       "learning_rate": 6.6983912169497005e-06,
-      "logits/chosen": -16.75,
-      "logits/rejected": -7.8125,
-      "logps/chosen": -748.0,
-      "logps/rejected": -4096.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.0625,
+      "logps/chosen": -744.0,
+      "logps/rejected": -4048.0,
+      "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.265625,
-      "rewards/margins": 22.625,
-      "rewards/rejected": -25.875,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 22.25,
+      "rewards/rejected": -25.5,
       "step": 2250
     },
     {
       "epoch": 2.3664921465968587,
-      "grad_norm": 0.01199700278811271,
+      "grad_norm": 0.004935784648324409,
       "learning_rate": 6.49221409929677e-06,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.6875,
-      "logps/chosen": -684.0,
-      "logps/rejected": -4576.0,
-      "loss": 0.0004,
+      "logits/chosen": -16.75,
+      "logits/rejected": -8.5,
+      "logps/chosen": -668.0,
+      "logps/rejected": -4320.0,
+      "loss": 0.0021,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 25.5,
-      "rewards/rejected": -28.875,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -26.5,
       "step": 2260
     },
     {
       "epoch": 2.3769633507853403,
-      "grad_norm": 0.005836743348294534,
+      "grad_norm": 0.02323936806019371,
       "learning_rate": 6.2887854010842075e-06,
-      "logits/chosen": -17.125,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -700.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0004,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -684.0,
+      "logps/rejected": -4096.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 24.25,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 22.625,
+      "rewards/rejected": -25.75,
       "step": 2270
     },
     {
       "epoch": 2.387434554973822,
-      "grad_norm": 0.0028185190298729737,
+      "grad_norm": 0.0076410889747321985,
       "learning_rate": 6.0881353316191385e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.375,
-      "logps/chosen": -792.0,
-      "logps/rejected": -4096.0,
+      "logits/chosen": -16.0,
+      "logits/rejected": -8.0,
+      "logps/chosen": -760.0,
+      "logps/rejected": -4000.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.78125,
-      "rewards/margins": 23.0,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 22.25,
+      "rewards/rejected": -25.75,
       "step": 2280
     },
     {
       "epoch": 2.3979057591623034,
-      "grad_norm": 0.010135143663081669,
+      "grad_norm": 0.0010312011867643664,
       "learning_rate": 5.890293687580381e-06,
-      "logits/chosen": -17.125,
-      "logits/rejected": -8.75,
-      "logps/chosen": -752.0,
-      "logps/rejected": -4256.0,
+      "logits/chosen": -16.5,
+      "logits/rejected": -8.5,
+      "logps/chosen": -744.0,
+      "logps/rejected": -4096.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 23.5,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 22.0,
+      "rewards/rejected": -25.5,
       "step": 2290
     },
     {
       "epoch": 2.4083769633507854,
-      "grad_norm": 0.02712555559000985,
+      "grad_norm": 0.05282876643502036,
       "learning_rate": 5.695289848593532e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -7.84375,
-      "logps/chosen": -712.0,
-      "logps/rejected": -4064.0,
-      "loss": 0.0004,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.0,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 22.25,
+      "rewards/rejected": -25.375,
       "step": 2300
     },
     {
       "epoch": 2.4083769633507854,
       "eval_logits/chosen": -16.375,
       "eval_logits/rejected": -16.25,
-      "eval_logps/chosen": -728.0,
-      "eval_logps/rejected": -740.0,
-      "eval_loss": 1.25767183303833,
-      "eval_rewards/accuracies": 0.494140625,
-      "eval_rewards/chosen": -4.09375,
-      "eval_rewards/margins": 0.42578125,
-      "eval_rewards/rejected": -4.53125,
-      "eval_runtime": 46.6039,
-      "eval_samples_per_second": 42.915,
-      "eval_steps_per_second": 0.687,
+      "eval_logps/chosen": -688.0,
+      "eval_logps/rejected": -700.0,
+      "eval_loss": 1.1278281211853027,
+      "eval_rewards/accuracies": 0.498046875,
+      "eval_rewards/chosen": -3.703125,
+      "eval_rewards/margins": 0.4140625,
+      "eval_rewards/rejected": -4.125,
+      "eval_runtime": 46.4866,
+      "eval_samples_per_second": 43.023,
+      "eval_steps_per_second": 0.688,
       "step": 2300
     },
     {
       "epoch": 2.418848167539267,
-      "grad_norm": 0.004218111546982863,
+      "grad_norm": 0.0013790280665224738,
       "learning_rate": 5.50315277286812e-06,
-      "logits/chosen": -17.0,
-      "logits/rejected": -8.25,
-      "logps/chosen": -692.0,
-      "logps/rejected": -4320.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.625,
+      "logits/rejected": -8.0,
+      "logps/chosen": -676.0,
+      "logps/rejected": -4160.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 24.625,
-      "rewards/rejected": -28.0,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -26.625,
       "step": 2310
     },
     {
       "epoch": 2.4293193717277486,
-      "grad_norm": 0.0009390086719263528,
+      "grad_norm": 0.0035022295937361873,
       "learning_rate": 5.313910992897281e-06,
-      "logits/chosen": -16.25,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -760.0,
-      "logps/rejected": -4288.0,
+      "logits/chosen": -16.0,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -720.0,
+      "logps/rejected": -4096.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.734375,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.375,
