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": 5.38308067689226,
+      "grad_norm": 5.383018034307613,
       "learning_rate": 1.7421602787456448e-07,
       "logits/chosen": -9.75,
       "logits/rejected": -8.6875,
@@ -25,43 +25,43 @@
     },
     {
       "epoch": 0.010471204188481676,
-      "grad_norm": 5.124685883317083,
+      "grad_norm": 5.12047887148474,
       "learning_rate": 1.7421602787456445e-06,
       "logits/chosen": -10.8125,
       "logits/rejected": -8.0,
       "logps/chosen": -510.0,
       "logps/rejected": -2080.0,
-      "loss": 0.6905,
-      "rewards/accuracies": 0.0902777761220932,
-      "rewards/chosen": -0.0002346038818359375,
-      "rewards/margins": 0.0002498626708984375,
-      "rewards/rejected": -0.000484466552734375,
+      "loss": 0.6903,
+      "rewards/accuracies": 0.0972222238779068,
+      "rewards/chosen": -0.00069427490234375,
+      "rewards/margins": 0.00119781494140625,
+      "rewards/rejected": -0.00189208984375,
       "step": 10
     },
     {
       "epoch": 0.020942408376963352,
-      "grad_norm": 5.266345890881455,
+      "grad_norm": 5.2561220267486695,
       "learning_rate": 3.484320557491289e-06,
       "logits/chosen": -10.9375,
       "logits/rejected": -8.375,
       "logps/chosen": -486.0,
       "logps/rejected": -2128.0,
       "loss": 0.6303,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.0067138671875,
-      "rewards/margins": 0.134765625,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.007171630859375,
+      "rewards/margins": 0.1337890625,
       "rewards/rejected": -0.1416015625,
       "step": 20
     },
     {
       "epoch": 0.031413612565445025,
-      "grad_norm": 1.7049065566770096,
+      "grad_norm": 1.6978262351476723,
       "learning_rate": 5.226480836236934e-06,
       "logits/chosen": -10.875,
       "logits/rejected": -8.1875,
       "logps/chosen": -448.0,
       "logps/rejected": -2272.0,
-      "loss": 0.3423,
+      "loss": 0.3421,
       "rewards/accuracies": 0.925000011920929,
       "rewards/chosen": -0.1044921875,
       "rewards/margins": 1.1640625,
@@ -70,43 +70,43 @@
     },
     {
       "epoch": 0.041884816753926704,
-      "grad_norm": 0.5243567019599547,
+      "grad_norm": 0.5221953002579055,
       "learning_rate": 6.968641114982578e-06,
       "logits/chosen": -11.1875,
       "logits/rejected": -8.9375,
       "logps/chosen": -576.0,
       "logps/rejected": -2544.0,
-      "loss": 0.2282,
+      "loss": 0.2286,
       "rewards/accuracies": 0.9125000238418579,
       "rewards/chosen": -0.68359375,
-      "rewards/margins": 3.765625,
-      "rewards/rejected": -4.4375,
+      "rewards/margins": 3.78125,
+      "rewards/rejected": -4.46875,
       "step": 40
     },
     {
       "epoch": 0.05235602094240838,
-      "grad_norm": 1.8665347307549,
+      "grad_norm": 1.8688713365109348,
       "learning_rate": 8.710801393728225e-06,
       "logits/chosen": -12.0,
       "logits/rejected": -9.875,
       "logps/chosen": -604.0,
       "logps/rejected": -2432.0,
-      "loss": 0.2274,
+      "loss": 0.2273,
       "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -0.7734375,
-      "rewards/margins": 3.78125,
+      "rewards/chosen": -0.76953125,
+      "rewards/margins": 3.765625,
       "rewards/rejected": -4.53125,
       "step": 50
     },
     {
       "epoch": 0.06282722513089005,
-      "grad_norm": 1.6133377477083573,
+      "grad_norm": 1.6195296149147707,
       "learning_rate": 1.0452961672473868e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.0625,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.125,
       "logps/chosen": -516.0,
       "logps/rejected": -2640.0,
-      "loss": 0.2086,
+      "loss": 0.2084,
       "rewards/accuracies": 0.9437500238418579,
       "rewards/chosen": -0.5625,
       "rewards/margins": 3.734375,
@@ -115,43 +115,43 @@
     },
     {
       "epoch": 0.07329842931937172,
-      "grad_norm": 2.758300742442637,
+      "grad_norm": 2.7781653554164896,
       "learning_rate": 1.2195121951219513e-05,
-      "logits/chosen": -14.4375,
-      "logits/rejected": -11.0625,
+      "logits/chosen": -14.5,
+      "logits/rejected": -11.125,
       "logps/chosen": -532.0,
       "logps/rejected": -2432.0,
-      "loss": 0.2306,
+      "loss": 0.2303,
       "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -0.65234375,
+      "rewards/chosen": -0.66015625,
       "rewards/margins": 3.21875,
       "rewards/rejected": -3.875,
       "step": 70
     },
     {
       "epoch": 0.08376963350785341,
-      "grad_norm": 1.4261185227440032,
+      "grad_norm": 1.4327439606952914,
       "learning_rate": 1.3937282229965156e-05,
-      "logits/chosen": -14.375,
-      "logits/rejected": -10.75,
+      "logits/chosen": -14.4375,
+      "logits/rejected": -10.8125,
       "logps/chosen": -516.0,
-      "logps/rejected": -2480.0,
+      "logps/rejected": -2496.0,
       "loss": 0.2193,
       "rewards/accuracies": 0.918749988079071,
       "rewards/chosen": -0.69921875,
       "rewards/margins": 3.03125,
-      "rewards/rejected": -3.71875,
+      "rewards/rejected": -3.734375,
       "step": 80
     },
     {
       "epoch": 0.09424083769633508,
-      "grad_norm": 1.5121621931310467,
+      "grad_norm": 1.4759866612339168,
       "learning_rate": 1.56794425087108e-05,
-      "logits/chosen": -14.5625,
+      "logits/chosen": -14.5,
       "logits/rejected": -10.875,
       "logps/chosen": -490.0,
       "logps/rejected": -2672.0,
-      "loss": 0.1808,
+      "loss": 0.1803,
       "rewards/accuracies": 0.9624999761581421,
       "rewards/chosen": -0.54296875,
       "rewards/margins": 3.515625,
@@ -160,13 +160,13 @@
     },
     {
       "epoch": 0.10471204188481675,
-      "grad_norm": 1.4750070586077515,
+      "grad_norm": 1.4701343266359446,
       "learning_rate": 1.742160278745645e-05,
       "logits/chosen": -14.25,
       "logits/rejected": -11.0,
       "logps/chosen": -500.0,
       "logps/rejected": -2640.0,
-      "loss": 0.1864,
+      "loss": 0.187,
       "rewards/accuracies": 0.9624999761581421,
       "rewards/chosen": -0.68359375,
       "rewards/margins": 3.984375,
@@ -176,255 +176,255 @@
     {
       "epoch": 0.10471204188481675,
       "eval_logits/chosen": -14.6875,
-      "eval_logits/rejected": -14.5,
+      "eval_logits/rejected": -14.5625,
       "eval_logps/chosen": -364.0,
       "eval_logps/rejected": -344.0,
-      "eval_loss": 0.676437497138977,
-      "eval_rewards/accuracies": 0.541015625,
-      "eval_rewards/chosen": -0.447265625,
+      "eval_loss": 0.6759999990463257,
+      "eval_rewards/accuracies": 0.546875,
+      "eval_rewards/chosen": -0.44921875,
       "eval_rewards/margins": 0.1083984375,
-      "eval_rewards/rejected": -0.5546875,
-      "eval_runtime": 48.6011,
-      "eval_samples_per_second": 41.151,
-      "eval_steps_per_second": 0.658,
+      "eval_rewards/rejected": -0.55859375,
+      "eval_runtime": 48.4993,
+      "eval_samples_per_second": 41.238,
+      "eval_steps_per_second": 0.66,
       "step": 100
     },
     {
       "epoch": 0.11518324607329843,
-      "grad_norm": 1.428472636565015,
+      "grad_norm": 1.448622836355066,
       "learning_rate": 1.9163763066202093e-05,
       "logits/chosen": -14.0625,
-      "logits/rejected": -11.25,
+      "logits/rejected": -11.3125,
       "logps/chosen": -576.0,
       "logps/rejected": -2480.0,
-      "loss": 0.2104,
+      "loss": 0.2096,
       "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -0.8046875,
-      "rewards/margins": 3.0625,
-      "rewards/rejected": -3.875,
+      "rewards/chosen": -0.8125,
+      "rewards/margins": 3.09375,
+      "rewards/rejected": -3.90625,
       "step": 110
     },
     {
       "epoch": 0.1256544502617801,
-      "grad_norm": 1.136327247749575,
+      "grad_norm": 1.1486948403190462,
       "learning_rate": 2.0905923344947736e-05,
       "logits/chosen": -12.125,
-      "logits/rejected": -10.125,
-      "logps/chosen": -556.0,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -560.0,
       "logps/rejected": -2608.0,
-      "loss": 0.1613,
+      "loss": 0.1618,
       "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -0.87109375,
+      "rewards/chosen": -0.875,
       "rewards/margins": 4.53125,
       "rewards/rejected": -5.40625,
       "step": 120
     },
     {
       "epoch": 0.13612565445026178,
-      "grad_norm": 1.2324544779713202,
+      "grad_norm": 1.2533802819160162,
       "learning_rate": 2.264808362369338e-05,
-      "logits/chosen": -11.9375,
-      "logits/rejected": -9.6875,
+      "logits/chosen": -12.1875,
+      "logits/rejected": -9.875,
       "logps/chosen": -512.0,
       "logps/rejected": -2624.0,
-      "loss": 0.1672,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -0.66796875,
-      "rewards/margins": 4.3125,
-      "rewards/rejected": -4.96875,
+      "loss": 0.1667,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.67578125,
+      "rewards/margins": 4.25,
+      "rewards/rejected": -4.90625,
       "step": 130
     },
     {
       "epoch": 0.14659685863874344,
-      "grad_norm": 1.5984891624397473,
+      "grad_norm": 1.7177662297335345,
       "learning_rate": 2.4390243902439026e-05,
-      "logits/chosen": -11.6875,
-      "logits/rejected": -9.3125,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -9.625,
       "logps/chosen": -456.0,
       "logps/rejected": -2560.0,
-      "loss": 0.2005,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -0.5546875,
-      "rewards/margins": 3.78125,
-      "rewards/rejected": -4.34375,
+      "loss": 0.1787,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -0.54296875,
+      "rewards/margins": 3.890625,
+      "rewards/rejected": -4.4375,
       "step": 140
     },
     {
       "epoch": 0.15706806282722513,
-      "grad_norm": 1.2606827932294156,
+      "grad_norm": 1.2078665737013297,
       "learning_rate": 2.6132404181184672e-05,
-      "logits/chosen": -11.625,
-      "logits/rejected": -9.125,
-      "logps/chosen": -488.0,
-      "logps/rejected": -2512.0,
-      "loss": 0.1755,
+      "logits/chosen": -12.25,
+      "logits/rejected": -9.625,
+      "logps/chosen": -492.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.1713,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -0.64453125,
-      "rewards/margins": 3.75,
-      "rewards/rejected": -4.40625,
+      "rewards/chosen": -0.69140625,
+      "rewards/margins": 3.828125,
+      "rewards/rejected": -4.53125,
       "step": 150
     },
     {
       "epoch": 0.16753926701570682,
-      "grad_norm": 3.093909888170432,
+      "grad_norm": 3.4807603308040584,
       "learning_rate": 2.7874564459930312e-05,
-      "logits/chosen": -12.6875,
-      "logits/rejected": -10.125,
-      "logps/chosen": -510.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.1924,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -1.0078125,
-      "rewards/margins": 4.09375,
-      "rewards/rejected": -5.09375,
+      "logits/chosen": -12.875,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -524.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.1858,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -1.1484375,
+      "rewards/margins": 4.21875,
+      "rewards/rejected": -5.375,
       "step": 160
     },
     {
       "epoch": 0.17801047120418848,
-      "grad_norm": 1.1896686461642088,
+      "grad_norm": 1.208294635292511,
       "learning_rate": 2.9616724738675962e-05,
-      "logits/chosen": -13.375,
-      "logits/rejected": -11.0,
-      "logps/chosen": -620.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.1624,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -1.2890625,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -10.625,
+      "logps/chosen": -660.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.1725,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -1.6875,
       "rewards/margins": 4.28125,
-      "rewards/rejected": -5.5625,
+      "rewards/rejected": -5.96875,
       "step": 170
     },
     {
       "epoch": 0.18848167539267016,
-      "grad_norm": 1.786162172320875,
+      "grad_norm": 1.389184071600737,
       "learning_rate": 3.13588850174216e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.6875,
-      "logps/chosen": -540.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.0998,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -1.0,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -5.6875,
+      "logits/chosen": -11.875,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -576.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.1086,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -1.3359375,
+      "rewards/margins": 4.59375,
+      "rewards/rejected": -5.9375,
       "step": 180
     },
     {
       "epoch": 0.19895287958115182,
-      "grad_norm": 4.125706219478258,
+      "grad_norm": 1.2821087081049185,
       "learning_rate": 3.310104529616725e-05,
-      "logits/chosen": -13.0,
-      "logits/rejected": -10.6875,
-      "logps/chosen": -572.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.2826,
+      "logits/chosen": -12.75,
+      "logits/rejected": -10.0,
+      "logps/chosen": -576.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.2036,
       "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -1.296875,
-      "rewards/margins": 5.53125,
-      "rewards/rejected": -6.8125,
+      "rewards/chosen": -1.34375,
+      "rewards/margins": 5.25,
+      "rewards/rejected": -6.59375,
       "step": 190
     },
     {
       "epoch": 0.2094240837696335,
-      "grad_norm": 1.5835108506389313,
+      "grad_norm": 1.1535483494131622,
       "learning_rate": 3.48432055749129e-05,
       "logits/chosen": -12.75,
-      "logits/rejected": -10.5,
+      "logits/rejected": -10.125,
       "logps/chosen": -494.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.1156,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -0.83984375,
-      "rewards/margins": 4.5625,
-      "rewards/rejected": -5.40625,
+      "logps/rejected": -2624.0,
+      "loss": 0.1162,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -0.81640625,
+      "rewards/margins": 4.40625,
+      "rewards/rejected": -5.25,
       "step": 200
     },
     {
       "epoch": 0.2094240837696335,
-      "eval_logits/chosen": -11.6875,
-      "eval_logits/rejected": -11.4375,
-      "eval_logps/chosen": -398.0,
-      "eval_logps/rejected": -378.0,
-      "eval_loss": 0.6999062299728394,
-      "eval_rewards/accuracies": 0.5234375,
-      "eval_rewards/chosen": -0.80078125,
-      "eval_rewards/margins": 0.09912109375,
-      "eval_rewards/rejected": -0.8984375,
-      "eval_runtime": 46.6479,
-      "eval_samples_per_second": 42.874,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -12.0625,
+      "eval_logits/rejected": -11.6875,
+      "eval_logps/chosen": -396.0,
+      "eval_logps/rejected": -376.0,
+      "eval_loss": 0.6879374980926514,
+      "eval_rewards/accuracies": 0.541015625,
+      "eval_rewards/chosen": -0.7734375,
+      "eval_rewards/margins": 0.1015625,
+      "eval_rewards/rejected": -0.875,
+      "eval_runtime": 46.5928,
+      "eval_samples_per_second": 42.925,
+      "eval_steps_per_second": 0.687,
       "step": 200
     },
     {
       "epoch": 0.2198952879581152,
-      "grad_norm": 0.987453416665501,
+      "grad_norm": 2.3780448338226745,
       "learning_rate": 3.6585365853658535e-05,
-      "logits/chosen": -11.75,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -524.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.1272,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -1.0234375,
-      "rewards/margins": 5.125,
-      "rewards/rejected": -6.125,
+      "logits/chosen": -11.8125,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -528.0,
+      "logps/rejected": -2768.0,
+      "loss": 0.1247,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -1.046875,
+      "rewards/margins": 5.84375,
+      "rewards/rejected": -6.90625,
       "step": 210
     },
     {
       "epoch": 0.23036649214659685,
-      "grad_norm": 1.0889613168882084,
+      "grad_norm": 1.144526845514364,
       "learning_rate": 3.8327526132404185e-05,
-      "logits/chosen": -12.625,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -532.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.1713,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -0.96875,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -5.375,
+      "logits/chosen": -12.3125,
+      "logits/rejected": -9.5,
+      "logps/chosen": -516.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.172,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -0.80078125,
+      "rewards/margins": 4.125,
+      "rewards/rejected": -4.9375,
       "step": 220
     },
     {
       "epoch": 0.24083769633507854,
-      "grad_norm": 1.6920828920147442,
+      "grad_norm": 1.8668957975082823,
       "learning_rate": 4.006968641114983e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -532.0,
-      "logps/rejected": -2944.0,
-      "loss": 0.1088,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -1.125,
-      "rewards/margins": 5.3125,
-      "rewards/rejected": -6.4375,
+      "logits/chosen": -12.125,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -524.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.11,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.078125,
+      "rewards/margins": 5.46875,
+      "rewards/rejected": -6.5625,
       "step": 230
     },
     {
       "epoch": 0.2513089005235602,
-      "grad_norm": 0.7898867619854981,
+      "grad_norm": 0.8786932577019532,
       "learning_rate": 4.181184668989547e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.4375,
-      "logps/chosen": -604.0,
-      "logps/rejected": -2800.0,
-      "loss": 0.1776,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -1.2109375,
-      "rewards/margins": 4.25,
-      "rewards/rejected": -5.46875,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -596.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.1747,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.140625,
+      "rewards/margins": 4.28125,
+      "rewards/rejected": -5.4375,
       "step": 240
     },
     {
       "epoch": 0.2617801047120419,
-      "grad_norm": 2.123153422041998,
+      "grad_norm": 2.1158278431724704,
       "learning_rate": 4.3554006968641115e-05,
-      "logits/chosen": -14.125,
-      "logits/rejected": -11.25,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -10.8125,
       "logps/chosen": -696.0,
       "logps/rejected": -2704.0,
-      "loss": 0.192,
-      "rewards/accuracies": 0.918749988079071,
+      "loss": 0.1825,
+      "rewards/accuracies": 0.925000011920929,
       "rewards/chosen": -1.3671875,
       "rewards/margins": 4.125,
       "rewards/rejected": -5.5,
@@ -432,4342 +432,4342 @@
     },
     {
       "epoch": 0.27225130890052357,
-      "grad_norm": 1.1472354666715698,
+      "grad_norm": 1.0397790565612381,
       "learning_rate": 4.529616724738676e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.9375,
+      "logits/chosen": -13.75,
+      "logits/rejected": -11.25,
       "logps/chosen": -592.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.1599,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -1.2578125,
-      "rewards/margins": 4.15625,
-      "rewards/rejected": -5.4375,
+      "logps/rejected": -2768.0,
+      "loss": 0.1714,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -1.28125,
+      "rewards/margins": 4.40625,
+      "rewards/rejected": -5.6875,
       "step": 260
     },
     {
       "epoch": 0.28272251308900526,
-      "grad_norm": 1.5959333305446395,
+      "grad_norm": 1.7559458008861224,
       "learning_rate": 4.703832752613241e-05,
-      "logits/chosen": -12.1875,
-      "logits/rejected": -10.3125,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -10.6875,
       "logps/chosen": -560.0,
-      "logps/rejected": -2848.0,
-      "loss": 0.1573,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -1.203125,
-      "rewards/margins": 5.8125,
-      "rewards/rejected": -7.03125,
+      "logps/rejected": -2768.0,
+      "loss": 0.1504,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -1.1953125,
+      "rewards/margins": 5.0,
+      "rewards/rejected": -6.1875,
       "step": 270
     },
     {
       "epoch": 0.2931937172774869,
-      "grad_norm": 1.185815344909836,
+      "grad_norm": 1.071278913730053,
       "learning_rate": 4.878048780487805e-05,
-      "logits/chosen": -12.8125,
-      "logits/rejected": -10.25,
-      "logps/chosen": -568.0,
-      "logps/rejected": -2512.0,
-      "loss": 0.2706,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -1.0546875,
-      "rewards/margins": 3.71875,
-      "rewards/rejected": -4.78125,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.625,
+      "logps/chosen": -612.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.2854,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -1.4765625,
+      "rewards/margins": 4.34375,
+      "rewards/rejected": -5.8125,
       "step": 280
     },
     {
       "epoch": 0.3036649214659686,
-      "grad_norm": 1.0364545571263137,
+      "grad_norm": 1.1458674592015403,
       "learning_rate": 4.999983293481418e-05,
-      "logits/chosen": -12.25,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -544.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.2109,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -0.8828125,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.625,
+      "logps/chosen": -624.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.2421,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -1.65625,
       "rewards/margins": 4.03125,
-      "rewards/rejected": -4.90625,
+      "rewards/rejected": -5.6875,
       "step": 290
     },
     {
       "epoch": 0.31413612565445026,
-      "grad_norm": 0.703025557252931,
+      "grad_norm": 1.0904473144529527,
       "learning_rate": 4.9996862949181595e-05,
-      "logits/chosen": -11.6875,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -556.0,
-      "logps/rejected": -2928.0,
-      "loss": 0.1465,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -628.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.1436,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -1.2890625,
-      "rewards/margins": 6.09375,
-      "rewards/rejected": -7.375,
+      "rewards/chosen": -1.9921875,
+      "rewards/margins": 5.71875,
+      "rewards/rejected": -7.6875,
       "step": 300
     },
     {
       "epoch": 0.31413612565445026,
-      "eval_logits/chosen": -11.375,
-      "eval_logits/rejected": -10.9375,
-      "eval_logps/chosen": -424.0,
-      "eval_logps/rejected": -406.0,
-      "eval_loss": 0.7544375061988831,
-      "eval_rewards/accuracies": 0.48828125,
-      "eval_rewards/chosen": -1.0546875,
-      "eval_rewards/margins": 0.10791015625,
-      "eval_rewards/rejected": -1.1640625,
-      "eval_runtime": 46.6412,
-      "eval_samples_per_second": 42.881,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -14.0625,
+      "eval_logits/rejected": -13.875,
+      "eval_logps/chosen": -484.0,
+      "eval_logps/rejected": -468.0,
+      "eval_loss": 0.7669844031333923,
+      "eval_rewards/accuracies": 0.494140625,
+      "eval_rewards/chosen": -1.65625,
+      "eval_rewards/margins": 0.1376953125,
+      "eval_rewards/rejected": -1.796875,
+      "eval_runtime": 46.5962,
+      "eval_samples_per_second": 42.922,
+      "eval_steps_per_second": 0.687,
       "step": 300
     },
     {
       "epoch": 0.32460732984293195,
-      "grad_norm": 1.3095123966398965,
+      "grad_norm": 1.2885530701828156,
       "learning_rate": 4.9990180911526896e-05,
-      "logits/chosen": -11.5,
-      "logits/rejected": -9.5,
-      "logps/chosen": -580.0,
-      "logps/rejected": -2880.0,
-      "loss": 0.1313,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -1.3984375,
-      "rewards/margins": 5.875,
-      "rewards/rejected": -7.28125,
+      "logits/chosen": -13.75,
+      "logits/rejected": -10.5,
+      "logps/chosen": -648.0,
+      "logps/rejected": -2976.0,
+      "loss": 0.1444,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -2.09375,
+      "rewards/margins": 6.15625,
+      "rewards/rejected": -8.25,
       "step": 310
     },
     {
       "epoch": 0.33507853403141363,
-      "grad_norm": 2.142337171370044,
+      "grad_norm": 2.4396582315170985,
       "learning_rate": 4.997978781413745e-05,
-      "logits/chosen": -12.5,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -664.0,
-      "logps/rejected": -2976.0,
-      "loss": 0.1775,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -2.171875,
-      "rewards/margins": 6.0625,
-      "rewards/rejected": -8.25,
+      "logits/chosen": -15.0,
+      "logits/rejected": -11.375,
+      "logps/chosen": -676.0,
+      "logps/rejected": -2928.0,
+      "loss": 0.1706,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -2.28125,
+      "rewards/margins": 5.5,
+      "rewards/rejected": -7.8125,
       "step": 320
     },
     {
       "epoch": 0.34554973821989526,
-      "grad_norm": 1.1180308740605138,
+      "grad_norm": 0.7610172277555747,
       "learning_rate": 4.996568520039565e-05,
-      "logits/chosen": -12.0,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -664.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.1794,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -11.375,
+      "logps/chosen": -744.0,
+      "logps/rejected": -2992.0,
+      "loss": 0.2232,
       "rewards/accuracies": 0.9375,
-      "rewards/chosen": -1.625,
-      "rewards/margins": 4.65625,
-      "rewards/rejected": -6.28125,
+      "rewards/chosen": -2.421875,
+      "rewards/margins": 5.90625,
+      "rewards/rejected": -8.3125,
       "step": 330
     },
     {
       "epoch": 0.35602094240837695,
-      "grad_norm": 0.7734569833149393,
+      "grad_norm": 0.792309299570663,
       "learning_rate": 4.994787516454973e-05,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -10.625,
-      "logps/chosen": -624.0,
+      "logits/chosen": -15.125,
+      "logits/rejected": -11.5,
+      "logps/chosen": -640.0,
       "logps/rejected": -2864.0,
-      "loss": 0.1475,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -1.453125,
-      "rewards/margins": 5.84375,
-      "rewards/rejected": -7.3125,
-      "step": 340
-    },
-    {
+      "loss": 0.221,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -1.640625,
+      "rewards/margins": 5.59375,
+      "rewards/rejected": -7.21875,
+      "step": 340
+    },
+    {
       "epoch": 0.36649214659685864,
-      "grad_norm": 1.5631652496307882,
+      "grad_norm": 1.3205949474721688,
       "learning_rate": 4.992636035140273e-05,
-      "logits/chosen": -14.1875,
-      "logits/rejected": -11.0,
-      "logps/chosen": -628.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.2169,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -10.875,
+      "logps/chosen": -668.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.3313,
       "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -1.7421875,
-      "rewards/margins": 3.984375,
-      "rewards/rejected": -5.75,
+      "rewards/chosen": -2.125,
+      "rewards/margins": 4.21875,
+      "rewards/rejected": -6.34375,
       "step": 350
     },
     {
       "epoch": 0.3769633507853403,
-      "grad_norm": 1.9544156123509902,
+      "grad_norm": 1.986141769222696,
       "learning_rate": 4.99011439559198e-05,
-      "logits/chosen": -14.75,
-      "logits/rejected": -11.25,
-      "logps/chosen": -616.0,
-      "logps/rejected": -2944.0,
-      "loss": 0.1645,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -1.65625,
-      "rewards/margins": 5.15625,
-      "rewards/rejected": -6.8125,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -11.125,
+      "logps/chosen": -988.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.2206,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -5.375,
+      "rewards/margins": 6.53125,
+      "rewards/rejected": -11.9375,
       "step": 360
     },
     {
       "epoch": 0.387434554973822,
-      "grad_norm": 2.2095468928684725,
+      "grad_norm": 1.9453847983219434,
