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": 2.1275672708481026,
+      "grad_norm": 2.1276682428831886,
       "learning_rate": 1.7421602787456448e-07,
       "logits/chosen": -11.9375,
       "logits/rejected": -10.5,
@@ -25,89 +25,89 @@
     },
     {
       "epoch": 0.010471204188481676,
-      "grad_norm": 1.8709024687890357,
+      "grad_norm": 1.8750029262029384,
       "learning_rate": 1.7421602787456445e-06,
       "logits/chosen": -11.1875,
       "logits/rejected": -10.125,
       "logps/chosen": -420.0,
       "logps/rejected": -924.0,
-      "loss": 0.691,
-      "rewards/accuracies": 0.2152777761220932,
-      "rewards/chosen": 0.00092315673828125,
-      "rewards/margins": 0.0026397705078125,
-      "rewards/rejected": -0.0017242431640625,
+      "loss": 0.6918,
+      "rewards/accuracies": 0.2013888955116272,
+      "rewards/chosen": 0.00089263916015625,
+      "rewards/margins": 0.00177764892578125,
+      "rewards/rejected": -0.000888824462890625,
       "step": 10
     },
     {
       "epoch": 0.020942408376963352,
-      "grad_norm": 2.107464740845283,
+      "grad_norm": 2.1248558631444343,
       "learning_rate": 3.484320557491289e-06,
       "logits/chosen": -11.0625,
       "logits/rejected": -10.5,
       "logps/chosen": -394.0,
       "logps/rejected": -752.0,
-      "loss": 0.6807,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": -0.001800537109375,
-      "rewards/margins": 0.01953125,
-      "rewards/rejected": -0.021240234375,
+      "loss": 0.6815,
+      "rewards/accuracies": 0.46875,
+      "rewards/chosen": -0.00194549560546875,
+      "rewards/margins": 0.0206298828125,
+      "rewards/rejected": -0.0224609375,
       "step": 20
     },
     {
       "epoch": 0.031413612565445025,
-      "grad_norm": 1.9212830322522017,
+      "grad_norm": 1.9210415969526244,
       "learning_rate": 5.226480836236934e-06,
       "logits/chosen": -11.8125,
       "logits/rejected": -11.0,
       "logps/chosen": -418.0,
       "logps/rejected": -796.0,
-      "loss": 0.6293,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -0.042236328125,
-      "rewards/margins": 0.1337890625,
-      "rewards/rejected": -0.17578125,
+      "loss": 0.6297,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -0.042724609375,
+      "rewards/margins": 0.1357421875,
+      "rewards/rejected": -0.1787109375,
       "step": 30
     },
     {
       "epoch": 0.041884816753926704,
-      "grad_norm": 1.9232137390501616,
+      "grad_norm": 1.924661414596908,
       "learning_rate": 6.968641114982578e-06,
       "logits/chosen": -12.25,
       "logits/rejected": -11.4375,
       "logps/chosen": -444.0,
       "logps/rejected": -896.0,
-      "loss": 0.5069,
-      "rewards/accuracies": 0.768750011920929,
+      "loss": 0.5074,
+      "rewards/accuracies": 0.762499988079071,
       "rewards/chosen": -0.32421875,
-      "rewards/margins": 0.6171875,
-      "rewards/rejected": -0.94140625,
+      "rewards/margins": 0.62109375,
+      "rewards/rejected": -0.9453125,
       "step": 40
     },
     {
       "epoch": 0.05235602094240838,
-      "grad_norm": 1.7487637151756203,
+      "grad_norm": 1.754615446701247,
       "learning_rate": 8.710801393728225e-06,
       "logits/chosen": -12.0625,
       "logits/rejected": -11.4375,
       "logps/chosen": -482.0,
       "logps/rejected": -1200.0,
-      "loss": 0.4419,
-      "rewards/accuracies": 0.8125,
+      "loss": 0.4423,
+      "rewards/accuracies": 0.8187500238418579,
       "rewards/chosen": -0.734375,
-      "rewards/margins": 2.0,
-      "rewards/rejected": -2.734375,
+      "rewards/margins": 1.9921875,
+      "rewards/rejected": -2.71875,
       "step": 50
     },
     {
       "epoch": 0.06282722513089005,
-      "grad_norm": 1.677537690526415,
+      "grad_norm": 1.6351575213568288,
       "learning_rate": 1.0452961672473868e-05,
       "logits/chosen": -13.75,
       "logits/rejected": -13.125,
       "logps/chosen": -416.0,
       "logps/rejected": -1016.0,
-      "loss": 0.3986,
-      "rewards/accuracies": 0.8062499761581421,
+      "loss": 0.3985,
+      "rewards/accuracies": 0.800000011920929,
       "rewards/chosen": -0.73828125,
       "rewards/margins": 1.9921875,
       "rewards/rejected": -2.734375,
@@ -115,4660 +115,4660 @@
     },
     {
       "epoch": 0.07329842931937172,
-      "grad_norm": 2.766104656622922,
+      "grad_norm": 2.74112848466254,
       "learning_rate": 1.2195121951219513e-05,
       "logits/chosen": -13.25,
       "logits/rejected": -12.875,
       "logps/chosen": -540.0,
       "logps/rejected": -1168.0,
-      "loss": 0.3694,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -1.1640625,
-      "rewards/margins": 2.625,
+      "loss": 0.3698,
+      "rewards/accuracies": 0.831250011920929,
+      "rewards/chosen": -1.171875,
+      "rewards/margins": 2.640625,
       "rewards/rejected": -3.796875,
       "step": 70
     },
     {
       "epoch": 0.08376963350785341,
-      "grad_norm": 2.390169749446497,
+      "grad_norm": 2.0451937172911188,
       "learning_rate": 1.3937282229965156e-05,
-      "logits/chosen": -13.25,
-      "logits/rejected": -12.8125,
-      "logps/chosen": -532.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -12.875,
+      "logps/chosen": -528.0,
       "logps/rejected": -1384.0,
-      "loss": 0.3693,
+      "loss": 0.3678,
       "rewards/accuracies": 0.84375,
-      "rewards/chosen": -1.5078125,
+      "rewards/chosen": -1.484375,
       "rewards/margins": 3.390625,
-      "rewards/rejected": -4.90625,
+      "rewards/rejected": -4.875,
       "step": 80
     },
     {
       "epoch": 0.09424083769633508,
-      "grad_norm": 2.5414529675284316,
+      "grad_norm": 2.6429876256364793,
       "learning_rate": 1.56794425087108e-05,
-      "logits/chosen": -11.4375,
-      "logits/rejected": -11.0625,
-      "logps/chosen": -520.0,
+      "logits/chosen": -11.625,
+      "logits/rejected": -11.25,
+      "logps/chosen": -516.0,
       "logps/rejected": -1128.0,
-      "loss": 0.3722,
-      "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": -1.7109375,
-      "rewards/margins": 2.828125,
+      "loss": 0.3701,
+      "rewards/accuracies": 0.831250011920929,
+      "rewards/chosen": -1.671875,
+      "rewards/margins": 2.859375,
       "rewards/rejected": -4.53125,
       "step": 90
     },
     {
       "epoch": 0.10471204188481675,
-      "grad_norm": 2.0638622069397927,
+      "grad_norm": 1.9326753753501016,
       "learning_rate": 1.742160278745645e-05,
-      "logits/chosen": -12.3125,
-      "logits/rejected": -11.625,
-      "logps/chosen": -684.0,
-      "logps/rejected": -1552.0,
-      "loss": 0.3697,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -2.25,
-      "rewards/margins": 3.78125,
-      "rewards/rejected": -6.03125,
+      "logits/chosen": -12.0,
+      "logits/rejected": -11.3125,
+      "logps/chosen": -680.0,
+      "logps/rejected": -1560.0,
+      "loss": 0.3683,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -2.203125,
+      "rewards/margins": 3.921875,
+      "rewards/rejected": -6.125,
       "step": 100
     },
     {
       "epoch": 0.10471204188481675,
-      "eval_logits/chosen": -13.1875,
-      "eval_logits/rejected": -12.875,
-      "eval_logps/chosen": -544.0,
-      "eval_logps/rejected": -548.0,
-      "eval_loss": 0.6860781311988831,
-      "eval_rewards/accuracies": 0.595703125,
-      "eval_rewards/chosen": -2.265625,
-      "eval_rewards/margins": 0.32421875,
-      "eval_rewards/rejected": -2.59375,
-      "eval_runtime": 48.6622,
-      "eval_samples_per_second": 41.1,
-      "eval_steps_per_second": 0.658,
+      "eval_logits/chosen": -12.4375,
+      "eval_logits/rejected": -12.125,
+      "eval_logps/chosen": -500.0,
+      "eval_logps/rejected": -506.0,
+      "eval_loss": 0.679937481880188,
+      "eval_rewards/accuracies": 0.6015625,
+      "eval_rewards/chosen": -1.8125,
+      "eval_rewards/margins": 0.35546875,
+      "eval_rewards/rejected": -2.171875,
+      "eval_runtime": 60.1197,
+      "eval_samples_per_second": 33.267,
+      "eval_steps_per_second": 0.532,
       "step": 100
     },
     {
       "epoch": 0.11518324607329843,
-      "grad_norm": 2.6015685064611267,
+      "grad_norm": 2.7418254058127096,
       "learning_rate": 1.9163763066202093e-05,
-      "logits/chosen": -13.0625,
-      "logits/rejected": -12.0625,
-      "logps/chosen": -612.0,
-      "logps/rejected": -1568.0,
-      "loss": 0.3155,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -2.015625,
-      "rewards/margins": 4.1875,
-      "rewards/rejected": -6.21875,
+      "logits/chosen": -12.8125,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -632.0,
+      "logps/rejected": -1632.0,
+      "loss": 0.31,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -2.203125,
+      "rewards/margins": 4.625,
+      "rewards/rejected": -6.8125,
       "step": 110
     },
     {
       "epoch": 0.1256544502617801,
-      "grad_norm": 3.3957294670018294,
+      "grad_norm": 2.398332257221043,
       "learning_rate": 2.0905923344947736e-05,
-      "logits/chosen": -11.875,
-      "logits/rejected": -11.375,
-      "logps/chosen": -616.0,
-      "logps/rejected": -1440.0,
-      "loss": 0.3465,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -2.546875,
-      "rewards/margins": 3.796875,
-      "rewards/rejected": -6.34375,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -12.1875,
+      "logps/chosen": -592.0,
+      "logps/rejected": -1416.0,
+      "loss": 0.3306,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -2.28125,
+      "rewards/margins": 3.875,
+      "rewards/rejected": -6.15625,
       "step": 120
     },
     {
       "epoch": 0.13612565445026178,
-      "grad_norm": 4.084766543736936,
+      "grad_norm": 3.403103987665289,
       "learning_rate": 2.264808362369338e-05,
-      "logits/chosen": -12.4375,
-      "logits/rejected": -12.0,
-      "logps/chosen": -616.0,
-      "logps/rejected": -1384.0,
-      "loss": 0.3722,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -2.171875,
-      "rewards/margins": 3.5625,
-      "rewards/rejected": -5.75,
+      "logits/chosen": -12.625,
+      "logits/rejected": -12.1875,
+      "logps/chosen": -620.0,
+      "logps/rejected": -1408.0,
+      "loss": 0.3639,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -2.1875,
+      "rewards/margins": 3.78125,
+      "rewards/rejected": -5.96875,
       "step": 130
     },
     {
       "epoch": 0.14659685863874344,
-      "grad_norm": 3.185355710699422,
+      "grad_norm": 3.903042233408565,
       "learning_rate": 2.4390243902439026e-05,
-      "logits/chosen": -14.0625,
-      "logits/rejected": -13.25,
-      "logps/chosen": -836.0,
-      "logps/rejected": -1616.0,
-      "loss": 0.3163,
-      "rewards/accuracies": 0.831250011920929,
-      "rewards/chosen": -4.03125,
-      "rewards/margins": 4.03125,
-      "rewards/rejected": -8.0625,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -14.0625,
+      "logps/chosen": -704.0,
+      "logps/rejected": -1424.0,
+      "loss": 0.3137,
+      "rewards/accuracies": 0.8687499761581421,
+      "rewards/chosen": -2.65625,
+      "rewards/margins": 3.453125,
+      "rewards/rejected": -6.09375,
       "step": 140
     },
     {
       "epoch": 0.15706806282722513,
-      "grad_norm": 3.3800270551568685,
+      "grad_norm": 6.418882409745814,
       "learning_rate": 2.6132404181184672e-05,
-      "logits/chosen": -11.6875,
-      "logits/rejected": -10.625,
-      "logps/chosen": -700.0,
-      "logps/rejected": -1464.0,
-      "loss": 0.3539,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -2.796875,
-      "rewards/margins": 3.171875,
-      "rewards/rejected": -5.96875,
+      "logits/chosen": -12.5625,
+      "logits/rejected": -11.5,
+      "logps/chosen": -684.0,
+      "logps/rejected": -1488.0,
+      "loss": 0.3395,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": -2.640625,
+      "rewards/margins": 3.625,
+      "rewards/rejected": -6.28125,
       "step": 150
     },
     {
       "epoch": 0.16753926701570682,
-      "grad_norm": 2.265952903747681,
+      "grad_norm": 3.0491594989339066,
       "learning_rate": 2.7874564459930312e-05,
-      "logits/chosen": -9.5625,
-      "logits/rejected": -8.875,
-      "logps/chosen": -652.0,
-      "logps/rejected": -1456.0,
-      "loss": 0.2772,
-      "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": -2.8125,
-      "rewards/margins": 3.8125,
-      "rewards/rejected": -6.625,
+      "logits/chosen": -9.75,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -640.0,
+      "logps/rejected": -1416.0,
+      "loss": 0.2877,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -2.703125,
+      "rewards/margins": 3.484375,
+      "rewards/rejected": -6.1875,
       "step": 160
     },
     {
       "epoch": 0.17801047120418848,
-      "grad_norm": 2.014823061831193,
+      "grad_norm": 1.8346193011766518,
       "learning_rate": 2.9616724738675962e-05,
-      "logits/chosen": -12.0,
-      "logits/rejected": -11.1875,
-      "logps/chosen": -732.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.2914,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -3.09375,
-      "rewards/margins": 5.28125,
-      "rewards/rejected": -8.375,
+      "logits/chosen": -9.0625,
+      "logits/rejected": -8.75,
+      "logps/chosen": -764.0,
+      "logps/rejected": -1736.0,
+      "loss": 0.2828,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 5.1875,
+      "rewards/rejected": -8.625,
       "step": 170
     },
     {
       "epoch": 0.18848167539267016,
-      "grad_norm": 2.2620708013800574,
+      "grad_norm": 2.0804876833074317,
       "learning_rate": 3.13588850174216e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -732.0,
-      "logps/rejected": -1632.0,
-      "loss": 0.2629,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 4.21875,
-      "rewards/rejected": -7.375,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -10.625,
+      "logps/chosen": -680.0,
+      "logps/rejected": -1520.0,
+      "loss": 0.2797,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -2.640625,
+      "rewards/margins": 3.640625,
+      "rewards/rejected": -6.28125,
       "step": 180
     },
     {
       "epoch": 0.19895287958115182,
-      "grad_norm": 4.2711285515371875,
+      "grad_norm": 4.423814467899269,
       "learning_rate": 3.310104529616725e-05,
-      "logits/chosen": -9.8125,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -868.0,
-      "logps/rejected": -1888.0,
-      "loss": 0.2894,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -4.875,
-      "rewards/margins": 5.1875,
-      "rewards/rejected": -10.0625,
+      "logits/chosen": -12.0,
+      "logits/rejected": -11.125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -1696.0,
+      "loss": 0.3007,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 4.90625,
+      "rewards/rejected": -8.125,
       "step": 190
     },
     {
       "epoch": 0.2094240837696335,
-      "grad_norm": 1.768027534048622,
+      "grad_norm": 1.6410440434525142,
       "learning_rate": 3.48432055749129e-05,
-      "logits/chosen": -9.0,
-      "logits/rejected": -8.3125,
-      "logps/chosen": -792.0,
-      "logps/rejected": -1616.0,
-      "loss": 0.2859,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -3.453125,
-      "rewards/margins": 4.0,
-      "rewards/rejected": -7.46875,
+      "logits/chosen": -11.4375,
+      "logits/rejected": -10.875,
+      "logps/chosen": -720.0,
+      "logps/rejected": -1576.0,
+      "loss": 0.2926,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -2.734375,
+      "rewards/margins": 4.3125,
+      "rewards/rejected": -7.0625,
       "step": 200
     },
     {
       "epoch": 0.2094240837696335,
-      "eval_logits/chosen": -7.90625,
-      "eval_logits/rejected": -7.125,
-      "eval_logps/chosen": -600.0,
-      "eval_logps/rejected": -620.0,
-      "eval_loss": 0.7472500205039978,
-      "eval_rewards/accuracies": 0.595703125,
-      "eval_rewards/chosen": -2.828125,
-      "eval_rewards/margins": 0.50390625,
-      "eval_rewards/rejected": -3.328125,
-      "eval_runtime": 46.6136,
-      "eval_samples_per_second": 42.906,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -10.5625,
+      "eval_logits/rejected": -10.0,
+      "eval_logps/chosen": -520.0,
+      "eval_logps/rejected": -540.0,
+      "eval_loss": 0.7127344012260437,
+      "eval_rewards/accuracies": 0.615234375,
+      "eval_rewards/chosen": -2.03125,
+      "eval_rewards/margins": 0.486328125,
+      "eval_rewards/rejected": -2.515625,
+      "eval_runtime": 58.158,
+      "eval_samples_per_second": 34.389,
+      "eval_steps_per_second": 0.55,
       "step": 200
     },
     {
       "epoch": 0.2198952879581152,
-      "grad_norm": 2.11950620788272,
+      "grad_norm": 2.1741794885260064,
       "learning_rate": 3.6585365853658535e-05,
-      "logits/chosen": -8.125,
-      "logits/rejected": -7.125,
-      "logps/chosen": -672.0,
-      "logps/rejected": -1392.0,
-      "loss": 0.3313,
+      "logits/chosen": -10.375,
+      "logits/rejected": -9.5,
+      "logps/chosen": -592.0,
+      "logps/rejected": -1320.0,
+      "loss": 0.29,
       "rewards/accuracies": 0.831250011920929,
-      "rewards/chosen": -3.203125,
-      "rewards/margins": 3.515625,
-      "rewards/rejected": -6.71875,
+      "rewards/chosen": -2.40625,
+      "rewards/margins": 3.578125,
+      "rewards/rejected": -6.0,
       "step": 210
     },
     {
       "epoch": 0.23036649214659685,
-      "grad_norm": 1.9674883328395525,
+      "grad_norm": 2.626176158470523,
       "learning_rate": 3.8327526132404185e-05,
-      "logits/chosen": -9.0,
-      "logits/rejected": -7.78125,
-      "logps/chosen": -652.0,
-      "logps/rejected": -1472.0,
-      "loss": 0.2898,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -2.46875,
-      "rewards/margins": 3.53125,
-      "rewards/rejected": -6.0,
+      "logits/chosen": -10.375,
+      "logits/rejected": -9.125,
+      "logps/chosen": -664.0,
+      "logps/rejected": -1520.0,
+      "loss": 0.3032,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -2.59375,
+      "rewards/margins": 3.859375,
+      "rewards/rejected": -6.4375,
       "step": 220
     },
     {
       "epoch": 0.24083769633507854,
-      "grad_norm": 5.412307877054619,
+      "grad_norm": 3.2918938468661696,
       "learning_rate": 4.006968641114983e-05,
-      "logits/chosen": -5.375,
-      "logits/rejected": -5.0625,
-      "logps/chosen": -760.0,
-      "logps/rejected": -1552.0,
-      "loss": 0.3269,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -3.859375,
-      "rewards/margins": 4.15625,
-      "rewards/rejected": -8.0,
+      "logits/chosen": -9.75,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -648.0,
+      "logps/rejected": -1336.0,
+      "loss": 0.2952,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -2.75,
+      "rewards/margins": 3.109375,
+      "rewards/rejected": -5.84375,
       "step": 230
     },
     {
       "epoch": 0.2513089005235602,
-      "grad_norm": 4.378793224788044,
+      "grad_norm": 2.3758475215042987,
       "learning_rate": 4.181184668989547e-05,
-      "logits/chosen": -7.59375,
-      "logits/rejected": -6.8125,
-      "logps/chosen": -792.0,
-      "logps/rejected": -1536.0,
-      "loss": 0.324,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -4.28125,
-      "rewards/margins": 3.703125,
-      "rewards/rejected": -7.96875,
+      "logits/chosen": -8.5,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -656.0,
+      "logps/rejected": -1408.0,
+      "loss": 0.3021,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 3.75,
+      "rewards/rejected": -6.65625,
       "step": 240
     },
     {
       "epoch": 0.2617801047120419,
-      "grad_norm": 2.8729764420490755,
+      "grad_norm": 3.528892969360077,
       "learning_rate": 4.3554006968641115e-05,
-      "logits/chosen": -8.0,
-      "logits/rejected": -7.0,
-      "logps/chosen": -736.0,
-      "logps/rejected": -1520.0,
-      "loss": 0.2574,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -3.0,
-      "rewards/margins": 3.46875,
-      "rewards/rejected": -6.46875,
+      "logits/chosen": -7.34375,
+      "logits/rejected": -6.4375,
+      "logps/chosen": -744.0,
+      "logps/rejected": -1624.0,