+      "rewards/chosen": -3.328125,
+      "rewards/margins": 22.25,
+      "rewards/rejected": -25.5,
       "step": 2320
     },
     {
       "epoch": 2.4397905759162306,
-      "grad_norm": 0.025368215341099357,
+      "grad_norm": 0.0019085019230278273,
       "learning_rate": 5.127592611220658e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.625,
-      "logps/chosen": -760.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.96875,
+      "logps/chosen": -744.0,
+      "logps/rejected": -4128.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.46875,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.0,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 22.625,
+      "rewards/rejected": -26.0,
       "step": 2330
     },
     {
       "epoch": 2.450261780104712,
-      "grad_norm": 0.00155160571170437,
+      "grad_norm": 0.0017893948983415213,
       "learning_rate": 4.944225296251159e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -688.0,
-      "logps/rejected": -4352.0,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.71875,
+      "logps/chosen": -660.0,
+      "logps/rejected": -4224.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 24.625,
-      "rewards/rejected": -28.0,
+      "rewards/chosen": -3.078125,
+      "rewards/margins": 23.5,
+      "rewards/rejected": -26.625,
       "step": 2340
     },
     {
       "epoch": 2.4607329842931938,
-      "grad_norm": 0.001116776304861321,
+      "grad_norm": 0.0022409957337437333,
       "learning_rate": 4.763836278166172e-06,
-      "logits/chosen": -16.5,
-      "logits/rejected": -7.90625,
-      "logps/chosen": -748.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0011,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -732.0,
+      "logps/rejected": -4000.0,
+      "loss": 0.0009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.640625,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 22.125,
+      "rewards/rejected": -25.625,
       "step": 2350
     },
     {
       "epoch": 2.4712041884816753,
-      "grad_norm": 0.0032054434286559095,
+      "grad_norm": 0.005797528285825235,
       "learning_rate": 4.586452344863876e-06,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -740.0,
-      "logps/rejected": -4048.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.5,
+      "logps/chosen": -720.0,
+      "logps/rejected": -3936.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.453125,
-      "rewards/margins": 22.875,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 21.875,
+      "rewards/rejected": -25.125,
       "step": 2360
     },
     {
       "epoch": 2.481675392670157,
-      "grad_norm": 0.006395704011350763,
+      "grad_norm": 0.003958962239142759,
       "learning_rate": 4.412099837985192e-06,
-      "logits/chosen": -17.125,
-      "logits/rejected": -7.78125,
-      "logps/chosen": -768.0,
-      "logps/rejected": -4416.0,
+      "logits/chosen": -16.375,
+      "logits/rejected": -6.96875,
+      "logps/chosen": -744.0,
+      "logps/rejected": -4320.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.609375,
-      "rewards/margins": 24.25,
-      "rewards/rejected": -27.875,
+      "rewards/chosen": -3.375,
+      "rewards/margins": 23.625,
+      "rewards/rejected": -27.0,
       "step": 2370
     },
     {
       "epoch": 2.492146596858639,
-      "grad_norm": 0.0003947640775256641,
+      "grad_norm": 0.0011354623055077945,
       "learning_rate": 4.240804649002089e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -708.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0007,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.875,
+      "logps/chosen": -680.0,
+      "logps/rejected": -4128.0,
+      "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.625,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.375,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 22.375,
+      "rewards/rejected": -25.75,
       "step": 2380
     },
     {
       "epoch": 2.5026178010471205,
-      "grad_norm": 0.017209425528467215,
+      "grad_norm": 0.03133261756225224,
       "learning_rate": 4.072592215372598e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -7.75,
-      "logps/chosen": -800.0,
-      "logps/rejected": -4224.0,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.0,
+      "logps/chosen": -780.0,
+      "logps/rejected": -4128.0,
       "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.78125,
-      "rewards/margins": 23.0,
-      "rewards/rejected": -26.875,
+      "rewards/chosen": -3.59375,
+      "rewards/margins": 22.25,
+      "rewards/rejected": -25.875,
       "step": 2390
     },
     {
       "epoch": 2.513089005235602,
-      "grad_norm": 0.0077483169044597145,
+      "grad_norm": 0.004554701382533362,
       "learning_rate": 3.907487516763389e-06,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.375,
-      "logps/chosen": -724.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.78125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -4048.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 24.25,
-      "rewards/rejected": -27.625,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 22.75,
+      "rewards/rejected": -25.875,
       "step": 2400
     },
     {
       "epoch": 2.513089005235602,
-      "eval_logits/chosen": -16.25,
-      "eval_logits/rejected": -16.0,
-      "eval_logps/chosen": -732.0,
-      "eval_logps/rejected": -744.0,
-      "eval_loss": 1.2620625495910645,
-      "eval_rewards/accuracies": 0.494140625,
-      "eval_rewards/chosen": -4.125,
-      "eval_rewards/margins": 0.435546875,
-      "eval_rewards/rejected": -4.5625,
-      "eval_runtime": 46.6298,