       "learning_rate": 4.987222972275369e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -12.5625,
-      "logps/chosen": -612.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.1457,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -1.640625,
-      "rewards/margins": 4.78125,
-      "rewards/rejected": -6.4375,
+      "logits/chosen": -15.75,
+      "logits/rejected": -12.125,
+      "logps/chosen": -600.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.1414,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -1.546875,
+      "rewards/margins": 4.125,
+      "rewards/rejected": -5.6875,
       "step": 370
     },
     {
       "epoch": 0.39790575916230364,
-      "grad_norm": 1.57655360572696,
+      "grad_norm": 1.9103647902776453,
       "learning_rate": 4.983962194568872e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -12.875,
-      "logps/chosen": -680.0,
-      "logps/rejected": -2800.0,
-      "loss": 0.1805,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -1.875,
-      "rewards/margins": 5.28125,
-      "rewards/rejected": -7.15625,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -12.75,
+      "logps/chosen": -688.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.182,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -1.9375,
+      "rewards/margins": 5.09375,
+      "rewards/rejected": -7.03125,
       "step": 380
     },
     {
       "epoch": 0.4083769633507853,
-      "grad_norm": 2.729095166731873,
+      "grad_norm": 1.358690062186765,
       "learning_rate": 4.9803325467003084e-05,
       "logits/chosen": -15.6875,
-      "logits/rejected": -11.625,
-      "logps/chosen": -608.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.1816,
+      "logits/rejected": -11.875,
+      "logps/chosen": -604.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.1844,
       "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -1.59375,
-      "rewards/margins": 4.0625,
-      "rewards/rejected": -5.6875,
+      "rewards/chosen": -1.5390625,
+      "rewards/margins": 4.09375,
+      "rewards/rejected": -5.625,
       "step": 390
     },
     {
       "epoch": 0.418848167539267,
-      "grad_norm": 1.0787492937668104,
+      "grad_norm": 1.07951306496438,
       "learning_rate": 4.9763345676749816e-05,
-      "logits/chosen": -15.0,
-      "logits/rejected": -11.75,
+      "logits/chosen": -15.875,
+      "logits/rejected": -12.3125,
       "logps/chosen": -624.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.1468,
+      "logps/rejected": -2736.0,
+      "loss": 0.1461,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -1.3828125,
-      "rewards/margins": 4.71875,
-      "rewards/rejected": -6.125,
+      "rewards/chosen": -1.375,
+      "rewards/margins": 4.5625,
+      "rewards/rejected": -5.9375,
       "step": 400
     },
     {
       "epoch": 0.418848167539267,
-      "eval_logits/chosen": -15.25,
-      "eval_logits/rejected": -15.1875,
-      "eval_logps/chosen": -428.0,
-      "eval_logps/rejected": -402.0,
-      "eval_loss": 0.7513124942779541,
-      "eval_rewards/accuracies": 0.470703125,
-      "eval_rewards/chosen": -1.1015625,
-      "eval_rewards/margins": 0.0322265625,
-      "eval_rewards/rejected": -1.1328125,
-      "eval_runtime": 46.6407,
-      "eval_samples_per_second": 42.881,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -16.625,
+      "eval_logits/rejected": -16.5,
+      "eval_logps/chosen": -422.0,
+      "eval_logps/rejected": -394.0,
+      "eval_loss": 0.7442499995231628,
+      "eval_rewards/accuracies": 0.50390625,
+      "eval_rewards/chosen": -1.046875,
+      "eval_rewards/margins": 0.0201416015625,
+      "eval_rewards/rejected": -1.0625,
+      "eval_runtime": 46.5948,
+      "eval_samples_per_second": 42.923,
+      "eval_steps_per_second": 0.687,
       "step": 400
     },
     {
       "epoch": 0.4293193717277487,
-      "grad_norm": 0.990800383756,
+      "grad_norm": 0.9608468048456311,
       "learning_rate": 4.9719688511956374e-05,
-      "logits/chosen": -14.75,
-      "logits/rejected": -11.625,
-      "logps/chosen": -664.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.2037,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -1.5625,
-      "rewards/margins": 5.03125,
-      "rewards/rejected": -6.59375,
+      "logits/chosen": -16.125,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -648.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.2054,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -1.40625,
+      "rewards/margins": 4.625,
+      "rewards/rejected": -6.03125,
       "step": 410
     },
     {
       "epoch": 0.4397905759162304,
-      "grad_norm": 0.9838504657021493,
+      "grad_norm": 0.8436840498493646,
       "learning_rate": 4.967236045574294e-05,
-      "logits/chosen": -14.875,
-      "logits/rejected": -11.625,
-      "logps/chosen": -580.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.1242,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -1.5390625,
-      "rewards/margins": 3.796875,
-      "rewards/rejected": -5.34375,
+      "logits/chosen": -16.5,
+      "logits/rejected": -12.25,
+      "logps/chosen": -544.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.133,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.2109375,
+      "rewards/margins": 3.84375,
+      "rewards/rejected": -5.0625,
       "step": 420
     },
     {
       "epoch": 0.450261780104712,
-      "grad_norm": 2.5137656623349414,
+      "grad_norm": 2.7346325044890354,
       "learning_rate": 4.96213685363597e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -12.3125,
-      "logps/chosen": -572.0,
-      "logps/rejected": -2848.0,
-      "loss": 0.1511,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -1.5234375,
-      "rewards/margins": 5.90625,
-      "rewards/rejected": -7.4375,
+      "logits/chosen": -16.875,
+      "logits/rejected": -12.625,
+      "logps/chosen": -584.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.1624,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -1.6328125,
+      "rewards/margins": 5.375,
+      "rewards/rejected": -7.0,
       "step": 430
     },
     {
       "epoch": 0.4607329842931937,
-      "grad_norm": 0.9069061490412824,
+      "grad_norm": 1.0495441672783739,
       "learning_rate": 4.956672032614315e-05,
-      "logits/chosen": -14.5625,
-      "logits/rejected": -11.5625,
-      "logps/chosen": -588.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.1128,
+      "logits/chosen": -16.75,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -600.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.1142,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -1.4921875,
-      "rewards/margins": 5.5,
-      "rewards/rejected": -7.0,
+      "rewards/chosen": -1.625,
+      "rewards/margins": 5.09375,
+      "rewards/rejected": -6.6875,
       "step": 440
     },
     {
       "epoch": 0.4712041884816754,
-      "grad_norm": 1.3568919349746313,
+      "grad_norm": 1.1366874068089412,
       "learning_rate": 4.950842394039157e-05,
-      "logits/chosen": -14.75,
-      "logits/rejected": -11.375,
+      "logits/chosen": -16.375,
+      "logits/rejected": -12.5,
       "logps/chosen": -572.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.1608,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -1.5078125,
-      "rewards/margins": 4.53125,
-      "rewards/rejected": -6.03125,
+      "logps/rejected": -2720.0,
+      "loss": 0.1455,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.546875,
+      "rewards/margins": 4.78125,
+      "rewards/rejected": -6.3125,
       "step": 450
     },
     {
       "epoch": 0.4816753926701571,
-      "grad_norm": 1.1441529193282705,
+      "grad_norm": 1.24591242794086,
       "learning_rate": 4.944648803615994e-05,
-      "logits/chosen": -15.0625,
-      "logits/rejected": -10.6875,
-      "logps/chosen": -556.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.1252,
+      "logits/chosen": -15.8125,
+      "logits/rejected": -11.6875,
+      "logps/chosen": -568.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.128,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -1.3671875,
-      "rewards/margins": 4.4375,
-      "rewards/rejected": -5.8125,
+      "rewards/chosen": -1.4765625,
+      "rewards/margins": 4.96875,
+      "rewards/rejected": -6.4375,
       "step": 460
     },
     {
       "epoch": 0.49214659685863876,
-      "grad_norm": 0.8364076445731514,
+      "grad_norm": 0.3411094125308126,
       "learning_rate": 4.938092181097433e-05,
-      "logits/chosen": -14.6875,
-      "logits/rejected": -11.1875,
-      "logps/chosen": -640.0,
-      "logps/rejected": -2848.0,
-      "loss": 0.11,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -1.7734375,
-      "rewards/margins": 5.84375,
-      "rewards/rejected": -7.625,
+      "logits/chosen": -14.375,
+      "logits/rejected": -11.5,
+      "logps/chosen": -648.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.1186,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -1.875,
+      "rewards/margins": 6.0625,
+      "rewards/rejected": -7.9375,
       "step": 470
     },
     {
       "epoch": 0.5026178010471204,
-      "grad_norm": 0.7167581392274818,
+      "grad_norm": 1.4164327506595464,
       "learning_rate": 4.931173500146607e-05,
-      "logits/chosen": -13.6875,
-      "logits/rejected": -10.625,
-      "logps/chosen": -676.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.1407,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -2.15625,
-      "rewards/margins": 7.90625,
-      "rewards/rejected": -10.0625,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -720.0,
+      "logps/rejected": -3328.0,
+      "loss": 0.1323,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -2.59375,
+      "rewards/margins": 8.4375,
+      "rewards/rejected": -11.0,
       "step": 480
     },
     {
       "epoch": 0.5130890052356021,
-      "grad_norm": 1.02493568676847,
+      "grad_norm": 1.0940496543628702,
       "learning_rate": 4.923893788192582e-05,
-      "logits/chosen": -13.0625,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -676.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.5,
+      "logps/chosen": -680.0,
       "logps/rejected": -2992.0,
-      "loss": 0.1668,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -2.03125,
-      "rewards/margins": 5.0,
-      "rewards/rejected": -7.03125,
+      "loss": 0.1636,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -2.078125,
+      "rewards/margins": 4.90625,
+      "rewards/rejected": -6.96875,
       "step": 490
     },
     {
       "epoch": 0.5235602094240838,
-      "grad_norm": 1.2468386390469541,
+      "grad_norm": 1.349187034794465,
       "learning_rate": 4.9162541262777906e-05,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -11.0625,
-      "logps/chosen": -616.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.1267,
+      "logits/chosen": -15.125,
+      "logits/rejected": -12.0625,
+      "logps/chosen": -632.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.1352,
       "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -1.84375,
-      "rewards/margins": 4.75,
-      "rewards/rejected": -6.59375,
+      "rewards/chosen": -2.015625,
+      "rewards/margins": 5.09375,
+      "rewards/rejected": -7.09375,
       "step": 500
     },
     {
       "epoch": 0.5235602094240838,
-      "eval_logits/chosen": -13.375,
-      "eval_logits/rejected": -13.125,
-      "eval_logps/chosen": -460.0,
-      "eval_logps/rejected": -436.0,
-      "eval_loss": 0.7807812690734863,
-      "eval_rewards/accuracies": 0.517578125,
-      "eval_rewards/chosen": -1.421875,
-      "eval_rewards/margins": 0.057373046875,
-      "eval_rewards/rejected": -1.484375,
-      "eval_runtime": 46.6389,
-      "eval_samples_per_second": 42.883,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -15.6875,
+      "eval_logits/rejected": -15.5625,
+      "eval_logps/chosen": -482.0,
+      "eval_logps/rejected": -460.0,
+      "eval_loss": 0.8131250143051147,
+      "eval_rewards/accuracies": 0.501953125,
+      "eval_rewards/chosen": -1.640625,
+      "eval_rewards/margins": 0.06298828125,
+      "eval_rewards/rejected": -1.703125,
+      "eval_runtime": 46.5907,
+      "eval_samples_per_second": 42.927,
+      "eval_steps_per_second": 0.687,
       "step": 500
     },
     {
       "epoch": 0.5340314136125655,
-      "grad_norm": 0.6155954698882383,
+      "grad_norm": 0.6385420996025591,
       "learning_rate": 4.90825564889749e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -11.0625,
-      "logps/chosen": -624.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.1299,
+      "logits/chosen": -15.5,
+      "logits/rejected": -12.1875,
+      "logps/chosen": -640.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.1357,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -1.8125,
-      "rewards/margins": 5.5,
-      "rewards/rejected": -7.3125,
+      "rewards/chosen": -1.953125,
+      "rewards/margins": 5.6875,
+      "rewards/rejected": -7.625,
       "step": 510
     },
     {
       "epoch": 0.5445026178010471,
-      "grad_norm": 1.8958326659247546,
+      "grad_norm": 1.857683250892685,
       "learning_rate": 4.899899543831288e-05,
-      "logits/chosen": -14.9375,
-      "logits/rejected": -12.0,
-      "logps/chosen": -688.0,
-      "logps/rejected": -2960.0,
-      "loss": 0.2037,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -2.140625,
-      "rewards/margins": 5.34375,
-      "rewards/rejected": -7.5,
+      "logits/chosen": -17.25,
+      "logits/rejected": -12.875,
+      "logps/chosen": -656.0,
+      "logps/rejected": -2912.0,
+      "loss": 0.2118,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -1.8359375,
+      "rewards/margins": 5.15625,
+      "rewards/rejected": -7.0,
       "step": 520
     },
     {
       "epoch": 0.5549738219895288,
-      "grad_norm": 1.3822359805479507,
+      "grad_norm": 1.4407418413237034,
       "learning_rate": 4.891187051966765e-05,
-      "logits/chosen": -15.375,
-      "logits/rejected": -11.8125,
-      "logps/chosen": -572.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.1362,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -1.625,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -6.03125,
+      "logits/chosen": -17.0,
+      "logits/rejected": -12.625,
+      "logps/chosen": -552.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.134,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -1.40625,
+      "rewards/margins": 4.375,
+      "rewards/rejected": -5.78125,
       "step": 530
     },
     {
       "epoch": 0.5654450261780105,
-      "grad_norm": 0.8236894372265684,
+      "grad_norm": 0.8612126282798214,
       "learning_rate": 4.882119467115191e-05,
-      "logits/chosen": -15.25,
-      "logits/rejected": -12.0,
+      "logits/chosen": -17.5,
+      "logits/rejected": -13.0625,
       "logps/chosen": -688.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.1558,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -1.9921875,
-      "rewards/margins": 4.65625,
-      "rewards/rejected": -6.65625,
+      "logps/rejected": -2800.0,
+      "loss": 0.1621,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -1.953125,
+      "rewards/margins": 4.75,
+      "rewards/rejected": -6.6875,
       "step": 540
     },
     {
       "epoch": 0.5759162303664922,
-      "grad_norm": 1.8254673958497956,
+      "grad_norm": 2.2595048337087986,
       "learning_rate": 4.872698135819402e-05,
-      "logits/chosen": -14.9375,
-      "logits/rejected": -11.0625,
-      "logps/chosen": -580.0,
+      "logits/chosen": -17.5,
+      "logits/rejected": -12.625,
+      "logps/chosen": -588.0,
       "logps/rejected": -2880.0,
-      "loss": 0.1554,
+      "loss": 0.1552,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.734375,
-      "rewards/margins": 6.09375,
-      "rewards/rejected": -7.8125,
+      "rewards/chosen": -1.8046875,
+      "rewards/margins": 6.03125,
+      "rewards/rejected": -7.84375,
       "step": 550
     },
     {
       "epoch": 0.5863874345549738,
-      "grad_norm": 0.7193719838484375,
+      "grad_norm": 0.6324310663953722,
       "learning_rate": 4.862924457153831e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -608.0,
-      "logps/rejected": -2960.0,
-      "loss": 0.1212,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -1.8359375,
-      "rewards/margins": 5.59375,
-      "rewards/rejected": -7.4375,
+      "logits/chosen": -15.75,
+      "logits/rejected": -11.625,
+      "logps/chosen": -612.0,
+      "logps/rejected": -2992.0,
+      "loss": 0.13,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -1.875,
+      "rewards/margins": 5.90625,
+      "rewards/rejected": -7.78125,
       "step": 560
     },
     {
       "epoch": 0.5968586387434555,
-      "grad_norm": 1.018114281649203,
+      "grad_norm": 1.0688692346196789,
       "learning_rate": 4.8527998825167484e-05,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -672.0,
+      "logits/chosen": -15.625,
+      "logits/rejected": -11.3125,
+      "logps/chosen": -668.0,
       "logps/rejected": -3008.0,
-      "loss": 0.1167,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -2.109375,
-      "rewards/margins": 5.75,
-      "rewards/rejected": -7.875,
+      "loss": 0.1216,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -2.0625,
+      "rewards/margins": 5.78125,
+      "rewards/rejected": -7.84375,
       "step": 570
     },
     {
       "epoch": 0.6073298429319371,
-      "grad_norm": 1.5917670740171264,
+      "grad_norm": 1.432625766018211,
       "learning_rate": 4.842325915414728e-05,
-      "logits/chosen": -12.0625,
-      "logits/rejected": -9.625,
-      "logps/chosen": -652.0,
-      "logps/rejected": -2928.0,
-      "loss": 0.1567,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -1.8125,
-      "rewards/margins": 5.46875,
-      "rewards/rejected": -7.28125,
+      "logits/chosen": -15.375,
+      "logits/rejected": -11.375,
+      "logps/chosen": -676.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.1526,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -2.046875,
+      "rewards/margins": 5.34375,
+      "rewards/rejected": -7.375,
       "step": 580
     },
     {
       "epoch": 0.6178010471204188,
-      "grad_norm": 0.994837983138027,
+      "grad_norm": 1.0918627791745568,
       "learning_rate": 4.831504111239373e-05,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -10.0625,
+      "logits/chosen": -16.5,
+      "logits/rejected": -11.6875,
       "logps/chosen": -632.0,
-      "logps/rejected": -2976.0,
-      "loss": 0.1247,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -1.9375,
-      "rewards/margins": 5.28125,
-      "rewards/rejected": -7.21875,
+      "logps/rejected": -3008.0,
+      "loss": 0.1389,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -1.9453125,
+      "rewards/margins": 5.59375,
+      "rewards/rejected": -7.5625,
       "step": 590
     },
     {
       "epoch": 0.6282722513089005,
-      "grad_norm": 1.979701944672432,
+      "grad_norm": 1.9821585073925911,
       "learning_rate": 4.8203360770363425e-05,
-      "logits/chosen": -14.875,
-      "logits/rejected": -11.625,
-      "logps/chosen": -704.0,
-      "logps/rejected": -2880.0,
-      "loss": 0.1435,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -2.390625,
-      "rewards/margins": 6.53125,
-      "rewards/rejected": -8.9375,
+      "logits/chosen": -16.75,
+      "logits/rejected": -12.875,
+      "logps/chosen": -692.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.1507,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -2.265625,
+      "rewards/margins": 7.46875,
+      "rewards/rejected": -9.75,
       "step": 600
     },
     {
       "epoch": 0.6282722513089005,
-      "eval_logits/chosen": -15.6875,
-      "eval_logits/rejected": -15.625,
-      "eval_logps/chosen": -490.0,
-      "eval_logps/rejected": -462.0,
-      "eval_loss": 0.8511093854904175,
-      "eval_rewards/accuracies": 0.5,
-      "eval_rewards/chosen": -1.7109375,
-      "eval_rewards/margins": 0.03076171875,
-      "eval_rewards/rejected": -1.7421875,
-      "eval_runtime": 46.6343,
-      "eval_samples_per_second": 42.887,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -17.375,
+      "eval_logits/rejected": -17.25,
+      "eval_logps/chosen": -482.0,
+      "eval_logps/rejected": -452.0,
+      "eval_loss": 0.8541562557220459,
+      "eval_rewards/accuracies": 0.4765625,
+      "eval_rewards/chosen": -1.625,
+      "eval_rewards/margins": 0.00958251953125,
+      "eval_rewards/rejected": -1.6328125,
+      "eval_runtime": 46.5924,
+      "eval_samples_per_second": 42.925,
+      "eval_steps_per_second": 0.687,
       "step": 600
     },
     {
       "epoch": 0.6387434554973822,
-      "grad_norm": 2.1824077332605794,
+      "grad_norm": 2.3074769493789082,
       "learning_rate": 4.808823471266702e-05,
-      "logits/chosen": -14.4375,
-      "logits/rejected": -11.3125,
+      "logits/chosen": -16.375,
+      "logits/rejected": -12.25,
       "logps/chosen": -724.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.1763,
+      "logps/rejected": -2816.0,
+      "loss": 0.1671,
       "rewards/accuracies": 0.90625,
-      "rewards/chosen": -2.09375,
-      "rewards/margins": 5.8125,
-      "rewards/rejected": -7.90625,
+      "rewards/chosen": -2.109375,
+      "rewards/margins": 5.21875,
+      "rewards/rejected": -7.3125,
       "step": 610
     },
     {
       "epoch": 0.6492146596858639,
-      "grad_norm": 1.2315438748306793,
+      "grad_norm": 1.4088583582948173,
       "learning_rate": 4.796968003560639e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -596.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.1297,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -1.5859375,
-      "rewards/margins": 4.28125,
-      "rewards/rejected": -5.84375,
+      "logits/chosen": -14.875,
+      "logits/rejected": -11.0625,
+      "logps/chosen": -612.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.1228,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.7421875,
+      "rewards/margins": 4.40625,
+      "rewards/rejected": -6.125,
       "step": 620
     },
     {
       "epoch": 0.6596858638743456,
-      "grad_norm": 0.5749144136274057,
+      "grad_norm": 0.7335777682041571,
       "learning_rate": 4.7847714344635864e-05,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -592.0,
+      "logits/chosen": -15.0,
+      "logits/rejected": -11.5,
+      "logps/chosen": -596.0,
       "logps/rejected": -2960.0,
-      "loss": 0.148,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -1.5703125,
-      "rewards/margins": 6.09375,
-      "rewards/rejected": -7.65625,
+      "loss": 0.1567,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -1.625,
+      "rewards/margins": 6.0625,
+      "rewards/rejected": -7.6875,
       "step": 630
     },
     {
       "epoch": 0.6701570680628273,
-      "grad_norm": 0.8964568015584935,
+      "grad_norm": 1.0134096297152118,
       "learning_rate": 4.772235575174776e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -604.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.1514,
+      "logits/chosen": -14.75,
+      "logits/rejected": -11.0,
+      "logps/chosen": -612.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.1593,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -1.515625,
-      "rewards/margins": 5.0,
-      "rewards/rejected": -6.5,
+      "rewards/chosen": -1.5625,
+      "rewards/margins": 4.28125,
+      "rewards/rejected": -5.8125,
       "step": 640
     },
     {
       "epoch": 0.680628272251309,
-      "grad_norm": 0.7339962590693953,
+      "grad_norm": 0.7220166980784821,
       "learning_rate": 4.7593622872782754e-05,
-      "logits/chosen": -15.0,
-      "logits/rejected": -11.0625,
-      "logps/chosen": -536.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.1485,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -1.328125,
-      "rewards/margins": 5.34375,
-      "rewards/rejected": -6.65625,
+      "logits/chosen": -15.6875,
+      "logits/rejected": -11.875,
+      "logps/chosen": -564.0,
+      "logps/rejected": -2928.0,
+      "loss": 0.1423,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -1.625,
+      "rewards/margins": 5.96875,
+      "rewards/rejected": -7.59375,
       "step": 650
     },
     {
       "epoch": 0.6910994764397905,
-      "grad_norm": 0.7478148538177102,
+      "grad_norm": 0.8796306749870705,
       "learning_rate": 4.746153482466542e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -11.8125,
-      "logps/chosen": -564.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.1465,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -1.40625,
-      "rewards/margins": 5.46875,
-      "rewards/rejected": -6.875,
+      "logits/chosen": -16.0,
+      "logits/rejected": -11.875,
+      "logps/chosen": -596.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.1389,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -1.734375,
+      "rewards/margins": 5.625,
+      "rewards/rejected": -7.34375,
       "step": 660
     },
     {
       "epoch": 0.7015706806282722,
-      "grad_norm": 1.0545978843778105,
+      "grad_norm": 0.9102106852275065,
       "learning_rate": 4.732611122256537e-05,
-      "logits/chosen": -15.1875,
-      "logits/rejected": -12.3125,
-      "logps/chosen": -632.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.1335,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -1.9375,
-      "rewards/margins": 4.65625,
-      "rewards/rejected": -6.625,
+      "logits/chosen": -15.125,
+      "logits/rejected": -12.5,
+      "logps/chosen": -648.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.1181,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -2.125,
+      "rewards/margins": 4.84375,
+      "rewards/rejected": -6.96875,
       "step": 670
     },
     {
       "epoch": 0.7120418848167539,
-      "grad_norm": 2.17361778361498,
+      "grad_norm": 2.0847029246456703,
       "learning_rate": 4.718737217698435e-05,
-      "logits/chosen": -14.875,
-      "logits/rejected": -11.875,
+      "logits/chosen": -15.3125,
+      "logits/rejected": -12.1875,
       "logps/chosen": -696.0,
-      "logps/rejected": -2816.0,
-      "loss": 0.132,
+      "logps/rejected": -2896.0,
+      "loss": 0.1343,
       "rewards/accuracies": 0.9375,
-      "rewards/chosen": -2.15625,
-      "rewards/margins": 5.40625,
-      "rewards/rejected": -7.5625,
+      "rewards/chosen": -2.1875,
+      "rewards/margins": 6.0625,
+      "rewards/rejected": -8.25,
       "step": 680
     },
     {
       "epoch": 0.7225130890052356,
-      "grad_norm": 0.8547784490240291,
+      "grad_norm": 0.8785617975257317,
       "learning_rate": 4.704533829076982e-05,
-      "logits/chosen": -14.1875,
-      "logits/rejected": -11.5,
-      "logps/chosen": -664.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.1658,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -2.0625,
-      "rewards/margins": 6.125,
-      "rewards/rejected": -8.1875,
+      "logits/chosen": -14.5,
+      "logits/rejected": -11.3125,
+      "logps/chosen": -684.0,
+      "logps/rejected": -2976.0,
+      "loss": 0.158,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -2.265625,
+      "rewards/margins": 6.6875,
+      "rewards/rejected": -8.9375,
       "step": 690
     },
     {
       "epoch": 0.7329842931937173,
-      "grad_norm": 0.9289807794147451,
+      "grad_norm": 0.9961864806215508,
       "learning_rate": 4.6900030656055455e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -684.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.1255,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -716.0,
+      "logps/rejected": -2928.0,
+      "loss": 0.1278,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -1.8828125,
-      "rewards/margins": 4.84375,
-      "rewards/rejected": -6.71875,
+      "rewards/chosen": -2.234375,
+      "rewards/margins": 5.25,
+      "rewards/rejected": -7.5,
       "step": 700
     },
     {
       "epoch": 0.7329842931937173,
-      "eval_logits/chosen": -13.5625,
-      "eval_logits/rejected": -13.125,