+      "loss": 0.2505,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -3.078125,
+      "rewards/margins": 4.40625,
+      "rewards/rejected": -7.46875,
       "step": 250
     },
     {
       "epoch": 0.27225130890052357,
-      "grad_norm": 1.850007513298479,
+      "grad_norm": 1.8292864225434013,
       "learning_rate": 4.529616724738676e-05,
-      "logits/chosen": -10.5625,
-      "logits/rejected": -9.25,
-      "logps/chosen": -664.0,
-      "logps/rejected": -1600.0,
-      "loss": 0.2776,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -3.03125,
-      "rewards/margins": 4.71875,
-      "rewards/rejected": -7.75,
+      "logits/chosen": -7.59375,
+      "logits/rejected": -6.90625,
+      "logps/chosen": -600.0,
+      "logps/rejected": -1544.0,
+      "loss": 0.2695,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -2.40625,
+      "rewards/margins": 4.8125,
+      "rewards/rejected": -7.21875,
       "step": 260
     },
     {
       "epoch": 0.28272251308900526,
-      "grad_norm": 2.271952464084681,
+      "grad_norm": 2.065058180685675,
       "learning_rate": 4.703832752613241e-05,
-      "logits/chosen": -10.5,
-      "logits/rejected": -8.5,
-      "logps/chosen": -660.0,
-      "logps/rejected": -1520.0,
-      "loss": 0.3377,
-      "rewards/accuracies": 0.831250011920929,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -7.5625,
+      "logits/chosen": -8.375,
+      "logits/rejected": -7.4375,
+      "logps/chosen": -668.0,
+      "logps/rejected": -1568.0,
+      "loss": 0.3388,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 5.03125,
+      "rewards/rejected": -8.0,
       "step": 270
     },
     {
       "epoch": 0.2931937172774869,
-      "grad_norm": 3.299374903793561,
+      "grad_norm": 2.7322746051379525,
       "learning_rate": 4.878048780487805e-05,
-      "logits/chosen": -6.6875,
-      "logits/rejected": -6.125,
-      "logps/chosen": -632.0,
-      "logps/rejected": -1376.0,
-      "loss": 0.3016,
-      "rewards/accuracies": 0.8500000238418579,
-      "rewards/chosen": -2.765625,
-      "rewards/margins": 3.6875,
-      "rewards/rejected": -6.4375,
+      "logits/chosen": -8.125,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -644.0,
+      "logps/rejected": -1344.0,
+      "loss": 0.3191,
+      "rewards/accuracies": 0.831250011920929,
+      "rewards/chosen": -2.875,
+      "rewards/margins": 3.296875,
+      "rewards/rejected": -6.15625,
       "step": 280
     },
     {
       "epoch": 0.3036649214659686,
-      "grad_norm": 4.661528799521329,
+      "grad_norm": 2.7015676633858963,
       "learning_rate": 4.999983293481418e-05,
-      "logits/chosen": -7.625,
-      "logits/rejected": -6.875,
-      "logps/chosen": -860.0,
-      "logps/rejected": -2000.0,
-      "loss": 0.2479,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -4.875,
-      "rewards/margins": 6.65625,
-      "rewards/rejected": -11.5,
+      "logits/chosen": -9.4375,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -736.0,
+      "logps/rejected": -1640.0,
+      "loss": 0.2588,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -3.609375,
+      "rewards/margins": 4.3125,
+      "rewards/rejected": -7.9375,
       "step": 290
     },
     {
       "epoch": 0.31413612565445026,
-      "grad_norm": 3.7461152856782407,
+      "grad_norm": 2.7096302152822376,
       "learning_rate": 4.9996862949181595e-05,
-      "logits/chosen": -7.09375,
-      "logits/rejected": -6.3125,
-      "logps/chosen": -792.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.2457,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -4.0,
-      "rewards/margins": 5.625,
-      "rewards/rejected": -9.625,
+      "logits/chosen": -8.625,
+      "logits/rejected": -7.71875,
+      "logps/chosen": -832.0,
+      "logps/rejected": -1688.0,
+      "loss": 0.2695,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": -4.40625,
+      "rewards/margins": 4.4375,
+      "rewards/rejected": -8.8125,
       "step": 300
     },
     {
       "epoch": 0.31413612565445026,
-      "eval_logits/chosen": -5.875,
-      "eval_logits/rejected": -4.875,
-      "eval_logps/chosen": -744.0,
-      "eval_logps/rejected": -772.0,
-      "eval_loss": 0.89453125,
-      "eval_rewards/accuracies": 0.576171875,
-      "eval_rewards/chosen": -4.25,
+      "eval_logits/chosen": -8.1875,
+      "eval_logits/rejected": -7.28125,
+      "eval_logps/chosen": -776.0,
+      "eval_logps/rejected": -804.0,
+      "eval_loss": 0.7960312366485596,
+      "eval_rewards/accuracies": 0.580078125,
+      "eval_rewards/chosen": -4.59375,
       "eval_rewards/margins": 0.578125,
-      "eval_rewards/rejected": -4.84375,
-      "eval_runtime": 46.6043,
-      "eval_samples_per_second": 42.914,
-      "eval_steps_per_second": 0.687,
+      "eval_rewards/rejected": -5.15625,
+      "eval_runtime": 58.151,
+      "eval_samples_per_second": 34.393,
+      "eval_steps_per_second": 0.55,
       "step": 300
     },
     {
       "epoch": 0.32460732984293195,
-      "grad_norm": 1.474457676792537,
+      "grad_norm": 1.8544285474936952,
       "learning_rate": 4.9990180911526896e-05,
-      "logits/chosen": -9.0625,
-      "logits/rejected": -8.125,
-      "logps/chosen": -760.0,
-      "logps/rejected": -1576.0,
-      "loss": 0.2937,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -3.96875,
-      "rewards/margins": 3.90625,
-      "rewards/rejected": -7.875,
+      "logits/chosen": -10.125,
+      "logits/rejected": -9.25,
+      "logps/chosen": -680.0,
+      "logps/rejected": -1456.0,
+      "loss": 0.2941,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 3.5,
+      "rewards/rejected": -6.6875,
       "step": 310
     },
     {
       "epoch": 0.33507853403141363,
-      "grad_norm": 3.567820537682467,
+      "grad_norm": 2.6712645554320433,
       "learning_rate": 4.997978781413745e-05,
-      "logits/chosen": -13.75,
-      "logits/rejected": -12.5,
-      "logps/chosen": -940.0,
-      "logps/rejected": -1800.0,
-      "loss": 0.3499,
-      "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -5.09375,
-      "rewards/margins": 5.40625,
-      "rewards/rejected": -10.5,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -8.125,
+      "logps/chosen": -900.0,
+      "logps/rejected": -1648.0,
+      "loss": 0.3032,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 4.25,
+      "rewards/rejected": -9.0,
       "step": 320
     },
     {
       "epoch": 0.34554973821989526,
-      "grad_norm": 3.2355963440350206,
+      "grad_norm": 3.113540399408468,
       "learning_rate": 4.996568520039565e-05,
-      "logits/chosen": -12.6875,
-      "logits/rejected": -11.125,
-      "logps/chosen": -760.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.2891,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -3.65625,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -8.0625,
+      "logits/chosen": -10.625,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -708.0,
+      "logps/rejected": -1608.0,
+      "loss": 0.2688,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 3.890625,
+      "rewards/rejected": -7.0,
       "step": 330
     },
     {
       "epoch": 0.35602094240837695,
-      "grad_norm": 2.9976143195948883,
+      "grad_norm": 2.5354754268697475,
       "learning_rate": 4.994787516454973e-05,
-      "logits/chosen": -7.9375,
-      "logits/rejected": -6.96875,
-      "logps/chosen": -852.0,
-      "logps/rejected": -1592.0,
-      "loss": 0.2829,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -4.625,
-      "rewards/margins": 3.921875,
-      "rewards/rejected": -8.5625,
+      "logits/chosen": -10.75,
+      "logits/rejected": -9.625,
+      "logps/chosen": -860.0,
+      "logps/rejected": -1656.0,
+      "loss": 0.2577,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 4.5,
+      "rewards/rejected": -9.1875,
       "step": 340
     },
     {
       "epoch": 0.36649214659685864,
-      "grad_norm": 2.698035615458425,
+      "grad_norm": 2.855398523244536,
       "learning_rate": 4.992636035140273e-05,
-      "logits/chosen": -9.875,
-      "logits/rejected": -8.5,
-      "logps/chosen": -696.0,
-      "logps/rejected": -1600.0,
-      "loss": 0.2796,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -7.9375,
+      "logits/chosen": -10.875,
+      "logits/rejected": -9.625,
+      "logps/chosen": -800.0,
+      "logps/rejected": -1832.0,
+      "loss": 0.2449,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 5.71875,
+      "rewards/rejected": -10.25,
       "step": 350
     },
     {
       "epoch": 0.3769633507853403,
-      "grad_norm": 2.7435912987153515,
+      "grad_norm": 2.496255320356891,
       "learning_rate": 4.99011439559198e-05,
-      "logits/chosen": -9.5625,
-      "logits/rejected": -7.65625,
-      "logps/chosen": -760.0,
-      "logps/rejected": -1664.0,
-      "loss": 0.2541,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -3.671875,
-      "rewards/margins": 4.96875,
-      "rewards/rejected": -8.625,
+      "logits/chosen": -10.0625,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -840.0,
+      "logps/rejected": -1672.0,
+      "loss": 0.2687,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 4.1875,
+      "rewards/rejected": -8.6875,
       "step": 360
     },
     {
       "epoch": 0.387434554973822,
-      "grad_norm": 1.929738469164756,
+      "grad_norm": 2.0343418435957044,
       "learning_rate": 4.987222972275369e-05,
-      "logits/chosen": -7.40625,
-      "logits/rejected": -5.90625,
-      "logps/chosen": -1000.0,
-      "logps/rejected": -1936.0,
-      "loss": 0.2805,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -6.03125,
-      "rewards/margins": 5.5625,
-      "rewards/rejected": -11.5625,
+      "logits/chosen": -9.5625,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -928.0,
+      "logps/rejected": -1760.0,
+      "loss": 0.2634,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -5.28125,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -9.8125,
       "step": 370
     },
     {
       "epoch": 0.39790575916230364,
-      "grad_norm": 1.8888706025178095,
+      "grad_norm": 1.6194768372835004,
       "learning_rate": 4.983962194568872e-05,
-      "logits/chosen": -9.1875,
-      "logits/rejected": -7.875,
-      "logps/chosen": -912.0,
-      "logps/rejected": -1800.0,
-      "loss": 0.2974,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -5.15625,
-      "rewards/margins": 4.28125,
-      "rewards/rejected": -9.4375,
+      "logits/chosen": -11.6875,
+      "logits/rejected": -10.625,
+      "logps/chosen": -836.0,
+      "logps/rejected": -1752.0,
+      "loss": 0.2629,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -4.375,
+      "rewards/margins": 4.5,
+      "rewards/rejected": -8.875,
       "step": 380
     },
     {
       "epoch": 0.4083769633507853,
-      "grad_norm": 1.7925756470571075,
+      "grad_norm": 1.5097487520133928,
       "learning_rate": 4.9803325467003084e-05,
-      "logits/chosen": -9.625,
-      "logits/rejected": -7.5,
-      "logps/chosen": -848.0,
-      "logps/rejected": -1920.0,
-      "loss": 0.2474,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -4.5,
-      "rewards/margins": 5.90625,
-      "rewards/rejected": -10.375,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -816.0,
+      "logps/rejected": -1912.0,
+      "loss": 0.2537,
+      "rewards/accuracies": 0.8687499761581421,
+      "rewards/chosen": -4.125,
+      "rewards/margins": 6.1875,
+      "rewards/rejected": -10.3125,
       "step": 390
     },
     {
       "epoch": 0.418848167539267,
-      "grad_norm": 1.8335053688554408,
+      "grad_norm": 1.9270576364286796,
       "learning_rate": 4.9763345676749816e-05,
-      "logits/chosen": -7.3125,
-      "logits/rejected": -6.125,
-      "logps/chosen": -932.0,
-      "logps/rejected": -2008.0,
-      "loss": 0.2779,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -5.53125,
-      "rewards/margins": 6.25,
-      "rewards/rejected": -11.8125,
-      "step": 400
-    },
+      "logits/chosen": -10.375,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -828.0,
+      "logps/rejected": -1856.0,
+      "loss": 0.245,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 5.71875,
+      "rewards/rejected": -10.25,
+      "step": 400
+    },
     {
       "epoch": 0.418848167539267,
-      "eval_logits/chosen": -6.8125,
-      "eval_logits/rejected": -5.65625,
-      "eval_logps/chosen": -788.0,
-      "eval_logps/rejected": -816.0,
-      "eval_loss": 0.7848281264305115,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -4.6875,
-      "eval_rewards/margins": 0.57421875,
-      "eval_rewards/rejected": -5.28125,
-      "eval_runtime": 46.5999,
-      "eval_samples_per_second": 42.919,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -9.5625,
+      "eval_logits/rejected": -8.75,
+      "eval_logps/chosen": -784.0,
+      "eval_logps/rejected": -812.0,
+      "eval_loss": 0.7902812361717224,
+      "eval_rewards/accuracies": 0.580078125,
+      "eval_rewards/chosen": -4.65625,
+      "eval_rewards/margins": 0.59765625,
+      "eval_rewards/rejected": -5.25,
+      "eval_runtime": 58.1661,
+      "eval_samples_per_second": 34.384,
+      "eval_steps_per_second": 0.55,
       "step": 400
     },
     {
       "epoch": 0.4293193717277487,
-      "grad_norm": 3.2529526253697045,
+      "grad_norm": 3.261792224773926,
       "learning_rate": 4.9719688511956374e-05,
-      "logits/chosen": -8.1875,
-      "logits/rejected": -6.71875,
-      "logps/chosen": -808.0,
-      "logps/rejected": -1744.0,
-      "loss": 0.3213,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -3.984375,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -8.6875,
+      "logits/chosen": -9.75,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -872.0,
+      "logps/rejected": -1840.0,
+      "loss": 0.3182,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.625,
+      "rewards/margins": 4.9375,
+      "rewards/rejected": -9.5625,
       "step": 410
     },
     {
       "epoch": 0.4397905759162304,
-      "grad_norm": 2.7413944483443657,
+      "grad_norm": 1.9876906582054223,
       "learning_rate": 4.967236045574294e-05,
-      "logits/chosen": -8.5,
-      "logits/rejected": -6.5625,
+      "logits/chosen": -9.8125,
+      "logits/rejected": -8.5,
       "logps/chosen": -800.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.2742,
-      "rewards/accuracies": 0.9125000238418579,
+      "logps/rejected": -1704.0,
+      "loss": 0.2815,
+      "rewards/accuracies": 0.90625,
       "rewards/chosen": -4.0625,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -8.5,
+      "rewards/margins": 3.84375,
+      "rewards/rejected": -7.90625,
       "step": 420
     },
     {
       "epoch": 0.450261780104712,
-      "grad_norm": 2.050592676528898,
+      "grad_norm": 1.9322265940344776,
       "learning_rate": 4.96213685363597e-05,
-      "logits/chosen": -9.1875,
-      "logits/rejected": -7.09375,
-      "logps/chosen": -812.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.2631,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -4.21875,
-      "rewards/margins": 4.84375,
-      "rewards/rejected": -9.0625,
+      "logits/chosen": -9.9375,
+      "logits/rejected": -8.375,
+      "logps/chosen": -760.0,
+      "logps/rejected": -1648.0,
+      "loss": 0.2417,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -3.71875,
+      "rewards/margins": 4.6875,
+      "rewards/rejected": -8.375,
       "step": 430
     },
     {
       "epoch": 0.4607329842931937,
-      "grad_norm": 2.9949228565699872,
+      "grad_norm": 4.590129579421251,
       "learning_rate": 4.956672032614315e-05,
-      "logits/chosen": -10.6875,
-      "logits/rejected": -8.5625,
-      "logps/chosen": -792.0,
-      "logps/rejected": -1728.0,
-      "loss": 0.2862,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -3.859375,
-      "rewards/margins": 4.65625,
-      "rewards/rejected": -8.5,
+      "logits/chosen": -10.375,
+      "logits/rejected": -8.625,
+      "logps/chosen": -800.0,
+      "logps/rejected": -1800.0,
+      "loss": 0.2896,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -3.90625,
+      "rewards/margins": 5.3125,
+      "rewards/rejected": -9.1875,
       "step": 440
     },
     {
       "epoch": 0.4712041884816754,
-      "grad_norm": 2.3296318461612096,
+      "grad_norm": 2.089351246552762,
       "learning_rate": 4.950842394039157e-05,
-      "logits/chosen": -9.875,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -744.0,
-      "logps/rejected": -1480.0,
-      "loss": 0.242,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -3.703125,
-      "rewards/margins": 4.09375,
-      "rewards/rejected": -7.78125,
+      "logits/chosen": -10.5,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -692.0,
+      "logps/rejected": -1392.0,
+      "loss": 0.2445,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -3.1875,
+      "rewards/margins": 3.78125,
+      "rewards/rejected": -6.96875,
       "step": 450
     },
     {
       "epoch": 0.4816753926701571,
-      "grad_norm": 2.9583338756482775,
+      "grad_norm": 2.997373480371455,
       "learning_rate": 4.944648803615994e-05,
-      "logits/chosen": -6.09375,
-      "logits/rejected": -4.59375,
-      "logps/chosen": -912.0,
-      "logps/rejected": -1976.0,
-      "loss": 0.2353,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -5.125,
-      "rewards/margins": 6.09375,
-      "rewards/rejected": -11.25,
+      "logits/chosen": -8.375,
+      "logits/rejected": -6.59375,
+      "logps/chosen": -860.0,
+      "logps/rejected": -1888.0,
+      "loss": 0.245,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.59375,
+      "rewards/margins": 5.75,
+      "rewards/rejected": -10.375,
       "step": 460
     },
     {
       "epoch": 0.49214659685863876,
-      "grad_norm": 2.555382501732579,
+      "grad_norm": 1.9000507746917805,
       "learning_rate": 4.938092181097433e-05,
-      "logits/chosen": -7.125,
-      "logits/rejected": -5.5625,
-      "logps/chosen": -792.0,
-      "logps/rejected": -1776.0,
-      "loss": 0.2401,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -7.09375,
+      "logps/chosen": -772.0,
+      "logps/rejected": -1680.0,
+      "loss": 0.2203,
       "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -4.03125,
-      "rewards/margins": 5.625,
-      "rewards/rejected": -9.625,
+      "rewards/chosen": -3.84375,
+      "rewards/margins": 4.8125,
+      "rewards/rejected": -8.625,
       "step": 470
     },
     {
       "epoch": 0.5026178010471204,
-      "grad_norm": 4.199835522177415,
+      "grad_norm": 3.4930671362151555,
       "learning_rate": 4.931173500146607e-05,
-      "logits/chosen": -6.71875,
-      "logits/rejected": -5.25,
-      "logps/chosen": -744.0,
-      "logps/rejected": -1928.0,
-      "loss": 0.2661,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -3.84375,
-      "rewards/margins": 6.53125,
-      "rewards/rejected": -10.375,
+      "logits/chosen": -8.5,
+      "logits/rejected": -6.9375,
+      "logps/chosen": -804.0,
+      "logps/rejected": -1936.0,
+      "loss": 0.2675,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -4.40625,
+      "rewards/margins": 6.0,
+      "rewards/rejected": -10.4375,
       "step": 480
     },
     {
       "epoch": 0.5130890052356021,
-      "grad_norm": 2.0995951106401827,
+      "grad_norm": 2.4537958318669553,
       "learning_rate": 4.923893788192582e-05,
-      "logits/chosen": -7.6875,
-      "logits/rejected": -5.6875,
-      "logps/chosen": -812.0,
-      "logps/rejected": -1880.0,
-      "loss": 0.2943,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -4.03125,
-      "rewards/margins": 6.25,
-      "rewards/rejected": -10.3125,
+      "logits/chosen": -10.6875,
+      "logits/rejected": -9.0,
+      "logps/chosen": -808.0,
+      "logps/rejected": -1792.0,
+      "loss": 0.2604,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -4.0,
+      "rewards/margins": 5.4375,
+      "rewards/rejected": -9.4375,
       "step": 490
     },
     {
       "epoch": 0.5235602094240838,
-      "grad_norm": 4.015941422842941,
+      "grad_norm": 4.635578029693967,
       "learning_rate": 4.9162541262777906e-05,
-      "logits/chosen": -8.9375,
-      "logits/rejected": -7.34375,
-      "logps/chosen": -756.0,
-      "logps/rejected": -1640.0,