-      "eval_samples_per_second": 42.891,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -16.5,
+      "eval_logits/rejected": -16.25,
+      "eval_logps/chosen": -692.0,
+      "eval_logps/rejected": -704.0,
+      "eval_loss": 1.1338906288146973,
+      "eval_rewards/accuracies": 0.501953125,
+      "eval_rewards/chosen": -3.734375,
+      "eval_rewards/margins": 0.41796875,
+      "eval_rewards/rejected": -4.15625,
+      "eval_runtime": 46.4772,
+      "eval_samples_per_second": 43.032,
+      "eval_steps_per_second": 0.689,
       "step": 2400
     },
     {
       "epoch": 2.5235602094240837,
-      "grad_norm": 0.001979430043060832,
+      "grad_norm": 0.0009124340851263128,
       "learning_rate": 3.745515071340286e-06,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -784.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0002,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -7.78125,
+      "logps/chosen": -760.0,
+      "logps/rejected": -4128.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.828125,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -3.578125,
+      "rewards/margins": 22.5,
+      "rewards/rejected": -26.125,
       "step": 2410
     },
     {
       "epoch": 2.5340314136125652,
-      "grad_norm": 0.0029456701636552487,
+      "grad_norm": 0.0018270120566884292,
       "learning_rate": 3.586698932127236e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -748.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0004,
+      "logits/chosen": -16.0,
+      "logits/rejected": -7.8125,
+      "logps/chosen": -728.0,
+      "logps/rejected": -4064.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.640625,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 22.25,
+      "rewards/rejected": -25.75,
       "step": 2420
     },
     {
       "epoch": 2.5445026178010473,
-      "grad_norm": 0.32375203593521923,
+      "grad_norm": 0.011831796115066845,
       "learning_rate": 3.4310626834344744e-06,
-      "logits/chosen": -16.875,
-      "logits/rejected": -7.9375,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4016.0,
-      "loss": 0.0011,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.625,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5625,
-      "rewards/margins": 22.625,
-      "rewards/rejected": -26.125,
+      "rewards/chosen": -3.328125,
+      "rewards/margins": 21.625,
+      "rewards/rejected": -25.0,
       "step": 2430
     },
     {
       "epoch": 2.554973821989529,
-      "grad_norm": 0.005332402752057912,
+      "grad_norm": 0.000779458774775631,
       "learning_rate": 3.2786294373562343e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.4375,
-      "logps/chosen": -760.0,
-      "logps/rejected": -4384.0,
-      "loss": 0.0002,
+      "logits/chosen": -15.8125,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -736.0,
+      "logps/rejected": -4224.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.671875,
-      "rewards/margins": 24.5,
-      "rewards/rejected": -28.125,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 23.0,
+      "rewards/rejected": -26.5,
       "step": 2440
     },
     {
       "epoch": 2.5654450261780104,
-      "grad_norm": 0.0030963154818849908,
+      "grad_norm": 0.0014703433612537425,
       "learning_rate": 3.1294218303385526e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -780.0,
-      "logps/rejected": -4320.0,
-      "loss": 0.0001,
+      "logits/chosen": -15.6875,
+      "logits/rejected": -7.375,
+      "logps/chosen": -768.0,
+      "logps/rejected": -4192.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.65625,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.375,
+      "rewards/chosen": -3.546875,
+      "rewards/margins": 22.375,
+      "rewards/rejected": -26.0,
       "step": 2450
     },
     {
       "epoch": 2.5759162303664924,
-      "grad_norm": 0.046870064045400034,
+      "grad_norm": 0.014714159457927811,
       "learning_rate": 2.9834620198177483e-06,
-      "logits/chosen": -16.375,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -744.0,
-      "logps/rejected": -4192.0,
+      "logits/chosen": -15.75,
+      "logits/rejected": -7.625,
+      "logps/chosen": -728.0,
+      "logps/rejected": -4032.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 22.625,
+      "rewards/rejected": -26.0,
       "step": 2460
     },
     {
       "epoch": 2.5863874345549736,
-      "grad_norm": 0.0026558473322661584,
+      "grad_norm": 0.005819028594919864,
       "learning_rate": 2.840771680930068e-06,
-      "logits/chosen": -17.0,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -764.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.71875,
+      "logps/chosen": -760.0,
+      "logps/rejected": -4192.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -3.484375,
+      "rewards/margins": 23.25,
+      "rewards/rejected": -26.75,
       "step": 2470
     },
     {
       "epoch": 2.5968586387434556,
-      "grad_norm": 0.0008884569031893539,
+      "grad_norm": 0.0012281994734994403,
       "learning_rate": 2.701372003292868e-06,
-      "logits/chosen": -16.375,
-      "logits/rejected": -8.0,
-      "logps/chosen": -720.0,
-      "logps/rejected": -4416.0,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -7.53125,
+      "logps/chosen": -700.0,
+      "logps/rejected": -4192.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.46875,
-      "rewards/margins": 25.25,
-      "rewards/rejected": -28.75,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -26.625,
       "step": 2480
     },
     {
       "epoch": 2.607329842931937,
-      "grad_norm": 0.004212630077311793,
+      "grad_norm": 0.002351144618665887,
       "learning_rate": 2.56528368785795e-06,