-      "eval_logps/chosen": -502.0,
-      "eval_logps/rejected": -492.0,
-      "eval_loss": 0.8534374833106995,
-      "eval_rewards/accuracies": 0.513671875,
-      "eval_rewards/chosen": -1.84375,
-      "eval_rewards/margins": 0.1796875,
-      "eval_rewards/rejected": -2.03125,
-      "eval_runtime": 46.8422,
-      "eval_samples_per_second": 42.697,
-      "eval_steps_per_second": 0.683,
+      "eval_logits/chosen": -15.0,
+      "eval_logits/rejected": -14.8125,
+      "eval_logps/chosen": -496.0,
+      "eval_logps/rejected": -484.0,
+      "eval_loss": 0.8274375200271606,
+      "eval_rewards/accuracies": 0.498046875,
+      "eval_rewards/chosen": -1.7890625,
+      "eval_rewards/margins": 0.1591796875,
+      "eval_rewards/rejected": -1.9453125,
+      "eval_runtime": 46.5891,
+      "eval_samples_per_second": 42.928,
+      "eval_steps_per_second": 0.687,
       "step": 700
     },
     {
       "epoch": 0.743455497382199,
-      "grad_norm": 1.2303340065121269,
+      "grad_norm": 0.9934178347091964,
       "learning_rate": 4.6751470851128906e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -656.0,
-      "logps/rejected": -3184.0,
-      "loss": 0.1001,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -2.15625,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -11.25,
+      "logps/chosen": -648.0,
+      "logps/rejected": -3168.0,
+      "loss": 0.102,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -2.078125,
       "rewards/margins": 7.6875,
-      "rewards/rejected": -9.875,
+      "rewards/rejected": -9.75,
       "step": 710
     },
     {
       "epoch": 0.7539267015706806,
-      "grad_norm": 0.9622826276528129,
+      "grad_norm": 0.9267330821459564,
       "learning_rate": 4.6599680937227454e-05,
-      "logits/chosen": -13.0625,
-      "logits/rejected": -10.375,
-      "logps/chosen": -636.0,
-      "logps/rejected": -2848.0,
-      "loss": 0.158,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -2.1875,
-      "rewards/margins": 5.90625,
-      "rewards/rejected": -8.125,
+      "logits/chosen": -14.25,
+      "logits/rejected": -11.25,
+      "logps/chosen": -632.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.1623,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -2.15625,
+      "rewards/margins": 6.0625,
+      "rewards/rejected": -8.25,
       "step": 720
     },
     {
       "epoch": 0.7643979057591623,
-      "grad_norm": 1.37822733965358,
+      "grad_norm": 1.6392110167330947,
       "learning_rate": 4.644468345526183e-05,
-      "logits/chosen": -12.5,
-      "logits/rejected": -9.9375,
+      "logits/chosen": -14.8125,
+      "logits/rejected": -11.5,
       "logps/chosen": -664.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.1627,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -2.15625,
-      "rewards/margins": 5.5,
-      "rewards/rejected": -7.6875,
+      "logps/rejected": -2832.0,
+      "loss": 0.1678,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -2.140625,
+      "rewards/margins": 5.28125,
+      "rewards/rejected": -7.40625,
       "step": 730
     },
     {
       "epoch": 0.774869109947644,
-      "grad_norm": 1.0983096387223583,
+      "grad_norm": 1.2649248368680912,
       "learning_rate": 4.628650142246894e-05,
-      "logits/chosen": -12.3125,
-      "logits/rejected": -10.375,
-      "logps/chosen": -612.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.148,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -1.859375,
-      "rewards/margins": 5.0625,
-      "rewards/rejected": -6.9375,
+      "logits/chosen": -16.0,
+      "logits/rejected": -11.9375,
+      "logps/chosen": -652.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.1488,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -2.28125,
+      "rewards/margins": 5.3125,
+      "rewards/rejected": -7.59375,
       "step": 740
     },
     {
       "epoch": 0.7853403141361257,
-      "grad_norm": 1.9668590726339368,
+      "grad_norm": 2.4215321417410967,
       "learning_rate": 4.6125158328993765e-05,
-      "logits/chosen": -12.4375,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -768.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.1633,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -2.515625,
-      "rewards/margins": 4.65625,
-      "rewards/rejected": -7.1875,
+      "logits/chosen": -16.5,
+      "logits/rejected": -12.8125,
+      "logps/chosen": -812.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.1686,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 4.9375,
+      "rewards/rejected": -7.90625,
       "step": 750
     },
     {
       "epoch": 0.7958115183246073,
-      "grad_norm": 1.1683830322642343,
+      "grad_norm": 1.4193665310445067,
       "learning_rate": 4.5960678134401064e-05,
-      "logits/chosen": -12.0625,
-      "logits/rejected": -10.0,
-      "logps/chosen": -680.0,
-      "logps/rejected": -3136.0,
-      "loss": 0.1278,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -2.25,
-      "rewards/margins": 6.75,
-      "rewards/rejected": -9.0,
+      "logits/chosen": -16.125,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.1351,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -2.6875,
+      "rewards/margins": 7.03125,
+      "rewards/rejected": -9.6875,
       "step": 760
     },
     {
       "epoch": 0.806282722513089,
-      "grad_norm": 1.032006314354136,
+      "grad_norm": 1.0701159213366997,
       "learning_rate": 4.579308526411734e-05,
-      "logits/chosen": -11.375,
-      "logits/rejected": -9.625,
-      "logps/chosen": -588.0,
+      "logits/chosen": -15.4375,
+      "logits/rejected": -11.625,
+      "logps/chosen": -600.0,
       "logps/rejected": -2832.0,
-      "loss": 0.1559,
+      "loss": 0.1557,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -1.84375,
-      "rewards/margins": 5.28125,
-      "rewards/rejected": -7.125,
+      "rewards/chosen": -1.984375,
+      "rewards/margins": 5.1875,
+      "rewards/rejected": -7.15625,
       "step": 770
     },
     {
       "epoch": 0.8167539267015707,
-      "grad_norm": 1.3437384964915426,
+      "grad_norm": 1.1920847261597434,
       "learning_rate": 4.562240460580367e-05,
-      "logits/chosen": -11.4375,
-      "logits/rejected": -9.75,
-      "logps/chosen": -692.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.1536,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -2.09375,
-      "rewards/margins": 4.34375,
-      "rewards/rejected": -6.4375,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -11.6875,
+      "logps/chosen": -700.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.1522,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -2.1875,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -6.71875,
       "step": 780
     },
     {
       "epoch": 0.8272251308900523,
-      "grad_norm": 0.8735836619606299,
+      "grad_norm": 1.0074120196248715,
       "learning_rate": 4.544866150565985e-05,
-      "logits/chosen": -12.125,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -596.0,
-      "logps/rejected": -2944.0,
-      "loss": 0.1346,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -1.765625,
-      "rewards/margins": 5.0625,
-      "rewards/rejected": -6.84375,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -11.4375,
+      "logps/chosen": -612.0,
+      "logps/rejected": -2928.0,
+      "loss": 0.1383,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -1.890625,
+      "rewards/margins": 4.84375,
+      "rewards/rejected": -6.75,
       "step": 790
     },
     {
       "epoch": 0.837696335078534,
-      "grad_norm": 0.6435600678567915,
+      "grad_norm": 0.6399144182072822,
       "learning_rate": 4.527188176466051e-05,
-      "logits/chosen": -11.8125,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -584.0,
+      "logits/chosen": -15.25,
+      "logits/rejected": -11.9375,
+      "logps/chosen": -616.0,
       "logps/rejected": -2864.0,
-      "loss": 0.1142,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -1.7578125,
-      "rewards/margins": 5.4375,
+      "loss": 0.1303,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.078125,
+      "rewards/margins": 5.125,
       "rewards/rejected": -7.1875,
       "step": 800
     },
     {
       "epoch": 0.837696335078534,
-      "eval_logits/chosen": -12.375,
-      "eval_logits/rejected": -11.875,
-      "eval_logps/chosen": -478.0,
-      "eval_logps/rejected": -452.0,
-      "eval_loss": 0.8559062480926514,
-      "eval_rewards/accuracies": 0.4921875,
-      "eval_rewards/chosen": -1.59375,
-      "eval_rewards/margins": 0.047119140625,
-      "eval_rewards/rejected": -1.640625,
-      "eval_runtime": 46.6154,
-      "eval_samples_per_second": 42.904,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -16.5,
+      "eval_logits/rejected": -16.5,
+      "eval_logps/chosen": -496.0,
+      "eval_logps/rejected": -468.0,
+      "eval_loss": 0.8349062204360962,
+      "eval_rewards/accuracies": 0.51953125,
+      "eval_rewards/chosen": -1.7734375,
+      "eval_rewards/margins": 0.0272216796875,
+      "eval_rewards/rejected": -1.796875,
+      "eval_runtime": 46.5875,
+      "eval_samples_per_second": 42.93,
+      "eval_steps_per_second": 0.687,
       "step": 800
     },
     {
       "epoch": 0.8481675392670157,
-      "grad_norm": 1.3268621315118674,
+      "grad_norm": 1.2660776484959255,
       "learning_rate": 4.509209163472357e-05,
-      "logits/chosen": -12.625,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -676.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.1462,
+      "logits/chosen": -15.8125,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -684.0,
+      "logps/rejected": -2912.0,
+      "loss": 0.1561,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -1.96875,
-      "rewards/margins": 5.625,
-      "rewards/rejected": -7.59375,
+      "rewards/chosen": -2.0625,
+      "rewards/margins": 5.65625,
+      "rewards/rejected": -7.6875,
       "step": 810
     },
     {
       "epoch": 0.8586387434554974,
-      "grad_norm": 1.1138973929256597,
+      "grad_norm": 1.1476274348546545,
       "learning_rate": 4.4909317814811896e-05,
-      "logits/chosen": -13.375,
-      "logits/rejected": -10.875,
-      "logps/chosen": -608.0,
-      "logps/rejected": -3008.0,
-      "loss": 0.131,
+      "logits/chosen": -16.75,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -620.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.1338,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -1.8359375,
-      "rewards/margins": 6.21875,
-      "rewards/rejected": -8.0625,
+      "rewards/chosen": -1.9296875,
+      "rewards/margins": 5.59375,
+      "rewards/rejected": -7.53125,
       "step": 820
     },
     {
       "epoch": 0.8691099476439791,
-      "grad_norm": 1.0913026710199536,
+      "grad_norm": 1.2421645274270807,
       "learning_rate": 4.472358744696845e-05,
-      "logits/chosen": -12.875,
-      "logits/rejected": -10.8125,
-      "logps/chosen": -640.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.1549,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -2.03125,
-      "rewards/margins": 4.78125,
-      "rewards/rejected": -6.8125,
+      "logits/chosen": -16.125,
+      "logits/rejected": -12.125,
+      "logps/chosen": -644.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.1555,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -2.0625,
+      "rewards/margins": 4.90625,
+      "rewards/rejected": -6.96875,
       "step": 830
     },
     {
       "epoch": 0.8795811518324608,
-      "grad_norm": 1.0701708659411528,
+      "grad_norm": 1.1219748886086405,
       "learning_rate": 4.4534928112285645e-05,
-      "logits/chosen": -13.375,
-      "logits/rejected": -10.75,
+      "logits/chosen": -15.5,
+      "logits/rejected": -11.5625,
       "logps/chosen": -640.0,
       "logps/rejected": -2752.0,
-      "loss": 0.1353,
+      "loss": 0.1266,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -1.890625,
-      "rewards/margins": 4.71875,
-      "rewards/rejected": -6.59375,
+      "rewards/chosen": -1.8828125,
+      "rewards/margins": 4.78125,
+      "rewards/rejected": -6.65625,
       "step": 840
     },
     {
       "epoch": 0.8900523560209425,
-      "grad_norm": 1.0935244825784507,
+      "grad_norm": 0.9453841107915908,
       "learning_rate": 4.43433678268096e-05,
-      "logits/chosen": -13.375,
-      "logits/rejected": -10.8125,
+      "logits/chosen": -15.375,
+      "logits/rejected": -11.3125,
       "logps/chosen": -656.0,
-      "logps/rejected": -2880.0,
-      "loss": 0.1409,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -2.078125,
-      "rewards/margins": 5.375,
-      "rewards/rejected": -7.4375,
+      "logps/rejected": -2848.0,
+      "loss": 0.137,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.046875,
+      "rewards/margins": 5.09375,
+      "rewards/rejected": -7.15625,
       "step": 850
     },
     {
       "epoch": 0.900523560209424,
-      "grad_norm": 0.8458258396301148,
+      "grad_norm": 0.8038844001228589,
       "learning_rate": 4.414893503737968e-05,
-      "logits/chosen": -12.6875,
-      "logits/rejected": -10.125,
+      "logits/chosen": -15.625,
+      "logits/rejected": -11.4375,
       "logps/chosen": -612.0,
       "logps/rejected": -2912.0,
-      "loss": 0.1224,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -2.03125,
-      "rewards/margins": 5.4375,
-      "rewards/rejected": -7.4375,
+      "loss": 0.1239,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -2.046875,
+      "rewards/margins": 5.53125,
+      "rewards/rejected": -7.5625,
       "step": 860
     },
     {
       "epoch": 0.9109947643979057,
-      "grad_norm": 1.223796863993444,
+      "grad_norm": 1.3229394548968458,
       "learning_rate": 4.395165861740414e-05,
-      "logits/chosen": -12.8125,
-      "logits/rejected": -10.625,
-      "logps/chosen": -632.0,
-      "logps/rejected": -3040.0,
-      "loss": 0.1037,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -2.09375,
-      "rewards/margins": 6.8125,
-      "rewards/rejected": -8.875,
+      "logits/chosen": -15.5625,
+      "logits/rejected": -11.75,
+      "logps/chosen": -648.0,
+      "logps/rejected": -3024.0,
+      "loss": 0.1032,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -2.265625,
+      "rewards/margins": 6.40625,
+      "rewards/rejected": -8.6875,
       "step": 870
     },
     {
       "epoch": 0.9214659685863874,
-      "grad_norm": 1.6368484057945356,
+      "grad_norm": 1.281692643724604,
       "learning_rate": 4.375156786257241e-05,
-      "logits/chosen": -12.625,
-      "logits/rejected": -10.5,
-      "logps/chosen": -700.0,
+      "logits/chosen": -15.375,
+      "logits/rejected": -11.625,
+      "logps/chosen": -724.0,
       "logps/rejected": -3184.0,
-      "loss": 0.12,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -2.203125,
-      "rewards/margins": 7.28125,
+      "loss": 0.1221,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -2.46875,
+      "rewards/margins": 7.03125,
       "rewards/rejected": -9.5,
       "step": 880
     },
     {
       "epoch": 0.9319371727748691,
-      "grad_norm": 1.7310358088279334,
+      "grad_norm": 1.30817913750753,
       "learning_rate": 4.354869248650466e-05,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -10.875,
-      "logps/chosen": -728.0,
-      "logps/rejected": -2944.0,
-      "loss": 0.1523,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -2.5,
-      "rewards/margins": 5.8125,
-      "rewards/rejected": -8.3125,
+      "logits/chosen": -16.125,
+      "logits/rejected": -12.3125,
+      "logps/chosen": -760.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.1329,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -2.796875,
+      "rewards/margins": 5.6875,
+      "rewards/rejected": -8.5,
       "step": 890
     },
     {
       "epoch": 0.9424083769633508,
-      "grad_norm": 0.7170892614715749,
+      "grad_norm": 0.6477821488867574,
       "learning_rate": 4.334306261633931e-05,
-      "logits/chosen": -14.5,
-      "logits/rejected": -10.9375,
-      "logps/chosen": -644.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.1596,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -2.234375,
-      "rewards/margins": 4.96875,
-      "rewards/rejected": -7.21875,
+      "logits/chosen": -16.125,
+      "logits/rejected": -12.0625,
+      "logps/chosen": -692.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.1614,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 5.03125,
+      "rewards/rejected": -7.75,
       "step": 900
     },
     {
       "epoch": 0.9424083769633508,
-      "eval_logits/chosen": -15.0625,
-      "eval_logits/rejected": -14.875,
-      "eval_logps/chosen": -524.0,
-      "eval_logps/rejected": -498.0,
-      "eval_loss": 0.8357968926429749,
-      "eval_rewards/accuracies": 0.4921875,
-      "eval_rewards/chosen": -2.046875,
-      "eval_rewards/margins": 0.051025390625,
-      "eval_rewards/rejected": -2.09375,
-      "eval_runtime": 46.6162,
-      "eval_samples_per_second": 42.904,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -16.375,
+      "eval_logits/rejected": -16.375,
+      "eval_logps/chosen": -548.0,
+      "eval_logps/rejected": -540.0,
+      "eval_loss": 0.8078437447547913,
+      "eval_rewards/accuracies": 0.533203125,
+      "eval_rewards/chosen": -2.296875,
+      "eval_rewards/margins": 0.19921875,
+      "eval_rewards/rejected": -2.5,
+      "eval_runtime": 46.5933,
+      "eval_samples_per_second": 42.925,
+      "eval_steps_per_second": 0.687,
       "step": 900
     },
     {
       "epoch": 0.9528795811518325,
-      "grad_norm": 1.0608482317231847,
+      "grad_norm": 1.3245146646747508,
       "learning_rate": 4.313470878825911e-05,
-      "logits/chosen": -14.625,
-      "logits/rejected": -11.625,
-      "logps/chosen": -668.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.1358,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -2.109375,
-      "rewards/margins": 4.78125,
-      "rewards/rejected": -6.875,
+      "logits/chosen": -16.125,
+      "logits/rejected": -12.6875,
+      "logps/chosen": -700.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.155,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.453125,
+      "rewards/margins": 5.0,
+      "rewards/rejected": -7.46875,
       "step": 910
     },
     {
       "epoch": 0.9633507853403142,
-      "grad_norm": 1.8533029509980772,
+      "grad_norm": 1.9331726995426524,
       "learning_rate": 4.292366194295652e-05,
-      "logits/chosen": -13.8125,
-      "logits/rejected": -10.625,
-      "logps/chosen": -696.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.1337,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -2.015625,
-      "rewards/margins": 5.0625,
-      "rewards/rejected": -7.09375,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -11.875,
+      "logps/chosen": -716.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.1372,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.203125,
+      "rewards/margins": 5.46875,
+      "rewards/rejected": -7.6875,
       "step": 920
     },
     {
       "epoch": 0.9738219895287958,
-      "grad_norm": 1.5361849089317439,
+      "grad_norm": 1.6270218216222347,
       "learning_rate": 4.2709953421038964e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.375,
-      "logps/chosen": -576.0,
-      "logps/rejected": -3008.0,
-      "loss": 0.1116,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -1.8203125,
-      "rewards/margins": 5.8125,
-      "rewards/rejected": -7.625,
+      "logits/chosen": -15.3125,
+      "logits/rejected": -11.5625,
+      "logps/chosen": -592.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.12,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -1.984375,
+      "rewards/margins": 5.90625,
+      "rewards/rejected": -7.90625,
       "step": 930
     },
     {
       "epoch": 0.9842931937172775,
-      "grad_norm": 0.8184270360275369,
+      "grad_norm": 0.8635103773393156,
       "learning_rate": 4.249361495837478e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.375,
+      "logits/chosen": -15.5,
+      "logits/rejected": -11.6875,
       "logps/chosen": -632.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.1171,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -2.296875,
-      "rewards/margins": 6.59375,
-      "rewards/rejected": -8.875,
+      "logps/rejected": -3168.0,
+      "loss": 0.1172,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.3125,
+      "rewards/margins": 6.65625,
+      "rewards/rejected": -9.0,
       "step": 940
     },
     {
       "epoch": 0.9947643979057592,
-      "grad_norm": 0.5850180924962325,
+      "grad_norm": 0.6434898170007491,
       "learning_rate": 4.227467868138035e-05,
-      "logits/chosen": -13.25,
-      "logits/rejected": -10.6875,
-      "logps/chosen": -672.0,
+      "logits/chosen": -15.25,
+      "logits/rejected": -11.875,
+      "logps/chosen": -668.0,
       "logps/rejected": -3040.0,
-      "loss": 0.1354,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -2.125,
-      "rewards/margins": 5.65625,
+      "loss": 0.1401,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -2.078125,
+      "rewards/margins": 5.75,
       "rewards/rejected": -7.8125,
       "step": 950
     },
     {
       "epoch": 1.0052356020942408,
-      "grad_norm": 0.9212669460316035,
+      "grad_norm": 0.6946682957121975,
       "learning_rate": 4.205317710224938e-05,
-      "logits/chosen": -12.4375,
-      "logits/rejected": -10.125,
-      "logps/chosen": -568.0,
-      "logps/rejected": -2800.0,
-      "loss": 0.0981,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -1.5546875,
-      "rewards/margins": 5.21875,
-      "rewards/rejected": -6.78125,
+      "logits/chosen": -14.625,
+      "logits/rejected": -11.1875,
+      "logps/chosen": -576.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0974,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.6328125,
+      "rewards/margins": 5.375,
+      "rewards/rejected": -7.0,
       "step": 960
     },
     {
       "epoch": 1.0157068062827226,
-      "grad_norm": 0.6838241096137042,
+      "grad_norm": 0.8252499886984264,
       "learning_rate": 4.1829143114124733e-05,
-      "logits/chosen": -12.375,
-      "logits/rejected": -10.125,
-      "logps/chosen": -620.0,
-      "logps/rejected": -3024.0,
-      "loss": 0.0565,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.7890625,
+      "logits/chosen": -14.125,
+      "logits/rejected": -11.0,
+      "logps/chosen": -644.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.0547,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.03125,
       "rewards/margins": 6.0625,
-      "rewards/rejected": -7.84375,
+      "rewards/rejected": -8.0625,
       "step": 970
     },
     {
       "epoch": 1.0261780104712042,
-      "grad_norm": 0.17737987455409518,
+      "grad_norm": 0.2457453018022123,
       "learning_rate": 4.160260998621387e-05,
-      "logits/chosen": -12.125,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -624.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.041,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -1.875,
-      "rewards/margins": 7.09375,
-      "rewards/rejected": -8.9375,
+      "logits/chosen": -14.8125,
+      "logits/rejected": -11.3125,
+      "logps/chosen": -616.0,
+      "logps/rejected": -3104.0,
+      "loss": 0.0403,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -1.8203125,
+      "rewards/margins": 6.65625,
+      "rewards/rejected": -8.5,
       "step": 980
     },
     {
       "epoch": 1.036649214659686,
-      "grad_norm": 0.2550137332359494,
+      "grad_norm": 0.2005384885752845,
       "learning_rate": 4.137361135884824e-05,
-      "logits/chosen": -12.5,
-      "logits/rejected": -10.4375,
-      "logps/chosen": -616.0,
-      "logps/rejected": -3088.0,
-      "loss": 0.0626,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -1.8671875,
-      "rewards/margins": 7.34375,
-      "rewards/rejected": -9.1875,
+      "logits/chosen": -15.875,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -628.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.0479,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -1.953125,
+      "rewards/margins": 6.71875,
+      "rewards/rejected": -8.6875,
       "step": 990
     },
     {
       "epoch": 1.0471204188481675,
-      "grad_norm": 0.1161136630664759,
+      "grad_norm": 0.1160231806000512,
       "learning_rate": 4.114218123848777e-05,
-      "logits/chosen": -12.8125,
-      "logits/rejected": -10.25,
-      "logps/chosen": -628.0,
-      "logps/rejected": -3136.0,
-      "loss": 0.0231,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -1.9296875,
-      "rewards/margins": 6.53125,
-      "rewards/rejected": -8.5,
+      "logits/chosen": -16.25,
+      "logits/rejected": -12.3125,
+      "logps/chosen": -656.0,
+      "logps/rejected": -3168.0,
+      "loss": 0.0199,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.21875,
+      "rewards/margins": 6.5625,
+      "rewards/rejected": -8.8125,
       "step": 1000
     },
     {
       "epoch": 1.0471204188481675,
-      "eval_logits/chosen": -12.5,
-      "eval_logits/rejected": -12.0625,
-      "eval_logps/chosen": -506.0,
-      "eval_logps/rejected": -492.0,
-      "eval_loss": 0.8379687666893005,
-      "eval_rewards/accuracies": 0.5078125,
-      "eval_rewards/chosen": -1.8828125,
-      "eval_rewards/margins": 0.140625,
-      "eval_rewards/rejected": -2.03125,
-      "eval_runtime": 46.599,
-      "eval_samples_per_second": 42.919,
+      "eval_logits/chosen": -15.875,
+      "eval_logits/rejected": -15.4375,
+      "eval_logps/chosen": -544.0,
+      "eval_logps/rejected": -528.0,
+      "eval_loss": 0.8232656121253967,
+      "eval_rewards/accuracies": 0.486328125,
+      "eval_rewards/chosen": -2.265625,
+      "eval_rewards/margins": 0.1279296875,
+      "eval_rewards/rejected": -2.390625,
+      "eval_runtime": 46.5881,
+      "eval_samples_per_second": 42.929,
       "eval_steps_per_second": 0.687,
       "step": 1000
     },
     {
       "epoch": 1.057591623036649,
-      "grad_norm": 0.6095935446682996,
+      "grad_norm": 0.9238279034650745,
       "learning_rate": 4.0908353992670824e-05,
-      "logits/chosen": -12.5,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -604.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0274,
+      "logits/chosen": -15.875,
+      "logits/rejected": -12.9375,
+      "logps/chosen": -664.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0347,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -1.7265625,
-      "rewards/margins": 7.96875,
-      "rewards/rejected": -9.6875,
+      "rewards/chosen": -2.3125,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -10.625,
       "step": 1010
     },
     {
       "epoch": 1.068062827225131,
-      "grad_norm": 1.2134087825889897,
+      "grad_norm": 1.2108687370403999,
       "learning_rate": 4.067216434491059e-05,
-      "logits/chosen": -12.6875,
-      "logits/rejected": -10.375,
-      "logps/chosen": -640.0,
-      "logps/rejected": -3120.0,
-      "loss": 0.0548,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.9921875,
-      "rewards/margins": 7.1875,
-      "rewards/rejected": -9.1875,
+      "logits/chosen": -16.125,
+      "logits/rejected": -12.625,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0471,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.40625,
+      "rewards/margins": 7.5625,
+      "rewards/rejected": -10.0,
       "step": 1020
     },
     {
       "epoch": 1.0785340314136125,
-      "grad_norm": 0.4360694049816956,
+      "grad_norm": 0.30398444211824593,
       "learning_rate": 4.043364736953866e-05,
-      "logits/chosen": -12.25,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -688.0,
-      "logps/rejected": -3008.0,
-      "loss": 0.022,
+      "logits/chosen": -15.875,
+      "logits/rejected": -12.0625,
+      "logps/chosen": -684.0,
+      "logps/rejected": -2992.0,
+      "loss": 0.0233,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.1875,
-      "rewards/margins": 7.34375,
-      "rewards/rejected": -9.5,