-      "loss": 0.2378,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 5.0,
-      "rewards/rejected": -8.5,
+      "logits/chosen": -10.5,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -904.0,
+      "logps/rejected": -1880.0,
+      "loss": 0.2375,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -5.0,
+      "rewards/margins": 5.90625,
+      "rewards/rejected": -10.875,
       "step": 500
     },
     {
       "epoch": 0.5235602094240838,
-      "eval_logits/chosen": -5.71875,
-      "eval_logits/rejected": -4.4375,
-      "eval_logps/chosen": -860.0,
-      "eval_logps/rejected": -928.0,
-      "eval_loss": 0.8629999756813049,
-      "eval_rewards/accuracies": 0.6015625,
-      "eval_rewards/chosen": -5.40625,
-      "eval_rewards/margins": 1.015625,
-      "eval_rewards/rejected": -6.40625,
-      "eval_runtime": 46.6087,
-      "eval_samples_per_second": 42.91,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -8.6875,
+      "eval_logits/rejected": -7.46875,
+      "eval_logps/chosen": -992.0,
+      "eval_logps/rejected": -1080.0,
+      "eval_loss": 0.9612030982971191,
+      "eval_rewards/accuracies": 0.611328125,
+      "eval_rewards/chosen": -6.75,
+      "eval_rewards/margins": 1.125,
+      "eval_rewards/rejected": -7.875,
+      "eval_runtime": 58.1878,
+      "eval_samples_per_second": 34.371,
+      "eval_steps_per_second": 0.55,
       "step": 500
     },
     {
       "epoch": 0.5340314136125655,
-      "grad_norm": 3.0689778704359347,
+      "grad_norm": 2.592774326458581,
       "learning_rate": 4.90825564889749e-05,
-      "logits/chosen": -6.34375,
-      "logits/rejected": -5.1875,
-      "logps/chosen": -1024.0,
-      "logps/rejected": -2032.0,
-      "loss": 0.2709,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -6.03125,
-      "rewards/margins": 7.03125,
-      "rewards/rejected": -13.0625,
+      "logits/chosen": -9.6875,
+      "logits/rejected": -7.8125,
+      "logps/chosen": -1032.0,
+      "logps/rejected": -2016.0,
+      "loss": 0.2612,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -6.125,
+      "rewards/margins": 6.8125,
+      "rewards/rejected": -12.9375,
       "step": 510
     },
     {
       "epoch": 0.5445026178010471,
-      "grad_norm": 3.525801637307732,
+      "grad_norm": 2.1761488132123024,
       "learning_rate": 4.899899543831288e-05,
-      "logits/chosen": -8.375,
-      "logits/rejected": -6.46875,
-      "logps/chosen": -828.0,
+      "logits/chosen": -10.4375,
+      "logits/rejected": -8.25,
+      "logps/chosen": -788.0,
       "logps/rejected": -1792.0,
-      "loss": 0.2632,
+      "loss": 0.2564,
       "rewards/accuracies": 0.875,
-      "rewards/chosen": -3.984375,
-      "rewards/margins": 5.28125,
-      "rewards/rejected": -9.25,
+      "rewards/chosen": -3.5625,
+      "rewards/margins": 5.78125,
+      "rewards/rejected": -9.3125,
       "step": 520
     },
     {
       "epoch": 0.5549738219895288,
-      "grad_norm": 3.750495412924153,
+      "grad_norm": 3.570432885472526,
       "learning_rate": 4.891187051966765e-05,
-      "logits/chosen": -7.875,
-      "logits/rejected": -6.8125,
-      "logps/chosen": -848.0,
-      "logps/rejected": -1680.0,
-      "loss": 0.2438,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -4.375,
-      "rewards/margins": 4.71875,
-      "rewards/rejected": -9.125,
+      "logits/chosen": -10.375,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -920.0,
+      "logps/rejected": -1928.0,
+      "loss": 0.2581,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 6.4375,
+      "rewards/rejected": -11.5,
       "step": 530
     },
     {
       "epoch": 0.5654450261780105,
-      "grad_norm": 1.9426526974609004,
+      "grad_norm": 1.2261684608982484,
       "learning_rate": 4.882119467115191e-05,
-      "logits/chosen": -6.28125,
-      "logits/rejected": -5.4375,
-      "logps/chosen": -960.0,
-      "logps/rejected": -1872.0,
-      "loss": 0.2335,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -5.65625,
-      "rewards/margins": 5.65625,
-      "rewards/rejected": -11.3125,
+      "logits/chosen": -12.25,
+      "logits/rejected": -10.75,
+      "logps/chosen": -760.0,
+      "logps/rejected": -1648.0,
+      "loss": 0.2332,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -3.65625,
+      "rewards/margins": 5.40625,
+      "rewards/rejected": -9.0625,
       "step": 540
     },
     {
       "epoch": 0.5759162303664922,
-      "grad_norm": 2.255937546496114,
+      "grad_norm": 3.1644026124279474,
       "learning_rate": 4.872698135819402e-05,
-      "logits/chosen": -8.625,
-      "logits/rejected": -7.65625,
-      "logps/chosen": -796.0,
-      "logps/rejected": -1896.0,
-      "loss": 0.279,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -4.21875,
-      "rewards/margins": 5.53125,
-      "rewards/rejected": -9.75,
+      "logits/chosen": -9.375,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -852.0,
+      "logps/rejected": -2048.0,
+      "loss": 0.237,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -4.75,
+      "rewards/margins": 6.53125,
+      "rewards/rejected": -11.3125,
       "step": 550
     },
     {
       "epoch": 0.5863874345549738,
-      "grad_norm": 3.230717270627345,
+      "grad_norm": 3.4252245326994197,
       "learning_rate": 4.862924457153831e-05,
-      "logits/chosen": -8.1875,
-      "logits/rejected": -6.65625,
-      "logps/chosen": -872.0,
-      "logps/rejected": -2144.0,
-      "loss": 0.2334,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -5.375,
-      "rewards/margins": 6.71875,
-      "rewards/rejected": -12.125,
+      "logits/chosen": -8.125,
+      "logits/rejected": -6.6875,
+      "logps/chosen": -972.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.2426,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -6.40625,
+      "rewards/margins": 8.9375,
+      "rewards/rejected": -15.3125,
       "step": 560
     },
     {
       "epoch": 0.5968586387434555,
-      "grad_norm": 1.6238653771182336,
+      "grad_norm": 2.9767536742500424,
       "learning_rate": 4.8527998825167484e-05,
-      "logits/chosen": -7.46875,
-      "logits/rejected": -7.0625,
-      "logps/chosen": -944.0,
-      "logps/rejected": -1912.0,
-      "loss": 0.21,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -5.6875,
-      "rewards/margins": 5.8125,
-      "rewards/rejected": -11.5,
+      "logits/chosen": -9.4375,
+      "logits/rejected": -8.625,
+      "logps/chosen": -868.0,
+      "logps/rejected": -1928.0,
+      "loss": 0.1883,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 6.71875,
+      "rewards/rejected": -11.625,
       "step": 570
     },
     {
       "epoch": 0.6073298429319371,
-      "grad_norm": 3.0290874798581653,
+      "grad_norm": 2.9897601418873307,
       "learning_rate": 4.842325915414728e-05,
-      "logits/chosen": -8.625,
-      "logits/rejected": -6.96875,
-      "logps/chosen": -840.0,
-      "logps/rejected": -1808.0,
-      "loss": 0.2044,
+      "logits/chosen": -10.6875,
+      "logits/rejected": -8.75,
+      "logps/chosen": -740.0,
+      "logps/rejected": -1760.0,
+      "loss": 0.2216,
       "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -4.84375,
-      "rewards/margins": 5.875,
-      "rewards/rejected": -10.75,
+      "rewards/chosen": -3.84375,
+      "rewards/margins": 6.375,
+      "rewards/rejected": -10.25,
       "step": 580
     },
     {
       "epoch": 0.6178010471204188,
-      "grad_norm": 3.974281262703676,
+      "grad_norm": 3.7445306810760197,
       "learning_rate": 4.831504111239373e-05,
-      "logits/chosen": -8.6875,
-      "logits/rejected": -6.53125,
-      "logps/chosen": -1004.0,
-      "logps/rejected": -2176.0,
-      "loss": 0.2453,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -5.90625,
-      "rewards/margins": 7.625,
-      "rewards/rejected": -13.5625,
+      "logits/chosen": -10.3125,
+      "logits/rejected": -8.25,
+      "logps/chosen": -928.0,
+      "logps/rejected": -2016.0,
+      "loss": 0.2602,
+      "rewards/accuracies": 0.831250011920929,
+      "rewards/chosen": -5.125,
+      "rewards/margins": 6.75,
+      "rewards/rejected": -11.875,
       "step": 590
     },
     {
       "epoch": 0.6282722513089005,
-      "grad_norm": 2.6854647482095313,
+      "grad_norm": 3.4834197485465244,
       "learning_rate": 4.8203360770363425e-05,
-      "logits/chosen": -6.875,
-      "logits/rejected": -5.46875,
+      "logits/chosen": -10.75,
+      "logits/rejected": -9.0625,
       "logps/chosen": -944.0,
-      "logps/rejected": -1984.0,
-      "loss": 0.2478,
-      "rewards/accuracies": 0.856249988079071,
+      "logps/rejected": -2000.0,
+      "loss": 0.2534,
+      "rewards/accuracies": 0.862500011920929,
       "rewards/chosen": -5.4375,
-      "rewards/margins": 5.96875,
-      "rewards/rejected": -11.375,
+      "rewards/margins": 6.1875,
+      "rewards/rejected": -11.625,
       "step": 600
     },
     {
       "epoch": 0.6282722513089005,
-      "eval_logits/chosen": -4.78125,
-      "eval_logits/rejected": -3.5625,
-      "eval_logps/chosen": -860.0,
-      "eval_logps/rejected": -904.0,
-      "eval_loss": 0.830203115940094,
-      "eval_rewards/accuracies": 0.626953125,
-      "eval_rewards/chosen": -5.40625,
-      "eval_rewards/margins": 0.73828125,
-      "eval_rewards/rejected": -6.15625,
-      "eval_runtime": 46.609,
-      "eval_samples_per_second": 42.91,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -9.6875,
+      "eval_logits/rejected": -8.75,
+      "eval_logps/chosen": -884.0,
+      "eval_logps/rejected": -940.0,
+      "eval_loss": 0.857296884059906,
+      "eval_rewards/accuracies": 0.61328125,
+      "eval_rewards/chosen": -5.65625,
+      "eval_rewards/margins": 0.84375,
+      "eval_rewards/rejected": -6.5,
+      "eval_runtime": 58.1405,
+      "eval_samples_per_second": 34.399,
+      "eval_steps_per_second": 0.55,
       "step": 600
     },
     {
       "epoch": 0.6387434554973822,
-      "grad_norm": 1.6677110197203666,
+      "grad_norm": 2.414720551573577,
       "learning_rate": 4.808823471266702e-05,
-      "logits/chosen": -5.5625,
-      "logits/rejected": -4.25,
-      "logps/chosen": -884.0,
-      "logps/rejected": -1656.0,
-      "loss": 0.2449,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -4.96875,
-      "rewards/margins": 4.125,
-      "rewards/rejected": -9.0625,
+      "logits/chosen": -10.1875,
+      "logits/rejected": -8.25,
+      "logps/chosen": -936.0,
+      "logps/rejected": -1760.0,
+      "loss": 0.2579,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -5.4375,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -10.125,
       "step": 610
     },
     {
       "epoch": 0.6492146596858639,
-      "grad_norm": 2.344977623502758,
+      "grad_norm": 2.564861789835996,
       "learning_rate": 4.796968003560639e-05,
-      "logits/chosen": -5.8125,
-      "logits/rejected": -4.9375,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -2192.0,
-      "loss": 0.2634,
-      "rewards/accuracies": 0.831250011920929,
-      "rewards/chosen": -6.40625,
-      "rewards/margins": 5.8125,
-      "rewards/rejected": -12.1875,
+      "logits/chosen": -11.25,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -904.0,
+      "logps/rejected": -2008.0,
+      "loss": 0.2348,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 5.40625,
+      "rewards/rejected": -10.4375,
       "step": 620
     },
     {
       "epoch": 0.6596858638743456,
-      "grad_norm": 2.3428744736727385,
+      "grad_norm": 1.7244097187377714,
       "learning_rate": 4.7847714344635864e-05,
-      "logits/chosen": -9.125,
-      "logits/rejected": -7.21875,
-      "logps/chosen": -912.0,
-      "logps/rejected": -2048.0,
-      "loss": 0.2175,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -5.34375,
-      "rewards/margins": 6.75,
-      "rewards/rejected": -12.125,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -808.0,
+      "logps/rejected": -1816.0,
+      "loss": 0.2472,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.3125,
+      "rewards/margins": 5.5,
+      "rewards/rejected": -9.8125,
       "step": 630
     },
     {
       "epoch": 0.6701570680628273,
-      "grad_norm": 2.0010522304079994,
+      "grad_norm": 2.650305302223484,
       "learning_rate": 4.772235575174776e-05,
-      "logits/chosen": -9.8125,
-      "logits/rejected": -7.65625,
-      "logps/chosen": -956.0,
-      "logps/rejected": -1888.0,
-      "loss": 0.2644,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -5.375,
-      "rewards/margins": 5.90625,
-      "rewards/rejected": -11.25,
+      "logits/chosen": -11.5,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -864.0,
+      "logps/rejected": -1712.0,
+      "loss": 0.2601,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": -4.46875,
+      "rewards/margins": 5.0625,
+      "rewards/rejected": -9.5,
       "step": 640
     },
     {
       "epoch": 0.680628272251309,
-      "grad_norm": 3.1451700792686994,
+      "grad_norm": 3.1339359862687197,
       "learning_rate": 4.7593622872782754e-05,
-      "logits/chosen": -8.875,
-      "logits/rejected": -7.8125,
-      "logps/chosen": -952.0,
-      "logps/rejected": -2000.0,
-      "loss": 0.2478,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -5.9375,
-      "rewards/margins": 5.9375,
-      "rewards/rejected": -11.875,
+      "logits/chosen": -9.875,
+      "logits/rejected": -8.5,
+      "logps/chosen": -856.0,
+      "logps/rejected": -1816.0,
+      "loss": 0.2374,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 5.125,
+      "rewards/rejected": -10.0625,
       "step": 650
     },
     {
       "epoch": 0.6910994764397905,
-      "grad_norm": 2.4110315416472585,
+      "grad_norm": 2.722803859256615,
       "learning_rate": 4.746153482466542e-05,
-      "logits/chosen": -9.75,
-      "logits/rejected": -7.78125,
-      "logps/chosen": -916.0,
-      "logps/rejected": -1808.0,
-      "loss": 0.2378,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -5.375,
-      "rewards/margins": 5.28125,
-      "rewards/rejected": -10.625,
+      "logits/chosen": -9.25,
+      "logits/rejected": -7.15625,
+      "logps/chosen": -832.0,
+      "logps/rejected": -1784.0,
+      "loss": 0.2349,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 5.90625,
+      "rewards/rejected": -10.4375,
       "step": 660
     },
     {
       "epoch": 0.7015706806282722,
-      "grad_norm": 2.0102627059492577,
+      "grad_norm": 1.7904765397805,
       "learning_rate": 4.732611122256537e-05,
-      "logits/chosen": -8.5625,
-      "logits/rejected": -6.90625,
-      "logps/chosen": -1020.0,
-      "logps/rejected": -2016.0,
-      "loss": 0.2009,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -5.875,
-      "rewards/margins": 5.84375,
-      "rewards/rejected": -11.6875,
+      "logits/chosen": -10.5625,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -844.0,
+      "logps/rejected": -1800.0,
+      "loss": 0.2194,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -4.125,
+      "rewards/margins": 5.375,
+      "rewards/rejected": -9.5,
       "step": 670
     },
     {
       "epoch": 0.7120418848167539,
-      "grad_norm": 2.2942723063930823,
+      "grad_norm": 2.053285506394441,
       "learning_rate": 4.718737217698435e-05,
-      "logits/chosen": -6.96875,
-      "logits/rejected": -5.78125,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.3075,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -7.59375,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -15.75,
+      "logits/chosen": -9.375,
+      "logits/rejected": -7.15625,
+      "logps/chosen": -904.0,
+      "logps/rejected": -2080.0,
+      "loss": 0.2695,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.3125,
+      "rewards/margins": 6.59375,
+      "rewards/rejected": -11.875,
       "step": 680
     },
     {
       "epoch": 0.7225130890052356,
-      "grad_norm": 1.8403280138916578,
+      "grad_norm": 1.8931481012227234,
       "learning_rate": 4.704533829076982e-05,
-      "logits/chosen": -9.75,
-      "logits/rejected": -7.6875,
-      "logps/chosen": -872.0,
+      "logits/chosen": -9.25,
+      "logits/rejected": -6.0625,
+      "logps/chosen": -864.0,
       "logps/rejected": -1728.0,
-      "loss": 0.2155,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -4.75,
-      "rewards/margins": 5.0625,
+      "loss": 0.2129,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.65625,
+      "rewards/margins": 5.15625,
       "rewards/rejected": -9.8125,
       "step": 690
     },
     {
       "epoch": 0.7329842931937173,
-      "grad_norm": 1.8068514714766295,
+      "grad_norm": 1.9376687805225403,
       "learning_rate": 4.6900030656055455e-05,
-      "logits/chosen": -10.5625,
-      "logits/rejected": -8.5625,
-      "logps/chosen": -800.0,
-      "logps/rejected": -1688.0,
-      "loss": 0.2346,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -3.9375,
-      "rewards/margins": 4.34375,
-      "rewards/rejected": -8.25,
+      "logits/chosen": -7.625,
+      "logits/rejected": -4.90625,
+      "logps/chosen": -852.0,
+      "logps/rejected": -1784.0,
+      "loss": 0.2213,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.4375,
+      "rewards/margins": 4.84375,
+      "rewards/rejected": -9.25,
       "step": 700
     },
     {
       "epoch": 0.7329842931937173,
-      "eval_logits/chosen": -9.0,
-      "eval_logits/rejected": -7.90625,
-      "eval_logps/chosen": -836.0,
-      "eval_logps/rejected": -884.0,
-      "eval_loss": 0.8087968826293945,
-      "eval_rewards/accuracies": 0.62109375,
-      "eval_rewards/chosen": -5.15625,
-      "eval_rewards/margins": 0.81640625,
-      "eval_rewards/rejected": -5.96875,
-      "eval_runtime": 46.6144,
-      "eval_samples_per_second": 42.905,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -7.3125,
+      "eval_logits/rejected": -5.75,
+      "eval_logps/chosen": -796.0,
+      "eval_logps/rejected": -860.0,
+      "eval_loss": 0.8133124709129333,
+      "eval_rewards/accuracies": 0.638671875,
+      "eval_rewards/chosen": -4.78125,
+      "eval_rewards/margins": 0.93359375,
+      "eval_rewards/rejected": -5.71875,
+      "eval_runtime": 58.1714,
+      "eval_samples_per_second": 34.381,
+      "eval_steps_per_second": 0.55,
       "step": 700
     },
     {
       "epoch": 0.743455497382199,
-      "grad_norm": 2.9027966264296503,
+      "grad_norm": 3.0899572912988447,
       "learning_rate": 4.6751470851128906e-05,
-      "logits/chosen": -11.4375,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -908.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.2151,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -4.84375,
-      "rewards/margins": 4.875,
-      "rewards/rejected": -9.75,
+      "logits/chosen": -9.9375,
+      "logits/rejected": -7.25,
+      "logps/chosen": -860.0,
+      "logps/rejected": -1688.0,
+      "loss": 0.2119,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -4.375,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -9.0625,
       "step": 710
     },
     {
       "epoch": 0.7539267015706806,
-      "grad_norm": 2.821021731817776,
+      "grad_norm": 1.9734603985133177,
       "learning_rate": 4.6599680937227454e-05,
-      "logits/chosen": -12.4375,
-      "logits/rejected": -10.25,
-      "logps/chosen": -880.0,
-      "logps/rejected": -1840.0,
-      "loss": 0.2409,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -4.78125,
-      "rewards/margins": 5.84375,
-      "rewards/rejected": -10.625,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -872.0,
+      "logps/rejected": -1864.0,
+      "loss": 0.245,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 6.1875,
+      "rewards/rejected": -10.875,
       "step": 720
     },
     {
       "epoch": 0.7643979057591623,
-      "grad_norm": 1.8951830619494674,
+      "grad_norm": 1.4742200819936666,
       "learning_rate": 4.644468345526183e-05,
-      "logits/chosen": -9.6875,
-      "logits/rejected": -7.75,
-      "logps/chosen": -944.0,
-      "logps/rejected": -1968.0,
-      "loss": 0.2216,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -5.4375,
-      "rewards/margins": 5.40625,
-      "rewards/rejected": -10.875,
+      "logits/chosen": -10.125,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -884.0,
+      "logps/rejected": -1896.0,