-      "logits/chosen": -17.0,
-      "logits/rejected": -7.53125,
-      "logps/chosen": -772.0,
-      "logps/rejected": -4512.0,
-      "loss": 0.001,
+      "logits/chosen": -16.375,
+      "logits/rejected": -6.65625,
+      "logps/chosen": -752.0,
+      "logps/rejected": -4352.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.65625,
-      "rewards/margins": 25.0,
-      "rewards/rejected": -28.625,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 23.75,
+      "rewards/rejected": -27.125,
       "step": 2490
     },
     {
       "epoch": 2.6178010471204187,
-      "grad_norm": 0.0033811055991562256,
+      "grad_norm": 0.0029470163486374477,
       "learning_rate": 2.432526943837471e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -7.6875,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4448.0,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -708.0,
+      "logps/rejected": -4288.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 24.875,
-      "rewards/rejected": -28.375,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -26.625,
       "step": 2500
     },
     {
       "epoch": 2.6178010471204187,
-      "eval_logits/chosen": -16.125,
-      "eval_logits/rejected": -15.9375,
-      "eval_logps/chosen": -736.0,
-      "eval_logps/rejected": -748.0,
-      "eval_loss": 1.2696406841278076,
-      "eval_rewards/accuracies": 0.49609375,
-      "eval_rewards/chosen": -4.15625,
-      "eval_rewards/margins": 0.4453125,
-      "eval_rewards/rejected": -4.59375,
+      "eval_logits/chosen": -16.375,
+      "eval_logits/rejected": -16.125,
+      "eval_logps/chosen": -696.0,
+      "eval_logps/rejected": -708.0,
+      "eval_loss": 1.137218713760376,
+      "eval_rewards/accuracies": 0.501953125,
+      "eval_rewards/chosen": -3.765625,
+      "eval_rewards/margins": 0.421875,
+      "eval_rewards/rejected": -4.1875,
       "eval_runtime": 46.6126,
       "eval_samples_per_second": 42.907,
       "eval_steps_per_second": 0.687,
@@ -4175,600 +4175,600 @@
     },
     {
       "epoch": 2.6282722513089007,
-      "grad_norm": 0.05710875519322326,
+      "grad_norm": 0.021049572280585016,
       "learning_rate": 2.3031214857028446e-06,
-      "logits/chosen": -16.375,
-      "logits/rejected": -8.625,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.125,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -728.0,
+      "logps/rejected": -4096.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 23.25,
-      "rewards/rejected": -26.875,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 22.5,
+      "rewards/rejected": -25.75,
       "step": 2510
     },
     {
       "epoch": 2.6387434554973823,
-      "grad_norm": 0.5368181041542199,
+      "grad_norm": 0.0038953650704224227,
       "learning_rate": 2.177086530257139e-06,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.875,
-      "logps/chosen": -736.0,
-      "logps/rejected": -4416.0,
-      "loss": 0.0005,
+      "logits/chosen": -16.25,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -712.0,
+      "logps/rejected": -4256.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.640625,
-      "rewards/margins": 25.125,
-      "rewards/rejected": -28.75,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 23.625,
+      "rewards/rejected": -27.0,
       "step": 2520
     },
     {
       "epoch": 2.649214659685864,
-      "grad_norm": 0.004930823198659643,
+      "grad_norm": 0.0008682966918469032,
       "learning_rate": 2.054440793781354e-06,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.9375,
-      "logps/chosen": -724.0,
-      "logps/rejected": -4352.0,
-      "loss": 0.0001,
+      "logits/chosen": -15.875,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -692.0,
+      "logps/rejected": -4192.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.46875,
-      "rewards/margins": 24.25,
-      "rewards/rejected": -27.75,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 23.0,
+      "rewards/rejected": -26.125,
       "step": 2530
     },
     {
       "epoch": 2.6596858638743455,
-      "grad_norm": 0.0012299117360699222,
+      "grad_norm": 0.0032419448953130156,
       "learning_rate": 1.9352024892550587e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.3125,
-      "logps/chosen": -740.0,
-      "logps/rejected": -4416.0,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -724.0,
+      "logps/rejected": -4224.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 24.875,
-      "rewards/rejected": -28.375,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 23.125,
+      "rewards/rejected": -26.5,
       "step": 2540
     },
     {
       "epoch": 2.670157068062827,
-      "grad_norm": 0.00127782739596471,
+      "grad_norm": 0.0010393041547883505,
       "learning_rate": 1.8193893236517379e-06,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.4375,
-      "logps/chosen": -744.0,
-      "logps/rejected": -4416.0,
-      "loss": 0.0001,
+      "logits/chosen": -15.8125,
+      "logits/rejected": -7.84375,
+      "logps/chosen": -728.0,
+      "logps/rejected": -4288.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 24.875,
-      "rewards/rejected": -28.375,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 23.875,
+      "rewards/rejected": -27.25,
       "step": 2550
     },
     {
       "epoch": 2.680628272251309,
-      "grad_norm": 0.011889511025777197,
+      "grad_norm": 0.004341799282556766,
       "learning_rate": 1.7070184953092956e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.375,
-      "logps/chosen": -736.0,
-      "logps/rejected": -4320.0,
+      "logits/chosen": -16.0,
+      "logits/rejected": -7.1875,
+      "logps/chosen": -720.0,