+      "rewards/chosen": -2.15625,
+      "rewards/margins": 7.25,
+      "rewards/rejected": -9.375,
       "step": 1030
     },
     {
       "epoch": 1.0890052356020943,
-      "grad_norm": 0.8980156145363332,
+      "grad_norm": 0.7986767564426122,
       "learning_rate": 4.019283848649643e-05,
-      "logits/chosen": -12.9375,
-      "logits/rejected": -10.625,
+      "logits/chosen": -17.0,
+      "logits/rejected": -13.0625,
       "logps/chosen": -696.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0351,
+      "logps/rejected": -3216.0,
+      "loss": 0.035,
       "rewards/accuracies": 0.987500011920929,
       "rewards/chosen": -2.375,
-      "rewards/margins": 7.75,
-      "rewards/rejected": -10.125,
+      "rewards/margins": 7.96875,
+      "rewards/rejected": -10.375,
       "step": 1040
     },
     {
       "epoch": 1.0994764397905759,
-      "grad_norm": 0.9939934691456076,
+      "grad_norm": 0.902228012435995,
       "learning_rate": 3.9949773456075215e-05,
-      "logits/chosen": -12.875,
-      "logits/rejected": -10.3125,
+      "logits/chosen": -16.625,
+      "logits/rejected": -12.875,
       "logps/chosen": -764.0,
-      "logps/rejected": -3120.0,
-      "loss": 0.0218,
-      "rewards/accuracies": 0.9937499761581421,
+      "logps/rejected": -3184.0,
+      "loss": 0.0285,
+      "rewards/accuracies": 0.981249988079071,
       "rewards/chosen": -2.34375,
-      "rewards/margins": 7.1875,
-      "rewards/rejected": -9.5,
+      "rewards/margins": 7.8125,
+      "rewards/rejected": -10.1875,
       "step": 1050
     },
     {
       "epoch": 1.1099476439790577,
-      "grad_norm": 0.6896214407859734,
+      "grad_norm": 1.1714954583655073,
       "learning_rate": 3.970448837360585e-05,
-      "logits/chosen": -12.625,
-      "logits/rejected": -10.625,
-      "logps/chosen": -728.0,
-      "logps/rejected": -3168.0,
-      "loss": 0.0232,
+      "logits/chosen": -16.125,
+      "logits/rejected": -12.75,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3184.0,
+      "loss": 0.0297,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.296875,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -10.5,
+      "rewards/chosen": -2.203125,
+      "rewards/margins": 8.375,
+      "rewards/rejected": -10.5625,
       "step": 1060
     },
     {
       "epoch": 1.1204188481675392,
-      "grad_norm": 1.3521974326991437,
+      "grad_norm": 1.0329891295255025,
       "learning_rate": 3.945701966409846e-05,
-      "logits/chosen": -11.9375,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -744.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.0346,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.734375,
-      "rewards/margins": 8.0625,
-      "rewards/rejected": -10.75,
+      "logits/chosen": -15.8125,
+      "logits/rejected": -12.125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3184.0,
+      "loss": 0.0258,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.34375,
+      "rewards/margins": 7.96875,
+      "rewards/rejected": -10.3125,
       "step": 1070
     },
     {
       "epoch": 1.130890052356021,
-      "grad_norm": 0.18880758895126787,
+      "grad_norm": 0.15507833490314601,
       "learning_rate": 3.9207404076833374e-05,
-      "logits/chosen": -12.0,
-      "logits/rejected": -9.75,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.0297,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -12.125,
+      "logps/chosen": -628.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0296,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.796875,
-      "rewards/margins": 7.40625,
-      "rewards/rejected": -10.1875,
+      "rewards/chosen": -2.25,
+      "rewards/margins": 7.59375,
+      "rewards/rejected": -9.8125,
       "step": 1080
     },
     {
       "epoch": 1.1413612565445026,
-      "grad_norm": 0.6157827170820976,
+      "grad_norm": 0.5857948566665748,
       "learning_rate": 3.895567867990379e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -10.625,
-      "logps/chosen": -660.0,
-      "logps/rejected": -3168.0,
-      "loss": 0.0192,
+      "logits/chosen": -17.0,
+      "logits/rejected": -12.6875,
+      "logps/chosen": -628.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.0209,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.375,
-      "rewards/margins": 7.21875,
-      "rewards/rejected": -9.625,
+      "rewards/chosen": -2.0625,
+      "rewards/margins": 7.34375,
+      "rewards/rejected": -9.375,
       "step": 1090
     },
     {
       "epoch": 1.1518324607329844,
-      "grad_norm": 0.8362615091972094,
+      "grad_norm": 1.3593198392642105,
       "learning_rate": 3.870188085471114e-05,
-      "logits/chosen": -12.5625,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -752.0,
-      "logps/rejected": -3216.0,
-      "loss": 0.0293,
+      "logits/chosen": -16.25,
+      "logits/rejected": -12.5625,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3184.0,
+      "loss": 0.0348,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.390625,
-      "rewards/margins": 7.9375,
-      "rewards/rejected": -10.3125,
+      "rewards/chosen": -2.1875,
+      "rewards/margins": 7.78125,
+      "rewards/rejected": -10.0,
       "step": 1100
     },
     {
       "epoch": 1.1518324607329844,
-      "eval_logits/chosen": -12.875,
-      "eval_logits/rejected": -12.4375,
-      "eval_logps/chosen": -544.0,
-      "eval_logps/rejected": -528.0,
-      "eval_loss": 0.8317187428474426,
-      "eval_rewards/accuracies": 0.548828125,
-      "eval_rewards/chosen": -2.265625,
-      "eval_rewards/margins": 0.1328125,
-      "eval_rewards/rejected": -2.390625,
-      "eval_runtime": 46.599,
-      "eval_samples_per_second": 42.919,
+      "eval_logits/chosen": -17.125,
+      "eval_logits/rejected": -17.0,
+      "eval_logps/chosen": -524.0,
+      "eval_logps/rejected": -504.0,
+      "eval_loss": 0.8451562523841858,
+      "eval_rewards/accuracies": 0.50390625,
+      "eval_rewards/chosen": -2.046875,
+      "eval_rewards/margins": 0.11865234375,
+      "eval_rewards/rejected": -2.15625,
+      "eval_runtime": 46.6009,
+      "eval_samples_per_second": 42.918,
       "eval_steps_per_second": 0.687,
       "step": 1100
     },
     {
       "epoch": 1.162303664921466,
-      "grad_norm": 0.09931936518209031,
+      "grad_norm": 0.537944913415695,
       "learning_rate": 3.844604829041395e-05,
-      "logits/chosen": -12.875,
-      "logits/rejected": -10.75,
-      "logps/chosen": -688.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.026,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.421875,
-      "rewards/margins": 8.25,
-      "rewards/rejected": -10.6875,
+      "logits/chosen": -17.125,
+      "logits/rejected": -13.375,
+      "logps/chosen": -668.0,
+      "logps/rejected": -3072.0,
+      "loss": 0.0356,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.234375,
+      "rewards/margins": 7.6875,
+      "rewards/rejected": -9.875,
       "step": 1110
     },
     {
       "epoch": 1.1727748691099475,
-      "grad_norm": 0.18941580523761176,
+      "grad_norm": 0.22857409243781537,
       "learning_rate": 3.818821897833095e-05,
-      "logits/chosen": -13.0625,
-      "logits/rejected": -10.375,
-      "logps/chosen": -768.0,
-      "logps/rejected": -3216.0,
-      "loss": 0.029,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.625,
-      "rewards/margins": 7.59375,
-      "rewards/rejected": -10.1875,
+      "logits/chosen": -17.0,
+      "logits/rejected": -12.9375,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3104.0,
+      "loss": 0.0268,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.328125,
+      "rewards/margins": 6.78125,
+      "rewards/rejected": -9.125,
       "step": 1120
     },
     {
       "epoch": 1.1832460732984293,
-      "grad_norm": 0.531097510461554,
+      "grad_norm": 0.39686055716544205,
       "learning_rate": 3.792843120629935e-05,
-      "logits/chosen": -12.5,
-      "logits/rejected": -10.375,
-      "logps/chosen": -600.0,
+      "logits/chosen": -16.875,
+      "logits/rejected": -13.0,
+      "logps/chosen": -616.0,
       "logps/rejected": -3072.0,
-      "loss": 0.0324,
+      "loss": 0.0321,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.125,
-      "rewards/margins": 7.625,
-      "rewards/rejected": -9.75,
+      "rewards/chosen": -2.28125,
+      "rewards/margins": 7.4375,
+      "rewards/rejected": -9.6875,
       "step": 1130
     },
     {
       "epoch": 1.193717277486911,
-      "grad_norm": 0.12176198767356863,
+      "grad_norm": 0.06700605967417687,
       "learning_rate": 3.766672355298907e-05,
-      "logits/chosen": -12.375,
-      "logits/rejected": -10.125,
-      "logps/chosen": -588.0,
-      "logps/rejected": -3120.0,
-      "loss": 0.0484,
+      "logits/chosen": -16.5,
+      "logits/rejected": -12.875,
+      "logps/chosen": -612.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.0544,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -1.890625,
-      "rewards/margins": 7.375,
-      "rewards/rejected": -9.25,
+      "rewards/chosen": -2.109375,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -10.3125,
       "step": 1140
     },
     {
       "epoch": 1.2041884816753927,
-      "grad_norm": 0.3992045723599849,
+      "grad_norm": 0.3382984798459006,
       "learning_rate": 3.740313488217373e-05,
-      "logits/chosen": -12.375,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -676.0,
-      "logps/rejected": -2960.0,
-      "loss": 0.0301,
+      "logits/chosen": -16.25,
+      "logits/rejected": -12.6875,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3072.0,
+      "loss": 0.0277,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.03125,
-      "rewards/margins": 6.4375,
-      "rewards/rejected": -8.5,
+      "rewards/chosen": -2.390625,
+      "rewards/margins": 7.09375,
+      "rewards/rejected": -9.5,
       "step": 1150
     },
     {
       "epoch": 1.2146596858638743,
-      "grad_norm": 0.19131973731463828,
+      "grad_norm": 0.18286782123955309,
       "learning_rate": 3.7137704336959464e-05,
-      "logits/chosen": -12.8125,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -676.0,
-      "logps/rejected": -3040.0,
-      "loss": 0.0167,
+      "logits/chosen": -16.0,
+      "logits/rejected": -12.8125,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3104.0,
+      "loss": 0.0171,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.078125,
-      "rewards/margins": 6.75,
-      "rewards/rejected": -8.8125,
+      "rewards/chosen": -2.421875,
+      "rewards/margins": 7.0625,
+      "rewards/rejected": -9.5,
       "step": 1160
     },
     {
       "epoch": 1.225130890052356,
-      "grad_norm": 0.11470142938828308,
+      "grad_norm": 0.17254871363358448,
       "learning_rate": 3.687047133397201e-05,
-      "logits/chosen": -12.4375,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -676.0,
-      "logps/rejected": -3088.0,
-      "loss": 0.0234,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.5,
-      "rewards/margins": 7.59375,
-      "rewards/rejected": -10.125,
+      "logits/chosen": -15.5,
+      "logits/rejected": -12.625,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3120.0,
+      "loss": 0.031,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.625,
+      "rewards/margins": 7.71875,
+      "rewards/rejected": -10.375,
       "step": 1170
     },
     {
       "epoch": 1.2356020942408377,
-      "grad_norm": 0.08832261340319174,
+      "grad_norm": 0.12298230999745197,
       "learning_rate": 3.6601475557503415e-05,
-      "logits/chosen": -12.5,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -708.0,
-      "logps/rejected": -3312.0,
-      "loss": 0.0239,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.515625,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -10.6875,
+      "logits/chosen": -15.5,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.0329,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 8.125,
+      "rewards/rejected": -11.0,
       "step": 1180
     },
     {
       "epoch": 1.2460732984293195,
-      "grad_norm": 0.3449797385252149,
+      "grad_norm": 0.1439991372386876,
       "learning_rate": 3.633075695361882e-05,
-      "logits/chosen": -12.9375,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3072.0,
-      "loss": 0.0346,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.359375,
-      "rewards/margins": 7.5,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3088.0,
+      "loss": 0.0326,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.671875,
+      "rewards/margins": 7.21875,
       "rewards/rejected": -9.875,
       "step": 1190
     },
     {
       "epoch": 1.256544502617801,
-      "grad_norm": 0.06528359676207203,
+      "grad_norm": 0.11298691628285298,
       "learning_rate": 3.6058355724224477e-05,
-      "logits/chosen": -12.5625,
-      "logits/rejected": -9.875,
-      "logps/chosen": -632.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0171,
+      "logits/chosen": -15.0,
+      "logits/rejected": -12.375,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0186,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.234375,
-      "rewards/margins": 7.875,
-      "rewards/rejected": -10.125,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 7.1875,
+      "rewards/rejected": -10.0,
       "step": 1200
     },
     {
       "epoch": 1.256544502617801,
-      "eval_logits/chosen": -12.625,
-      "eval_logits/rejected": -12.1875,
-      "eval_logps/chosen": -556.0,
-      "eval_logps/rejected": -540.0,
-      "eval_loss": 0.8362187743186951,
-      "eval_rewards/accuracies": 0.533203125,
-      "eval_rewards/chosen": -2.375,
-      "eval_rewards/margins": 0.13671875,
-      "eval_rewards/rejected": -2.515625,
-      "eval_runtime": 46.5864,
-      "eval_samples_per_second": 42.931,
-      "eval_steps_per_second": 0.687,
-      "step": 1200
-    },
+      "eval_logits/chosen": -15.0625,
+      "eval_logits/rejected": -14.75,
+      "eval_logps/chosen": -612.0,
+      "eval_logps/rejected": -592.0,
+      "eval_loss": 0.8787500262260437,
+      "eval_rewards/accuracies": 0.509765625,
+      "eval_rewards/chosen": -2.921875,
+      "eval_rewards/margins": 0.107421875,
+      "eval_rewards/rejected": -3.03125,
+      "eval_runtime": 46.5902,
+      "eval_samples_per_second": 42.927,
+      "eval_steps_per_second": 0.687,
+      "step": 1200
+    },
     {
       "epoch": 1.2670157068062826,
-      "grad_norm": 0.13409407316838193,
+      "grad_norm": 0.15836352966572764,
       "learning_rate": 3.5784312321097744e-05,
-      "logits/chosen": -12.6875,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -744.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.0197,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.53125,
-      "rewards/margins": 7.5625,
-      "rewards/rejected": -10.0625,
+      "logits/chosen": -15.625,
+      "logits/rejected": -12.75,
+      "logps/chosen": -788.0,
+      "logps/rejected": -3328.0,
+      "loss": 0.0239,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 7.90625,
+      "rewards/rejected": -10.875,
       "step": 1210
     },
     {
       "epoch": 1.2774869109947644,
-      "grad_norm": 0.2086770903320873,
+      "grad_norm": 0.16103379347883745,
       "learning_rate": 3.550866743987994e-05,
-      "logits/chosen": -12.3125,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -764.0,
-      "logps/rejected": -3184.0,
-      "loss": 0.0256,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.890625,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -11.0625,
+      "logits/chosen": -15.5625,
+      "logits/rejected": -13.0,
+      "logps/chosen": -812.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0173,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -12.125,
       "step": 1220
     },
     {
       "epoch": 1.2879581151832462,
-      "grad_norm": 4.179359658448476,
+      "grad_norm": 0.7559289682710013,
       "learning_rate": 3.5231462014033024e-05,
-      "logits/chosen": -12.375,
-      "logits/rejected": -10.125,
-      "logps/chosen": -812.0,
-      "logps/rejected": -3104.0,
-      "loss": 0.0614,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -3.125,
-      "rewards/margins": 7.875,
-      "rewards/rejected": -11.0,
+      "logits/chosen": -15.8125,
+      "logits/rejected": -13.1875,
+      "logps/chosen": -848.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.0474,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.5,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -12.3125,
       "step": 1230
     },
     {
       "epoch": 1.2984293193717278,
-      "grad_norm": 0.3993866482932888,
+      "grad_norm": 0.21935268402345665,
       "learning_rate": 3.495273720876095e-05,
-      "logits/chosen": -12.1875,
-      "logits/rejected": -9.375,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.019,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.75,
-      "rewards/margins": 7.96875,
-      "rewards/rejected": -10.6875,
+      "logits/chosen": -14.75,
+      "logits/rejected": -12.0,
+      "logps/chosen": -768.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.031,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 7.78125,
+      "rewards/rejected": -11.125,
       "step": 1240
     },
     {
       "epoch": 1.3089005235602094,
-      "grad_norm": 1.3227606871810966,
+      "grad_norm": 1.3055420949319623,
       "learning_rate": 3.4672534414896616e-05,
-      "logits/chosen": -12.75,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -768.0,
+      "logits/chosen": -15.25,
+      "logits/rejected": -12.75,
+      "logps/chosen": -796.0,
       "logps/rejected": -3168.0,
-      "loss": 0.0182,
+      "loss": 0.019,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.84375,
-      "rewards/margins": 8.1875,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 7.875,
       "rewards/rejected": -11.0,
       "step": 1250
     },
     {
       "epoch": 1.3193717277486912,
-      "grad_norm": 0.09648882989672453,
+      "grad_norm": 0.19067392109508402,
       "learning_rate": 3.439089524275526e-05,
-      "logits/chosen": -12.625,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -716.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.0256,
+      "logits/chosen": -15.625,
+      "logits/rejected": -13.0,
+      "logps/chosen": -748.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0259,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.65625,
-      "rewards/margins": 8.5,
-      "rewards/rejected": -11.1875,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 7.96875,
+      "rewards/rejected": -10.9375,
       "step": 1260
     },
     {
       "epoch": 1.3298429319371727,
-      "grad_norm": 0.8207102260091197,
+      "grad_norm": 0.838253335726383,
       "learning_rate": 3.410786151595533e-05,
-      "logits/chosen": -12.6875,
-      "logits/rejected": -10.0,
-      "logps/chosen": -748.0,
-      "logps/rejected": -3168.0,
-      "loss": 0.0352,
+      "logits/chosen": -15.3125,
+      "logits/rejected": -12.3125,
+      "logps/chosen": -800.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0374,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.765625,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -10.9375,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 7.875,
+      "rewards/rejected": -11.1875,
       "step": 1270
     },
     {
       "epoch": 1.3403141361256545,
-      "grad_norm": 0.49631322875828615,
+      "grad_norm": 0.45058217299939207,
       "learning_rate": 3.382347526520765e-05,
-      "logits/chosen": -12.0625,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -728.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.0183,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.71875,
-      "rewards/margins": 7.8125,
-      "rewards/rejected": -10.5625,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -12.0,
+      "logps/chosen": -764.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0208,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 7.96875,
+      "rewards/rejected": -11.0625,
       "step": 1280
     },
     {
       "epoch": 1.350785340314136,
-      "grad_norm": 0.4694601062335547,
+      "grad_norm": 0.7503178140917749,
       "learning_rate": 3.353777872207381e-05,
-      "logits/chosen": -12.125,
-      "logits/rejected": -9.75,
-      "logps/chosen": -792.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0274,
+      "logits/chosen": -14.375,
+      "logits/rejected": -12.0,
+      "logps/chosen": -804.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0276,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.859375,
-      "rewards/margins": 7.90625,
-      "rewards/rejected": -10.75,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 7.5625,
+      "rewards/rejected": -10.5625,
       "step": 1290
     },
     {
       "epoch": 1.3612565445026177,
-      "grad_norm": 0.4906816240186705,
+      "grad_norm": 0.3934320814876777,
       "learning_rate": 3.325081431269475e-05,
-      "logits/chosen": -12.0625,
-      "logits/rejected": -9.875,
-      "logps/chosen": -700.0,
-      "logps/rejected": -3280.0,
-      "loss": 0.0174,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.546875,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -10.6875,
+      "logits/chosen": -14.4375,
+      "logits/rejected": -12.0,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0277,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.484375,
+      "rewards/margins": 7.40625,
+      "rewards/rejected": -9.875,
       "step": 1300
     },
     {
       "epoch": 1.3612565445026177,
-      "eval_logits/chosen": -12.25,
-      "eval_logits/rejected": -11.75,
-      "eval_logps/chosen": -612.0,
-      "eval_logps/rejected": -600.0,
-      "eval_loss": 0.8684218525886536,
-      "eval_rewards/accuracies": 0.51953125,
-      "eval_rewards/chosen": -2.9375,
-      "eval_rewards/margins": 0.166015625,
-      "eval_rewards/rejected": -3.09375,
-      "eval_runtime": 46.6049,
-      "eval_samples_per_second": 42.914,
+      "eval_logits/chosen": -14.5,
+      "eval_logits/rejected": -14.25,
+      "eval_logps/chosen": -600.0,
+      "eval_logps/rejected": -576.0,
+      "eval_loss": 0.8304218649864197,
+      "eval_rewards/accuracies": 0.513671875,
+      "eval_rewards/chosen": -2.796875,
+      "eval_rewards/margins": 0.0927734375,
+      "eval_rewards/rejected": -2.890625,
+      "eval_runtime": 46.5924,
+      "eval_samples_per_second": 42.925,
       "eval_steps_per_second": 0.687,
       "step": 1300
     },
     {
       "epoch": 1.3717277486910995,
-      "grad_norm": 0.5658517944851932,
+      "grad_norm": 0.1450116292670592,
       "learning_rate": 3.296262465149046e-05,
-      "logits/chosen": -12.1875,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -776.0,
-      "logps/rejected": -3216.0,
-      "loss": 0.0334,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.8125,
-      "rewards/margins": 8.3125,
-      "rewards/rejected": -11.125,
+      "logits/chosen": -14.4375,
+      "logits/rejected": -11.9375,
+      "logps/chosen": -760.0,
+      "logps/rejected": -3120.0,
+      "loss": 0.0236,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.65625,
+      "rewards/margins": 7.53125,
+      "rewards/rejected": -10.1875,
       "step": 1310
     },
     {
       "epoch": 1.3821989528795813,
-      "grad_norm": 0.37353285828126176,
+      "grad_norm": 0.1692858634638375,
       "learning_rate": 3.2673252534831687e-05,
-      "logits/chosen": -11.9375,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -708.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0359,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -12.0625,
+      "logps/chosen": -720.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0306,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.78125,
-      "rewards/margins": 7.65625,
-      "rewards/rejected": -10.4375,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 7.8125,
+      "rewards/rejected": -10.6875,
       "step": 1320
     },
     {
       "epoch": 1.3926701570680629,
-      "grad_norm": 0.17740342198350953,
+      "grad_norm": 0.48850670672294366,
       "learning_rate": 3.2382740934684696e-05,
-      "logits/chosen": -11.6875,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -716.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0213,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.984375,
-      "rewards/margins": 8.125,
-      "rewards/rejected": -11.0625,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -12.1875,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0241,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.140625,
+      "rewards/margins": 7.875,
+      "rewards/rejected": -11.0,
       "step": 1330
     },
     {
       "epoch": 1.4031413612565444,
-      "grad_norm": 1.2932760724936367,
+      "grad_norm": 0.6326802557797059,
       "learning_rate": 3.209113299222982e-05,
-      "logits/chosen": -12.6875,
-      "logits/rejected": -10.4375,
-      "logps/chosen": -780.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0406,
+      "logits/chosen": -15.0625,
+      "logits/rejected": -12.5,
+      "logps/chosen": -788.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0315,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -3.046875,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -11.875,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 8.125,
+      "rewards/rejected": -11.25,
       "step": 1340
     },
     {
       "epoch": 1.4136125654450262,
-      "grad_norm": 0.677287622358092,
+      "grad_norm": 0.9230993101330381,
       "learning_rate": 3.179847201145505e-05,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -10.25,
-      "logps/chosen": -720.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0371,
+      "logits/chosen": -15.5,
+      "logits/rejected": -12.3125,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0356,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.625,
-      "rewards/margins": 7.84375,
-      "rewards/rejected": -10.5,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 7.96875,
+      "rewards/rejected": -10.875,
       "step": 1350
     },
     {
       "epoch": 1.4240837696335078,
-      "grad_norm": 0.16190054322062475,
+      "grad_norm": 0.14927661366159814,
       "learning_rate": 3.150480145272528e-05,
-      "logits/chosen": -12.625,
-      "logits/rejected": -10.125,
-      "logps/chosen": -760.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0159,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -12.1875,
+      "logps/chosen": -788.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.0144,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.609375,
-      "rewards/margins": 6.65625,
-      "rewards/rejected": -9.25,
+      "rewards/chosen": -2.875,
+      "rewards/margins": 6.78125,
+      "rewards/rejected": -9.6875,
       "step": 1360
     },
     {
       "epoch": 1.4345549738219896,
-      "grad_norm": 0.2308589290080698,
+      "grad_norm": 0.5816031584106096,
       "learning_rate": 3.121016492632848e-05,
-      "logits/chosen": -12.6875,
-      "logits/rejected": -10.0,
-      "logps/chosen": -712.0,
+      "logits/chosen": -14.8125,
+      "logits/rejected": -12.1875,
+      "logps/chosen": -740.0,
       "logps/rejected": -3136.0,
-      "loss": 0.0173,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.28125,
-      "rewards/margins": 7.71875,
+      "loss": 0.023,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.546875,
+      "rewards/margins": 7.4375,
       "rewards/rejected": -10.0,
       "step": 1370
     },
     {
       "epoch": 1.4450261780104712,
-      "grad_norm": 0.22275028531554653,
+      "grad_norm": 0.3382648730097446,
       "learning_rate": 3.091460618599951e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -9.75,
-      "logps/chosen": -688.0,
+      "logits/chosen": -15.3125,
+      "logits/rejected": -11.875,
+      "logps/chosen": -708.0,
       "logps/rejected": -3360.0,
-      "loss": 0.0199,
+      "loss": 0.0225,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.5,
-      "rewards/margins": 8.75,