+      "loss": 0.2303,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.84375,
+      "rewards/margins": 5.3125,
+      "rewards/rejected": -10.125,
       "step": 730
     },
     {
       "epoch": 0.774869109947644,
-      "grad_norm": 2.9850595623161644,
+      "grad_norm": 2.3380083347340475,
       "learning_rate": 4.628650142246894e-05,
-      "logits/chosen": -10.5,
-      "logits/rejected": -7.1875,
-      "logps/chosen": -856.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.2184,
+      "logits/chosen": -9.75,
+      "logits/rejected": -6.34375,
+      "logps/chosen": -848.0,
+      "logps/rejected": -1800.0,
+      "loss": 0.2105,
       "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -4.78125,
-      "rewards/margins": 5.34375,
-      "rewards/rejected": -10.125,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 5.5625,
+      "rewards/rejected": -10.25,
       "step": 740
     },
     {
       "epoch": 0.7853403141361257,
-      "grad_norm": 4.473480668644008,
+      "grad_norm": 2.2806092614776188,
       "learning_rate": 4.6125158328993765e-05,
-      "logits/chosen": -11.375,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -880.0,
-      "logps/rejected": -1840.0,
-      "loss": 0.2427,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -5.0625,
-      "rewards/margins": 5.71875,
-      "rewards/rejected": -10.75,
+      "logits/chosen": -9.5625,
+      "logits/rejected": -6.71875,
+      "logps/chosen": -888.0,
+      "logps/rejected": -1816.0,
+      "loss": 0.2386,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -5.15625,
+      "rewards/margins": 5.375,
+      "rewards/rejected": -10.5,
       "step": 750
     },
     {
       "epoch": 0.7958115183246073,
-      "grad_norm": 3.378009781672457,
+      "grad_norm": 3.953367203437828,
       "learning_rate": 4.5960678134401064e-05,
-      "logits/chosen": -12.1875,
-      "logits/rejected": -10.125,
-      "logps/chosen": -876.0,
-      "logps/rejected": -1896.0,
-      "loss": 0.2509,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -4.78125,
-      "rewards/margins": 6.0,
-      "rewards/rejected": -10.8125,
+      "logits/chosen": -10.5,
+      "logits/rejected": -8.0,
+      "logps/chosen": -960.0,
+      "logps/rejected": -2024.0,
+      "loss": 0.2293,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.59375,
+      "rewards/margins": 6.53125,
+      "rewards/rejected": -12.125,
       "step": 760
     },
     {
       "epoch": 0.806282722513089,
-      "grad_norm": 1.5325479598421954,
+      "grad_norm": 1.666529707381326,
       "learning_rate": 4.579308526411734e-05,
-      "logits/chosen": -10.25,
-      "logits/rejected": -8.3125,
-      "logps/chosen": -900.0,
-      "logps/rejected": -2112.0,
-      "loss": 0.1821,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -5.125,
-      "rewards/margins": 7.71875,
-      "rewards/rejected": -12.875,
+      "logits/chosen": -10.375,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -928.0,
+      "logps/rejected": -2144.0,
+      "loss": 0.1964,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.375,
+      "rewards/margins": 7.6875,
+      "rewards/rejected": -13.0625,
       "step": 770
     },
     {
       "epoch": 0.8167539267015707,
-      "grad_norm": 3.161054609254887,
+      "grad_norm": 3.054254850631056,
       "learning_rate": 4.562240460580367e-05,
-      "logits/chosen": -8.3125,
-      "logits/rejected": -6.625,
-      "logps/chosen": -944.0,
-      "logps/rejected": -2096.0,
-      "loss": 0.2234,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -5.5,
-      "rewards/margins": 7.1875,
-      "rewards/rejected": -12.6875,
+      "logits/chosen": -10.6875,
+      "logits/rejected": -8.5,
+      "logps/chosen": -820.0,
+      "logps/rejected": -1896.0,
+      "loss": 0.23,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.28125,
+      "rewards/margins": 6.46875,
+      "rewards/rejected": -10.75,
       "step": 780
     },
     {
       "epoch": 0.8272251308900523,
-      "grad_norm": 2.283000301220118,
+      "grad_norm": 1.6125439763176364,
       "learning_rate": 4.544866150565985e-05,
-      "logits/chosen": -6.0,
-      "logits/rejected": -4.40625,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -2288.0,
-      "loss": 0.2214,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -6.5,
-      "rewards/margins": 7.59375,
-      "rewards/rejected": -14.0625,
+      "logits/chosen": -8.0625,
+      "logits/rejected": -5.8125,
+      "logps/chosen": -992.0,
+      "logps/rejected": -2256.0,
+      "loss": 0.1931,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.0625,
+      "rewards/margins": 7.65625,
+      "rewards/rejected": -13.75,
       "step": 790
     },
     {
       "epoch": 0.837696335078534,
-      "grad_norm": 2.8679246428944856,
+      "grad_norm": 3.0844822397711407,
       "learning_rate": 4.527188176466051e-05,
-      "logits/chosen": -7.125,
-      "logits/rejected": -5.78125,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -2080.0,
-      "loss": 0.2222,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -6.21875,
-      "rewards/margins": 6.65625,
-      "rewards/rejected": -12.875,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -7.125,
+      "logps/chosen": -1032.0,
+      "logps/rejected": -2024.0,
+      "loss": 0.2342,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -6.125,
+      "rewards/margins": 6.125,
+      "rewards/rejected": -12.25,
       "step": 800
     },
     {
       "epoch": 0.837696335078534,
-      "eval_logits/chosen": -7.0625,
-      "eval_logits/rejected": -5.6875,
-      "eval_logps/chosen": -936.0,
-      "eval_logps/rejected": -984.0,
-      "eval_loss": 0.9451718926429749,
-      "eval_rewards/accuracies": 0.564453125,
-      "eval_rewards/chosen": -6.15625,
-      "eval_rewards/margins": 0.7734375,
-      "eval_rewards/rejected": -6.9375,
-      "eval_runtime": 46.6011,
-      "eval_samples_per_second": 42.917,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -8.5625,
+      "eval_logits/rejected": -7.28125,
+      "eval_logps/chosen": -872.0,
+      "eval_logps/rejected": -936.0,
+      "eval_loss": 0.8573750257492065,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -5.5625,
+      "eval_rewards/margins": 0.93359375,
+      "eval_rewards/rejected": -6.46875,
+      "eval_runtime": 58.1382,
+      "eval_samples_per_second": 34.401,
+      "eval_steps_per_second": 0.55,
       "step": 800
     },
     {
       "epoch": 0.8481675392670157,
-      "grad_norm": 2.6260057445413483,
+      "grad_norm": 1.664117022534246,
       "learning_rate": 4.509209163472357e-05,
-      "logits/chosen": -9.3125,
-      "logits/rejected": -7.125,
-      "logps/chosen": -980.0,
-      "logps/rejected": -2224.0,
-      "loss": 0.2219,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -5.78125,
-      "rewards/margins": 7.90625,
-      "rewards/rejected": -13.6875,
+      "logits/chosen": -10.0625,
+      "logits/rejected": -7.90625,
+      "logps/chosen": -912.0,
+      "logps/rejected": -2008.0,
+      "loss": 0.21,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -5.0625,
+      "rewards/margins": 6.46875,
+      "rewards/rejected": -11.5,
       "step": 810
     },
     {
       "epoch": 0.8586387434554974,
-      "grad_norm": 1.4475168181923295,
+      "grad_norm": 1.9077941825727036,
       "learning_rate": 4.4909317814811896e-05,
-      "logits/chosen": -10.0,
-      "logits/rejected": -8.25,
-      "logps/chosen": -932.0,
-      "logps/rejected": -1952.0,
-      "loss": 0.2633,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -5.28125,
-      "rewards/margins": 6.0625,
-      "rewards/rejected": -11.3125,
+      "logits/chosen": -10.0625,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -908.0,
+      "logps/rejected": -1984.0,
+      "loss": 0.2437,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -5.0625,
+      "rewards/margins": 6.53125,
+      "rewards/rejected": -11.625,
       "step": 820
     },
     {
       "epoch": 0.8691099476439791,
-      "grad_norm": 1.618716481251612,
+      "grad_norm": 1.4369435923839038,
       "learning_rate": 4.472358744696845e-05,
-      "logits/chosen": -9.5625,
-      "logits/rejected": -7.96875,
-      "logps/chosen": -972.0,
-      "logps/rejected": -2144.0,
-      "loss": 0.2244,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -5.8125,
-      "rewards/margins": 7.28125,
-      "rewards/rejected": -13.0625,
+      "logits/chosen": -9.125,
+      "logits/rejected": -7.3125,
+      "logps/chosen": -888.0,
+      "logps/rejected": -1976.0,
+      "loss": 0.2373,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 6.34375,
+      "rewards/rejected": -11.3125,
       "step": 830
     },
     {
       "epoch": 0.8795811518324608,
-      "grad_norm": 3.9327410558152267,
+      "grad_norm": 3.641233593017983,
       "learning_rate": 4.4534928112285645e-05,
       "logits/chosen": -8.5,
-      "logits/rejected": -7.03125,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -2384.0,
-      "loss": 0.2291,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -6.78125,
-      "rewards/margins": 8.6875,
-      "rewards/rejected": -15.4375,
+      "logits/rejected": -6.46875,
+      "logps/chosen": -900.0,
+      "logps/rejected": -2048.0,
+      "loss": 0.2102,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.15625,
+      "rewards/margins": 7.0625,
+      "rewards/rejected": -12.1875,
       "step": 840
     },
     {
       "epoch": 0.8900523560209425,
-      "grad_norm": 1.9370449857749532,
+      "grad_norm": 2.1420710878561473,
       "learning_rate": 4.43433678268096e-05,
-      "logits/chosen": -9.9375,
-      "logits/rejected": -8.0,
-      "logps/chosen": -888.0,
-      "logps/rejected": -1816.0,
-      "loss": 0.2321,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -5.03125,
-      "rewards/margins": 5.75,
-      "rewards/rejected": -10.75,
+      "logits/chosen": -8.5625,
+      "logits/rejected": -6.40625,
+      "logps/chosen": -896.0,
+      "logps/rejected": -1856.0,
+      "loss": 0.2429,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.125,
+      "rewards/margins": 6.09375,
+      "rewards/rejected": -11.1875,
       "step": 850
     },
     {
       "epoch": 0.900523560209424,
-      "grad_norm": 3.1219081933901927,
+      "grad_norm": 2.1523545908822754,
       "learning_rate": 4.414893503737968e-05,
-      "logits/chosen": -7.71875,
-      "logits/rejected": -6.40625,
-      "logps/chosen": -864.0,
-      "logps/rejected": -1744.0,
-      "loss": 0.223,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -5.15625,
-      "rewards/margins": 4.96875,
-      "rewards/rejected": -10.125,
+      "logits/chosen": -7.34375,
+      "logits/rejected": -5.625,
+      "logps/chosen": -800.0,
+      "logps/rejected": -1680.0,
+      "loss": 0.2063,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 5.03125,
+      "rewards/rejected": -9.5,
       "step": 860
     },
     {
       "epoch": 0.9109947643979057,
-      "grad_norm": 2.3615091656933607,
+      "grad_norm": 4.415171004708555,
       "learning_rate": 4.395165861740414e-05,
-      "logits/chosen": -9.125,
-      "logits/rejected": -7.21875,
-      "logps/chosen": -844.0,
-      "logps/rejected": -2000.0,
-      "loss": 0.1856,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -4.8125,
-      "rewards/margins": 6.4375,
-      "rewards/rejected": -11.25,
+      "logits/chosen": -8.375,
+      "logits/rejected": -6.15625,
+      "logps/chosen": -932.0,
+      "logps/rejected": -2160.0,
+      "loss": 0.1808,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -5.71875,
+      "rewards/margins": 7.21875,
+      "rewards/rejected": -12.9375,
       "step": 870
     },
     {
       "epoch": 0.9214659685863874,
-      "grad_norm": 3.1003306344435884,
+      "grad_norm": 3.971169499671966,
       "learning_rate": 4.375156786257241e-05,
-      "logits/chosen": -11.6875,
-      "logits/rejected": -9.25,
-      "logps/chosen": -940.0,
-      "logps/rejected": -2024.0,
-      "loss": 0.2646,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -5.1875,
-      "rewards/margins": 6.5625,
-      "rewards/rejected": -11.75,
+      "logits/chosen": -10.625,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -924.0,
+      "logps/rejected": -2040.0,
+      "loss": 0.2641,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 6.875,
+      "rewards/rejected": -11.9375,
       "step": 880
     },
     {
       "epoch": 0.9319371727748691,
-      "grad_norm": 2.1017611512381547,
+      "grad_norm": 1.525652424843547,
       "learning_rate": 4.354869248650466e-05,
-      "logits/chosen": -11.0,
-      "logits/rejected": -8.125,
-      "logps/chosen": -860.0,
-      "logps/rejected": -1816.0,
-      "loss": 0.2358,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -4.59375,
-      "rewards/margins": 5.5625,
-      "rewards/rejected": -10.125,
-      "step": 890
-    },
-    {
+      "logits/chosen": -10.4375,
+      "logits/rejected": -8.25,
+      "logps/chosen": -804.0,
+      "logps/rejected": -1744.0,
+      "loss": 0.2288,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.03125,
+      "rewards/margins": 5.4375,
+      "rewards/rejected": -9.4375,
+      "step": 890
+    },
+    {
       "epoch": 0.9424083769633508,
-      "grad_norm": 3.0151000884727255,
+      "grad_norm": 3.0728127553372993,
       "learning_rate": 4.334306261633931e-05,
-      "logits/chosen": -8.875,
-      "logits/rejected": -6.9375,
-      "logps/chosen": -968.0,
-      "logps/rejected": -2048.0,
-      "loss": 0.2077,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -5.46875,
+      "logits/chosen": -8.0625,
+      "logits/rejected": -6.15625,
+      "logps/chosen": -996.0,
+      "logps/rejected": -2080.0,
+      "loss": 0.199,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -5.71875,
       "rewards/margins": 6.21875,
-      "rewards/rejected": -11.6875,
+      "rewards/rejected": -11.9375,
       "step": 900
     },
     {
       "epoch": 0.9424083769633508,
-      "eval_logits/chosen": -7.75,
-      "eval_logits/rejected": -6.65625,
-      "eval_logps/chosen": -1016.0,
-      "eval_logps/rejected": -1064.0,
-      "eval_loss": 0.9077343940734863,
-      "eval_rewards/accuracies": 0.591796875,
-      "eval_rewards/chosen": -6.96875,
-      "eval_rewards/margins": 0.8125,
-      "eval_rewards/rejected": -7.78125,
-      "eval_runtime": 46.5947,
-      "eval_samples_per_second": 42.923,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -6.09375,
+      "eval_logits/rejected": -4.65625,
+      "eval_logps/chosen": -1040.0,
+      "eval_logps/rejected": -1104.0,
+      "eval_loss": 0.8852656483650208,
+      "eval_rewards/accuracies": 0.607421875,
+      "eval_rewards/chosen": -7.1875,
+      "eval_rewards/margins": 0.95703125,
+      "eval_rewards/rejected": -8.1875,
+      "eval_runtime": 58.1504,
+      "eval_samples_per_second": 34.394,
+      "eval_steps_per_second": 0.55,
       "step": 900
     },
     {
       "epoch": 0.9528795811518325,
-      "grad_norm": 2.0873636477308213,
+      "grad_norm": 2.3572084722084314,
       "learning_rate": 4.313470878825911e-05,
-      "logits/chosen": -9.25,
-      "logits/rejected": -7.78125,
-      "logps/chosen": -972.0,
+      "logits/chosen": -8.3125,
+      "logits/rejected": -6.34375,
+      "logps/chosen": -1000.0,
       "logps/rejected": -2528.0,
-      "loss": 0.2498,
+      "loss": 0.2172,
       "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -5.8125,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -14.625,
+      "rewards/chosen": -6.0625,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -14.6875,
       "step": 910
     },
     {
       "epoch": 0.9633507853403142,
-      "grad_norm": 2.1474194842517833,
+      "grad_norm": 2.252055800400268,
       "learning_rate": 4.292366194295652e-05,
-      "logits/chosen": -8.1875,
-      "logits/rejected": -6.34375,
-      "logps/chosen": -916.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.2206,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -5.21875,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -9.6875,
+      "logits/chosen": -9.5625,
+      "logits/rejected": -6.5625,
+      "logps/chosen": -876.0,
+      "logps/rejected": -1768.0,
+      "loss": 0.2231,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -4.78125,
+      "rewards/margins": 5.5,
+      "rewards/rejected": -10.25,
       "step": 920
     },
     {
       "epoch": 0.9738219895287958,
-      "grad_norm": 1.9854364707495304,
+      "grad_norm": 1.8869531862077833,
       "learning_rate": 4.2709953421038964e-05,
-      "logits/chosen": -8.375,
-      "logits/rejected": -6.65625,
-      "logps/chosen": -904.0,
-      "logps/rejected": -1696.0,
-      "loss": 0.2448,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -5.28125,
-      "rewards/margins": 4.21875,
-      "rewards/rejected": -9.5,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -6.21875,
+      "logps/chosen": -916.0,
+      "logps/rejected": -1784.0,
+      "loss": 0.2274,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -5.375,
+      "rewards/margins": 4.9375,
+      "rewards/rejected": -10.3125,
       "step": 930
     },
     {
       "epoch": 0.9842931937172775,
-      "grad_norm": 1.6231068998327571,
+      "grad_norm": 1.6489456510951601,
       "learning_rate": 4.249361495837478e-05,
-      "logits/chosen": -9.125,
-      "logits/rejected": -7.09375,
-      "logps/chosen": -924.0,
-      "logps/rejected": -1944.0,
-      "loss": 0.2321,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -5.53125,
-      "rewards/margins": 5.0625,
-      "rewards/rejected": -10.625,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -6.6875,
+      "logps/chosen": -940.0,
+      "logps/rejected": -2008.0,
+      "loss": 0.229,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -5.65625,
+      "rewards/margins": 5.5625,
+      "rewards/rejected": -11.25,
       "step": 940
     },
     {
       "epoch": 0.9947643979057592,
-      "grad_norm": 2.8144657670716953,
+      "grad_norm": 2.69887663284038,
       "learning_rate": 4.227467868138035e-05,
-      "logits/chosen": -8.6875,
-      "logits/rejected": -6.5,
-      "logps/chosen": -1012.0,
+      "logits/chosen": -9.125,
+      "logits/rejected": -6.78125,
+      "logps/chosen": -1000.0,
       "logps/rejected": -2000.0,
-      "loss": 0.2197,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -6.125,
-      "rewards/margins": 5.53125,
-      "rewards/rejected": -11.625,
+      "loss": 0.2163,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -6.0,
+      "rewards/margins": 5.6875,
+      "rewards/rejected": -11.6875,
       "step": 950
     },
     {
       "epoch": 1.0052356020942408,
-      "grad_norm": 0.9702130956040784,
+      "grad_norm": 1.2705076759097176,
       "learning_rate": 4.205317710224938e-05,
       "logits/chosen": -9.75,
-      "logits/rejected": -7.34375,
-      "logps/chosen": -920.0,
-      "logps/rejected": -2192.0,
-      "loss": 0.1492,
+      "logits/rejected": -7.3125,
+      "logps/chosen": -896.0,
+      "logps/rejected": -2096.0,
+      "loss": 0.1407,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -5.09375,
-      "rewards/margins": 6.9375,
-      "rewards/rejected": -12.0625,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 6.21875,
+      "rewards/rejected": -11.125,
       "step": 960
     },
     {
       "epoch": 1.0157068062827226,
-      "grad_norm": 0.7481766283682068,
+      "grad_norm": 0.7354946540629796,
       "learning_rate": 4.1829143114124733e-05,
-      "logits/chosen": -8.75,
-      "logits/rejected": -6.78125,
-      "logps/chosen": -1064.0,
+      "logits/chosen": -9.125,
+      "logits/rejected": -6.40625,
+      "logps/chosen": -1032.0,
       "logps/rejected": -2288.0,
-      "loss": 0.0744,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.6875,
-      "rewards/margins": 8.875,
+      "loss": 0.0703,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.375,
+      "rewards/margins": 9.1875,
       "rewards/rejected": -15.5625,
       "step": 970
     },
     {
       "epoch": 1.0261780104712042,
-      "grad_norm": 0.7946875440307771,
+      "grad_norm": 0.842772959119336,
       "learning_rate": 4.160260998621387e-05,
-      "logits/chosen": -8.875,
-      "logits/rejected": -6.3125,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0622,