+      "logps/rejected": -4192.0,
       "loss": 0.0006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 24.75,
-      "rewards/rejected": -28.125,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 23.5,
+      "rewards/rejected": -26.75,
       "step": 2560
     },
     {
       "epoch": 2.6910994764397906,
-      "grad_norm": 0.0015964958531365266,
+      "grad_norm": 0.005435379348900859,
       "learning_rate": 1.5981066913760918e-06,
-      "logits/chosen": -16.5,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -728.0,
-      "logps/rejected": -4320.0,
-      "loss": 0.0005,
+      "logits/chosen": -16.0,
+      "logits/rejected": -7.34375,
+      "logps/chosen": -700.0,
+      "logps/rejected": -4160.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 22.75,
+      "rewards/rejected": -26.0,
       "step": 2570
     },
     {
       "epoch": 2.701570680628272,
-      "grad_norm": 0.09387666593826648,
+      "grad_norm": 0.024030911993250866,
       "learning_rate": 1.492670085332884e-06,
-      "logits/chosen": -17.0,
-      "logits/rejected": -8.3125,
-      "logps/chosen": -760.0,
-      "logps/rejected": -4384.0,
-      "loss": 0.0005,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.625,
+      "logps/chosen": -732.0,
+      "logps/rejected": -4320.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.6875,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.375,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 23.25,
+      "rewards/rejected": -26.625,
       "step": 2580
     },
     {
       "epoch": 2.712041884816754,
-      "grad_norm": 0.0020036689784118274,
+      "grad_norm": 0.00018046818904815312,
       "learning_rate": 1.3907243345910787e-06,
-      "logits/chosen": -17.0,
-      "logits/rejected": -8.125,
-      "logps/chosen": -752.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.375,
+      "logits/rejected": -6.71875,
+      "logps/chosen": -732.0,
+      "logps/rejected": -4128.0,
+      "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 23.0,
-      "rewards/rejected": -26.375,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 22.0,
+      "rewards/rejected": -25.25,
       "step": 2590
     },
     {
       "epoch": 2.7225130890052354,
-      "grad_norm": 0.16195373650337483,
+      "grad_norm": 0.040882019611321145,
       "learning_rate": 1.292284578167538e-06,
-      "logits/chosen": -16.375,
-      "logits/rejected": -8.125,
-      "logps/chosen": -728.0,
-      "logps/rejected": -4096.0,
-      "loss": 0.0073,
+      "logits/chosen": -16.0,
+      "logits/rejected": -7.09375,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3984.0,
+      "loss": 0.0017,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 23.25,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 22.5,
+      "rewards/rejected": -25.75,
       "step": 2600
     },
     {
       "epoch": 2.7225130890052354,
       "eval_logits/chosen": -16.375,
       "eval_logits/rejected": -16.125,
-      "eval_logps/chosen": -732.0,
-      "eval_logps/rejected": -744.0,
-      "eval_loss": 1.2632499933242798,
+      "eval_logps/chosen": -696.0,
+      "eval_logps/rejected": -708.0,
+      "eval_loss": 1.1388906240463257,
       "eval_rewards/accuracies": 0.501953125,
-      "eval_rewards/chosen": -4.125,
-      "eval_rewards/margins": 0.4375,
-      "eval_rewards/rejected": -4.5625,
-      "eval_runtime": 46.616,
-      "eval_samples_per_second": 42.904,
+      "eval_rewards/chosen": -3.78125,
+      "eval_rewards/margins": 0.421875,
+      "eval_rewards/rejected": -4.21875,
+      "eval_runtime": 46.6241,
+      "eval_samples_per_second": 42.896,
       "eval_steps_per_second": 0.686,
       "step": 2600
     },
     {
       "epoch": 2.7329842931937174,
-      "grad_norm": 0.0001973715744176083,
+      "grad_norm": 0.0005143719186004678,
       "learning_rate": 1.1973654344364925e-06,
-      "logits/chosen": -16.625,
-      "logits/rejected": -7.0,
-      "logps/chosen": -764.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.0,
+      "logits/rejected": -6.09375,
+      "logps/chosen": -760.0,
+      "logps/rejected": -4128.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5625,
-      "rewards/margins": 23.125,
-      "rewards/rejected": -26.625,
+      "rewards/chosen": -3.5,
+      "rewards/margins": 22.75,
+      "rewards/rejected": -26.25,
       "step": 2610
     },
     {
       "epoch": 2.743455497382199,
-      "grad_norm": 0.023453314362470456,
+      "grad_norm": 0.07464060497383562,
       "learning_rate": 1.1059809989586583e-06,
-      "logits/chosen": -17.125,
-      "logits/rejected": -8.5,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4320.0,
-      "loss": 0.0047,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.59375,
+      "logps/chosen": -716.0,
+      "logps/rejected": -4256.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.390625,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -26.625,
       "step": 2620
     },
     {
       "epoch": 2.7539267015706805,
-      "grad_norm": 0.02422759165762764,
+      "grad_norm": 0.008137072832960543,
       "learning_rate": 1.018144842388044e-06,
       "logits/chosen": -16.25,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -776.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0006,
+      "logits/rejected": -6.96875,
+      "logps/chosen": -784.0,
+      "logps/rejected": -4160.0,
+      "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 23.5,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -3.609375,
+      "rewards/margins": 22.625,
+      "rewards/rejected": -26.25,
       "step": 2630
     },
     {
       "epoch": 2.7643979057591626,
-      "grad_norm": 0.033148049216260855,