-      "rewards/rejected": -11.25,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 8.4375,
+      "rewards/rejected": -11.1875,
       "step": 1380
     },
     {
       "epoch": 1.4554973821989527,
-      "grad_norm": 0.4557083934840219,
+      "grad_norm": 0.5839459270999949,
       "learning_rate": 3.061816912242265e-05,
-      "logits/chosen": -12.625,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -708.0,
+      "logits/chosen": -14.875,
+      "logits/rejected": -11.9375,
+      "logps/chosen": -728.0,
       "logps/rejected": -3456.0,
-      "loss": 0.0196,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.4375,
-      "rewards/margins": 9.5,
+      "loss": 0.0197,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.6875,
+      "rewards/margins": 9.25,
       "rewards/rejected": -11.9375,
       "step": 1390
     },
     {
       "epoch": 1.4659685863874345,
-      "grad_norm": 1.0326552236672093,
+      "grad_norm": 0.682121594552336,
       "learning_rate": 3.0320897756713785e-05,
-      "logits/chosen": -12.4375,
-      "logits/rejected": -10.125,
-      "logps/chosen": -676.0,
-      "logps/rejected": -3328.0,
-      "loss": 0.0191,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -12.375,
+      "logps/chosen": -696.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0212,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.5,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -11.4375,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 8.375,
+      "rewards/rejected": -11.125,
       "step": 1400
     },
     {
       "epoch": 1.4659685863874345,
-      "eval_logits/chosen": -12.0,
-      "eval_logits/rejected": -11.5625,
-      "eval_logps/chosen": -588.0,
-      "eval_logps/rejected": -572.0,
-      "eval_loss": 0.8702499866485596,
-      "eval_rewards/accuracies": 0.515625,
-      "eval_rewards/chosen": -2.6875,
-      "eval_rewards/margins": 0.150390625,
-      "eval_rewards/rejected": -2.84375,
-      "eval_runtime": 46.5939,
-      "eval_samples_per_second": 42.924,
+      "eval_logits/chosen": -14.4375,
+      "eval_logits/rejected": -14.25,
+      "eval_logps/chosen": -600.0,
+      "eval_logps/rejected": -580.0,
+      "eval_loss": 0.8989999890327454,
+      "eval_rewards/accuracies": 0.5,
+      "eval_rewards/chosen": -2.796875,
+      "eval_rewards/margins": 0.10986328125,
+      "eval_rewards/rejected": -2.90625,
+      "eval_runtime": 46.596,
+      "eval_samples_per_second": 42.922,
       "eval_steps_per_second": 0.687,
       "step": 1400
     },
     {
       "epoch": 1.4764397905759163,
-      "grad_norm": 0.6840147426394543,
+      "grad_norm": 0.6770117039766996,
       "learning_rate": 3.002283623388332e-05,
-      "logits/chosen": -11.5625,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -724.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0255,
+      "logits/chosen": -14.5,
+      "logits/rejected": -11.5625,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3264.0,
+      "loss": 0.0196,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.59375,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -11.4375,
+      "rewards/chosen": -2.734375,
+      "rewards/margins": 8.4375,
+      "rewards/rejected": -11.1875,
       "step": 1410
     },
     {
       "epoch": 1.486910994764398,
-      "grad_norm": 0.09885297804480472,
+      "grad_norm": 0.08203713754271427,
       "learning_rate": 2.9724028816280507e-05,
-      "logits/chosen": -11.6875,
-      "logits/rejected": -9.625,
-      "logps/chosen": -704.0,
-      "logps/rejected": -3392.0,
-      "loss": 0.0139,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.5625,
-      "rewards/margins": 9.4375,
-      "rewards/rejected": -12.0,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -12.0,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3424.0,
+      "loss": 0.0178,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.859375,
+      "rewards/margins": 9.3125,
+      "rewards/rejected": -12.1875,
       "step": 1420
     },
     {
       "epoch": 1.4973821989528795,
-      "grad_norm": 0.5447267084878693,
+      "grad_norm": 0.5633947514858227,
       "learning_rate": 2.9424519877020524e-05,
-      "logits/chosen": -12.6875,
-      "logits/rejected": -9.875,
-      "logps/chosen": -664.0,
-      "logps/rejected": -3216.0,
-      "loss": 0.0351,
+      "logits/chosen": -15.3125,
+      "logits/rejected": -12.125,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0336,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.515625,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -11.375,
+      "rewards/chosen": -2.734375,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -11.25,
       "step": 1430
     },
     {
       "epoch": 1.5078534031413613,
-      "grad_norm": 1.7612852542771738,
+      "grad_norm": 1.4559278105607552,
       "learning_rate": 2.912435389339504e-05,
-      "logits/chosen": -12.0,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -636.0,
-      "logps/rejected": -3360.0,
-      "loss": 0.033,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.546875,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -11.75,
+      "logits/chosen": -15.125,
+      "logits/rejected": -12.1875,
+      "logps/chosen": -652.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0313,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.703125,
+      "rewards/margins": 8.4375,
+      "rewards/rejected": -11.125,
       "step": 1440
     },
     {
       "epoch": 1.518324607329843,
-      "grad_norm": 0.21797196230306862,
+      "grad_norm": 0.20527634038282278,
       "learning_rate": 2.8823575440267226e-05,
-      "logits/chosen": -12.5,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.019,
+      "logits/chosen": -15.5,
+      "logits/rejected": -12.25,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0167,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.359375,
-      "rewards/margins": 8.3125,
-      "rewards/rejected": -10.625,
+      "rewards/chosen": -2.625,
+      "rewards/margins": 7.90625,
+      "rewards/rejected": -10.5625,
       "step": 1450
     },
     {
       "epoch": 1.5287958115183247,
-      "grad_norm": 0.1398275390233516,
+      "grad_norm": 0.1542059716546675,
       "learning_rate": 2.8522229183452465e-05,
-      "logits/chosen": -12.4375,
-      "logits/rejected": -9.5,
-      "logps/chosen": -676.0,
-      "logps/rejected": -3376.0,
-      "loss": 0.0184,
+      "logits/chosen": -15.4375,
+      "logits/rejected": -11.6875,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3392.0,
+      "loss": 0.017,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.359375,
-      "rewards/margins": 8.4375,
-      "rewards/rejected": -10.75,
+      "rewards/chosen": -2.609375,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -10.875,
       "step": 1460
     },
     {
       "epoch": 1.5392670157068062,
-      "grad_norm": 0.07480740506899823,
+      "grad_norm": 0.3043884663797174,
       "learning_rate": 2.8220359873085373e-05,
-      "logits/chosen": -12.0,
-      "logits/rejected": -9.75,
-      "logps/chosen": -752.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0397,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -12.125,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0371,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.78125,
-      "rewards/margins": 7.90625,
-      "rewards/rejected": -10.6875,
+      "rewards/chosen": -3.0625,
+      "rewards/margins": 8.125,
+      "rewards/rejected": -11.125,
       "step": 1470
     },
     {
       "epoch": 1.5497382198952878,
-      "grad_norm": 0.242912155955285,
+      "grad_norm": 0.268350331146206,
       "learning_rate": 2.7918012336974388e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -9.125,
-      "logps/chosen": -816.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0257,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -11.5,
+      "logps/chosen": -856.0,
+      "logps/rejected": -3328.0,
+      "loss": 0.0281,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.75,
-      "rewards/margins": 7.875,
-      "rewards/rejected": -10.625,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 7.75,
+      "rewards/rejected": -10.9375,
       "step": 1480
     },
     {
       "epoch": 1.5602094240837696,
-      "grad_norm": 0.7594533324298329,
+      "grad_norm": 0.5240126747864844,
       "learning_rate": 2.7615231473944815e-05,
-      "logits/chosen": -11.5625,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3280.0,
-      "loss": 0.0282,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.484375,
-      "rewards/margins": 7.96875,
-      "rewards/rejected": -10.4375,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -11.3125,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.0311,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 8.125,
+      "rewards/rejected": -11.125,
       "step": 1490
     },
     {
       "epoch": 1.5706806282722514,
-      "grad_norm": 1.6055276260680513,
+      "grad_norm": 1.305965808275632,
       "learning_rate": 2.7312062247171327e-05,
-      "logits/chosen": -11.6875,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -736.0,
-      "logps/rejected": -3312.0,
-      "loss": 0.036,
+      "logits/chosen": -13.25,
+      "logits/rejected": -11.1875,
+      "logps/chosen": -792.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0333,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.734375,
-      "rewards/margins": 7.4375,
-      "rewards/rejected": -10.125,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 7.71875,
+      "rewards/rejected": -11.0625,
       "step": 1500
     },
     {
       "epoch": 1.5706806282722514,
-      "eval_logits/chosen": -11.8125,
-      "eval_logits/rejected": -11.3125,
-      "eval_logps/chosen": -576.0,
-      "eval_logps/rejected": -560.0,
-      "eval_loss": 0.9228906035423279,
-      "eval_rewards/accuracies": 0.509765625,
-      "eval_rewards/chosen": -2.578125,
-      "eval_rewards/margins": 0.134765625,
-      "eval_rewards/rejected": -2.71875,
-      "eval_runtime": 46.5961,
-      "eval_samples_per_second": 42.922,
+      "eval_logits/chosen": -12.625,
+      "eval_logits/rejected": -12.1875,
+      "eval_logps/chosen": -640.0,
+      "eval_logps/rejected": -620.0,
+      "eval_loss": 0.9111406207084656,
+      "eval_rewards/accuracies": 0.521484375,
+      "eval_rewards/chosen": -3.203125,
+      "eval_rewards/margins": 0.09814453125,
+      "eval_rewards/rejected": -3.296875,
+      "eval_runtime": 46.6037,
+      "eval_samples_per_second": 42.915,
       "eval_steps_per_second": 0.687,
       "step": 1500
     },
     {
       "epoch": 1.581151832460733,
-      "grad_norm": 0.44937458978019706,
+      "grad_norm": 1.2317858890956381,
       "learning_rate": 2.7008549677500876e-05,
-      "logits/chosen": -12.4375,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -756.0,
-      "logps/rejected": -3280.0,
-      "loss": 0.0131,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.84375,
-      "rewards/margins": 7.625,
-      "rewards/rejected": -10.5,
+      "logits/chosen": -14.125,
+      "logits/rejected": -11.0,
+      "logps/chosen": -812.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.0211,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 7.59375,
+      "rewards/rejected": -11.0,
       "step": 1510
     },
     {
       "epoch": 1.5916230366492146,
-      "grad_norm": 0.13225735658196988,
+      "grad_norm": 0.2922125183150869,
       "learning_rate": 2.670473883676709e-05,
-      "logits/chosen": -12.25,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0257,
+      "logits/chosen": -13.5,
+      "logits/rejected": -11.4375,
+      "logps/chosen": -772.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0251,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.734375,
-      "rewards/margins": 8.3125,
-      "rewards/rejected": -11.0,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 7.78125,
+      "rewards/rejected": -11.125,
       "step": 1520
     },
     {
       "epoch": 1.6020942408376964,
-      "grad_norm": 0.3470754497933429,
+      "grad_norm": 0.17791350356699734,
       "learning_rate": 2.6400674841097074e-05,
-      "logits/chosen": -12.0,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -732.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0232,
+      "logits/chosen": -13.375,
+      "logits/rejected": -11.1875,
+      "logps/chosen": -788.0,
+      "logps/rejected": -3360.0,
+      "loss": 0.0243,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.75,
-      "rewards/margins": 8.25,
-      "rewards/rejected": -11.0,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -11.5,
       "step": 1530
     },
     {
       "epoch": 1.6125654450261782,
-      "grad_norm": 0.31904725186480803,
+      "grad_norm": 0.14156715699962813,
       "learning_rate": 2.609640284421158e-05,
-      "logits/chosen": -11.75,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -732.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.0205,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.96875,
-      "rewards/margins": 7.6875,
-      "rewards/rejected": -10.625,
+      "logits/chosen": -13.5,
+      "logits/rejected": -11.3125,
+      "logps/chosen": -776.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.0215,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -11.6875,
       "step": 1540
     },
     {
       "epoch": 1.6230366492146597,
-      "grad_norm": 0.3882924340654773,
+      "grad_norm": 0.3362210410623208,
       "learning_rate": 2.5791968030719726e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -9.375,
-      "logps/chosen": -768.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0295,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.171875,
-      "rewards/margins": 8.0,
-      "rewards/rejected": -11.1875,
+      "logits/chosen": -13.875,
+      "logits/rejected": -11.4375,
+      "logps/chosen": -776.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.0229,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.25,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -12.25,
       "step": 1550
     },
     {
       "epoch": 1.6335078534031413,
-      "grad_norm": 0.1078678678208724,
+      "grad_norm": 0.12146646291861969,
       "learning_rate": 2.548741560940902e-05,
-      "logits/chosen": -12.0625,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -856.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0299,
+      "logits/chosen": -14.0,
+      "logits/rejected": -11.75,
+      "logps/chosen": -888.0,
+      "logps/rejected": -3328.0,
+      "loss": 0.0324,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.25,
-      "rewards/margins": 8.3125,
-      "rewards/rejected": -11.5625,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -12.25,
       "step": 1560
     },
     {
       "epoch": 1.6439790575916229,
-      "grad_norm": 1.6353071104324566,
+      "grad_norm": 1.352388630631102,
       "learning_rate": 2.5182790806531787e-05,
-      "logits/chosen": -12.3125,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -768.0,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -11.0,
+      "logps/chosen": -776.0,
       "logps/rejected": -3392.0,
-      "loss": 0.0166,
+      "loss": 0.0136,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.203125,
+      "rewards/chosen": -3.28125,
       "rewards/margins": 8.5,
-      "rewards/rejected": -11.6875,
+      "rewards/rejected": -11.8125,
       "step": 1570
     },
     {
       "epoch": 1.6544502617801047,
-      "grad_norm": 0.5004458601183499,
+      "grad_norm": 0.6419426883018355,
       "learning_rate": 2.4878138859089074e-05,
-      "logits/chosen": -11.5625,
-      "logits/rejected": -9.125,
-      "logps/chosen": -724.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.0272,
+      "logits/chosen": -12.8125,
+      "logits/rejected": -10.875,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0296,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.984375,
-      "rewards/margins": 8.25,
-      "rewards/rejected": -11.25,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 8.5625,
+      "rewards/rejected": -11.75,
       "step": 1580
     },
     {
       "epoch": 1.6649214659685865,
-      "grad_norm": 0.2045399227444437,
+      "grad_norm": 0.34294359111612727,
       "learning_rate": 2.4573505008112924e-05,
       "logits/chosen": -11.9375,
-      "logits/rejected": -8.625,
-      "logps/chosen": -764.0,
-      "logps/rejected": -3408.0,
-      "loss": 0.0183,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -3.03125,
-      "rewards/margins": 8.4375,
-      "rewards/rejected": -11.4375,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -820.0,
+      "logps/rejected": -3488.0,
+      "loss": 0.026,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.609375,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -12.25,
       "step": 1590
     },
     {
       "epoch": 1.675392670157068,
-      "grad_norm": 0.44927815319300485,
+      "grad_norm": 0.3626272247771784,
       "learning_rate": 2.4268934491948032e-05,
-      "logits/chosen": -12.1875,
-      "logits/rejected": -8.9375,
-      "logps/chosen": -756.0,
-      "logps/rejected": -3088.0,
-      "loss": 0.018,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.6875,
-      "rewards/margins": 8.0,
-      "rewards/rejected": -10.6875,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -800.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.0163,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -11.375,
       "step": 1600
     },
     {
       "epoch": 1.675392670157068,
-      "eval_logits/chosen": -12.875,
-      "eval_logits/rejected": -12.4375,
-      "eval_logps/chosen": -600.0,
-      "eval_logps/rejected": -588.0,
-      "eval_loss": 0.9153749942779541,
-      "eval_rewards/accuracies": 0.505859375,
-      "eval_rewards/chosen": -2.828125,
-      "eval_rewards/margins": 0.15234375,
-      "eval_rewards/rejected": -2.984375,
-      "eval_runtime": 46.5864,
-      "eval_samples_per_second": 42.931,
+      "eval_logits/chosen": -11.5,
+      "eval_logits/rejected": -11.0625,
+      "eval_logps/chosen": -640.0,
+      "eval_logps/rejected": -620.0,
+      "eval_loss": 0.9214531183242798,
+      "eval_rewards/accuracies": 0.494140625,
+      "eval_rewards/chosen": -3.21875,
+      "eval_rewards/margins": 0.1103515625,
+      "eval_rewards/rejected": -3.328125,
+      "eval_runtime": 46.6059,
+      "eval_samples_per_second": 42.913,
       "eval_steps_per_second": 0.687,
       "step": 1600
     },
     {
       "epoch": 1.6858638743455496,
-      "grad_norm": 0.4102090929634209,
+      "grad_norm": 0.6670959237114126,
       "learning_rate": 2.396447253953385e-05,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -724.0,
-      "logps/rejected": -3328.0,
-      "loss": 0.0129,
+      "logits/chosen": -12.5,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -760.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0138,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.875,
-      "rewards/margins": 8.6875,
-      "rewards/rejected": -11.5625,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -12.375,
       "step": 1610
     },
     {
       "epoch": 1.6963350785340314,
-      "grad_norm": 0.03945129756196896,
+      "grad_norm": 0.038636999190201346,
       "learning_rate": 2.3660164363687998e-05,
-      "logits/chosen": -13.375,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -800.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.0575,
+      "logits/chosen": -12.8125,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -832.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0221,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.84375,
-      "rewards/margins": 8.6875,
-      "rewards/rejected": -11.5,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -12.1875,
       "step": 1620
     },
     {
       "epoch": 1.7068062827225132,
-      "grad_norm": 0.9283966646541102,
+      "grad_norm": 1.1629865250114928,
       "learning_rate": 2.33560551543922e-05,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -10.125,
-      "logps/chosen": -672.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0496,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.703125,
+      "logits/chosen": -12.4375,
+      "logits/rejected": -10.875,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0491,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.96875,
       "rewards/margins": 8.8125,
-      "rewards/rejected": -11.5,
+      "rewards/rejected": -11.8125,
       "step": 1630
     },
     {
       "epoch": 1.7172774869109948,
-      "grad_norm": 1.3535331917454998,
+      "grad_norm": 1.2074175179942512,
       "learning_rate": 2.3052190072081493e-05,
-      "logits/chosen": -13.25,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -772.0,
+      "logits/chosen": -13.0,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -776.0,
       "logps/rejected": -3264.0,
-      "loss": 0.0401,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -2.921875,
-      "rewards/margins": 7.9375,
-      "rewards/rejected": -10.875,
+      "loss": 0.033,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.9375,
+      "rewards/margins": 8.0,
+      "rewards/rejected": -10.9375,
       "step": 1640
     },
     {
       "epoch": 1.7277486910994764,
-      "grad_norm": 1.3273003659694136,
+      "grad_norm": 1.4014335094165082,
       "learning_rate": 2.2748614240937864e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -748.0,
-      "logps/rejected": -3072.0,
-      "loss": 0.0446,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.6875,
-      "rewards/margins": 7.53125,
-      "rewards/rejected": -10.1875,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -10.875,
+      "logps/chosen": -768.0,
+      "logps/rejected": -3120.0,
+      "loss": 0.0461,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 7.8125,
+      "rewards/rejected": -10.75,
       "step": 1650
     },
     {
       "epoch": 1.738219895287958,
-      "grad_norm": 0.23019286697417038,
+      "grad_norm": 0.1784985381016052,
       "learning_rate": 2.2445372742189334e-05,
-      "logits/chosen": -13.0,
-      "logits/rejected": -9.5,
-      "logps/chosen": -716.0,
-      "logps/rejected": -3216.0,
-      "loss": 0.0139,
+      "logits/chosen": -12.125,
+      "logits/rejected": -10.0,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.0146,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.609375,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -10.75,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 8.75,
+      "rewards/rejected": -11.75,
       "step": 1660
     },
     {
       "epoch": 1.7486910994764397,
-      "grad_norm": 0.43006415236324225,
+      "grad_norm": 0.18175038740904031,
       "learning_rate": 2.214251060741528e-05,
-      "logits/chosen": -13.0625,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0427,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.75,
-      "rewards/margins": 7.90625,
-      "rewards/rejected": -10.625,
+      "logits/chosen": -12.25,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -772.0,
+      "logps/rejected": -3472.0,
+      "loss": 0.0483,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -12.5,
       "step": 1670
     },
     {
       "epoch": 1.7591623036649215,
-      "grad_norm": 0.2619475218293731,
+      "grad_norm": 0.27552385766246257,
       "learning_rate": 2.1840072811859302e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -9.625,
-      "logps/chosen": -748.0,
-      "logps/rejected": -3040.0,
-      "loss": 0.0285,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -10.25,
+      "logps/chosen": -812.0,
+      "logps/rejected": -3168.0,
+      "loss": 0.0263,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.5,
-      "rewards/margins": 6.9375,
-      "rewards/rejected": -9.4375,
+      "rewards/chosen": -3.140625,
+      "rewards/margins": 7.53125,
+      "rewards/rejected": -10.6875,
       "step": 1680
     },
     {
       "epoch": 1.7696335078534031,
-      "grad_norm": 0.11053365547409333,
+      "grad_norm": 0.09953153932019926,
       "learning_rate": 2.1538104267750287e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -9.5,
-      "logps/chosen": -632.0,
-      "logps/rejected": -3312.0,
-      "loss": 0.0115,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3424.0,
+      "loss": 0.0113,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.265625,
-      "rewards/margins": 7.875,
-      "rewards/rejected": -10.125,
+      "rewards/chosen": -2.734375,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -11.25,
       "step": 1690
     },
     {
       "epoch": 1.7801047120418847,
-      "grad_norm": 0.25584214657604637,
+      "grad_norm": 0.3285612639151191,
       "learning_rate": 2.1236649817632952e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.125,
-      "logps/chosen": -680.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0304,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.515625,
-      "rewards/margins": 7.96875,
-      "rewards/rejected": -10.5,
+      "logits/chosen": -13.8125,
+      "logits/rejected": -10.625,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0309,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -12.0,
       "step": 1700
     },
     {
       "epoch": 1.7801047120418847,
-      "eval_logits/chosen": -13.75,
-      "eval_logits/rejected": -13.5,
-      "eval_logps/chosen": -556.0,
-      "eval_logps/rejected": -536.0,
-      "eval_loss": 0.9087187647819519,
-      "eval_rewards/accuracies": 0.490234375,
-      "eval_rewards/chosen": -2.375,
-      "eval_rewards/margins": 0.08203125,
-      "eval_rewards/rejected": -2.453125,
-      "eval_runtime": 46.5775,
-      "eval_samples_per_second": 42.939,
+      "eval_logits/chosen": -13.8125,
+      "eval_logits/rejected": -13.5625,
+      "eval_logps/chosen": -584.0,
+      "eval_logps/rejected": -560.0,
+      "eval_loss": 0.9203125238418579,
+      "eval_rewards/accuracies": 0.505859375,
+      "eval_rewards/chosen": -2.671875,
+      "eval_rewards/margins": 0.0634765625,
+      "eval_rewards/rejected": -2.734375,
+      "eval_runtime": 46.598,
+      "eval_samples_per_second": 42.92,
       "eval_steps_per_second": 0.687,
       "step": 1700
     },
     {
       "epoch": 1.7905759162303665,
-      "grad_norm": 0.9648087990007245,
+      "grad_norm": 1.2180447907947058,
       "learning_rate": 2.093575422770872e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -744.0,
-      "logps/rejected": -3168.0,
-      "loss": 0.0212,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.53125,
-      "rewards/margins": 7.96875,
-      "rewards/rejected": -10.5,
+      "logits/chosen": -13.75,
+      "logits/rejected": -10.875,
+      "logps/chosen": -764.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0227,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.75,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -11.375,
       "step": 1710
     },
     {
       "epoch": 1.8010471204188483,
-      "grad_norm": 1.0081394718938281,
+      "grad_norm": 0.5821164930474221,
       "learning_rate": 2.0635462181187827e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -728.0,
-      "logps/rejected": -3360.0,
-      "loss": 0.0172,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -748.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0184,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.59375,
+      "rewards/chosen": -2.796875,
       "rewards/margins": 8.625,
-      "rewards/rejected": -11.25,
+      "rewards/rejected": -11.375,
       "step": 1720
     },
     {
       "epoch": 1.8115183246073299,
-      "grad_norm": 0.18797543737430802,
+      "grad_norm": 0.1116732114948823,
       "learning_rate": 2.0335818271653917e-05,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -648.0,
-      "logps/rejected": -3328.0,
-      "loss": 0.0221,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.484375,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -11.5,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -652.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0366,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.515625,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -11.125,