+      "logits/chosen": -9.6875,
+      "logits/rejected": -6.25,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.066,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.78125,
-      "rewards/margins": 10.75,
-      "rewards/rejected": -18.5,
+      "rewards/chosen": -7.5625,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -20.375,
       "step": 980
     },
     {
       "epoch": 1.036649214659686,
-      "grad_norm": 0.7321348458888588,
+      "grad_norm": 0.7664175822646269,
       "learning_rate": 4.137361135884824e-05,
-      "logits/chosen": -7.90625,
-      "logits/rejected": -6.0625,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -2464.0,
-      "loss": 0.0696,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.0625,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -16.375,
+      "logits/chosen": -9.625,
+      "logits/rejected": -7.125,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0676,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -6.59375,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -18.0,
       "step": 990
     },
     {
       "epoch": 1.0471204188481675,
-      "grad_norm": 2.613912125598333,
+      "grad_norm": 1.4921957452405281,
       "learning_rate": 4.114218123848777e-05,
-      "logits/chosen": -6.46875,
-      "logits/rejected": -4.46875,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0616,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -6.125,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0529,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.25,
-      "rewards/margins": 10.375,
-      "rewards/rejected": -17.625,
+      "rewards/chosen": -6.71875,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -18.25,
       "step": 1000
     },
     {
       "epoch": 1.0471204188481675,
-      "eval_logits/chosen": -5.21875,
-      "eval_logits/rejected": -3.4375,
-      "eval_logps/chosen": -1120.0,
-      "eval_logps/rejected": -1208.0,
-      "eval_loss": 1.1158593893051147,
-      "eval_rewards/accuracies": 0.587890625,
-      "eval_rewards/chosen": -8.0,
-      "eval_rewards/margins": 1.1953125,
-      "eval_rewards/rejected": -9.1875,
-      "eval_runtime": 46.5996,
-      "eval_samples_per_second": 42.919,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -6.21875,
+      "eval_logits/rejected": -4.40625,
+      "eval_logps/chosen": -1168.0,
+      "eval_logps/rejected": -1264.0,
+      "eval_loss": 1.1147187948226929,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -8.5,
+      "eval_rewards/margins": 1.2734375,
+      "eval_rewards/rejected": -9.75,
+      "eval_runtime": 58.1694,
+      "eval_samples_per_second": 34.382,
+      "eval_steps_per_second": 0.55,
       "step": 1000
     },
     {
       "epoch": 1.057591623036649,
-      "grad_norm": 2.250311865273712,
+      "grad_norm": 1.645850086763686,
       "learning_rate": 4.0908353992670824e-05,
-      "logits/chosen": -7.53125,
-      "logits/rejected": -4.71875,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -2336.0,
-      "loss": 0.0727,
+      "logits/chosen": -7.625,
+      "logits/rejected": -4.9375,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0614,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.6875,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -16.0,
+      "rewards/chosen": -8.0,
+      "rewards/margins": 10.5,
+      "rewards/rejected": -18.5,
       "step": 1010
     },
     {
       "epoch": 1.068062827225131,
-      "grad_norm": 0.8296847282219908,
+      "grad_norm": 0.7093406641171098,
       "learning_rate": 4.067216434491059e-05,
-      "logits/chosen": -6.625,
-      "logits/rejected": -4.71875,
-      "logps/chosen": -1120.0,
-      "logps/rejected": -2400.0,
-      "loss": 0.0558,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.40625,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -16.5,
+      "logits/chosen": -9.0,
+      "logits/rejected": -6.65625,
+      "logps/chosen": -1080.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0639,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -6.9375,
+      "rewards/margins": 10.0625,
+      "rewards/rejected": -17.0,
       "step": 1020
     },
     {
       "epoch": 1.0785340314136125,
-      "grad_norm": 1.1364008183611616,
+      "grad_norm": 3.4584374650751,
       "learning_rate": 4.043364736953866e-05,
-      "logits/chosen": -6.96875,
-      "logits/rejected": -4.25,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.1024,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.5,
-      "rewards/margins": 10.5625,
-      "rewards/rejected": -18.0,
+      "logits/chosen": -9.25,
+      "logits/rejected": -6.15625,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0757,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.125,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -18.625,
       "step": 1030
     },
     {
       "epoch": 1.0890052356020943,
-      "grad_norm": 3.62151391472933,
+      "grad_norm": 2.200119537866104,
       "learning_rate": 4.019283848649643e-05,
-      "logits/chosen": -8.5,
-      "logits/rejected": -5.84375,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -2224.0,
-      "loss": 0.0695,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.625,
-      "rewards/margins": 7.75,
-      "rewards/rejected": -14.375,
+      "logits/chosen": -9.0,
+      "logits/rejected": -5.71875,
+      "logps/chosen": -1224.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0437,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 10.6875,
+      "rewards/rejected": -19.0,
       "step": 1040
     },
     {
       "epoch": 1.0994764397905759,
-      "grad_norm": 0.38700583540094974,
+      "grad_norm": 0.7485478222096574,
       "learning_rate": 3.9949773456075215e-05,
-      "logits/chosen": -6.78125,
-      "logits/rejected": -4.3125,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0495,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.78125,
-      "rewards/margins": 10.0,
-      "rewards/rejected": -16.875,
+      "logits/chosen": -8.1875,
+      "logits/rejected": -5.03125,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -2928.0,
+      "loss": 0.0398,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.78125,
+      "rewards/margins": 12.75,
+      "rewards/rejected": -20.5,
       "step": 1050
     },
     {
       "epoch": 1.1099476439790577,
-      "grad_norm": 3.6965432258570097,
+      "grad_norm": 0.4983746852146091,
       "learning_rate": 3.970448837360585e-05,
-      "logits/chosen": -7.75,
-      "logits/rejected": -5.46875,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -2512.0,
-      "loss": 0.0509,
+      "logits/chosen": -9.25,
+      "logits/rejected": -6.71875,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0567,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.03125,
-      "rewards/margins": 9.75,
-      "rewards/rejected": -16.75,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 10.25,
+      "rewards/rejected": -17.875,
       "step": 1060
     },
     {
       "epoch": 1.1204188481675392,
-      "grad_norm": 0.6255778585853107,
+      "grad_norm": 1.5775047761554017,
       "learning_rate": 3.945701966409846e-05,
-      "logits/chosen": -7.375,
-      "logits/rejected": -4.5625,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.0523,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.03125,
-      "rewards/margins": 12.4375,
-      "rewards/rejected": -19.5,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -6.375,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.048,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -7.1875,
+      "rewards/margins": 12.125,
+      "rewards/rejected": -19.25,
       "step": 1070
     },
     {
       "epoch": 1.130890052356021,
-      "grad_norm": 0.7174968129472442,
+      "grad_norm": 1.4791922288176875,
       "learning_rate": 3.9207404076833374e-05,
-      "logits/chosen": -5.71875,
-      "logits/rejected": -2.734375,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.0659,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.5,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -18.875,
+      "logits/chosen": -7.40625,
+      "logits/rejected": -4.4375,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.0531,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 11.75,
+      "rewards/rejected": -20.0,
       "step": 1080
     },
     {
       "epoch": 1.1413612565445026,
-      "grad_norm": 3.1972433211307876,
+      "grad_norm": 2.928108106533324,
       "learning_rate": 3.895567867990379e-05,
-      "logits/chosen": -6.15625,
-      "logits/rejected": -3.296875,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0611,
+      "logits/chosen": -8.75,
+      "logits/rejected": -5.78125,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.0626,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.75,
-      "rewards/margins": 10.375,
-      "rewards/rejected": -17.125,
+      "rewards/chosen": -7.3125,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -18.75,
       "step": 1090
     },
     {
       "epoch": 1.1518324607329844,
-      "grad_norm": 1.192747500225899,
+      "grad_norm": 0.7466073261652407,
       "learning_rate": 3.870188085471114e-05,
-      "logits/chosen": -5.96875,
-      "logits/rejected": -4.15625,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -2336.0,
-      "loss": 0.0692,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.0625,
+      "logits/chosen": -9.625,
+      "logits/rejected": -7.25,
+      "logps/chosen": -984.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.058,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -5.9375,
       "rewards/margins": 8.875,
-      "rewards/rejected": -15.875,
+      "rewards/rejected": -14.8125,
       "step": 1100
     },
     {
       "epoch": 1.1518324607329844,
-      "eval_logits/chosen": -5.65625,
-      "eval_logits/rejected": -4.1875,
-      "eval_logps/chosen": -1080.0,
-      "eval_logps/rejected": -1152.0,
-      "eval_loss": 1.0923594236373901,
-      "eval_rewards/accuracies": 0.5859375,
-      "eval_rewards/chosen": -7.59375,
-      "eval_rewards/margins": 1.046875,
-      "eval_rewards/rejected": -8.625,
-      "eval_runtime": 46.6083,
-      "eval_samples_per_second": 42.911,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -9.1875,
+      "eval_logits/rejected": -7.9375,
+      "eval_logps/chosen": -940.0,
+      "eval_logps/rejected": -1012.0,
+      "eval_loss": 1.0443124771118164,
+      "eval_rewards/accuracies": 0.59765625,
+      "eval_rewards/chosen": -6.25,
+      "eval_rewards/margins": 1.0,
+      "eval_rewards/rejected": -7.25,
+      "eval_runtime": 58.1299,
+      "eval_samples_per_second": 34.406,
+      "eval_steps_per_second": 0.55,
       "step": 1100
     },
     {
       "epoch": 1.162303664921466,
-      "grad_norm": 0.8704059696073393,
+      "grad_norm": 1.37175607536578,
       "learning_rate": 3.844604829041395e-05,
-      "logits/chosen": -6.90625,
-      "logits/rejected": -4.59375,
-      "logps/chosen": -996.0,
-      "logps/rejected": -2128.0,
-      "loss": 0.0589,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -6.3125,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -14.5,
-      "step": 1110
+      "logits/chosen": -9.625,
+      "logits/rejected": -6.8125,
+      "logps/chosen": -960.0,
+      "logps/rejected": -2144.0,
+      "loss": 0.0634,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -5.96875,
+      "rewards/margins": 8.75,
+      "rewards/rejected": -14.6875,
+      "step": 1110
     },
     {
       "epoch": 1.1727748691099475,
-      "grad_norm": 2.2334242222601994,
+      "grad_norm": 1.9639954551481416,
       "learning_rate": 3.818821897833095e-05,
-      "logits/chosen": -6.8125,
-      "logits/rejected": -4.4375,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -2224.0,
-      "loss": 0.0812,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.5625,
-      "rewards/margins": 7.8125,
-      "rewards/rejected": -14.375,
+      "logits/chosen": -9.1875,
+      "logits/rejected": -6.8125,
+      "logps/chosen": -1032.0,
+      "logps/rejected": -2288.0,
+      "loss": 0.0754,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -6.25,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -15.0,
       "step": 1120
     },
     {
       "epoch": 1.1832460732984293,
-      "grad_norm": 1.5384646967292843,
+      "grad_norm": 1.5250127222503982,
       "learning_rate": 3.792843120629935e-05,
-      "logits/chosen": -6.96875,
-      "logits/rejected": -4.21875,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -2416.0,
-      "loss": 0.0653,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.59375,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -15.875,
+      "logits/chosen": -8.3125,
+      "logits/rejected": -5.46875,
+      "logps/chosen": -1020.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0534,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -6.4375,
+      "rewards/margins": 10.4375,
+      "rewards/rejected": -16.875,
       "step": 1130
     },
     {
       "epoch": 1.193717277486911,
-      "grad_norm": 0.6829541621617344,
+      "grad_norm": 0.674211227096609,
       "learning_rate": 3.766672355298907e-05,
-      "logits/chosen": -5.71875,
-      "logits/rejected": -2.953125,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0491,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.75,
-      "rewards/margins": 10.25,
-      "rewards/rejected": -18.0,
+      "logits/chosen": -6.03125,
+      "logits/rejected": -3.609375,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.0636,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -8.75,
+      "rewards/margins": 11.875,
+      "rewards/rejected": -20.625,
       "step": 1140
     },
     {
       "epoch": 1.2041884816753927,
-      "grad_norm": 1.6791668363908208,
+      "grad_norm": 1.825317268090979,
       "learning_rate": 3.740313488217373e-05,
-      "logits/chosen": -7.71875,
-      "logits/rejected": -5.34375,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0681,
+      "logits/chosen": -7.78125,
+      "logits/rejected": -5.8125,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.0619,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.1875,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -17.0,
+      "rewards/chosen": -7.03125,
+      "rewards/margins": 9.3125,
+      "rewards/rejected": -16.375,
       "step": 1150
     },
     {
       "epoch": 1.2146596858638743,
-      "grad_norm": 2.808293698761363,
+      "grad_norm": 2.301976315167313,
       "learning_rate": 3.7137704336959464e-05,
-      "logits/chosen": -4.6875,
-      "logits/rejected": -2.90625,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -3088.0,
-      "loss": 0.0385,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -9.3125,
-      "rewards/margins": 13.125,
-      "rewards/rejected": -22.5,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -5.84375,
+      "logps/chosen": -976.0,
+      "logps/rejected": -2400.0,
+      "loss": 0.0403,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -6.15625,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -15.5,
       "step": 1160
     },
     {
       "epoch": 1.225130890052356,
-      "grad_norm": 3.2289248744024586,
+      "grad_norm": 4.018640963530922,
       "learning_rate": 3.687047133397201e-05,
-      "logits/chosen": -5.75,
-      "logits/rejected": -3.0625,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -3104.0,
-      "loss": 0.0863,
+      "logits/chosen": -7.59375,
+      "logits/rejected": -4.875,
+      "logps/chosen": -1080.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.0809,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 14.4375,
-      "rewards/rejected": -23.375,
+      "rewards/chosen": -7.0625,
+      "rewards/margins": 10.875,
+      "rewards/rejected": -18.0,
       "step": 1170
     },
     {
       "epoch": 1.2356020942408377,
-      "grad_norm": 0.48779647543819155,
+      "grad_norm": 0.9672650144514118,
       "learning_rate": 3.6601475557503415e-05,
-      "logits/chosen": -7.53125,
-      "logits/rejected": -4.34375,
+      "logits/chosen": -7.84375,
+      "logits/rejected": -4.6875,
       "logps/chosen": -944.0,
-      "logps/rejected": -2304.0,
-      "loss": 0.0627,
-      "rewards/accuracies": 0.96875,
+      "logps/rejected": -2272.0,
+      "loss": 0.0741,
+      "rewards/accuracies": 0.987500011920929,
       "rewards/chosen": -5.46875,
-      "rewards/margins": 9.4375,
-      "rewards/rejected": -14.875,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -14.5,
       "step": 1180
     },
     {
       "epoch": 1.2460732984293195,
-      "grad_norm": 1.782667456157701,
+      "grad_norm": 1.7208220700781585,
       "learning_rate": 3.633075695361882e-05,
-      "logits/chosen": -4.96875,
-      "logits/rejected": -3.1875,
-      "logps/chosen": -1024.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0593,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.5,
-      "rewards/margins": 9.9375,
-      "rewards/rejected": -16.375,
+      "logits/chosen": -7.0,
+      "logits/rejected": -4.3125,
+      "logps/chosen": -952.0,
+      "logps/rejected": -2320.0,
+      "loss": 0.0582,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -5.78125,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -14.875,
       "step": 1190
     },
     {
       "epoch": 1.256544502617801,
-      "grad_norm": 0.7656024005239067,
+      "grad_norm": 0.7735105728914586,
       "learning_rate": 3.6058355724224477e-05,
-      "logits/chosen": -3.921875,
-      "logits/rejected": -2.515625,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -2976.0,
-      "loss": 0.0488,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -8.5625,
-      "rewards/margins": 12.0625,
-      "rewards/rejected": -20.625,
+      "logits/chosen": -4.6875,
+      "logits/rejected": -2.875,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0436,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.875,
+      "rewards/margins": 11.125,
+      "rewards/rejected": -19.0,
       "step": 1200
     },
     {
       "epoch": 1.256544502617801,
-      "eval_logits/chosen": -1.796875,
-      "eval_logits/rejected": -0.3046875,
-      "eval_logps/chosen": -1456.0,
-      "eval_logps/rejected": -1552.0,
-      "eval_loss": 1.3851250410079956,
-      "eval_rewards/accuracies": 0.59375,
-      "eval_rewards/chosen": -11.375,
-      "eval_rewards/margins": 1.234375,
-      "eval_rewards/rejected": -12.625,
-      "eval_runtime": 46.6254,
-      "eval_samples_per_second": 42.895,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -3.09375,
+      "eval_logits/rejected": -1.3125,
+      "eval_logps/chosen": -1272.0,
+      "eval_logps/rejected": -1376.0,
+      "eval_loss": 1.1756093502044678,
+      "eval_rewards/accuracies": 0.61328125,
+      "eval_rewards/chosen": -9.5625,
+      "eval_rewards/margins": 1.34375,
+      "eval_rewards/rejected": -10.875,
+      "eval_runtime": 58.1595,
+      "eval_samples_per_second": 34.388,
+      "eval_steps_per_second": 0.55,
       "step": 1200
     },
     {
       "epoch": 1.2670157068062826,
-      "grad_norm": 3.7896355502770493,
+      "grad_norm": 3.0011996914577352,
       "learning_rate": 3.5784312321097744e-05,
-      "logits/chosen": -3.984375,
-      "logits/rejected": -2.515625,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.0682,
+      "logits/chosen": -5.6875,
+      "logits/rejected": -3.390625,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0534,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 11.5625,
-      "rewards/rejected": -20.375,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 11.6875,
+      "rewards/rejected": -20.0,
       "step": 1210
     },
     {
       "epoch": 1.2774869109947644,
-      "grad_norm": 1.5613932842171248,
+      "grad_norm": 1.6410079992224549,
       "learning_rate": 3.550866743987994e-05,
-      "logits/chosen": -4.625,
-      "logits/rejected": -2.59375,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0658,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.84375,
-      "rewards/margins": 9.875,
-      "rewards/rejected": -17.75,
+      "logits/chosen": -7.34375,
+      "logits/rejected": -4.3125,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0649,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.90625,