+      "grad_norm": 0.012507634336175034,
       "learning_rate": 9.338700084567109e-07,
-      "logits/chosen": -16.625,
-      "logits/rejected": -9.375,
-      "logps/chosen": -720.0,
-      "logps/rejected": -4352.0,
+      "logits/chosen": -15.875,
+      "logits/rejected": -8.25,
+      "logps/chosen": -724.0,
+      "logps/rejected": -4288.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
       "rewards/chosen": -3.4375,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.5,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -26.75,
       "step": 2640
     },
     {
       "epoch": 2.774869109947644,
-      "grad_norm": 0.009903980366383526,
+      "grad_norm": 0.01322583395828459,
       "learning_rate": 8.531690120377606e-07,
-      "logits/chosen": -16.75,
-      "logits/rejected": -7.75,
-      "logps/chosen": -788.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.125,
+      "logits/rejected": -6.84375,
+      "logps/chosen": -772.0,
+      "logps/rejected": -4000.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.609375,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.375,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 22.5,
+      "rewards/rejected": -26.0,
       "step": 2650
     },
     {
       "epoch": 2.7853403141361257,
-      "grad_norm": 0.001721319806903957,
+      "grad_norm": 0.0014478973267603604,
       "learning_rate": 7.760538372868636e-07,
-      "logits/chosen": -17.0,
-      "logits/rejected": -8.875,
-      "logps/chosen": -712.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0039,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.46875,
+      "logps/chosen": -704.0,
+      "logps/rejected": -4224.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 22.875,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 22.25,
+      "rewards/rejected": -25.5,
       "step": 2660
     },
     {
       "epoch": 2.7958115183246073,
-      "grad_norm": 0.011997608455097605,
+      "grad_norm": 0.021289923820027503,
       "learning_rate": 7.025359358626166e-07,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.125,
-      "logps/chosen": -760.0,
-      "logps/rejected": -4352.0,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.4375,
+      "logps/chosen": -740.0,
+      "logps/rejected": -4288.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 23.625,
+      "rewards/rejected": -27.0,
       "step": 2670
     },
     {
       "epoch": 2.806282722513089,
-      "grad_norm": 0.004651446398726752,
+      "grad_norm": 0.002222248187323922,
       "learning_rate": 6.326262252259463e-07,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.25,
-      "logps/chosen": -768.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0003,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.640625,
-      "rewards/margins": 23.0,
-      "rewards/rejected": -26.625,
+      "logits/chosen": -16.0,
+      "logits/rejected": -7.34375,
+      "logps/chosen": -748.0,
+      "logps/rejected": -4192.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 22.5,
+      "rewards/rejected": -26.0,
       "step": 2680
     },
     {
       "epoch": 2.816753926701571,
-      "grad_norm": 0.0068208538806739105,
+      "grad_norm": 0.00384428313255218,
       "learning_rate": 5.66335087018871e-07,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.3125,
-      "logps/chosen": -736.0,
-      "logps/rejected": -4192.0,
-      "loss": 0.0009,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -728.0,
+      "logps/rejected": -4128.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.296875,
-      "rewards/margins": 23.125,
-      "rewards/rejected": -26.375,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 22.75,
+      "rewards/rejected": -25.875,
       "step": 2690
     },
     {
       "epoch": 2.8272251308900525,
-      "grad_norm": 0.00020105586016302157,
+      "grad_norm": 8.980126674495042e-05,
       "learning_rate": 5.036723655228226e-07,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -728.0,
-      "logps/rejected": -4352.0,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.46875,
+      "logps/chosen": -716.0,
+      "logps/rejected": -4256.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 24.25,
-      "rewards/rejected": -27.75,
+      "rewards/chosen": -3.421875,
+      "rewards/margins": 23.125,
+      "rewards/rejected": -26.5,
       "step": 2700
     },
     {
       "epoch": 2.8272251308900525,
       "eval_logits/chosen": -16.375,
       "eval_logits/rejected": -16.125,
-      "eval_logps/chosen": -728.0,
-      "eval_logps/rejected": -740.0,
-      "eval_loss": 1.2519999742507935,
-      "eval_rewards/accuracies": 0.4921875,
-      "eval_rewards/chosen": -4.09375,
-      "eval_rewards/margins": 0.42578125,
-      "eval_rewards/rejected": -4.5,
-      "eval_runtime": 46.6288,
-      "eval_samples_per_second": 42.892,
+      "eval_logps/chosen": -696.0,
+      "eval_logps/rejected": -712.0,
+      "eval_loss": 1.1391093730926514,
+      "eval_rewards/accuracies": 0.505859375,
+      "eval_rewards/chosen": -3.78125,
+      "eval_rewards/margins": 0.423828125,
+      "eval_rewards/rejected": -4.21875,
+      "eval_runtime": 46.6253,
+      "eval_samples_per_second": 42.895,
       "eval_steps_per_second": 0.686,
       "step": 2700
     },
     {
       "epoch": 2.837696335078534,
-      "grad_norm": 0.0005103911018979473,
+      "grad_norm": 0.000741557712324482,
       "learning_rate": 4.446473661967432e-07,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.125,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.0625,
       "logps/chosen": -720.0,
-      "logps/rejected": -4080.0,