       "step": 1730
     },
     {
       "epoch": 1.8219895287958114,
-      "grad_norm": 0.16860467639320212,
+      "grad_norm": 0.18271538824926176,
       "learning_rate": 2.0036866996441815e-05,
-      "logits/chosen": -13.6875,
-      "logits/rejected": -9.9375,
+      "logits/chosen": -12.9375,
+      "logits/rejected": -10.5,
       "logps/chosen": -708.0,
-      "logps/rejected": -3280.0,
-      "loss": 0.025,
+      "logps/rejected": -3232.0,
+      "loss": 0.0199,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.65625,
-      "rewards/margins": 8.375,
-      "rewards/rejected": -11.0625,
+      "rewards/chosen": -2.640625,
+      "rewards/margins": 8.0,
+      "rewards/rejected": -10.625,
       "step": 1740
     },
     {
       "epoch": 1.8324607329842932,
-      "grad_norm": 0.9329828424413206,
+      "grad_norm": 0.35164637675373916,
       "learning_rate": 1.9738652750029677e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -764.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.625,
+      "logps/chosen": -780.0,
       "logps/rejected": -3136.0,
-      "loss": 0.0179,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.828125,
-      "rewards/margins": 7.59375,
-      "rewards/rejected": -10.4375,
+      "loss": 0.0154,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 7.53125,
+      "rewards/rejected": -10.5,
       "step": 1750
     },
     {
       "epoch": 1.8429319371727748,
-      "grad_norm": 0.09630025226284178,
+      "grad_norm": 0.06338743723517574,
       "learning_rate": 1.944121981744631e-05,
-      "logits/chosen": -13.375,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -692.0,
-      "logps/rejected": -3216.0,
-      "loss": 0.0186,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.5,
-      "rewards/margins": 8.625,
-      "rewards/rejected": -11.125,
+      "logits/chosen": -13.75,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.018,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.59375,
+      "rewards/margins": 8.875,
+      "rewards/rejected": -11.4375,
       "step": 1760
     },
     {
       "epoch": 1.8534031413612566,
-      "grad_norm": 0.29770614579765814,
+      "grad_norm": 0.13470039125294114,
       "learning_rate": 1.9144612367694887e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -692.0,
+      "logits/chosen": -13.875,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -700.0,
       "logps/rejected": -3104.0,
-      "loss": 0.0231,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.5,
-      "rewards/margins": 8.0,
-      "rewards/rejected": -10.5,
+      "loss": 0.0183,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.578125,
+      "rewards/margins": 8.125,
+      "rewards/rejected": -10.6875,
       "step": 1770
     },
     {
       "epoch": 1.8638743455497382,
-      "grad_norm": 0.5425820005935259,
+      "grad_norm": 0.5690893079610777,
       "learning_rate": 1.8848874447193803e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -704.0,
+      "logits/chosen": -14.0625,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -720.0,
       "logps/rejected": -3232.0,
-      "loss": 0.0115,
+      "loss": 0.011,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.59375,
-      "rewards/margins": 8.5625,
-      "rewards/rejected": -11.125,
+      "rewards/chosen": -2.75,
+      "rewards/margins": 8.4375,
+      "rewards/rejected": -11.1875,
       "step": 1780
     },
     {
       "epoch": 1.8743455497382198,
-      "grad_norm": 0.647072199849212,
+      "grad_norm": 0.70094028821332,
       "learning_rate": 1.855404997323571e-05,
-      "logits/chosen": -13.75,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -736.0,
+      "logits/chosen": -14.25,
+      "logits/rejected": -10.9375,
+      "logps/chosen": -756.0,
       "logps/rejected": -3280.0,
-      "loss": 0.0163,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.78125,
-      "rewards/margins": 8.9375,
+      "loss": 0.019,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 8.75,
       "rewards/rejected": -11.75,
       "step": 1790
     },
     {
       "epoch": 1.8848167539267016,
-      "grad_norm": 0.5988047084665278,
+      "grad_norm": 0.7072182976177557,
       "learning_rate": 1.82601827274658e-05,
-      "logits/chosen": -13.0,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -716.0,
-      "logps/rejected": -3440.0,
-      "loss": 0.0189,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.84375,
-      "rewards/margins": 8.4375,
-      "rewards/rejected": -11.25,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.125,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0228,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 8.375,
+      "rewards/rejected": -11.4375,
       "step": 1800
     },
     {
       "epoch": 1.8848167539267016,
-      "eval_logits/chosen": -13.375,
-      "eval_logits/rejected": -13.0,
-      "eval_logps/chosen": -584.0,
-      "eval_logps/rejected": -568.0,
-      "eval_loss": 0.8958749771118164,
-      "eval_rewards/accuracies": 0.50390625,
-      "eval_rewards/chosen": -2.671875,
-      "eval_rewards/margins": 0.13671875,
-      "eval_rewards/rejected": -2.8125,
-      "eval_runtime": 46.5907,
-      "eval_samples_per_second": 42.927,
+      "eval_logits/chosen": -13.5625,
+      "eval_logits/rejected": -13.3125,
+      "eval_logps/chosen": -604.0,
+      "eval_logps/rejected": -584.0,
+      "eval_loss": 0.9032187461853027,
+      "eval_rewards/accuracies": 0.494140625,
+      "eval_rewards/chosen": -2.859375,
+      "eval_rewards/margins": 0.09716796875,
+      "eval_rewards/rejected": -2.953125,
+      "eval_runtime": 46.5916,
+      "eval_samples_per_second": 42.926,
       "eval_steps_per_second": 0.687,
       "step": 1800
     },
     {
       "epoch": 1.8952879581151834,
-      "grad_norm": 0.17098314735531286,
+      "grad_norm": 0.16241128394468046,
       "learning_rate": 1.7967316349380213e-05,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -744.0,
-      "logps/rejected": -3120.0,
-      "loss": 0.0213,
+      "logits/chosen": -13.75,
+      "logits/rejected": -10.375,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3136.0,
+      "loss": 0.0197,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.796875,
-      "rewards/margins": 7.625,
-      "rewards/rejected": -10.4375,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 7.65625,
+      "rewards/rejected": -10.5625,
       "step": 1810
     },
     {
       "epoch": 1.905759162303665,
-      "grad_norm": 0.06935447293402892,
+      "grad_norm": 0.1459047088845035,
       "learning_rate": 1.7675494329845516e-05,
-      "logits/chosen": -12.875,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -712.0,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -720.0,
       "logps/rejected": -3360.0,
-      "loss": 0.029,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.921875,
-      "rewards/margins": 8.625,
+      "loss": 0.0289,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 8.5625,
       "rewards/rejected": -11.5,
       "step": 1820
     },
     {
       "epoch": 1.9162303664921465,
-      "grad_norm": 0.11739092594291739,
+      "grad_norm": 0.4154611855038144,
       "learning_rate": 1.738476000464026e-05,
-      "logits/chosen": -12.8125,
-      "logits/rejected": -9.375,
-      "logps/chosen": -724.0,
-      "logps/rejected": -3376.0,
-      "loss": 0.0351,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.890625,
-      "rewards/margins": 8.75,
-      "rewards/rejected": -11.625,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.0415,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -11.3125,
       "step": 1830
     },
     {
       "epoch": 1.9267015706806283,
-      "grad_norm": 0.1838408720062665,
+      "grad_norm": 0.19310026323172322,
       "learning_rate": 1.7095156548019646e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -9.9375,
+      "logits/chosen": -13.8125,
+      "logits/rejected": -10.625,
       "logps/chosen": -760.0,
-      "logps/rejected": -3328.0,
-      "loss": 0.0229,
+      "logps/rejected": -3312.0,
+      "loss": 0.0221,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.84375,
-      "rewards/margins": 8.6875,
-      "rewards/rejected": -11.5,
+      "rewards/chosen": -2.859375,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -11.375,
       "step": 1840
     },
     {
       "epoch": 1.93717277486911,
-      "grad_norm": 0.28910958447459284,
+      "grad_norm": 0.47604811123596286,
       "learning_rate": 1.6806726966304064e-05,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.0405,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.71875,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -10.875,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.25,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.0426,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 7.875,
+      "rewards/rejected": -10.8125,
       "step": 1850
     },
     {
       "epoch": 1.9476439790575917,
-      "grad_norm": 1.2674093527891848,
+      "grad_norm": 1.3694428440934308,
       "learning_rate": 1.6519514091492625e-05,
-      "logits/chosen": -12.875,
-      "logits/rejected": -9.75,
-      "logps/chosen": -740.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0258,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.65625,
-      "rewards/margins": 7.84375,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -768.0,
+      "logps/rejected": -3360.0,
+      "loss": 0.0265,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 7.59375,
       "rewards/rejected": -10.5,
       "step": 1860
     },
     {
       "epoch": 1.9581151832460733,
-      "grad_norm": 0.9413467962900552,
+      "grad_norm": 1.3018516747243323,
       "learning_rate": 1.6233560574902627e-05,
-      "logits/chosen": -13.375,
-      "logits/rejected": -10.125,
-      "logps/chosen": -768.0,
-      "logps/rejected": -3040.0,
-      "loss": 0.0235,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.78125,
-      "rewards/margins": 7.71875,
-      "rewards/rejected": -10.5,
+      "logits/chosen": -13.875,
+      "logits/rejected": -10.75,
+      "logps/chosen": -792.0,
+      "logps/rejected": -3072.0,
+      "loss": 0.0194,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 7.75,
+      "rewards/rejected": -10.8125,
       "step": 1870
     },
     {
       "epoch": 1.9685863874345548,
-      "grad_norm": 0.6240464515354742,
+      "grad_norm": 0.6533888871003399,
       "learning_rate": 1.5948908880835753e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3424.0,
-      "loss": 0.0383,
+      "logits/chosen": -13.8125,
+      "logits/rejected": -10.5,
+      "logps/chosen": -748.0,
+      "logps/rejected": -3440.0,
+      "loss": 0.0419,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 8.5,
-      "rewards/rejected": -11.375,
+      "rewards/chosen": -3.25,
+      "rewards/margins": 8.4375,
+      "rewards/rejected": -11.6875,
       "step": 1880
     },
     {
       "epoch": 1.9790575916230366,
-      "grad_norm": 0.05542960533750751,
+      "grad_norm": 0.0729279059739014,
       "learning_rate": 1.5665601280272124e-05,
-      "logits/chosen": -13.375,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3552.0,
-      "loss": 0.0089,
+      "logits/chosen": -13.8125,
+      "logits/rejected": -10.5,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0118,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.828125,
+      "rewards/chosen": -3.125,
       "rewards/margins": 9.125,
-      "rewards/rejected": -12.0,
+      "rewards/rejected": -12.1875,
       "step": 1890
     },
     {
       "epoch": 1.9895287958115184,
-      "grad_norm": 0.06820617401303433,
+      "grad_norm": 0.08217785034476056,
       "learning_rate": 1.538367984459298e-05,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -9.75,
-      "logps/chosen": -736.0,
-      "logps/rejected": -3104.0,
-      "loss": 0.0113,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.375,
+      "logps/chosen": -760.0,
+      "logps/rejected": -3136.0,
+      "loss": 0.0116,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.875,
-      "rewards/margins": 8.6875,
-      "rewards/rejected": -11.5625,
+      "rewards/chosen": -3.140625,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -11.8125,
       "step": 1900
     },
     {
       "epoch": 1.9895287958115184,
       "eval_logits/chosen": -13.625,
       "eval_logits/rejected": -13.375,
-      "eval_logps/chosen": -596.0,
-      "eval_logps/rejected": -584.0,
-      "eval_loss": 0.909375011920929,
-      "eval_rewards/accuracies": 0.51171875,
-      "eval_rewards/chosen": -2.78125,
-      "eval_rewards/margins": 0.173828125,
-      "eval_rewards/rejected": -2.953125,
-      "eval_runtime": 46.5809,
-      "eval_samples_per_second": 42.936,
+      "eval_logps/chosen": -620.0,
+      "eval_logps/rejected": -600.0,
+      "eval_loss": 0.9122968912124634,
+      "eval_rewards/accuracies": 0.5,
+      "eval_rewards/chosen": -3.015625,
+      "eval_rewards/margins": 0.11865234375,
+      "eval_rewards/rejected": -3.125,
+      "eval_runtime": 46.6038,
+      "eval_samples_per_second": 42.915,
       "eval_steps_per_second": 0.687,
       "step": 1900
     },
     {
       "epoch": 2.0,
-      "grad_norm": 0.7923736138071442,
+      "grad_norm": 0.7599649688859563,
       "learning_rate": 1.5103186439333133e-05,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -9.625,
-      "logps/chosen": -676.0,
-      "logps/rejected": -3440.0,
-      "loss": 0.0141,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.125,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3392.0,
+      "loss": 0.0164,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.546875,
-      "rewards/margins": 10.0625,
-      "rewards/rejected": -12.625,
+      "rewards/chosen": -2.859375,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -12.25,
       "step": 1910
     },
     {
       "epoch": 2.0104712041884816,
-      "grad_norm": 0.03797413952811105,
+      "grad_norm": 0.044651384034995086,
       "learning_rate": 1.4824162717963829e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -772.0,
-      "logps/rejected": -3280.0,
-      "loss": 0.0021,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -792.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0024,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.765625,
-      "rewards/margins": 9.625,
-      "rewards/rejected": -12.375,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -12.0625,
       "step": 1920
     },
     {
       "epoch": 2.020942408376963,
-      "grad_norm": 0.08445522476020367,
+      "grad_norm": 0.044442715180427216,
       "learning_rate": 1.4546650115707192e-05,
       "logits/chosen": -13.5625,
-      "logits/rejected": -10.25,
-      "logps/chosen": -740.0,
-      "logps/rejected": -3456.0,
-      "loss": 0.0024,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3440.0,
+      "loss": 0.0022,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.640625,
-      "rewards/margins": 10.375,
-      "rewards/rejected": -13.0625,
+      "rewards/chosen": -2.828125,
+      "rewards/margins": 10.0625,
+      "rewards/rejected": -12.875,
       "step": 1930
     },
     {
       "epoch": 2.031413612565445,
-      "grad_norm": 0.008763710192524561,
+      "grad_norm": 0.02192458111072301,
       "learning_rate": 1.4270689843383111e-05,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -764.0,
-      "logps/rejected": -3568.0,
-      "loss": 0.0015,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.75,
+      "logps/chosen": -800.0,
+      "logps/rejected": -3552.0,
+      "loss": 0.0017,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.734375,
-      "rewards/margins": 10.8125,
-      "rewards/rejected": -13.5,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 10.25,
+      "rewards/rejected": -13.375,
       "step": 1940
     },
     {
       "epoch": 2.0418848167539267,
-      "grad_norm": 0.010871862750517299,
+      "grad_norm": 0.009715391050262126,
       "learning_rate": 1.3996322881289347e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.25,
-      "logps/chosen": -756.0,
-      "logps/rejected": -3456.0,
-      "loss": 0.0012,
+      "logits/chosen": -13.75,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -784.0,
+      "logps/rejected": -3440.0,
+      "loss": 0.0013,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 10.625,
-      "rewards/rejected": -13.5,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 10.25,
+      "rewards/rejected": -13.4375,
       "step": 1950
     },
     {
       "epoch": 2.0523560209424083,
-      "grad_norm": 0.0355862679650303,
+      "grad_norm": 0.08209293066141186,
       "learning_rate": 1.3723589973115963e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.125,
-      "logps/chosen": -740.0,
-      "logps/rejected": -3600.0,
-      "loss": 0.001,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -10.5,
+      "logps/chosen": -760.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0011,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.96875,
-      "rewards/margins": 11.375,
-      "rewards/rejected": -14.3125,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 10.875,
+      "rewards/rejected": -14.125,
       "step": 1960
     },
     {
       "epoch": 2.06282722513089,
-      "grad_norm": 0.012331372934647332,
+      "grad_norm": 0.013554623707018636,
       "learning_rate": 1.3452531619894837e-05,
-      "logits/chosen": -13.25,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -752.0,
-      "logps/rejected": -3536.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -776.0,
+      "logps/rejected": -3520.0,
       "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.0,
-      "rewards/margins": 11.25,
-      "rewards/rejected": -14.25,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 10.875,
+      "rewards/rejected": -14.0625,
       "step": 1970
     },
     {
       "epoch": 2.073298429319372,
-      "grad_norm": 0.03788379705316475,
+      "grad_norm": 0.008632996353291053,
       "learning_rate": 1.3183188073985245e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -10.125,
-      "logps/chosen": -816.0,
+      "logits/chosen": -13.125,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -848.0,
       "logps/rejected": -3568.0,
-      "loss": 0.0017,
+      "loss": 0.0011,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.890625,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -14.3125,
+      "rewards/chosen": -3.1875,
+      "rewards/margins": 11.25,
+      "rewards/rejected": -14.375,
       "step": 1980
     },
     {
       "epoch": 2.0837696335078535,
-      "grad_norm": 0.020425162583961123,
+      "grad_norm": 0.02578721551162634,
       "learning_rate": 1.2915599333096351e-05,
-      "logits/chosen": -13.6875,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -732.0,
-      "logps/rejected": -3440.0,
-      "loss": 0.0013,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.0,
-      "rewards/margins": 10.875,
-      "rewards/rejected": -13.875,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 10.8125,
+      "rewards/rejected": -14.0625,
       "step": 1990
     },
     {
       "epoch": 2.094240837696335,
-      "grad_norm": 0.043482431208357666,
+      "grad_norm": 0.03430472824141071,
       "learning_rate": 1.264980513434752e-05,
-      "logits/chosen": -13.6875,
-      "logits/rejected": -10.125,
-      "logps/chosen": -732.0,
-      "logps/rejected": -3600.0,
-      "loss": 0.0013,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3632.0,
+      "loss": 0.0011,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.109375,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -14.5,
+      "rewards/chosen": -3.359375,
+      "rewards/margins": 11.625,
+      "rewards/rejected": -14.9375,
       "step": 2000
     },
     {
       "epoch": 2.094240837696335,
-      "eval_logits/chosen": -13.75,
-      "eval_logits/rejected": -13.5,
-      "eval_logps/chosen": -616.0,
-      "eval_logps/rejected": -608.0,
-      "eval_loss": 0.9535468816757202,
-      "eval_rewards/accuracies": 0.513671875,
-      "eval_rewards/chosen": -2.96875,
-      "eval_rewards/margins": 0.220703125,
-      "eval_rewards/rejected": -3.1875,
-      "eval_runtime": 46.5716,
-      "eval_samples_per_second": 42.945,
+      "eval_logits/chosen": -13.3125,
+      "eval_logits/rejected": -13.0625,
+      "eval_logps/chosen": -644.0,
+      "eval_logps/rejected": -636.0,
+      "eval_loss": 0.9715155959129333,
+      "eval_rewards/accuracies": 0.498046875,
+      "eval_rewards/chosen": -3.265625,
+      "eval_rewards/margins": 0.1865234375,
+      "eval_rewards/rejected": -3.453125,
+      "eval_runtime": 46.602,
+      "eval_samples_per_second": 42.917,
       "eval_steps_per_second": 0.687,
       "step": 2000
     },
     {
       "epoch": 2.1047120418848166,
-      "grad_norm": 0.021990040282395953,
+      "grad_norm": 0.01811075419731086,
       "learning_rate": 1.2385844948367324e-05,
-      "logits/chosen": -13.75,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -736.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0011,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -768.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.953125,
-      "rewards/margins": 11.75,
-      "rewards/rejected": -14.75,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 12.0,
+      "rewards/rejected": -15.25,
       "step": 2010
     },
     {
       "epoch": 2.115183246073298,
-      "grad_norm": 0.02681732808537171,
+      "grad_norm": 0.016051391089347427,
       "learning_rate": 1.2123757973432114e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.375,
-      "logps/chosen": -708.0,
-      "logps/rejected": -3488.0,
-      "loss": 0.0049,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.875,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3536.0,
+      "loss": 0.0019,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.84375,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -14.25,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -14.6875,
       "step": 2020
     },
     {
       "epoch": 2.1256544502617802,
-      "grad_norm": 0.02804504925376927,
+      "grad_norm": 0.2443017249322444,
       "learning_rate": 1.1863583129645106e-05,
-      "logits/chosen": -12.9375,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -800.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0015,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.078125,
-      "rewards/margins": 11.75,
-      "rewards/rejected": -14.875,
+      "logits/chosen": -12.875,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -844.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0036,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 11.875,
+      "rewards/rejected": -15.4375,
       "step": 2030
     },
     {
       "epoch": 2.136125654450262,
-      "grad_norm": 0.8075496941248711,
+      "grad_norm": 0.07615052792658578,
       "learning_rate": 1.1605359053156607e-05,
-      "logits/chosen": -13.75,
-      "logits/rejected": -10.25,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3424.0,
-      "loss": 0.0017,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -760.0,
+      "logps/rejected": -3472.0,
+      "loss": 0.0022,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.796875,
-      "rewards/margins": 11.375,
-      "rewards/rejected": -14.1875,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -14.75,
       "step": 2040
     },
     {
       "epoch": 2.1465968586387434,
-      "grad_norm": 0.01902900455234704,
+      "grad_norm": 0.015605865580488496,
       "learning_rate": 1.1349124090426644e-05,
       "logits/chosen": -13.3125,
-      "logits/rejected": -9.875,
-      "logps/chosen": -816.0,
-      "logps/rejected": -3808.0,
-      "loss": 0.0016,
+      "logits/rejected": -10.25,
+      "logps/chosen": -860.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.0,
-      "rewards/margins": 12.25,
-      "rewards/rejected": -15.1875,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -15.75,
       "step": 2050
     },
     {
       "epoch": 2.157068062827225,
-      "grad_norm": 0.0283077420771619,
+      "grad_norm": 0.013992533654090781,
       "learning_rate": 1.1094916292530406e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3760.0,
-      "loss": 0.0018,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.796875,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -768.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0044,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.375,
       "rewards/margins": 12.1875,
-      "rewards/rejected": -15.0,
+      "rewards/rejected": -15.5625,
       "step": 2060
     },
     {
       "epoch": 2.167539267015707,
-      "grad_norm": 0.012130048267476778,
+      "grad_norm": 0.011845455993297177,
       "learning_rate": 1.0842773409507623e-05,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -10.5,
-      "logps/chosen": -704.0,
-      "logps/rejected": -3600.0,
-      "loss": 0.001,
+      "logits/chosen": -13.0,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -764.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.6875,
-      "rewards/margins": 12.75,
-      "rewards/rejected": -15.4375,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -15.875,
       "step": 2070
     },
     {
       "epoch": 2.1780104712041886,
-      "grad_norm": 0.11186142873174292,
+      "grad_norm": 0.0813868167012562,
       "learning_rate": 1.0592732884756754e-05,
       "logits/chosen": -13.5,
-      "logits/rejected": -10.0,
-      "logps/chosen": -740.0,
-      "logps/rejected": -3760.0,
-      "loss": 0.0014,
+      "logits/rejected": -10.25,
+      "logps/chosen": -788.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.96875,
-      "rewards/margins": 11.9375,
-      "rewards/rejected": -14.9375,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 12.375,
+      "rewards/rejected": -15.8125,
       "step": 2080
     },
     {
       "epoch": 2.18848167539267,
-      "grad_norm": 0.022020673007143294,
+      "grad_norm": 0.017334523321388594,
       "learning_rate": 1.0344831849474507e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.0,
-      "logps/chosen": -740.0,
-      "logps/rejected": -3648.0,
-      "loss": 0.0009,
+      "logits/chosen": -13.125,
+      "logits/rejected": -10.25,
+      "logps/chosen": -800.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.0,
-      "rewards/margins": 12.25,
-      "rewards/rejected": -15.25,
+      "rewards/chosen": -3.578125,
+      "rewards/margins": 12.125,
+      "rewards/rejected": -15.75,
       "step": 2090
     },
     {
       "epoch": 2.1989528795811517,
-      "grad_norm": 0.031190943948052354,
+      "grad_norm": 0.01981870785374481,
       "learning_rate": 1.0099107117141879e-05,
-      "logits/chosen": -13.5625,
+      "logits/chosen": -13.1875,
       "logits/rejected": -10.1875,
-      "logps/chosen": -796.0,
-      "logps/rejected": -3648.0,
-      "loss": 0.0014,
+      "logps/chosen": -852.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0019,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.265625,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -15.625,
+      "rewards/chosen": -3.796875,
+      "rewards/margins": 12.5,
+      "rewards/rejected": -16.25,
       "step": 2100
     },
     {
       "epoch": 2.1989528795811517,
-      "eval_logits/chosen": -13.5625,
-      "eval_logits/rejected": -13.3125,
-      "eval_logps/chosen": -632.0,
-      "eval_logps/rejected": -628.0,
-      "eval_loss": 0.979812502861023,
-      "eval_rewards/accuracies": 0.517578125,
-      "eval_rewards/chosen": -3.15625,
-      "eval_rewards/margins": 0.228515625,
-      "eval_rewards/rejected": -3.375,
-      "eval_runtime": 46.5669,
-      "eval_samples_per_second": 42.949,
+      "eval_logits/chosen": -12.8125,
+      "eval_logits/rejected": -12.5,
+      "eval_logps/chosen": -684.0,