+      "rewards/margins": 10.75,
+      "rewards/rejected": -18.625,
       "step": 1220
     },
     {
       "epoch": 1.2879581151832462,
-      "grad_norm": 0.8993753920436516,
+      "grad_norm": 1.0413549086047504,
       "learning_rate": 3.5231462014033024e-05,
-      "logits/chosen": -5.03125,
-      "logits/rejected": -3.75,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0342,
+      "logits/chosen": -7.75,
+      "logits/rejected": -5.125,
+      "logps/chosen": -1008.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.0289,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.875,
-      "rewards/margins": 11.125,
-      "rewards/rejected": -18.0,
+      "rewards/chosen": -6.375,
+      "rewards/margins": 10.625,
+      "rewards/rejected": -17.0,
       "step": 1230
     },
     {
       "epoch": 1.2984293193717278,
-      "grad_norm": 1.3878763124879798,
+      "grad_norm": 2.022905900205324,
       "learning_rate": 3.495273720876095e-05,
-      "logits/chosen": -5.6875,
-      "logits/rejected": -3.921875,
+      "logits/chosen": -6.125,
+      "logits/rejected": -4.09375,
       "logps/chosen": -1184.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0335,
+      "logps/rejected": -2512.0,
+      "loss": 0.0349,
       "rewards/accuracies": 0.9750000238418579,
       "rewards/chosen": -7.90625,
-      "rewards/margins": 10.5625,
-      "rewards/rejected": -18.5,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -17.375,
       "step": 1240
     },
     {
       "epoch": 1.3089005235602094,
-      "grad_norm": 2.198687745840975,
+      "grad_norm": 1.8279987961065252,
       "learning_rate": 3.4672534414896616e-05,
-      "logits/chosen": -6.0625,
-      "logits/rejected": -3.078125,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -3008.0,
-      "loss": 0.0804,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 12.8125,
-      "rewards/rejected": -21.75,
+      "logits/chosen": -6.46875,
+      "logits/rejected": -3.375,
+      "logps/chosen": -1312.0,
+      "logps/rejected": -2912.0,
+      "loss": 0.0516,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -9.0,
+      "rewards/margins": 11.875,
+      "rewards/rejected": -21.0,
       "step": 1250
     },
     {
       "epoch": 1.3193717277486912,
-      "grad_norm": 3.905889305560073,
+      "grad_norm": 3.9359301350097913,
       "learning_rate": 3.439089524275526e-05,
-      "logits/chosen": -5.59375,
-      "logits/rejected": -2.9375,
+      "logits/chosen": -7.53125,
+      "logits/rejected": -4.71875,
       "logps/chosen": -1112.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0815,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.40625,
-      "rewards/margins": 11.3125,
-      "rewards/rejected": -18.75,
+      "logps/rejected": -2704.0,
+      "loss": 0.0798,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.4375,
+      "rewards/margins": 11.1875,
+      "rewards/rejected": -18.625,
       "step": 1260
     },
     {
       "epoch": 1.3298429319371727,
-      "grad_norm": 1.5676392537945478,
+      "grad_norm": 1.0478734273621086,
       "learning_rate": 3.410786151595533e-05,
-      "logits/chosen": -6.75,
-      "logits/rejected": -3.921875,
-      "logps/chosen": -968.0,
-      "logps/rejected": -2304.0,
-      "loss": 0.0785,
+      "logits/chosen": -8.1875,
+      "logits/rejected": -5.21875,
+      "logps/chosen": -1000.0,
+      "logps/rejected": -2400.0,
+      "loss": 0.0632,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -5.90625,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -15.25,
+      "rewards/chosen": -6.25,
+      "rewards/margins": 9.9375,
+      "rewards/rejected": -16.125,
       "step": 1270
     },
     {
       "epoch": 1.3403141361256545,
-      "grad_norm": 0.9297919358283999,
+      "grad_norm": 1.0643794784727114,
       "learning_rate": 3.382347526520765e-05,
-      "logits/chosen": -7.375,
-      "logits/rejected": -4.3125,
-      "logps/chosen": -956.0,
-      "logps/rejected": -2272.0,
-      "loss": 0.059,
+      "logits/chosen": -7.53125,
+      "logits/rejected": -4.15625,
+      "logps/chosen": -1008.0,
+      "logps/rejected": -2496.0,
+      "loss": 0.0486,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -5.625,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -14.625,
+      "rewards/chosen": -6.1875,
+      "rewards/margins": 10.625,
+      "rewards/rejected": -16.875,
       "step": 1280
     },
     {
       "epoch": 1.350785340314136,
-      "grad_norm": 3.303599827326764,
+      "grad_norm": 3.1756999445682537,
       "learning_rate": 3.353777872207381e-05,
-      "logits/chosen": -6.4375,
-      "logits/rejected": -4.21875,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0737,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -6.78125,
-      "rewards/margins": 10.3125,
-      "rewards/rejected": -17.125,
+      "logits/chosen": -7.15625,
+      "logits/rejected": -4.3125,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0571,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.46875,
+      "rewards/margins": 11.875,
+      "rewards/rejected": -19.375,
       "step": 1290
     },
     {
       "epoch": 1.3612565445026177,
-      "grad_norm": 0.4673430814495325,
+      "grad_norm": 0.5526071424430825,
       "learning_rate": 3.325081431269475e-05,
-      "logits/chosen": -6.4375,
-      "logits/rejected": -3.890625,
-      "logps/chosen": -984.0,
-      "logps/rejected": -2400.0,
-      "loss": 0.0483,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -6.25,
-      "rewards/margins": 10.0625,
-      "rewards/rejected": -16.25,
+      "logits/chosen": -6.46875,
+      "logits/rejected": -3.421875,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0353,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.5625,
+      "rewards/margins": 12.5,
+      "rewards/rejected": -20.125,
       "step": 1300
     },
     {
       "epoch": 1.3612565445026177,
-      "eval_logits/chosen": -6.75,
-      "eval_logits/rejected": -5.03125,
-      "eval_logps/chosen": -1104.0,
-      "eval_logps/rejected": -1192.0,
-      "eval_loss": 1.1342500448226929,
-      "eval_rewards/accuracies": 0.609375,
-      "eval_rewards/chosen": -7.875,
-      "eval_rewards/margins": 1.1640625,
-      "eval_rewards/rejected": -9.0,
-      "eval_runtime": 46.6221,
-      "eval_samples_per_second": 42.898,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -7.0625,
+      "eval_logits/rejected": -5.28125,
+      "eval_logps/chosen": -1192.0,
+      "eval_logps/rejected": -1328.0,
+      "eval_loss": 1.2986562252044678,
+      "eval_rewards/accuracies": 0.5859375,
+      "eval_rewards/chosen": -8.75,
+      "eval_rewards/margins": 1.6875,
+      "eval_rewards/rejected": -10.4375,
+      "eval_runtime": 58.1526,
+      "eval_samples_per_second": 34.392,
+      "eval_steps_per_second": 0.55,
       "step": 1300
     },
     {
       "epoch": 1.3717277486910995,
-      "grad_norm": 1.0270682894768532,
+      "grad_norm": 2.9237917772834217,
       "learning_rate": 3.296262465149046e-05,
-      "logits/chosen": -7.96875,
-      "logits/rejected": -4.96875,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.0625,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.28125,
-      "rewards/margins": 11.25,
-      "rewards/rejected": -18.5,
+      "logits/chosen": -8.75,
+      "logits/rejected": -5.5,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -3088.0,
+      "loss": 0.0534,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.84375,
+      "rewards/margins": 14.125,
+      "rewards/rejected": -22.0,
       "step": 1310
     },
     {
       "epoch": 1.3821989528795813,
-      "grad_norm": 1.9340098984342564,
+      "grad_norm": 1.1100976109406426,
       "learning_rate": 3.2673252534831687e-05,
-      "logits/chosen": -6.78125,
-      "logits/rejected": -4.125,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0562,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.84375,
-      "rewards/margins": 10.8125,
-      "rewards/rejected": -18.625,
+      "logits/chosen": -8.375,
+      "logits/rejected": -5.1875,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0507,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.5,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -20.375,
       "step": 1320
     },
     {
       "epoch": 1.3926701570680629,
-      "grad_norm": 1.7709355470935333,
+      "grad_norm": 2.166445270573239,
       "learning_rate": 3.2382740934684696e-05,
-      "logits/chosen": -5.1875,
-      "logits/rejected": -3.1875,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0873,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 10.6875,
-      "rewards/rejected": -18.75,
-      "step": 1330
+      "logits/chosen": -7.375,
+      "logits/rejected": -4.65625,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0982,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.9375,
+      "rewards/margins": 11.1875,
+      "rewards/rejected": -19.125,
+      "step": 1330
     },
     {
       "epoch": 1.4031413612565444,
-      "grad_norm": 0.7096763726744558,
+      "grad_norm": 0.5740335171132326,
       "learning_rate": 3.209113299222982e-05,
-      "logits/chosen": -4.15625,
-      "logits/rejected": -2.703125,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0618,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 9.875,
-      "rewards/rejected": -18.25,
+      "logits/chosen": -7.65625,
+      "logits/rejected": -5.53125,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.064,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.125,
+      "rewards/margins": 10.125,
+      "rewards/rejected": -17.25,
       "step": 1340
     },
     {
       "epoch": 1.4136125654450262,
-      "grad_norm": 1.7941777306082054,
+      "grad_norm": 0.6161437767157479,
       "learning_rate": 3.179847201145505e-05,
-      "logits/chosen": -5.375,
-      "logits/rejected": -2.46875,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0674,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -16.5,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -5.4375,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0622,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -6.6875,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -16.0,
       "step": 1350
     },
     {
       "epoch": 1.4240837696335078,
-      "grad_norm": 0.6058485356735879,
+      "grad_norm": 0.7673435070348803,
       "learning_rate": 3.150480145272528e-05,
-      "logits/chosen": -6.375,
-      "logits/rejected": -3.296875,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -2320.0,
-      "loss": 0.0585,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.59375,
-      "rewards/margins": 8.0,
-      "rewards/rejected": -15.625,
+      "logits/chosen": -8.25,
+      "logits/rejected": -5.125,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0555,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.78125,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -16.875,
       "step": 1360
     },
     {
       "epoch": 1.4345549738219896,
-      "grad_norm": 2.4096662560659254,
+      "grad_norm": 8.29187454766096,
       "learning_rate": 3.121016492632848e-05,
-      "logits/chosen": -6.375,
-      "logits/rejected": -3.21875,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0754,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 10.5,
-      "rewards/rejected": -18.625,
+      "logits/chosen": -7.25,
+      "logits/rejected": -4.1875,
+      "logps/chosen": -1304.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.0753,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -9.0625,
+      "rewards/margins": 12.1875,
+      "rewards/rejected": -21.25,
       "step": 1370
     },
     {
       "epoch": 1.4450261780104712,
-      "grad_norm": 2.1989874208465063,
+      "grad_norm": 2.566261754701602,
       "learning_rate": 3.091460618599951e-05,
-      "logits/chosen": -5.5,
-      "logits/rejected": -2.34375,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0502,
+      "logits/chosen": -8.0,
+      "logits/rejected": -4.53125,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.0476,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.9375,
-      "rewards/margins": 10.1875,
-      "rewards/rejected": -18.125,
+      "rewards/chosen": -8.0,
+      "rewards/margins": 11.875,
+      "rewards/rejected": -19.875,
       "step": 1380
     },
     {
       "epoch": 1.4554973821989527,
-      "grad_norm": 2.1021393633763403,
+      "grad_norm": 1.29701468983871,
       "learning_rate": 3.061816912242265e-05,
-      "logits/chosen": -4.625,
-      "logits/rejected": -1.3671875,
+      "logits/chosen": -7.75,
+      "logits/rejected": -4.125,
       "logps/chosen": -1168.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0593,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.71875,
-      "rewards/margins": 10.1875,
-      "rewards/rejected": -17.875,
+      "logps/rejected": -2784.0,
+      "loss": 0.0558,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -7.6875,
+      "rewards/margins": 12.5,
+      "rewards/rejected": -20.125,
       "step": 1390
     },
     {
       "epoch": 1.4659685863874345,
-      "grad_norm": 0.9301325395364118,
+      "grad_norm": 0.8486548844390419,
       "learning_rate": 3.0320897756713785e-05,
-      "logits/chosen": -5.0625,
-      "logits/rejected": -2.515625,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.072,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.0,
-      "rewards/margins": 11.1875,
-      "rewards/rejected": -18.25,
+      "logits/chosen": -7.71875,
+      "logits/rejected": -5.125,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.0576,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.03125,
+      "rewards/margins": 13.9375,
+      "rewards/rejected": -21.0,
       "step": 1400
     },
     {
       "epoch": 1.4659685863874345,
-      "eval_logits/chosen": -3.984375,
-      "eval_logits/rejected": -2.046875,
-      "eval_logps/chosen": -1152.0,
-      "eval_logps/rejected": -1256.0,
-      "eval_loss": 1.1480156183242798,
-      "eval_rewards/accuracies": 0.623046875,
-      "eval_rewards/chosen": -8.375,
-      "eval_rewards/margins": 1.34375,
-      "eval_rewards/rejected": -9.6875,
-      "eval_runtime": 46.6057,
-      "eval_samples_per_second": 42.913,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -6.46875,
+      "eval_logits/rejected": -4.625,
+      "eval_logps/chosen": -1128.0,
+      "eval_logps/rejected": -1240.0,
+      "eval_loss": 1.0486406087875366,
+      "eval_rewards/accuracies": 0.6171875,
+      "eval_rewards/chosen": -8.0625,
+      "eval_rewards/margins": 1.4609375,
+      "eval_rewards/rejected": -9.5625,
+      "eval_runtime": 58.2331,
+      "eval_samples_per_second": 34.345,
+      "eval_steps_per_second": 0.55,
       "step": 1400
     },
     {
       "epoch": 1.4764397905759163,
-      "grad_norm": 1.26013180124848,
+      "grad_norm": 2.841555230805402,
       "learning_rate": 3.002283623388332e-05,
-      "logits/chosen": -5.71875,
-      "logits/rejected": -2.875,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.052,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.21875,
-      "rewards/margins": 10.9375,
+      "logits/chosen": -7.4375,
+      "logits/rejected": -4.46875,
+      "logps/chosen": -1064.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.0529,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.03125,
+      "rewards/margins": 11.0625,
       "rewards/rejected": -18.125,
       "step": 1410
     },
     {
       "epoch": 1.486910994764398,
-      "grad_norm": 2.769761673223266,
+      "grad_norm": 2.9263432339170783,
       "learning_rate": 2.9724028816280507e-05,
-      "logits/chosen": -6.5625,
-      "logits/rejected": -3.375,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0481,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.3125,
+      "logits/chosen": -6.90625,
+      "logits/rejected": -3.796875,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.0578,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.59375,
       "rewards/margins": 11.3125,
-      "rewards/rejected": -18.625,
+      "rewards/rejected": -18.875,
       "step": 1420
     },
     {
       "epoch": 1.4973821989528795,
-      "grad_norm": 1.1372036302643795,
+      "grad_norm": 0.5173327869110749,
       "learning_rate": 2.9424519877020524e-05,
-      "logits/chosen": -7.78125,
-      "logits/rejected": -3.78125,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -3136.0,
-      "loss": 0.0509,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.40625,
-      "rewards/margins": 14.1875,
-      "rewards/rejected": -21.625,
+      "logits/chosen": -7.625,
+      "logits/rejected": -4.03125,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -3104.0,
+      "loss": 0.0681,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.4375,
+      "rewards/margins": 13.6875,
+      "rewards/rejected": -21.125,
       "step": 1430
     },
     {
       "epoch": 1.5078534031413613,
-      "grad_norm": 0.9495770122802057,
+      "grad_norm": 0.3111962702045739,
       "learning_rate": 2.912435389339504e-05,
-      "logits/chosen": -6.25,
-      "logits/rejected": -2.109375,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0435,
+      "logits/chosen": -6.8125,
+      "logits/rejected": -2.625,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -2496.0,
+      "loss": 0.0458,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.90625,
-      "rewards/margins": 13.125,
-      "rewards/rejected": -21.0,
+      "rewards/chosen": -6.75,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -18.25,
       "step": 1440
     },
     {
       "epoch": 1.518324607329843,
-      "grad_norm": 1.4612179008054074,
+      "grad_norm": 1.7820063113384021,
       "learning_rate": 2.8823575440267226e-05,
-      "logits/chosen": -5.96875,
-      "logits/rejected": -3.828125,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -2960.0,
-      "loss": 0.0732,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 12.875,
-      "rewards/rejected": -21.125,
+      "logits/chosen": -5.84375,
+      "logits/rejected": -3.984375,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.0568,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.71875,
+      "rewards/margins": 10.875,
+      "rewards/rejected": -18.625,
       "step": 1450
     },
     {
       "epoch": 1.5287958115183247,
-      "grad_norm": 0.9076788509055936,
+      "grad_norm": 1.1192498724631175,
       "learning_rate": 2.8522229183452465e-05,
-      "logits/chosen": -7.34375,
-      "logits/rejected": -4.1875,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -2448.0,
-      "loss": 0.096,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.75,
-      "rewards/margins": 10.0625,
-      "rewards/rejected": -16.75,
+      "logits/chosen": -5.96875,
+      "logits/rejected": -3.03125,
+      "logps/chosen": -1160.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0871,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 11.125,
+      "rewards/rejected": -18.75,
       "step": 1460
     },
     {
       "epoch": 1.5392670157068062,
-      "grad_norm": 1.4140324849573487,
+      "grad_norm": 1.2751663046944213,
       "learning_rate": 2.8220359873085373e-05,
-      "logits/chosen": -6.75,
-      "logits/rejected": -4.03125,
-      "logps/chosen": -1020.0,
-      "logps/rejected": -2416.0,
-      "loss": 0.0748,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.25,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -15.0,
+      "logits/chosen": -5.4375,
+      "logits/rejected": -3.03125,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.0506,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.40625,
+      "rewards/margins": 11.0,
+      "rewards/rejected": -18.375,
       "step": 1470
     },
     {
       "epoch": 1.5497382198952878,
-      "grad_norm": 0.6632910658098241,
+      "grad_norm": 0.9623343935382528,
       "learning_rate": 2.7918012336974388e-05,
-      "logits/chosen": -6.8125,
-      "logits/rejected": -4.46875,
-      "logps/chosen": -972.0,
-      "logps/rejected": -2432.0,
-      "loss": 0.0569,
+      "logits/chosen": -6.21875,
+      "logits/rejected": -4.0625,
+      "logps/chosen": -1024.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0468,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -5.90625,