+      "logps/rejected": -4032.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 22.875,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -3.328125,
+      "rewards/margins": 22.375,
+      "rewards/rejected": -25.75,
       "step": 2710
     },
     {
       "epoch": 2.8481675392670156,
-      "grad_norm": 0.005987108960235917,
+      "grad_norm": 0.013572476389035557,
       "learning_rate": 3.892688542952411e-07,
-      "logits/chosen": -16.875,
-      "logits/rejected": -7.78125,
-      "logps/chosen": -696.0,
-      "logps/rejected": -4096.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.375,
+      "logits/rejected": -6.9375,
+      "logps/chosen": -700.0,
+      "logps/rejected": -4064.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.234375,
-      "rewards/margins": 23.0,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -3.25,
+      "rewards/margins": 22.5,
+      "rewards/rejected": -25.75,
       "step": 2720
     },
     {
       "epoch": 2.858638743455497,
-      "grad_norm": 0.006281466101341092,
+      "grad_norm": 0.008722036390874205,
       "learning_rate": 3.3754505356694e-07,
-      "logits/chosen": -17.25,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4416.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.3125,
+      "logps/chosen": -728.0,
+      "logps/rejected": -4320.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 24.875,
-      "rewards/rejected": -28.25,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 24.0,
+      "rewards/rejected": -27.375,
       "step": 2730
     },
     {
       "epoch": 2.869109947643979,
-      "grad_norm": 0.00402088081841525,
+      "grad_norm": 0.0037292810494409107,
       "learning_rate": 2.8948364503322276e-07,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.125,
-      "logps/chosen": -728.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0013,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.125,
+      "logps/chosen": -708.0,
+      "logps/rejected": -4128.0,
+      "loss": 0.0014,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.375,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 23.5,
+      "rewards/rejected": -26.875,
       "step": 2740
     },
     {
       "epoch": 2.8795811518324608,
-      "grad_norm": 0.023060097709138708,
+      "grad_norm": 0.018108985337783587,
       "learning_rate": 2.4509176584762706e-07,
-      "logits/chosen": -17.0,
-      "logits/rejected": -7.46875,
-      "logps/chosen": -724.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.375,
+      "logits/rejected": -6.5625,
+      "logps/chosen": -728.0,
+      "logps/rejected": -4160.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -3.421875,
+      "rewards/margins": 22.625,
+      "rewards/rejected": -26.0,
       "step": 2750
     },
     {
       "epoch": 2.8900523560209423,
-      "grad_norm": 0.0009307438950141762,
+      "grad_norm": 0.0020691645958474834,
       "learning_rate": 2.0437600823595694e-07,
-      "logits/chosen": -16.375,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -792.0,
-      "logps/rejected": -4352.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.0,
+      "logits/rejected": -7.25,
+      "logps/chosen": -764.0,
+      "logps/rejected": -4256.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.75,
-      "rewards/margins": 23.5,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -3.5,
+      "rewards/margins": 23.0,
+      "rewards/rejected": -26.5,
       "step": 2760
     },
     {
       "epoch": 2.900523560209424,
-      "grad_norm": 0.0042895354887642215,
+      "grad_norm": 0.024752726552903972,
       "learning_rate": 1.6734241851733e-07,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.0,
-      "logps/chosen": -792.0,
-      "logps/rejected": -4288.0,
+      "logits/chosen": -16.25,
+      "logits/rejected": -6.875,
+      "logps/chosen": -784.0,
+      "logps/rejected": -4224.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.65625,
-      "rewards/margins": 23.5,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -3.5625,
+      "rewards/margins": 23.0,
+      "rewards/rejected": -26.625,
       "step": 2770
     },
     {
       "epoch": 2.9109947643979055,
-      "grad_norm": 7.551128328899704e-05,
+      "grad_norm": 0.00082376291909267,
       "learning_rate": 1.339964962062984e-07,
-      "logits/chosen": -17.125,
-      "logits/rejected": -9.125,
-      "logps/chosen": -688.0,
-      "logps/rejected": -4480.0,
+      "logits/chosen": -16.125,
+      "logits/rejected": -8.125,
+      "logps/chosen": -676.0,
+      "logps/rejected": -4352.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.328125,
-      "rewards/margins": 24.875,
-      "rewards/rejected": -28.25,
+      "rewards/chosen": -3.1875,
+      "rewards/margins": 23.75,
+      "rewards/rejected": -27.0,
       "step": 2780
     },
     {
       "epoch": 2.9214659685863875,
-      "grad_norm": 0.006255058759031103,
+      "grad_norm": 0.004437345957753471,
       "learning_rate": 1.0434319319617136e-07,
-      "logits/chosen": -17.0,
-      "logits/rejected": -9.25,
-      "logps/chosen": -704.0,
-      "logps/rejected": -4384.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.125,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -712.0,
+      "logps/rejected": -4320.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 24.25,
-      "rewards/rejected": -27.625,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 23.75,
+      "rewards/rejected": -27.125,
       "step": 2790
     },
     {
       "epoch": 2.931937172774869,
-      "grad_norm": 0.009042087853764426,
+      "grad_norm": 0.0012400324547334493,