+      "eval_logps/rejected": -680.0,
+      "eval_loss": 1.037750005722046,
+      "eval_rewards/accuracies": 0.509765625,
+      "eval_rewards/chosen": -3.671875,
+      "eval_rewards/margins": 0.2392578125,
+      "eval_rewards/rejected": -3.90625,
+      "eval_runtime": 46.5983,
+      "eval_samples_per_second": 42.92,
       "eval_steps_per_second": 0.687,
       "step": 2100
     },
     {
       "epoch": 2.2094240837696333,
-      "grad_norm": 0.02086347682769286,
+      "grad_norm": 0.050126684437074596,
       "learning_rate": 9.855595178057334e-06,
-      "logits/chosen": -13.375,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -848.0,
-      "logps/rejected": -3664.0,
-      "loss": 0.0015,
+      "logits/chosen": -13.0,
+      "logits/rejected": -9.875,
+      "logps/chosen": -904.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0017,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.109375,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -15.4375,
+      "rewards/chosen": -3.65625,
+      "rewards/margins": 12.4375,
+      "rewards/rejected": -16.125,
       "step": 2110
     },
-    {
-      "epoch": 2.2198952879581153,
-      "grad_norm": 0.0714355062071931,
-      "learning_rate": 9.614332193917949e-06,
-      "logits/chosen": -13.25,
-      "logits/rejected": -10.125,
-      "logps/chosen": -836.0,
-      "logps/rejected": -3648.0,
-      "loss": 0.0011,
+    {
+      "epoch": 2.2198952879581153,
+      "grad_norm": 0.050815367567544255,
+      "learning_rate": 9.614332193917949e-06,
+      "logits/chosen": -13.0,
+      "logits/rejected": -10.375,
+      "logps/chosen": -884.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 12.0625,
-      "rewards/rejected": -15.375,
+      "rewards/chosen": -3.75,
+      "rewards/margins": 12.0,
+      "rewards/rejected": -15.75,
       "step": 2120
     },
     {
       "epoch": 2.230366492146597,
-      "grad_norm": 0.010214424707746925,
+      "grad_norm": 0.013227994573835587,
       "learning_rate": 9.375353992449385e-06,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -9.75,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3664.0,
+      "logits/chosen": -12.8125,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3696.0,
       "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.015625,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -15.3125,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -15.6875,
       "step": 2130
     },
     {
       "epoch": 2.2408376963350785,
-      "grad_norm": 0.009449610139048667,
+      "grad_norm": 0.01791797799118319,
       "learning_rate": 9.138696062085442e-06,
-      "logits/chosen": -13.25,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -752.0,
-      "logps/rejected": -3760.0,
-      "loss": 0.0013,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -10.125,
+      "logps/chosen": -804.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0023,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 13.0,
-      "rewards/rejected": -16.125,
+      "rewards/chosen": -3.671875,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -16.5,
       "step": 2140
     },
     {
       "epoch": 2.25130890052356,
-      "grad_norm": 0.030870851684990906,
+      "grad_norm": 0.035056158924740774,
       "learning_rate": 8.904393546698007e-06,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -820.0,
-      "logps/rejected": -3744.0,
-      "loss": 0.0008,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -10.125,
+      "logps/chosen": -856.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.296875,
-      "rewards/margins": 12.625,
-      "rewards/rejected": -15.9375,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 12.5625,
+      "rewards/rejected": -16.25,
       "step": 2150
     },
     {
       "epoch": 2.261780104712042,
-      "grad_norm": 0.017928103149379482,
+      "grad_norm": 0.018600593502528,
       "learning_rate": 8.672481240378142e-06,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -856.0,
-      "logps/rejected": -3664.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -892.0,
+      "logps/rejected": -3648.0,
       "loss": 0.0016,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.296875,
-      "rewards/margins": 12.375,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 12.0,
       "rewards/rejected": -15.6875,
       "step": 2160
     },
     {
       "epoch": 2.2722513089005236,
-      "grad_norm": 0.020979844872565743,
+      "grad_norm": 0.0176691760449618,
       "learning_rate": 8.44299358226919e-06,
       "logits/chosen": -13.4375,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -836.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0019,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -872.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0016,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 13.3125,
+      "rewards/chosen": -3.71875,
+      "rewards/margins": 12.875,
       "rewards/rejected": -16.625,
       "step": 2170
     },
     {
       "epoch": 2.282722513089005,
-      "grad_norm": 0.01132743031264989,
+      "grad_norm": 0.009540059086755108,
       "learning_rate": 8.215964651452457e-06,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -824.0,
-      "logps/rejected": -3776.0,
-      "loss": 0.0013,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -864.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.328125,
-      "rewards/margins": 13.375,
-      "rewards/rejected": -16.75,
+      "rewards/chosen": -3.71875,
+      "rewards/margins": 12.5625,
+      "rewards/rejected": -16.25,
       "step": 2180
     },
     {
       "epoch": 2.2931937172774868,
-      "grad_norm": 0.008195016345104533,
+      "grad_norm": 0.016240879547740093,
       "learning_rate": 7.991428161886502e-06,
       "logits/chosen": -13.375,
-      "logits/rejected": -10.25,
-      "logps/chosen": -820.0,
-      "logps/rejected": -3856.0,
-      "loss": 0.0009,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -860.0,
+      "logps/rejected": -3872.0,
+      "loss": 0.005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 13.0625,
-      "rewards/rejected": -16.375,
+      "rewards/chosen": -3.734375,
+      "rewards/margins": 12.8125,
+      "rewards/rejected": -16.5,
       "step": 2190
     },
     {
       "epoch": 2.303664921465969,
-      "grad_norm": 0.007235192617224554,
+      "grad_norm": 0.05904505689919353,
       "learning_rate": 7.76941745740061e-06,
-      "logits/chosen": -13.125,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -724.0,
+      "logits/chosen": -12.875,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -764.0,
       "logps/rejected": -3888.0,
-      "loss": 0.0013,
+      "loss": 0.0011,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.171875,
-      "rewards/margins": 13.125,
-      "rewards/rejected": -16.375,
+      "rewards/chosen": -3.578125,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -16.25,
       "step": 2200
     },
     {
       "epoch": 2.303664921465969,
-      "eval_logits/chosen": -13.5625,
-      "eval_logits/rejected": -13.3125,
-      "eval_logps/chosen": -652.0,
-      "eval_logps/rejected": -648.0,
-      "eval_loss": 0.9977812767028809,
-      "eval_rewards/accuracies": 0.513671875,
-      "eval_rewards/chosen": -3.328125,
-      "eval_rewards/margins": 0.240234375,
-      "eval_rewards/rejected": -3.578125,
-      "eval_runtime": 46.5722,
-      "eval_samples_per_second": 42.944,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -13.125,
+      "eval_logits/rejected": -12.8125,
+      "eval_logps/chosen": -692.0,
+      "eval_logps/rejected": -684.0,
+      "eval_loss": 1.0456249713897705,
+      "eval_rewards/accuracies": 0.501953125,
+      "eval_rewards/chosen": -3.71875,
+      "eval_rewards/margins": 0.22265625,
+      "eval_rewards/rejected": -3.9375,
+      "eval_runtime": 46.8087,
+      "eval_samples_per_second": 42.727,
+      "eval_steps_per_second": 0.684,
       "step": 2200
     },
     {
       "epoch": 2.3141361256544504,
-      "grad_norm": 0.019723181322208762,
+      "grad_norm": 0.009905225960300586,
       "learning_rate": 7.549965506743123e-06,
-      "logits/chosen": -13.5,
-      "logits/rejected": -9.75,
-      "logps/chosen": -708.0,
-      "logps/rejected": -3712.0,
-      "loss": 0.0012,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -10.125,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0014,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.234375,
-      "rewards/margins": 13.0,
-      "rewards/rejected": -16.25,
+      "rewards/chosen": -3.484375,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -16.125,
       "step": 2210
     },
     {
       "epoch": 2.324607329842932,
-      "grad_norm": 0.014077934151234522,
+      "grad_norm": 0.030851753470104073,
       "learning_rate": 7.333104898685631e-06,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -728.0,
-      "logps/rejected": -3680.0,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -10.75,
+      "logps/chosen": -760.0,
+      "logps/rejected": -3664.0,
       "loss": 0.0013,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.046875,
-      "rewards/margins": 13.0625,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 12.6875,
       "rewards/rejected": -16.125,
       "step": 2220
     },
     {
       "epoch": 2.3350785340314135,
-      "grad_norm": 0.02530387524099349,
+      "grad_norm": 0.02286184359289322,
       "learning_rate": 7.118867837183518e-06,
-      "logits/chosen": -13.5,
-      "logits/rejected": -9.625,
-      "logps/chosen": -696.0,
-      "logps/rejected": -3856.0,
-      "loss": 0.0012,
+      "logits/chosen": -13.375,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.015625,
-      "rewards/margins": 13.5625,
-      "rewards/rejected": -16.625,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 13.1875,
+      "rewards/rejected": -16.5,
       "step": 2230
     },
     {
       "epoch": 2.345549738219895,
-      "grad_norm": 0.010256460710128305,
+      "grad_norm": 0.012778253381178623,
       "learning_rate": 6.9072861365936045e-06,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.0,
-      "logps/chosen": -780.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0014,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -820.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.203125,
-      "rewards/margins": 13.0,
-      "rewards/rejected": -16.25,
+      "rewards/chosen": -3.59375,
+      "rewards/margins": 12.5,
+      "rewards/rejected": -16.0,
       "step": 2240
     },
     {
       "epoch": 2.356020942408377,
-      "grad_norm": 0.06519330938521245,
+      "grad_norm": 0.024679574080629697,
       "learning_rate": 6.6983912169497005e-06,
-      "logits/chosen": -13.375,
-      "logits/rejected": -10.125,
-      "logps/chosen": -796.0,
-      "logps/rejected": -3824.0,
-      "loss": 0.0008,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -828.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0013,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 12.5625,
-      "rewards/rejected": -16.0,
+      "rewards/chosen": -3.84375,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -16.125,
       "step": 2250
     },
     {
       "epoch": 2.3664921465968587,
-      "grad_norm": 0.02535199174674999,
+      "grad_norm": 0.027204839168028027,
       "learning_rate": 6.49221409929677e-06,
-      "logits/chosen": -13.375,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -816.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -848.0,
       "logps/rejected": -3840.0,
-      "loss": 0.0009,
+      "loss": 0.0016,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 12.9375,
+      "rewards/chosen": -3.671875,
+      "rewards/margins": 12.625,
       "rewards/rejected": -16.25,
       "step": 2260
     },
     {
       "epoch": 2.3769633507853403,
-      "grad_norm": 0.008646543365720447,
+      "grad_norm": 0.007154529652714207,
       "learning_rate": 6.2887854010842075e-06,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -756.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -788.0,
       "logps/rejected": -3808.0,
-      "loss": 0.0018,
+      "loss": 0.0015,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.3125,
-      "rewards/margins": 13.25,
+      "rewards/chosen": -3.625,
+      "rewards/margins": 13.0,
       "rewards/rejected": -16.625,
       "step": 2270
     },
     {
       "epoch": 2.387434554973822,
-      "grad_norm": 0.048063864804168376,
+      "grad_norm": 0.01483717813637708,
       "learning_rate": 6.0881353316191385e-06,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -804.0,
-      "logps/rejected": -3824.0,
-      "loss": 0.0007,
+      "logits/chosen": -13.125,
+      "logits/rejected": -10.25,
+      "logps/chosen": -836.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.609375,
-      "rewards/margins": 12.6875,
-      "rewards/rejected": -16.375,
+      "rewards/chosen": -3.90625,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -16.25,
       "step": 2280
     },
     {
       "epoch": 2.3979057591623034,
-      "grad_norm": 0.02029365870305144,
+      "grad_norm": 0.06804402139371733,
       "learning_rate": 5.890293687580381e-06,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.125,
-      "logps/chosen": -744.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0008,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.75,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3872.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 13.125,
-      "rewards/rejected": -16.625,
+      "rewards/chosen": -3.8125,
+      "rewards/margins": 12.375,
+      "rewards/rejected": -16.25,
       "step": 2290
     },
     {
       "epoch": 2.4083769633507854,
-      "grad_norm": 0.021633310444226825,
+      "grad_norm": 0.02836136746888331,
       "learning_rate": 5.695289848593532e-06,
       "logits/chosen": -13.3125,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -824.0,
-      "logps/rejected": -3664.0,
-      "loss": 0.0011,
+      "logits/rejected": -10.375,
+      "logps/chosen": -848.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 12.25,
-      "rewards/rejected": -15.75,
+      "rewards/chosen": -3.75,
+      "rewards/margins": 11.875,
+      "rewards/rejected": -15.625,
       "step": 2300
     },
     {
       "epoch": 2.4083769633507854,
-      "eval_logits/chosen": -13.5625,
-      "eval_logits/rejected": -13.3125,
-      "eval_logps/chosen": -664.0,
-      "eval_logps/rejected": -660.0,
-      "eval_loss": 1.008062481880188,
-      "eval_rewards/accuracies": 0.513671875,
-      "eval_rewards/chosen": -3.453125,
-      "eval_rewards/margins": 0.2470703125,
-      "eval_rewards/rejected": -3.6875,
-      "eval_runtime": 46.5711,
-      "eval_samples_per_second": 42.945,
+      "eval_logits/chosen": -13.25,
+      "eval_logits/rejected": -12.9375,
+      "eval_logps/chosen": -692.0,
+      "eval_logps/rejected": -684.0,
+      "eval_loss": 1.0567030906677246,
+      "eval_rewards/accuracies": 0.501953125,
+      "eval_rewards/chosen": -3.75,
+      "eval_rewards/margins": 0.2216796875,
+      "eval_rewards/rejected": -3.96875,
+      "eval_runtime": 46.6042,
+      "eval_samples_per_second": 42.915,
       "eval_steps_per_second": 0.687,
       "step": 2300
     },
     {
       "epoch": 2.418848167539267,
-      "grad_norm": 0.038924291142115805,
+      "grad_norm": 0.01400737289004599,
       "learning_rate": 5.50315277286812e-06,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -9.875,
-      "logps/chosen": -780.0,
-      "logps/rejected": -3872.0,
-      "loss": 0.0019,
+      "logits/chosen": -13.125,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -804.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 13.25,
+      "rewards/chosen": -3.671875,
+      "rewards/margins": 12.9375,
       "rewards/rejected": -16.625,
       "step": 2310
     },
     {
       "epoch": 2.4293193717277486,
-      "grad_norm": 0.027840950067718296,
+      "grad_norm": 0.28639660159668534,
       "learning_rate": 5.313910992897281e-06,
       "logits/chosen": -13.5,
-      "logits/rejected": -9.875,
-      "logps/chosen": -756.0,
-      "logps/rejected": -3712.0,
-      "loss": 0.0019,
+      "logits/rejected": -10.5,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0018,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 12.5,
-      "rewards/rejected": -15.75,
+      "rewards/chosen": -3.515625,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -15.875,
       "step": 2320
     },
     {
       "epoch": 2.4397905759162306,
-      "grad_norm": 0.013668323394273506,
+      "grad_norm": 0.0063499060921826615,
       "learning_rate": 5.127592611220658e-06,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -784.0,
-      "logps/rejected": -3920.0,
-      "loss": 0.0011,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -812.0,
+      "logps/rejected": -3952.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
+      "rewards/chosen": -3.765625,
       "rewards/margins": 12.5625,
-      "rewards/rejected": -16.0,
+      "rewards/rejected": -16.375,
       "step": 2330
     },
     {
       "epoch": 2.450261780104712,
-      "grad_norm": 0.015624084426943893,
+      "grad_norm": 0.011148849738373098,
       "learning_rate": 4.944225296251159e-06,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -9.5,
-      "logps/chosen": -800.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0009,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.0,
+      "logps/chosen": -828.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 12.5,
-      "rewards/rejected": -15.9375,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -16.25,
       "step": 2340
     },
     {
       "epoch": 2.4607329842931938,
-      "grad_norm": 0.0330934045068677,
+      "grad_norm": 0.08575689598010647,
       "learning_rate": 4.763836278166172e-06,
-      "logits/chosen": -13.25,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -792.0,
-      "logps/rejected": -3760.0,
-      "loss": 0.0012,
+      "logits/chosen": -13.125,
+      "logits/rejected": -10.5,
+      "logps/chosen": -836.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0029,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 12.6875,
-      "rewards/rejected": -16.125,
+      "rewards/chosen": -3.859375,
+      "rewards/margins": 12.75,
+      "rewards/rejected": -16.625,
       "step": 2350
     },
     {
       "epoch": 2.4712041884816753,
-      "grad_norm": 0.21612560380681003,
+      "grad_norm": 0.047874626017986886,
       "learning_rate": 4.586452344863876e-06,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -788.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0013,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -820.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.234375,
-      "rewards/margins": 13.125,
+      "rewards/chosen": -3.5625,
+      "rewards/margins": 12.875,
       "rewards/rejected": -16.375,
       "step": 2360
     },
     {
       "epoch": 2.481675392670157,
-      "grad_norm": 0.038060395714203235,
+      "grad_norm": 0.03647074829847085,
       "learning_rate": 4.412099837985192e-06,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -9.5,
-      "logps/chosen": -724.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0015,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.1875,
-      "rewards/margins": 12.9375,
-      "rewards/rejected": -16.125,
+      "rewards/chosen": -3.359375,
+      "rewards/margins": 12.4375,
+      "rewards/rejected": -15.8125,
       "step": 2370
     },
     {
       "epoch": 2.492146596858639,
-      "grad_norm": 0.01808326965074876,
+      "grad_norm": 0.01838269526556759,
       "learning_rate": 4.240804649002089e-06,
-      "logits/chosen": -13.25,
-      "logits/rejected": -9.625,
-      "logps/chosen": -748.0,
-      "logps/rejected": -3888.0,
-      "loss": 0.0011,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -772.0,
+      "logps/rejected": -3872.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 13.0,
-      "rewards/rejected": -16.375,
+      "rewards/chosen": -3.671875,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -16.25,
       "step": 2380
     },
     {
       "epoch": 2.5026178010471205,
-      "grad_norm": 0.02466175839503429,
+      "grad_norm": 0.024885877025228643,
       "learning_rate": 4.072592215372598e-06,
-      "logits/chosen": -13.8125,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -812.0,
-      "logps/rejected": -3664.0,
-      "loss": 0.0012,
+      "logits/chosen": -14.0,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -816.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0013,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 12.4375,
-      "rewards/rejected": -15.75,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -15.625,
       "step": 2390
     },
     {
       "epoch": 2.513089005235602,
-      "grad_norm": 0.0070712892991022045,
+      "grad_norm": 0.015280863517852504,
       "learning_rate": 3.907487516763389e-06,
-      "logits/chosen": -13.5,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -796.0,
-      "logps/rejected": -3808.0,
-      "loss": 0.0017,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.328125,
-      "rewards/margins": 13.0625,
-      "rewards/rejected": -16.375,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.5,
+      "logps/chosen": -824.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0022,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.609375,
+      "rewards/margins": 12.6875,
+      "rewards/rejected": -16.25,
       "step": 2400
     },
     {
       "epoch": 2.513089005235602,
-      "eval_logits/chosen": -13.5625,
-      "eval_logits/rejected": -13.3125,
-      "eval_logps/chosen": -660.0,
-      "eval_logps/rejected": -656.0,
-      "eval_loss": 1.0045312643051147,
-      "eval_rewards/accuracies": 0.51171875,
-      "eval_rewards/chosen": -3.421875,
-      "eval_rewards/margins": 0.24609375,
-      "eval_rewards/rejected": -3.671875,
-      "eval_runtime": 46.5818,
-      "eval_samples_per_second": 42.935,
+      "eval_logits/chosen": -13.3125,
+      "eval_logits/rejected": -13.0,
+      "eval_logps/chosen": -692.0,
+      "eval_logps/rejected": -680.0,
+      "eval_loss": 1.0449531078338623,
+      "eval_rewards/accuracies": 0.49609375,
+      "eval_rewards/chosen": -3.71875,
+      "eval_rewards/margins": 0.1953125,
+      "eval_rewards/rejected": -3.90625,
+      "eval_runtime": 46.6032,
+      "eval_samples_per_second": 42.916,
       "eval_steps_per_second": 0.687,
       "step": 2400
     },
     {
       "epoch": 2.5235602094240837,
-      "grad_norm": 0.051153708522702936,
+      "grad_norm": 0.011394438211547137,
       "learning_rate": 3.745515071340286e-06,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -10.5,
-      "logps/chosen": -864.0,
-      "logps/rejected": -3696.0,
-      "loss": 0.0014,
+      "logits/chosen": -13.625,
+      "logits/rejected": -11.125,
+      "logps/chosen": -888.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.0009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 12.9375,
-      "rewards/rejected": -16.375,
+      "rewards/chosen": -3.65625,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -16.0,
       "step": 2410
     },
     {
       "epoch": 2.5340314136125652,
-      "grad_norm": 0.006987118197520675,
+      "grad_norm": 0.005963379985835373,
       "learning_rate": 3.586698932127236e-06,
-      "logits/chosen": -13.625,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -800.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0008,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -840.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 13.0,
-      "rewards/rejected": -16.375,
+      "rewards/chosen": -3.796875,
+      "rewards/margins": 12.375,
+      "rewards/rejected": -16.25,
       "step": 2420
     },
     {
       "epoch": 2.5445026178010473,
-      "grad_norm": 0.02454198188363595,
+      "grad_norm": 0.019549430825252706,
       "learning_rate": 3.4310626834344744e-06,
-      "logits/chosen": -13.0625,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -864.0,
-      "logps/rejected": -3616.0,
-      "loss": 0.008,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -15.5625,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -10.125,
+      "logps/chosen": -880.0,
+      "logps/rejected": -3600.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.46875,
+      "rewards/margins": 11.9375,
+      "rewards/rejected": -15.4375,
       "step": 2430
     },
     {
       "epoch": 2.554973821989529,
-      "grad_norm": 0.03616404967795384,
+      "grad_norm": 0.03411308812933091,
       "learning_rate": 3.2786294373562343e-06,
-      "logits/chosen": -13.625,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -748.0,
-      "logps/rejected": -3744.0,
-      "loss": 0.0013,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.625,
+      "logps/chosen": -776.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 12.75,
-      "rewards/rejected": -16.0,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -15.875,
       "step": 2440
     },
     {
       "epoch": 2.5654450261780104,
-      "grad_norm": 0.018868867279177436,
+      "grad_norm": 0.018466467887488104,
       "learning_rate": 3.1294218303385526e-06,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -784.0,
-      "logps/rejected": -3744.0,
-      "loss": 0.0009,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -808.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -15.6875,
+      "rewards/chosen": -3.59375,
+      "rewards/margins": 12.375,
+      "rewards/rejected": -15.9375,
       "step": 2450
     },
     {
       "epoch": 2.5759162303664924,
-      "grad_norm": 0.09786363961612284,
+      "grad_norm": 0.004624965876436088,
       "learning_rate": 2.9834620198177483e-06,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -776.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -812.0,
       "logps/rejected": -3744.0,
-      "loss": 0.0013,
+      "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.203125,
-      "rewards/margins": 12.5,
+      "rewards/chosen": -3.59375,
+      "rewards/margins": 12.1875,
       "rewards/rejected": -15.75,
       "step": 2460
     },
     {
       "epoch": 2.5863874345549736,
-      "grad_norm": 0.0425125160908712,
+      "grad_norm": 0.065610167737027,
       "learning_rate": 2.840771680930068e-06,
-      "logits/chosen": -13.125,
-      "logits/rejected": -9.75,
-      "logps/chosen": -796.0,
-      "logps/rejected": -3552.0,
-      "loss": 0.001,
+      "logits/chosen": -13.25,
+      "logits/rejected": -10.375,
+      "logps/chosen": -828.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -15.625,
+      "rewards/chosen": -3.625,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -15.875,
       "step": 2470
     },
     {
       "epoch": 2.5968586387434556,
-      "grad_norm": 0.016020161645640844,
+      "grad_norm": 0.015089549160948276,
       "learning_rate": 2.701372003292868e-06,
-      "logits/chosen": -13.5,
-      "logits/rejected": -9.75,
-      "logps/chosen": -800.0,
-      "logps/rejected": -3792.0,
-      "loss": 0.0006,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.375,
+      "logps/chosen": -824.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 12.375,
-      "rewards/rejected": -15.8125,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 12.4375,
+      "rewards/rejected": -16.125,
       "step": 2480
     },
     {
       "epoch": 2.607329842931937,
-      "grad_norm": 0.05485646786669231,
+      "grad_norm": 0.017436324470715074,
       "learning_rate": 2.56528368785795e-06,
-      "logits/chosen": -13.125,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -808.0,
-      "logps/rejected": -3808.0,
-      "loss": 0.0015,
+      "logits/chosen": -13.25,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -840.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
+      "rewards/chosen": -3.703125,
       "rewards/margins": 12.4375,
-      "rewards/rejected": -15.8125,
+      "rewards/rejected": -16.125,
       "step": 2490