-      "rewards/margins": 9.75,
-      "rewards/rejected": -15.625,
+      "rewards/chosen": -6.40625,
+      "rewards/margins": 10.6875,
+      "rewards/rejected": -17.125,
       "step": 1480
     },
     {
       "epoch": 1.5602094240837696,
-      "grad_norm": 1.0734424775843388,
+      "grad_norm": 0.9481724666076792,
       "learning_rate": 2.7615231473944815e-05,
-      "logits/chosen": -6.15625,
+      "logits/chosen": -6.1875,
       "logits/rejected": -2.78125,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -2464.0,
-      "loss": 0.0436,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.125,
-      "rewards/margins": 10.1875,
-      "rewards/rejected": -17.25,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.0481,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -6.625,
+      "rewards/margins": 10.25,
+      "rewards/rejected": -16.875,
       "step": 1490
     },
     {
       "epoch": 1.5706806282722514,
-      "grad_norm": 2.5913540759892877,
+      "grad_norm": 1.6846898048300776,
       "learning_rate": 2.7312062247171327e-05,
-      "logits/chosen": -5.75,
-      "logits/rejected": -2.65625,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -2912.0,
-      "loss": 0.0465,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 12.6875,
-      "rewards/rejected": -21.0,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -3.59375,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0444,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -18.625,
       "step": 1500
     },
     {
       "epoch": 1.5706806282722514,
-      "eval_logits/chosen": -4.375,
-      "eval_logits/rejected": -2.4375,
-      "eval_logps/chosen": -1280.0,
-      "eval_logps/rejected": -1416.0,
-      "eval_loss": 1.353421926498413,
-      "eval_rewards/accuracies": 0.59375,
-      "eval_rewards/chosen": -9.625,
-      "eval_rewards/margins": 1.6015625,
-      "eval_rewards/rejected": -11.25,
-      "eval_runtime": 46.606,
-      "eval_samples_per_second": 42.913,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -3.921875,
+      "eval_logits/rejected": -1.9140625,
+      "eval_logps/chosen": -1208.0,
+      "eval_logps/rejected": -1344.0,
+      "eval_loss": 1.145937442779541,
+      "eval_rewards/accuracies": 0.611328125,
+      "eval_rewards/chosen": -8.875,
+      "eval_rewards/margins": 1.6484375,
+      "eval_rewards/rejected": -10.5,
+      "eval_runtime": 58.1651,
+      "eval_samples_per_second": 34.385,
+      "eval_steps_per_second": 0.55,
       "step": 1500
     },
     {
       "epoch": 1.581151832460733,
-      "grad_norm": 0.7865525275979149,
+      "grad_norm": 0.5124867571683146,
       "learning_rate": 2.7008549677500876e-05,
-      "logits/chosen": -7.3125,
-      "logits/rejected": -3.375,
+      "logits/chosen": -6.65625,
+      "logits/rejected": -3.234375,
       "logps/chosen": -1216.0,
-      "logps/rejected": -2800.0,
-      "loss": 0.0593,
+      "logps/rejected": -2752.0,
+      "loss": 0.05,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.90625,
-      "rewards/margins": 11.8125,
-      "rewards/rejected": -19.75,
+      "rewards/chosen": -7.875,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -19.25,
       "step": 1510
     },
     {
       "epoch": 1.5916230366492146,
-      "grad_norm": 0.6506735099144141,
+      "grad_norm": 0.7172133670967106,
       "learning_rate": 2.670473883676709e-05,
-      "logits/chosen": -6.625,
-      "logits/rejected": -3.453125,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -2464.0,
-      "loss": 0.0559,
+      "logits/chosen": -6.25,
+      "logits/rejected": -3.203125,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0449,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.90625,
-      "rewards/margins": 9.6875,
-      "rewards/rejected": -16.625,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 10.625,
+      "rewards/rejected": -17.375,
       "step": 1520
     },
     {
       "epoch": 1.6020942408376964,
-      "grad_norm": 2.673506824832795,
+      "grad_norm": 2.7201634145406626,
       "learning_rate": 2.6400674841097074e-05,
-      "logits/chosen": -7.40625,
-      "logits/rejected": -4.5,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0484,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.0,
-      "rewards/margins": 10.125,
-      "rewards/rejected": -17.125,
+      "logits/chosen": -6.59375,
+      "logits/rejected": -3.859375,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.0521,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.4375,
+      "rewards/margins": 12.1875,
+      "rewards/rejected": -19.625,
       "step": 1530
     },
     {
       "epoch": 1.6125654450261782,
-      "grad_norm": 1.7574785088990774,
+      "grad_norm": 2.0792769813940417,
       "learning_rate": 2.609640284421158e-05,
-      "logits/chosen": -7.0,
-      "logits/rejected": -3.421875,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -2816.0,
-      "loss": 0.0555,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -19.125,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -2.984375,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -3136.0,
+      "loss": 0.0657,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -8.4375,
+      "rewards/margins": 13.9375,
+      "rewards/rejected": -22.375,
       "step": 1540
     },
     {
       "epoch": 1.6230366492146597,
-      "grad_norm": 1.6912586794549878,
+      "grad_norm": 0.4894322605123542,
       "learning_rate": 2.5791968030719726e-05,
-      "logits/chosen": -6.78125,
-      "logits/rejected": -3.75,
-      "logps/chosen": -1144.0,
+      "logits/chosen": -7.3125,
+      "logits/rejected": -4.375,
+      "logps/chosen": -1064.0,
       "logps/rejected": -2592.0,
-      "loss": 0.0515,
+      "loss": 0.047,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.25,
-      "rewards/margins": 10.5625,
+      "rewards/chosen": -6.4375,
+      "rewards/margins": 11.4375,
       "rewards/rejected": -17.875,
       "step": 1550
     },
     {
       "epoch": 1.6335078534031413,
-      "grad_norm": 1.5135802721981377,
+      "grad_norm": 1.470584261341943,
       "learning_rate": 2.548741560940902e-05,
-      "logits/chosen": -6.375,
-      "logits/rejected": -3.375,
-      "logps/chosen": -1120.0,
-      "logps/rejected": -2800.0,
-      "loss": 0.0575,
+      "logits/chosen": -7.5,
+      "logits/rejected": -4.4375,
+      "logps/chosen": -948.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.05,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.40625,
-      "rewards/margins": 11.5625,
-      "rewards/rejected": -19.0,
+      "rewards/chosen": -5.65625,
+      "rewards/margins": 11.625,
+      "rewards/rejected": -17.25,
       "step": 1560
     },
     {
       "epoch": 1.6439790575916229,
-      "grad_norm": 0.6368449013841351,
+      "grad_norm": 0.8317325081744855,
       "learning_rate": 2.5182790806531787e-05,
-      "logits/chosen": -6.34375,
-      "logits/rejected": -4.15625,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0693,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.90625,
-      "rewards/margins": 10.6875,
-      "rewards/rejected": -17.625,
+      "logits/chosen": -6.59375,
+      "logits/rejected": -4.375,
+      "logps/chosen": -1064.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.0674,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -6.59375,
+      "rewards/margins": 11.75,
+      "rewards/rejected": -18.25,
       "step": 1570
     },
     {
       "epoch": 1.6544502617801047,
-      "grad_norm": 1.244376273929856,
+      "grad_norm": 0.8836441018147038,
       "learning_rate": 2.4878138859089074e-05,
-      "logits/chosen": -6.78125,
-      "logits/rejected": -3.84375,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -2368.0,
-      "loss": 0.0518,
+      "logits/chosen": -6.375,
+      "logits/rejected": -3.15625,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0407,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.5625,
-      "rewards/margins": 9.6875,
-      "rewards/rejected": -16.25,
+      "rewards/chosen": -7.09375,
+      "rewards/margins": 11.3125,
+      "rewards/rejected": -18.375,
       "step": 1580
     },
     {
       "epoch": 1.6649214659685865,
-      "grad_norm": 1.6074627897892773,
+      "grad_norm": 1.3155143434735408,
       "learning_rate": 2.4573505008112924e-05,
-      "logits/chosen": -6.5625,
-      "logits/rejected": -3.609375,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0617,
+      "logits/chosen": -5.78125,
+      "logits/rejected": -2.5625,
+      "logps/chosen": -1224.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.0559,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.5625,
-      "rewards/margins": 10.25,
-      "rewards/rejected": -17.875,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 11.8125,
+      "rewards/rejected": -20.0,
       "step": 1590
     },
     {
       "epoch": 1.675392670157068,
-      "grad_norm": 0.9931518764126477,
+      "grad_norm": 1.1572568481551615,
       "learning_rate": 2.4268934491948032e-05,
-      "logits/chosen": -6.125,
-      "logits/rejected": -3.703125,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0541,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 9.5,
-      "rewards/rejected": -17.125,
+      "logits/chosen": -6.25,
+      "logits/rejected": -3.359375,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0475,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.46875,
+      "rewards/margins": 11.125,
+      "rewards/rejected": -18.625,
       "step": 1600
     },
     {
       "epoch": 1.675392670157068,
-      "eval_logits/chosen": -4.9375,
-      "eval_logits/rejected": -3.171875,
-      "eval_logps/chosen": -1168.0,
-      "eval_logps/rejected": -1264.0,
-      "eval_loss": 1.1638906002044678,
-      "eval_rewards/accuracies": 0.59375,
-      "eval_rewards/chosen": -8.5,
-      "eval_rewards/margins": 1.2265625,
-      "eval_rewards/rejected": -9.75,
-      "eval_runtime": 46.6295,
-      "eval_samples_per_second": 42.891,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -4.5625,
+      "eval_logits/rejected": -2.59375,
+      "eval_logps/chosen": -1176.0,
+      "eval_logps/rejected": -1304.0,
+      "eval_loss": 1.181796908378601,
+      "eval_rewards/accuracies": 0.591796875,
+      "eval_rewards/chosen": -8.5625,
+      "eval_rewards/margins": 1.5546875,
+      "eval_rewards/rejected": -10.125,
+      "eval_runtime": 58.1645,
+      "eval_samples_per_second": 34.385,
+      "eval_steps_per_second": 0.55,
       "step": 1600
     },
     {
       "epoch": 1.6858638743455496,
-      "grad_norm": 2.6677960230129685,
+      "grad_norm": 2.497544615164656,
       "learning_rate": 2.396447253953385e-05,
-      "logits/chosen": -7.90625,
-      "logits/rejected": -4.75,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0451,
+      "logits/chosen": -6.96875,
+      "logits/rejected": -3.796875,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -2912.0,
+      "loss": 0.0447,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.5625,
-      "rewards/margins": 11.0625,
-      "rewards/rejected": -17.625,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 13.125,
+      "rewards/rejected": -20.375,
       "step": 1610
     },
     {
       "epoch": 1.6963350785340314,
-      "grad_norm": 1.956488565372064,
+      "grad_norm": 1.6115239981635492,
       "learning_rate": 2.3660164363687998e-05,
-      "logits/chosen": -7.4375,
-      "logits/rejected": -4.28125,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.054,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.90625,
-      "rewards/margins": 10.6875,
-      "rewards/rejected": -17.5,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -3.078125,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -3168.0,
+      "loss": 0.0477,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.96875,
+      "rewards/margins": 13.75,
+      "rewards/rejected": -21.75,
       "step": 1620
     },
     {
       "epoch": 1.7068062827225132,
-      "grad_norm": 1.0190542871004828,
+      "grad_norm": 0.9078838064235139,
       "learning_rate": 2.33560551543922e-05,
-      "logits/chosen": -6.75,
-      "logits/rejected": -3.984375,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -2416.0,
-      "loss": 0.0446,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.1875,
-      "rewards/margins": 9.625,
-      "rewards/rejected": -16.75,
+      "logits/chosen": -5.6875,
+      "logits/rejected": -3.171875,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0524,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 12.4375,
+      "rewards/rejected": -20.625,
       "step": 1630
     },
     {
       "epoch": 1.7172774869109948,
-      "grad_norm": 0.6002033751460811,
+      "grad_norm": 1.5445664509625014,
       "learning_rate": 2.3052190072081493e-05,
-      "logits/chosen": -6.75,
-      "logits/rejected": -3.90625,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0504,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.75,
-      "rewards/margins": 11.5,
-      "rewards/rejected": -19.25,
+      "logits/chosen": -5.78125,
+      "logits/rejected": -2.859375,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -3056.0,
+      "loss": 0.0494,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 13.4375,
+      "rewards/rejected": -21.875,
       "step": 1640
     },
     {
       "epoch": 1.7277486910994764,
-      "grad_norm": 0.9271446946488918,
+      "grad_norm": 0.759656956011693,
       "learning_rate": 2.2748614240937864e-05,
-      "logits/chosen": -6.125,
-      "logits/rejected": -3.578125,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -2848.0,
-      "loss": 0.0623,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.15625,
-      "rewards/margins": 12.0625,
-      "rewards/rejected": -19.25,
+      "logits/chosen": -4.9375,
+      "logits/rejected": -2.53125,
+      "logps/chosen": -1160.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.0588,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.9375,
+      "rewards/margins": 13.25,
+      "rewards/rejected": -21.125,
       "step": 1650
     },
     {
       "epoch": 1.738219895287958,
-      "grad_norm": 0.3180868890161927,
+      "grad_norm": 0.8359741374732479,
       "learning_rate": 2.2445372742189334e-05,
-      "logits/chosen": -7.09375,
-      "logits/rejected": -4.59375,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.0385,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.03125,
-      "rewards/margins": 12.4375,
-      "rewards/rejected": -19.5,
+      "logits/chosen": -6.46875,
+      "logits/rejected": -3.8125,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -2976.0,
+      "loss": 0.0456,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -7.40625,
+      "rewards/margins": 12.8125,
+      "rewards/rejected": -20.25,
       "step": 1660
     },
     {
       "epoch": 1.7486910994764397,
-      "grad_norm": 2.0956106466164863,
+      "grad_norm": 2.5836673486582553,
       "learning_rate": 2.214251060741528e-05,
-      "logits/chosen": -7.375,
-      "logits/rejected": -4.28125,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0464,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.15625,
-      "rewards/margins": 12.1875,
-      "rewards/rejected": -19.375,
+      "logits/chosen": -6.9375,
+      "logits/rejected": -3.703125,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.0412,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.53125,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -19.875,
       "step": 1670
     },
     {
       "epoch": 1.7591623036649215,
-      "grad_norm": 2.572078432294397,
+      "grad_norm": 1.0460127711190368,
       "learning_rate": 2.1840072811859302e-05,
-      "logits/chosen": -6.5,
-      "logits/rejected": -4.0,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -2944.0,
-      "loss": 0.0566,
+      "logits/chosen": -6.6875,
+      "logits/rejected": -4.09375,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -2912.0,
+      "loss": 0.0453,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.21875,
-      "rewards/margins": 12.4375,
-      "rewards/rejected": -19.75,
+      "rewards/chosen": -7.1875,
+      "rewards/margins": 12.0625,
+      "rewards/rejected": -19.25,
       "step": 1680
     },
     {
       "epoch": 1.7696335078534031,
-      "grad_norm": 0.6634526549558009,
+      "grad_norm": 1.4784677768335746,
       "learning_rate": 2.1538104267750287e-05,
-      "logits/chosen": -4.96875,
-      "logits/rejected": -3.265625,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.035,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.5625,
-      "rewards/margins": 11.6875,
-      "rewards/rejected": -19.25,
+      "logits/chosen": -5.53125,
+      "logits/rejected": -3.59375,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0378,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -7.46875,
+      "rewards/margins": 11.0625,
+      "rewards/rejected": -18.5,
       "step": 1690
     },
     {
       "epoch": 1.7801047120418847,
-      "grad_norm": 0.5921501795028989,
+      "grad_norm": 0.8629209408093159,
       "learning_rate": 2.1236649817632952e-05,
-      "logits/chosen": -4.40625,
-      "logits/rejected": -2.8125,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -2912.0,
-      "loss": 0.0826,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -20.625,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -3.59375,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0644,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.96875,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -19.375,
       "step": 1700
     },
     {
       "epoch": 1.7801047120418847,
-      "eval_logits/chosen": -3.640625,
-      "eval_logits/rejected": -1.8203125,
-      "eval_logps/chosen": -1272.0,
-      "eval_logps/rejected": -1392.0,
-      "eval_loss": 1.2758437395095825,
-      "eval_rewards/accuracies": 0.59765625,
-      "eval_rewards/chosen": -9.5,
-      "eval_rewards/margins": 1.4921875,
-      "eval_rewards/rejected": -11.0,
-      "eval_runtime": 46.6071,
-      "eval_samples_per_second": 42.912,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -4.59375,
+      "eval_logits/rejected": -2.734375,
+      "eval_logps/chosen": -1280.0,
+      "eval_logps/rejected": -1416.0,
+      "eval_loss": 1.2222031354904175,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -9.625,
+      "eval_rewards/margins": 1.65625,
+      "eval_rewards/rejected": -11.25,
+      "eval_runtime": 58.2435,
+      "eval_samples_per_second": 34.339,
+      "eval_steps_per_second": 0.549,
       "step": 1700
     },
     {
       "epoch": 1.7905759162303665,
-      "grad_norm": 0.6237010728675902,
+      "grad_norm": 1.2453196882202302,
       "learning_rate": 2.093575422770872e-05,
-      "logits/chosen": -5.0625,
-      "logits/rejected": -2.25,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -2512.0,
-      "loss": 0.0346,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.375,
-      "rewards/margins": 10.3125,
-      "rewards/rejected": -17.75,
+      "logits/chosen": -6.28125,
+      "logits/rejected": -3.296875,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.0427,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.4375,
+      "rewards/margins": 10.0625,
+      "rewards/rejected": -17.5,
       "step": 1710
     },
     {
       "epoch": 1.8010471204188483,
-      "grad_norm": 0.6060475944082778,
+      "grad_norm": 1.641672404777744,
       "learning_rate": 2.0635462181187827e-05,
-      "logits/chosen": -5.75,
-      "logits/rejected": -3.390625,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -3104.0,
-      "loss": 0.0478,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 13.375,
-      "rewards/rejected": -21.5,
+      "logits/chosen": -7.9375,
+      "logits/rejected": -5.5,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.0501,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.71875,
+      "rewards/margins": 11.625,
+      "rewards/rejected": -19.375,
       "step": 1720
     },
     {
       "epoch": 1.8115183246073299,
-      "grad_norm": 2.855243261816164,
+      "grad_norm": 1.7018517645065043,
       "learning_rate": 2.0335818271653917e-05,
-      "logits/chosen": -5.3125,