       "learning_rate": 7.838691302365363e-08,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.25,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.53125,
+      "logps/chosen": -720.0,
+      "logps/rejected": -4160.0,
+      "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 23.375,
-      "rewards/rejected": -26.875,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 22.875,
+      "rewards/rejected": -26.125,
       "step": 2800
     },
     {
       "epoch": 2.931937172774869,
       "eval_logits/chosen": -16.375,
       "eval_logits/rejected": -16.125,
-      "eval_logps/chosen": -728.0,
-      "eval_logps/rejected": -740.0,
-      "eval_loss": 1.2507343292236328,
-      "eval_rewards/accuracies": 0.49609375,
-      "eval_rewards/chosen": -4.09375,
-      "eval_rewards/margins": 0.427734375,
-      "eval_rewards/rejected": -4.5,
-      "eval_runtime": 46.6176,
-      "eval_samples_per_second": 42.902,
+      "eval_logps/chosen": -696.0,
+      "eval_logps/rejected": -712.0,
+      "eval_loss": 1.1391406059265137,
+      "eval_rewards/accuracies": 0.5,
+      "eval_rewards/chosen": -3.796875,
+      "eval_rewards/margins": 0.423828125,
+      "eval_rewards/rejected": -4.21875,
+      "eval_runtime": 46.6303,
+      "eval_samples_per_second": 42.891,
       "eval_steps_per_second": 0.686,
       "step": 2800
     },
     {
       "epoch": 2.9424083769633507,
-      "grad_norm": 0.011726198213756842,
+      "grad_norm": 0.00463776794818808,
       "learning_rate": 5.613151021490459e-08,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.375,
-      "logps/chosen": -656.0,
-      "logps/rejected": -4416.0,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.625,
+      "logps/chosen": -632.0,
+      "logps/rejected": -4320.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.375,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 23.5,
+      "rewards/rejected": -26.5,
       "step": 2810
     },
     {
       "epoch": 2.9528795811518327,
-      "grad_norm": 0.012432187717223604,
+      "grad_norm": 0.0071741489363227025,
       "learning_rate": 3.7580289713159965e-08,
-      "logits/chosen": -16.75,
-      "logits/rejected": -8.25,
-      "logps/chosen": -740.0,
-      "logps/rejected": -3936.0,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.5,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3888.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 22.25,
-      "rewards/rejected": -25.5,
+      "rewards/chosen": -3.375,
+      "rewards/margins": 21.75,
+      "rewards/rejected": -25.125,
       "step": 2820
     },
     {
       "epoch": 2.9633507853403143,
-      "grad_norm": 0.003866882320718451,
+      "grad_norm": 0.006790637062876803,
       "learning_rate": 2.2736006387927277e-08,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.625,
-      "logps/chosen": -748.0,
-      "logps/rejected": -4320.0,
+      "logits/chosen": -16.5,
+      "logits/rejected": -8.0,
+      "logps/chosen": -740.0,
+      "logps/rejected": -4288.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.46875,
-      "rewards/margins": 23.875,
-      "rewards/rejected": -27.375,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 23.625,
+      "rewards/rejected": -27.0,
       "step": 2830
     },
     {
       "epoch": 2.973821989528796,
-      "grad_norm": 0.00255639454782368,
+      "grad_norm": 0.0015034207715979204,
       "learning_rate": 1.1600864625893604e-08,
-      "logits/chosen": -16.625,
-      "logits/rejected": -8.125,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4352.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.125,
+      "logits/rejected": -7.0625,
+      "logps/chosen": -704.0,
+      "logps/rejected": -4192.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.453125,
-      "rewards/margins": 24.5,
-      "rewards/rejected": -28.0,
+      "rewards/chosen": -3.1875,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -26.5,
       "step": 2840
     },
     {
       "epoch": 2.9842931937172774,
-      "grad_norm": 0.0004020623983245948,
+      "grad_norm": 0.014818109525900496,
       "learning_rate": 4.176518003573549e-09,
-      "logits/chosen": -16.875,
-      "logits/rejected": -8.0,
-      "logps/chosen": -748.0,
-      "logps/rejected": -4416.0,
-      "loss": 0.0019,
+      "logits/chosen": -16.25,
+      "logits/rejected": -6.9375,
+      "logps/chosen": -724.0,
+      "logps/rejected": -4320.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.65625,
-      "rewards/margins": 24.5,
-      "rewards/rejected": -28.125,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 23.625,
+      "rewards/rejected": -27.0,
       "step": 2850
     },
     {
       "epoch": 2.994764397905759,
-      "grad_norm": 0.0061148040302435175,
+      "grad_norm": 0.014656029663605687,
       "learning_rate": 4.640690417528903e-10,
-      "logits/chosen": -16.375,
-      "logits/rejected": -8.25,
-      "logps/chosen": -716.0,
-      "logps/rejected": -4352.0,
+      "logits/chosen": -16.125,
+      "logits/rejected": -7.84375,
+      "logps/chosen": -704.0,
+      "logps/rejected": -4256.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 24.75,
-      "rewards/rejected": -28.25,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 23.875,
+      "rewards/rejected": -27.25,
       "step": 2860
     },
     {
       "epoch": 3.0,
       "step": 2865,
       "total_flos": 0.0,
-      "train_loss": 0.02219091602174458,
-      "train_runtime": 12517.1932,
-      "train_samples_per_second": 14.648,
-      "train_steps_per_second": 0.229
+      "train_loss": 0.022635619988056292,
+      "train_runtime": 12425.7414,
+      "train_samples_per_second": 14.756,
+      "train_steps_per_second": 0.231
     }
   ],
   "logging_steps": 10,