     },
     {
       "epoch": 2.6178010471204187,
-      "grad_norm": 0.029447658857205895,
+      "grad_norm": 0.022946925319352728,
       "learning_rate": 2.432526943837471e-06,
-      "logits/chosen": -13.4375,
-      "logits/rejected": -9.875,
-      "logps/chosen": -720.0,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -10.625,
+      "logps/chosen": -744.0,
       "logps/rejected": -3968.0,
       "loss": 0.0013,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.265625,
-      "rewards/margins": 13.5625,
+      "rewards/chosen": -3.515625,
+      "rewards/margins": 13.375,
       "rewards/rejected": -16.875,
       "step": 2500
     },
     {
       "epoch": 2.6178010471204187,
-      "eval_logits/chosen": -13.5625,
-      "eval_logits/rejected": -13.3125,
-      "eval_logps/chosen": -660.0,
-      "eval_logps/rejected": -656.0,
-      "eval_loss": 1.0018750429153442,
-      "eval_rewards/accuracies": 0.51171875,
-      "eval_rewards/chosen": -3.40625,
-      "eval_rewards/margins": 0.2431640625,
-      "eval_rewards/rejected": -3.65625,
-      "eval_runtime": 46.5655,
-      "eval_samples_per_second": 42.95,
+      "eval_logits/chosen": -13.25,
+      "eval_logits/rejected": -12.9375,
+      "eval_logps/chosen": -696.0,
+      "eval_logps/rejected": -684.0,
+      "eval_loss": 1.0499062538146973,
+      "eval_rewards/accuracies": 0.501953125,
+      "eval_rewards/chosen": -3.765625,
+      "eval_rewards/margins": 0.2080078125,
+      "eval_rewards/rejected": -3.96875,
+      "eval_runtime": 46.5994,
+      "eval_samples_per_second": 42.919,
       "eval_steps_per_second": 0.687,
       "step": 2500
     },
     {
       "epoch": 2.6282722513089007,
-      "grad_norm": 0.005800482150465527,
+      "grad_norm": 0.007284446731722391,
       "learning_rate": 2.3031214857028446e-06,
       "logits/chosen": -13.375,
-      "logits/rejected": -9.875,
-      "logps/chosen": -772.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.002,
+      "logits/rejected": -10.5,
+      "logps/chosen": -808.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 12.375,
-      "rewards/rejected": -15.75,
+      "rewards/chosen": -3.734375,
+      "rewards/margins": 12.0625,
+      "rewards/rejected": -15.8125,
       "step": 2510
     },
     {
       "epoch": 2.6387434554973823,
-      "grad_norm": 0.013866432950815213,
+      "grad_norm": 0.007606983108438646,
       "learning_rate": 2.177086530257139e-06,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -752.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0007,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.125,
+      "logps/chosen": -776.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.296875,
-      "rewards/margins": 12.4375,
-      "rewards/rejected": -15.75,
+      "rewards/chosen": -3.5625,
+      "rewards/margins": 12.5625,
+      "rewards/rejected": -16.125,
       "step": 2520
     },
     {
       "epoch": 2.649214659685864,
-      "grad_norm": 0.05500628156454972,
+      "grad_norm": 0.036126127718753606,
       "learning_rate": 2.054440793781354e-06,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -820.0,
-      "logps/rejected": -3776.0,
-      "loss": 0.0014,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -848.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.578125,
-      "rewards/margins": 12.4375,
-      "rewards/rejected": -16.0,
+      "rewards/chosen": -3.84375,
+      "rewards/margins": 12.75,
+      "rewards/rejected": -16.625,
       "step": 2530
     },
     {
       "epoch": 2.6596858638743455,
-      "grad_norm": 0.004091710513700706,
+      "grad_norm": 0.006594036638146291,
       "learning_rate": 1.9352024892550587e-06,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -812.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0019,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -844.0,
+      "logps/rejected": -3920.0,
+      "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 12.875,
-      "rewards/rejected": -16.375,
+      "rewards/chosen": -3.828125,
+      "rewards/margins": 12.8125,
+      "rewards/rejected": -16.625,
       "step": 2540
     },
     {
       "epoch": 2.670157068062827,
-      "grad_norm": 0.6674184407955883,
+      "grad_norm": 0.33805006129163256,
       "learning_rate": 1.8193893236517379e-06,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -9.875,
-      "logps/chosen": -772.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0037,
+      "logits/chosen": -13.0,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -804.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0015,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 12.5,
-      "rewards/rejected": -15.875,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 12.5625,
+      "rewards/rejected": -16.25,
       "step": 2550
     },
     {
       "epoch": 2.680628272251309,
-      "grad_norm": 0.01364424265323226,
+      "grad_norm": 0.01086983853481333,
       "learning_rate": 1.7070184953092956e-06,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.0,
-      "logps/chosen": -824.0,
-      "logps/rejected": -3664.0,
-      "loss": 0.001,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.5,
+      "logps/chosen": -856.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.390625,
-      "rewards/margins": 12.25,
-      "rewards/rejected": -15.625,
+      "rewards/chosen": -3.71875,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -16.0,
       "step": 2560
     },
     {
       "epoch": 2.6910994764397906,
-      "grad_norm": 0.046943393549352876,
+      "grad_norm": 0.023036521635480284,
       "learning_rate": 1.5981066913760918e-06,
-      "logits/chosen": -13.125,
-      "logits/rejected": -9.5,
-      "logps/chosen": -780.0,
-      "logps/rejected": -3584.0,
-      "loss": 0.0007,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -9.875,
+      "logps/chosen": -816.0,
+      "logps/rejected": -3632.0,
+      "loss": 0.0006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 11.875,
-      "rewards/rejected": -15.125,
+      "rewards/chosen": -3.5625,
+      "rewards/margins": 12.0,
+      "rewards/rejected": -15.625,
       "step": 2570
     },
     {
       "epoch": 2.701570680628272,
-      "grad_norm": 0.014697967818783861,
+      "grad_norm": 0.011260646237113996,
       "learning_rate": 1.492670085332884e-06,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -780.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0012,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.125,
+      "logps/chosen": -816.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 12.25,
-      "rewards/rejected": -15.6875,
+      "rewards/chosen": -3.734375,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -16.0,
       "step": 2580
     },
     {
       "epoch": 2.712041884816754,
-      "grad_norm": 0.2009031898381944,
+      "grad_norm": 0.12588057986423667,
       "learning_rate": 1.3907243345910787e-06,
       "logits/chosen": -13.25,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -796.0,
-      "logps/rejected": -3744.0,
-      "loss": 0.0014,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -844.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0011,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.3125,
-      "rewards/margins": 12.75,
-      "rewards/rejected": -16.0,
+      "rewards/chosen": -3.8125,
+      "rewards/margins": 12.5625,
+      "rewards/rejected": -16.375,
       "step": 2590
     },
     {
       "epoch": 2.7225130890052354,
-      "grad_norm": 0.0213273128699711,
+      "grad_norm": 0.027230290955720386,
       "learning_rate": 1.292284578167538e-06,
-      "logits/chosen": -13.375,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -716.0,
-      "logps/rejected": -3808.0,
-      "loss": 0.0038,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.625,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 12.8125,
-      "rewards/rejected": -16.0,
+      "rewards/chosen": -3.578125,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -16.25,
       "step": 2600
     },
     {
       "epoch": 2.7225130890052354,
-      "eval_logits/chosen": -13.625,
-      "eval_logits/rejected": -13.375,
-      "eval_logps/chosen": -660.0,
-      "eval_logps/rejected": -652.0,
-      "eval_loss": 1.0070937871932983,
-      "eval_rewards/accuracies": 0.505859375,
-      "eval_rewards/chosen": -3.40625,
-      "eval_rewards/margins": 0.2353515625,
-      "eval_rewards/rejected": -3.640625,
-      "eval_runtime": 46.5719,
-      "eval_samples_per_second": 42.944,
+      "eval_logits/chosen": -13.25,
+      "eval_logits/rejected": -12.9375,
+      "eval_logps/chosen": -696.0,
+      "eval_logps/rejected": -688.0,
+      "eval_loss": 1.0571562051773071,
+      "eval_rewards/accuracies": 0.49609375,
+      "eval_rewards/chosen": -3.78125,
+      "eval_rewards/margins": 0.2099609375,
+      "eval_rewards/rejected": -3.984375,
+      "eval_runtime": 46.5939,
+      "eval_samples_per_second": 42.924,
       "eval_steps_per_second": 0.687,
       "step": 2600
     },
     {
       "epoch": 2.7329842931937174,
-      "grad_norm": 0.028491461200941728,
+      "grad_norm": 0.032004474171721434,
       "learning_rate": 1.1973654344364925e-06,
-      "logits/chosen": -13.6875,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -788.0,
-      "logps/rejected": -3696.0,
-      "loss": 0.0012,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.375,
+      "logps/chosen": -816.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.234375,
+      "rewards/chosen": -3.53125,
       "rewards/margins": 12.25,
-      "rewards/rejected": -15.4375,
+      "rewards/rejected": -15.8125,
       "step": 2610
     },
     {
       "epoch": 2.743455497382199,
-      "grad_norm": 0.02339833657575842,
+      "grad_norm": 0.018529444258966386,
       "learning_rate": 1.1059809989586583e-06,
       "logits/chosen": -13.5625,
-      "logits/rejected": -10.125,
-      "logps/chosen": -728.0,
-      "logps/rejected": -3776.0,
-      "loss": 0.001,
+      "logits/rejected": -10.75,
+      "logps/chosen": -760.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.140625,
-      "rewards/margins": 12.875,
-      "rewards/rejected": -16.0,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 13.125,
+      "rewards/rejected": -16.625,
       "step": 2620
     },
     {
       "epoch": 2.7539267015706805,
-      "grad_norm": 0.03392001247428998,
+      "grad_norm": 0.019937213094845092,
       "learning_rate": 1.018144842388044e-06,
-      "logits/chosen": -13.375,
-      "logits/rejected": -9.875,
-      "logps/chosen": -740.0,
-      "logps/rejected": -3776.0,
-      "loss": 0.0007,
+      "logits/chosen": -13.125,
+      "logits/rejected": -10.375,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.265625,
-      "rewards/margins": 12.5625,
-      "rewards/rejected": -15.875,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -16.25,
       "step": 2630
     },
     {
       "epoch": 2.7643979057591626,
-      "grad_norm": 0.04739797262632349,
+      "grad_norm": 0.06258000332527579,
       "learning_rate": 9.338700084567109e-07,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -800.0,
-      "logps/rejected": -3824.0,
-      "loss": 0.0013,
+      "logits/chosen": -13.25,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -840.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.390625,
-      "rewards/margins": 12.875,
-      "rewards/rejected": -16.25,
+      "rewards/chosen": -3.796875,
+      "rewards/margins": 12.9375,
+      "rewards/rejected": -16.75,
       "step": 2640
     },
     {
       "epoch": 2.774869109947644,
-      "grad_norm": 0.06621260558475561,
+      "grad_norm": 0.046668411413204706,
       "learning_rate": 8.531690120377606e-07,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -808.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0017,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -848.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.125,
-      "rewards/margins": 13.0625,
-      "rewards/rejected": -16.125,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 12.8125,
+      "rewards/rejected": -16.375,
       "step": 2650
     },
     {
       "epoch": 2.7853403141361257,
-      "grad_norm": 0.019839730997041942,
+      "grad_norm": 0.06832244898536642,
       "learning_rate": 7.760538372868636e-07,
-      "logits/chosen": -13.625,
-      "logits/rejected": -10.125,
-      "logps/chosen": -704.0,
-      "logps/rejected": -3856.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3888.0,
       "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 13.1875,
-      "rewards/rejected": -16.5,
+      "rewards/chosen": -3.796875,
+      "rewards/margins": 13.125,
+      "rewards/rejected": -17.0,
       "step": 2660
     },
     {
       "epoch": 2.7958115183246073,
-      "grad_norm": 0.01559260064244627,
+      "grad_norm": 0.01187703063364007,
       "learning_rate": 7.025359358626166e-07,
-      "logits/chosen": -13.75,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -768.0,
-      "logps/rejected": -3744.0,
-      "loss": 0.0008,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.875,
+      "logps/chosen": -800.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 12.6875,
-      "rewards/rejected": -16.125,
+      "rewards/chosen": -3.703125,
+      "rewards/margins": 13.125,
+      "rewards/rejected": -16.75,
       "step": 2670
     },
-    {
-      "epoch": 2.806282722513089,
-      "grad_norm": 0.01055973669572818,
-      "learning_rate": 6.326262252259463e-07,
-      "logits/chosen": -13.6875,
-      "logits/rejected": -10.25,
-      "logps/chosen": -824.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0012,
+    {
+      "epoch": 2.806282722513089,
+      "grad_norm": 0.05649059788785088,
+      "learning_rate": 6.326262252259463e-07,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.9375,
+      "logps/chosen": -860.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 12.875,
-      "rewards/rejected": -16.125,
+      "rewards/chosen": -3.609375,
+      "rewards/margins": 12.9375,
+      "rewards/rejected": -16.625,
       "step": 2680
     },
     {
       "epoch": 2.816753926701571,
-      "grad_norm": 0.029388384625518307,
+      "grad_norm": 0.028528195853448034,
       "learning_rate": 5.66335087018871e-07,
-      "logits/chosen": -13.25,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -844.0,
-      "logps/rejected": -3824.0,
-      "loss": 0.0009,
+      "logits/chosen": -12.9375,
+      "logits/rejected": -10.25,
+      "logps/chosen": -876.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.46875,
-      "rewards/margins": 12.875,
-      "rewards/rejected": -16.375,
+      "rewards/chosen": -3.796875,
+      "rewards/margins": 13.125,
+      "rewards/rejected": -16.875,
       "step": 2690
     },
     {
       "epoch": 2.8272251308900525,
-      "grad_norm": 0.011543059380749308,
+      "grad_norm": 0.010474478856199,
       "learning_rate": 5.036723655228226e-07,
-      "logits/chosen": -13.375,
-      "logits/rejected": -9.5,
-      "logps/chosen": -792.0,
-      "logps/rejected": -3584.0,
-      "loss": 0.0009,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.0,
+      "logps/chosen": -836.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 11.75,
-      "rewards/rejected": -15.0625,
+      "rewards/chosen": -3.78125,
+      "rewards/margins": 12.0625,
+      "rewards/rejected": -15.875,
       "step": 2700
     },
     {
       "epoch": 2.8272251308900525,
-      "eval_logits/chosen": -13.625,
-      "eval_logits/rejected": -13.375,
-      "eval_logps/chosen": -660.0,
-      "eval_logps/rejected": -652.0,
-      "eval_loss": 1.008218765258789,
-      "eval_rewards/accuracies": 0.509765625,
-      "eval_rewards/chosen": -3.40625,
-      "eval_rewards/margins": 0.236328125,
-      "eval_rewards/rejected": -3.640625,
-      "eval_runtime": 46.5936,
-      "eval_samples_per_second": 42.924,
+      "eval_logits/chosen": -13.25,
+      "eval_logits/rejected": -12.9375,
+      "eval_logps/chosen": -700.0,
+      "eval_logps/rejected": -692.0,
+      "eval_loss": 1.0599687099456787,
+      "eval_rewards/accuracies": 0.501953125,
+      "eval_rewards/chosen": -3.796875,
+      "eval_rewards/margins": 0.216796875,
+      "eval_rewards/rejected": -4.0,
+      "eval_runtime": 46.6103,
+      "eval_samples_per_second": 42.909,
       "eval_steps_per_second": 0.687,
       "step": 2700
     },
     {
       "epoch": 2.837696335078534,
-      "grad_norm": 0.027333554374588787,
+      "grad_norm": 0.012358960745641006,
       "learning_rate": 4.446473661967432e-07,
-      "logits/chosen": -13.5,
-      "logits/rejected": -10.0,
-      "logps/chosen": -812.0,
-      "logps/rejected": -3760.0,
-      "loss": 0.001,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.75,
+      "logps/chosen": -852.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.453125,
-      "rewards/margins": 12.25,
-      "rewards/rejected": -15.75,
+      "rewards/chosen": -3.828125,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -16.125,
       "step": 2710
     },
     {
       "epoch": 2.8481675392670156,
-      "grad_norm": 0.015194606967890609,
+      "grad_norm": 0.016107187817336038,
       "learning_rate": 3.892688542952411e-07,
-      "logits/chosen": -13.625,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -844.0,
-      "logps/rejected": -3744.0,
-      "loss": 0.0008,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -872.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 12.5625,
-      "rewards/rejected": -16.0,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -16.5,
       "step": 2720
     },
     {
       "epoch": 2.858638743455497,
-      "grad_norm": 0.01950443993574235,
+      "grad_norm": 0.07607171066065449,
       "learning_rate": 3.3754505356694e-07,
-      "logits/chosen": -13.25,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -780.0,
-      "logps/rejected": -3664.0,
-      "loss": 0.0008,
+      "logits/chosen": -13.125,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -824.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0011,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.1875,
-      "rewards/margins": 12.875,
-      "rewards/rejected": -16.125,
+      "rewards/chosen": -3.609375,
+      "rewards/margins": 13.0,
+      "rewards/rejected": -16.625,
       "step": 2730
     },
     {
       "epoch": 2.869109947643979,
-      "grad_norm": 0.022576475096029308,
+      "grad_norm": 0.02410118962312118,
       "learning_rate": 2.8948364503322276e-07,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -9.75,
-      "logps/chosen": -856.0,
-      "logps/rejected": -3856.0,
-      "loss": 0.0014,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -10.375,
+      "logps/chosen": -896.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -15.8125,
+      "rewards/chosen": -3.890625,
+      "rewards/margins": 12.375,
+      "rewards/rejected": -16.25,
       "step": 2740
     },
     {
       "epoch": 2.8795811518324608,
-      "grad_norm": 0.013850195199160394,
+      "grad_norm": 0.007956080044025019,
       "learning_rate": 2.4509176584762706e-07,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -9.875,
-      "logps/chosen": -752.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0009,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.375,
+      "logps/chosen": -792.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 12.125,
-      "rewards/rejected": -15.375,
+      "rewards/chosen": -3.609375,
+      "rewards/margins": 12.1875,
+      "rewards/rejected": -15.8125,
       "step": 2750
     },
     {
       "epoch": 2.8900523560209423,
-      "grad_norm": 0.04347102184590706,
+      "grad_norm": 0.33030466704346784,
       "learning_rate": 2.0437600823595694e-07,
       "logits/chosen": -13.75,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -808.0,
-      "logps/rejected": -3968.0,
-      "loss": 0.0007,
+      "logits/rejected": -10.5,
+      "logps/chosen": -836.0,
+      "logps/rejected": -4000.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 13.25,
-      "rewards/rejected": -16.625,
+      "rewards/chosen": -3.703125,
+      "rewards/margins": 13.375,
+      "rewards/rejected": -17.0,
       "step": 2760
     },
     {
       "epoch": 2.900523560209424,
-      "grad_norm": 0.03540481134680033,
+      "grad_norm": 0.034380156740548166,
       "learning_rate": 1.6734241851733e-07,
       "logits/chosen": -13.5625,
-      "logits/rejected": -10.125,
-      "logps/chosen": -776.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0008,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -812.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.3125,
-      "rewards/margins": 12.625,
-      "rewards/rejected": -15.9375,
+      "rewards/chosen": -3.65625,
+      "rewards/margins": 12.75,
+      "rewards/rejected": -16.375,
       "step": 2770
     },
     {
       "epoch": 2.9109947643979055,
-      "grad_norm": 0.01574807485170882,
+      "grad_norm": 0.011547145623222973,
       "learning_rate": 1.339964962062984e-07,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -820.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.001,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -868.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0011,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.453125,
-      "rewards/margins": 12.875,
-      "rewards/rejected": -16.375,
+      "rewards/chosen": -3.921875,
+      "rewards/margins": 13.0625,
+      "rewards/rejected": -17.0,
       "step": 2780
     },
     {
       "epoch": 2.9214659685863875,
-      "grad_norm": 0.03463690015275206,
+      "grad_norm": 0.013750248878304804,
       "learning_rate": 1.0434319319617136e-07,
-      "logits/chosen": -13.875,
-      "logits/rejected": -10.0,
-      "logps/chosen": -768.0,
-      "logps/rejected": -3696.0,
-      "loss": 0.0006,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -812.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.265625,
+      "rewards/chosen": -3.671875,
       "rewards/margins": 12.5,
-      "rewards/rejected": -15.75,
+      "rewards/rejected": -16.125,
       "step": 2790
     },
     {
       "epoch": 2.931937172774869,
-      "grad_norm": 0.0080680306236487,
+      "grad_norm": 0.18045887484059875,
       "learning_rate": 7.838691302365363e-08,
-      "logits/chosen": -13.5,
-      "logits/rejected": -9.875,
-      "logps/chosen": -728.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0011,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -10.5,
+      "logps/chosen": -764.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.296875,
-      "rewards/margins": 12.625,
-      "rewards/rejected": -15.9375,
+      "rewards/chosen": -3.640625,
+      "rewards/margins": 12.6875,
+      "rewards/rejected": -16.375,
       "step": 2800
     },
     {
       "epoch": 2.931937172774869,
-      "eval_logits/chosen": -13.625,
-      "eval_logits/rejected": -13.375,
-      "eval_logps/chosen": -660.0,
-      "eval_logps/rejected": -652.0,
-      "eval_loss": 1.0087811946868896,
-      "eval_rewards/accuracies": 0.5078125,
-      "eval_rewards/chosen": -3.40625,
-      "eval_rewards/margins": 0.2353515625,
-      "eval_rewards/rejected": -3.640625,
-      "eval_runtime": 46.5682,
-      "eval_samples_per_second": 42.948,
+      "eval_logits/chosen": -13.25,
+      "eval_logits/rejected": -12.9375,
+      "eval_logps/chosen": -700.0,
+      "eval_logps/rejected": -692.0,
+      "eval_loss": 1.0608750581741333,
+      "eval_rewards/accuracies": 0.5,
+      "eval_rewards/chosen": -3.796875,
+      "eval_rewards/margins": 0.21484375,
+      "eval_rewards/rejected": -4.0,
+      "eval_runtime": 46.6093,
+      "eval_samples_per_second": 42.91,
       "eval_steps_per_second": 0.687,
       "step": 2800
     },
     {
       "epoch": 2.9424083769633507,
-      "grad_norm": 0.022797110033489408,
+      "grad_norm": 0.019399222258025355,
       "learning_rate": 5.613151021490459e-08,
-      "logits/chosen": -13.75,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -780.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0014,
+      "logits/chosen": -13.875,
+      "logits/rejected": -10.625,
+      "logps/chosen": -816.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 12.0,
-      "rewards/rejected": -15.4375,
+      "rewards/chosen": -3.78125,
+      "rewards/margins": 12.0625,
+      "rewards/rejected": -15.8125,
       "step": 2810
     },
     {
       "epoch": 2.9528795811518327,
-      "grad_norm": 0.012692747223129556,
+      "grad_norm": 0.016500443161887722,
       "learning_rate": 3.7580289713159965e-08,
       "logits/chosen": -13.625,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -736.0,
-      "logps/rejected": -3792.0,
-      "loss": 0.0009,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -776.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 12.6875,
-      "rewards/rejected": -15.8125,
+      "rewards/chosen": -3.5625,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -16.5,
       "step": 2820
     },
     {
       "epoch": 2.9633507853403143,
-      "grad_norm": 1.7651805279527109,
+      "grad_norm": 0.12894158601097658,
       "learning_rate": 2.2736006387927277e-08,
-      "logits/chosen": -13.0625,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -752.0,
-      "logps/rejected": -3584.0,
-      "loss": 0.0051,
+      "logits/chosen": -12.875,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -796.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 11.9375,
-      "rewards/rejected": -15.25,
+      "rewards/chosen": -3.703125,
+      "rewards/margins": 12.125,
+      "rewards/rejected": -15.8125,
       "step": 2830
     },
     {
       "epoch": 2.973821989528796,
-      "grad_norm": 0.04429116103492594,
+      "grad_norm": 0.05654352742444268,
       "learning_rate": 1.1600864625893604e-08,
-      "logits/chosen": -13.6875,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -832.0,
-      "logps/rejected": -3760.0,
-      "loss": 0.0011,
+      "logits/chosen": -13.875,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -880.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.265625,
-      "rewards/margins": 12.0,
-      "rewards/rejected": -15.3125,
+      "rewards/chosen": -3.75,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -16.0,
       "step": 2840
     },
     {
       "epoch": 2.9842931937172774,
-      "grad_norm": 0.011064137488642318,
+      "grad_norm": 0.009863553480644582,
       "learning_rate": 4.176518003573549e-09,
       "logits/chosen": -13.75,
-      "logits/rejected": -10.0,
-      "logps/chosen": -784.0,
-      "logps/rejected": -3712.0,
-      "loss": 0.0008,
+      "logits/rejected": -10.625,
+      "logps/chosen": -824.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 12.1875,
-      "rewards/rejected": -15.625,
+      "rewards/chosen": -3.78125,
+      "rewards/margins": 12.4375,
+      "rewards/rejected": -16.25,
       "step": 2850
     },
     {
       "epoch": 2.994764397905759,
-      "grad_norm": 0.03583142898507438,
+      "grad_norm": 0.030090935222444017,
       "learning_rate": 4.640690417528903e-10,
-      "logits/chosen": -13.6875,
-      "logits/rejected": -10.5,
-      "logps/chosen": -740.0,
-      "logps/rejected": -3584.0,
-      "loss": 0.0018,
+      "logits/chosen": -13.75,
+      "logits/rejected": -11.25,
+      "logps/chosen": -776.0,
+      "logps/rejected": -3632.0,
+      "loss": 0.0017,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.09375,
-      "rewards/margins": 12.1875,
-      "rewards/rejected": -15.3125,
+      "rewards/chosen": -3.46875,
+      "rewards/margins": 12.375,
+      "rewards/rejected": -15.875,
       "step": 2860
     },
     {
       "epoch": 3.0,
       "step": 2865,
       "total_flos": 0.0,
-      "train_loss": 0.06583595570358879,
-      "train_runtime": 12397.8277,
-      "train_samples_per_second": 14.789,
+      "train_loss": 0.06693936966889477,
+      "train_runtime": 12411.6425,
+      "train_samples_per_second": 14.773,
       "train_steps_per_second": 0.231
     }
   ],