-      "logits/rejected": -2.875,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -2928.0,
-      "loss": 0.0558,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.90625,
-      "rewards/margins": 12.5,
-      "rewards/rejected": -20.375,
+      "logits/chosen": -8.3125,
+      "logits/rejected": -5.84375,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.0488,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.0625,
+      "rewards/margins": 11.3125,
+      "rewards/rejected": -18.375,
       "step": 1730
     },
-    {
-      "epoch": 1.8219895287958114,
-      "grad_norm": 0.3835532669632143,
-      "learning_rate": 2.0036866996441815e-05,
-      "logits/chosen": -5.5625,
-      "logits/rejected": -2.265625,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -2928.0,
-      "loss": 0.0364,
+    {
+      "epoch": 1.8219895287958114,
+      "grad_norm": 1.0687032056125314,
+      "learning_rate": 2.0036866996441815e-05,
+      "logits/chosen": -8.5625,
+      "logits/rejected": -5.375,
+      "logps/chosen": -1160.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0337,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 12.625,
-      "rewards/rejected": -20.875,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 11.75,
+      "rewards/rejected": -19.375,
       "step": 1740
     },
     {
       "epoch": 1.8324607329842932,
-      "grad_norm": 2.378925026693113,
+      "grad_norm": 1.5811457375490994,
       "learning_rate": 1.9738652750029677e-05,
-      "logits/chosen": -4.6875,
-      "logits/rejected": -1.7890625,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -2928.0,
-      "loss": 0.0672,
+      "logits/chosen": -7.03125,
+      "logits/rejected": -4.25,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.0533,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -8.625,
-      "rewards/margins": 12.5,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 11.875,
+      "rewards/rejected": -20.375,
       "step": 1750
     },
     {
       "epoch": 1.8429319371727748,
-      "grad_norm": 2.440859821266978,
+      "grad_norm": 2.7473414695110807,
       "learning_rate": 1.944121981744631e-05,
-      "logits/chosen": -5.53125,
-      "logits/rejected": -1.9140625,
-      "logps/chosen": -1280.0,
+      "logits/chosen": -7.4375,
+      "logits/rejected": -4.28125,
+      "logps/chosen": -1328.0,
       "logps/rejected": -2768.0,
-      "loss": 0.0341,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 11.6875,
-      "rewards/rejected": -20.0,
+      "loss": 0.0387,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -8.8125,
+      "rewards/margins": 11.125,
+      "rewards/rejected": -19.875,
       "step": 1760
     },
     {
       "epoch": 1.8534031413612566,
-      "grad_norm": 1.5552848078280324,
+      "grad_norm": 1.0185906923037513,
       "learning_rate": 1.9144612367694887e-05,
-      "logits/chosen": -5.03125,
-      "logits/rejected": -2.109375,
+      "logits/chosen": -7.84375,
+      "logits/rejected": -4.84375,
       "logps/chosen": -1152.0,
       "logps/rejected": -2608.0,
-      "loss": 0.0641,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 10.875,
-      "rewards/rejected": -18.5,
+      "loss": 0.054,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -7.65625,
+      "rewards/margins": 10.75,
+      "rewards/rejected": -18.375,
       "step": 1770
     },
     {
       "epoch": 1.8638743455497382,
-      "grad_norm": 0.5457509976908292,
+      "grad_norm": 0.3105722891153544,
       "learning_rate": 1.8848874447193803e-05,
-      "logits/chosen": -4.5625,
-      "logits/rejected": -1.90625,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0405,
+      "logits/chosen": -7.25,
+      "logits/rejected": -4.15625,
+      "logps/chosen": -1080.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0445,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.375,
-      "rewards/margins": 10.375,
-      "rewards/rejected": -17.75,
+      "rewards/chosen": -7.0625,
+      "rewards/margins": 10.1875,
+      "rewards/rejected": -17.25,
       "step": 1780
     },
     {
       "epoch": 1.8743455497382198,
-      "grad_norm": 0.5938368335839197,
+      "grad_norm": 0.7819035983494802,
       "learning_rate": 1.855404997323571e-05,
-      "logits/chosen": -6.03125,
-      "logits/rejected": -3.28125,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0485,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 11.0625,
-      "rewards/rejected": -18.75,
+      "logits/chosen": -8.125,
+      "logits/rejected": -5.125,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0351,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 10.5625,
+      "rewards/rejected": -17.375,
       "step": 1790
     },
     {
       "epoch": 1.8848167539267016,
-      "grad_norm": 0.7839961493455135,
+      "grad_norm": 1.0794159743957012,
       "learning_rate": 1.82601827274658e-05,
-      "logits/chosen": -6.15625,
-      "logits/rejected": -3.375,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -2432.0,
-      "loss": 0.0429,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.96875,
-      "rewards/margins": 9.9375,
-      "rewards/rejected": -16.875,
+      "logits/chosen": -7.75,
+      "logits/rejected": -4.8125,
+      "logps/chosen": -1024.0,
+      "logps/rejected": -2384.0,
+      "loss": 0.0397,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -6.53125,
+      "rewards/margins": 9.875,
+      "rewards/rejected": -16.375,
       "step": 1800
     },
     {
       "epoch": 1.8848167539267016,
-      "eval_logits/chosen": -3.328125,
-      "eval_logits/rejected": -1.3671875,
-      "eval_logps/chosen": -1176.0,
-      "eval_logps/rejected": -1288.0,
-      "eval_loss": 1.1922187805175781,
-      "eval_rewards/accuracies": 0.59375,
-      "eval_rewards/chosen": -8.5625,
-      "eval_rewards/margins": 1.4375,
-      "eval_rewards/rejected": -10.0,
-      "eval_runtime": 46.6204,
-      "eval_samples_per_second": 42.9,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -5.375,
+      "eval_logits/rejected": -3.34375,
+      "eval_logps/chosen": -1096.0,
+      "eval_logps/rejected": -1224.0,
+      "eval_loss": 1.083234429359436,
+      "eval_rewards/accuracies": 0.6171875,
+      "eval_rewards/chosen": -7.8125,
+      "eval_rewards/margins": 1.546875,
+      "eval_rewards/rejected": -9.375,
+      "eval_runtime": 58.2434,
+      "eval_samples_per_second": 34.339,
+      "eval_steps_per_second": 0.549,
       "step": 1800
     },
     {
       "epoch": 1.8952879581151834,
-      "grad_norm": 3.549759264634964,
+      "grad_norm": 3.4450761905028844,
       "learning_rate": 1.7967316349380213e-05,
-      "logits/chosen": -4.78125,
-      "logits/rejected": -2.453125,
-      "logps/chosen": -1112.0,
+      "logits/chosen": -6.71875,
+      "logits/rejected": -3.546875,
+      "logps/chosen": -1064.0,
       "logps/rejected": -2624.0,
-      "loss": 0.0459,
+      "loss": 0.0508,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.46875,
-      "rewards/margins": 10.9375,
-      "rewards/rejected": -18.5,
+      "rewards/chosen": -7.03125,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -18.375,
       "step": 1810
     },
     {
       "epoch": 1.905759162303665,
-      "grad_norm": 1.5781247488712187,
+      "grad_norm": 1.983354531382712,
       "learning_rate": 1.7675494329845516e-05,
-      "logits/chosen": -3.96875,
-      "logits/rejected": -1.6796875,
-      "logps/chosen": -1216.0,
+      "logits/chosen": -4.96875,
+      "logits/rejected": -2.625,
+      "logps/chosen": -1224.0,
       "logps/rejected": -2848.0,
-      "loss": 0.0351,
-      "rewards/accuracies": 0.96875,
+      "loss": 0.0347,
+      "rewards/accuracies": 0.981249988079071,
       "rewards/chosen": -8.375,
-      "rewards/margins": 11.875,
-      "rewards/rejected": -20.25,
+      "rewards/margins": 11.75,
+      "rewards/rejected": -20.125,
       "step": 1820
     },
     {
       "epoch": 1.9162303664921465,
-      "grad_norm": 0.3682117033095076,
+      "grad_norm": 0.36892975908213066,
       "learning_rate": 1.738476000464026e-05,
-      "logits/chosen": -3.65625,
-      "logits/rejected": -1.515625,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.0531,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.65625,
-      "rewards/margins": 12.625,
-      "rewards/rejected": -20.375,
+      "logits/chosen": -4.625,
+      "logits/rejected": -2.234375,
+      "logps/chosen": -1160.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.0453,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -7.78125,
+      "rewards/margins": 12.375,
+      "rewards/rejected": -20.125,
       "step": 1830
     },
     {
       "epoch": 1.9267015706806283,
-      "grad_norm": 1.527332351896551,
+      "grad_norm": 1.2166511643244515,
       "learning_rate": 1.7095156548019646e-05,
-      "logits/chosen": -4.625,
-      "logits/rejected": -2.65625,
-      "logps/chosen": -1120.0,
+      "logits/chosen": -6.53125,
+      "logits/rejected": -3.875,
+      "logps/chosen": -1080.0,
       "logps/rejected": -2656.0,
-      "loss": 0.0483,
+      "loss": 0.0659,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.5,
-      "rewards/margins": 10.875,
-      "rewards/rejected": -18.375,
+      "rewards/chosen": -7.125,
+      "rewards/margins": 11.1875,
+      "rewards/rejected": -18.25,
       "step": 1840
     },
     {
       "epoch": 1.93717277486911,
-      "grad_norm": 1.9043489447219888,
+      "grad_norm": 2.1562017211875264,
       "learning_rate": 1.6806726966304064e-05,
-      "logits/chosen": -4.75,
-      "logits/rejected": -1.6640625,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -2848.0,
-      "loss": 0.0583,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.84375,
-      "rewards/margins": 11.875,
-      "rewards/rejected": -19.75,
+      "logits/chosen": -6.78125,
+      "logits/rejected": -3.203125,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0451,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -19.125,
       "step": 1850
     },
     {
       "epoch": 1.9476439790575917,
-      "grad_norm": 0.7121539779861119,
+      "grad_norm": 0.8958313858159986,
       "learning_rate": 1.6519514091492625e-05,
-      "logits/chosen": -4.90625,
-      "logits/rejected": -2.59375,
-      "logps/chosen": -1120.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.042,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.59375,
-      "rewards/margins": 11.75,
-      "rewards/rejected": -19.375,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -3.703125,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0356,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.6875,
+      "rewards/margins": 11.25,
+      "rewards/rejected": -18.875,
       "step": 1860
     },
     {
       "epoch": 1.9581151832460733,
-      "grad_norm": 0.8201059921921471,
+      "grad_norm": 2.146716799304083,
       "learning_rate": 1.6233560574902627e-05,
-      "logits/chosen": -5.125,
-      "logits/rejected": -2.6875,
-      "logps/chosen": -1064.0,
+      "logits/chosen": -6.40625,
+      "logits/rejected": -3.875,
+      "logps/chosen": -1104.0,
       "logps/rejected": -2800.0,
-      "loss": 0.067,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.1875,
-      "rewards/margins": 12.1875,
+      "loss": 0.0691,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -7.5625,
+      "rewards/margins": 11.75,
       "rewards/rejected": -19.375,
       "step": 1870
     },
     {
       "epoch": 1.9685863874345548,
-      "grad_norm": 1.124249061970188,
+      "grad_norm": 2.1142973322051652,
       "learning_rate": 1.5948908880835753e-05,
-      "logits/chosen": -6.09375,
-      "logits/rejected": -2.625,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -2336.0,
-      "loss": 0.0413,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.84375,
-      "rewards/margins": 10.375,
-      "rewards/rejected": -17.25,
+      "logits/chosen": -7.21875,
+      "logits/rejected": -3.921875,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -2352.0,
+      "loss": 0.0425,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.46875,
+      "rewards/margins": 9.875,
+      "rewards/rejected": -17.375,
       "step": 1880
     },
     {
       "epoch": 1.9790575916230366,
-      "grad_norm": 0.42217240435285186,
+      "grad_norm": 0.277197540929884,
       "learning_rate": 1.5665601280272124e-05,
-      "logits/chosen": -6.46875,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.0465,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.09375,
-      "rewards/margins": 12.0625,
-      "rewards/rejected": -19.125,
+      "logits/chosen": -7.59375,
+      "logits/rejected": -4.625,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0317,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.46875,
+      "rewards/margins": 11.0625,
+      "rewards/rejected": -18.5,
       "step": 1890
     },
     {
       "epoch": 1.9895287958115184,
-      "grad_norm": 2.3849339432534675,
+      "grad_norm": 2.379211956017916,
       "learning_rate": 1.538367984459298e-05,
-      "logits/chosen": -6.15625,
-      "logits/rejected": -3.625,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0316,
+      "logits/chosen": -6.96875,
+      "logits/rejected": -4.40625,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.0254,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.28125,
-      "rewards/margins": 12.5,
-      "rewards/rejected": -19.75,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 11.5625,
+      "rewards/rejected": -19.125,
       "step": 1900
     },
     {
       "epoch": 1.9895287958115184,
-      "eval_logits/chosen": -4.21875,
-      "eval_logits/rejected": -2.296875,
-      "eval_logps/chosen": -1240.0,
-      "eval_logps/rejected": -1384.0,
-      "eval_loss": 1.284999966621399,
-      "eval_rewards/accuracies": 0.59375,
-      "eval_rewards/chosen": -9.25,
-      "eval_rewards/margins": 1.6875,
-      "eval_rewards/rejected": -10.9375,
-      "eval_runtime": 47.0908,
-      "eval_samples_per_second": 42.471,
-      "eval_steps_per_second": 0.68,
+      "eval_logits/chosen": -5.46875,
+      "eval_logits/rejected": -3.734375,
+      "eval_logps/chosen": -1296.0,
+      "eval_logps/rejected": -1432.0,
+      "eval_loss": 1.1882187128067017,
+      "eval_rewards/accuracies": 0.619140625,
+      "eval_rewards/chosen": -9.8125,
+      "eval_rewards/margins": 1.671875,
+      "eval_rewards/rejected": -11.4375,
+      "eval_runtime": 58.2815,
+      "eval_samples_per_second": 34.316,
+      "eval_steps_per_second": 0.549,
       "step": 1900
     },
     {
       "epoch": 2.0,
-      "grad_norm": 0.8824324794803711,
+      "grad_norm": 0.9605758945135122,
       "learning_rate": 1.5103186439333133e-05,
-      "logits/chosen": -4.96875,
-      "logits/rejected": -2.359375,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.0657,
+      "logits/chosen": -6.625,
+      "logits/rejected": -3.8125,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0662,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.71875,
-      "rewards/margins": 13.0,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 11.625,
+      "rewards/rejected": -19.875,
       "step": 1910
     },
     {
       "epoch": 2.0104712041884816,
-      "grad_norm": 0.16171891440924113,
+      "grad_norm": 0.16514201654236546,
       "learning_rate": 1.4824162717963829e-05,
-      "logits/chosen": -5.21875,
-      "logits/rejected": -1.3359375,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.0073,
+      "logits/chosen": -7.15625,
+      "logits/rejected": -3.375,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0061,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.78125,
-      "rewards/margins": 13.25,
-      "rewards/rejected": -21.0,
+      "rewards/chosen": -8.0625,
+      "rewards/margins": 11.875,
+      "rewards/rejected": -20.0,
       "step": 1920
     },
     {
       "epoch": 2.020942408376963,
-      "grad_norm": 0.07515183746500824,
+      "grad_norm": 0.11006759849355995,
       "learning_rate": 1.4546650115707192e-05,
-      "logits/chosen": -4.84375,
-      "logits/rejected": -2.234375,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.0059,
+      "logits/chosen": -6.65625,
+      "logits/rejected": -3.921875,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -3184.0,
+      "loss": 0.0056,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.0,
-      "rewards/margins": 15.3125,
-      "rewards/rejected": -23.375,
+      "rewards/chosen": -8.5625,
+      "rewards/margins": 14.25,
+      "rewards/rejected": -22.75,
       "step": 1930
     },
     {
       "epoch": 2.031413612565445,
-      "grad_norm": 0.20491512087733108,
+      "grad_norm": 0.13890406005804348,
       "learning_rate": 1.4270689843383111e-05,
-      "logits/chosen": -4.5,
-      "logits/rejected": -2.3125,
-      "logps/chosen": -1208.0,
+      "logits/chosen": -6.21875,
+      "logits/rejected": -3.703125,
+      "logps/chosen": -1264.0,
       "logps/rejected": -3120.0,
-      "loss": 0.0141,
+      "loss": 0.0106,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 14.625,
-      "rewards/rejected": -22.75,
+      "rewards/chosen": -8.6875,
+      "rewards/margins": 13.9375,
+      "rewards/rejected": -22.625,
       "step": 1940
     },
     {
       "epoch": 2.0418848167539267,
-      "grad_norm": 0.04538128457119996,
+      "grad_norm": 0.08152356292363548,
       "learning_rate": 1.3996322881289347e-05,
-      "logits/chosen": -5.875,
-      "logits/rejected": -1.9375,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -2960.0,
-      "loss": 0.0103,
+      "logits/chosen": -6.875,
+      "logits/rejected": -3.046875,
+      "logps/chosen": -1288.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 13.875,
-      "rewards/rejected": -21.875,
+      "rewards/chosen": -9.3125,
+      "rewards/margins": 13.375,
+      "rewards/rejected": -22.75,
       "step": 1950
     },
     {
       "epoch": 2.0523560209424083,
-      "grad_norm": 0.17247519701512967,
+      "grad_norm": 0.2898384338780262,
       "learning_rate": 1.3723589973115963e-05,
-      "logits/chosen": -5.65625,
-      "logits/rejected": -3.09375,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -2992.0,
-      "loss": 0.0102,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.65625,
-      "rewards/margins": 14.0,
-      "rewards/rejected": -21.75,
+      "logits/chosen": -6.28125,
+      "logits/rejected": -3.703125,
+      "logps/chosen": -1296.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.0057,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -9.25,
+      "rewards/margins": 14.125,
+      "rewards/rejected": -23.25,
       "step": 1960
     },
     {
       "epoch": 2.06282722513089,
-      "grad_norm": 0.07365972800935087,
+      "grad_norm": 0.9349821886869782,
       "learning_rate": 1.3452531619894837e-05,
-      "logits/chosen": -5.125,
-      "logits/rejected": -2.078125,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.0065,
+      "logits/chosen": -6.4375,
+      "logits/rejected": -3.15625,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.96875,
-      "rewards/margins": 14.8125,
-      "rewards/rejected": -22.75,
+      "rewards/chosen": -8.8125,
+      "rewards/margins": 14.75,
+      "rewards/rejected": -23.5,
       "step": 1970
     },
     {
       "epoch": 2.073298429319372,
-      "grad_norm": 0.30648299768145193,
+      "grad_norm": 0.7792160486597904,
       "learning_rate": 1.3183188073985245e-05,
-      "logits/chosen": -4.96875,
-      "logits/rejected": -1.90625,
-      "logps/chosen": -1232.0,
+      "logits/chosen": -6.34375,
+      "logits/rejected": -3.1875,
+      "logps/chosen": -1296.0,
       "logps/rejected": -3216.0,
-      "loss": 0.007,
+      "loss": 0.0074,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 14.75,
-      "rewards/rejected": -23.375,
+      "rewards/chosen": -9.3125,
+      "rewards/margins": 14.1875,
+      "rewards/rejected": -23.5,
       "step": 1980
     },
     {
       "epoch": 2.0837696335078535,
-      "grad_norm": 0.05717844466127662,
+      "grad_norm": 0.2870832241324423,
       "learning_rate": 1.2915599333096351e-05,
-      "logits/chosen": -5.5625,
-      "logits/rejected": -2.625,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -3056.0,
-      "loss": 0.0106,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.625,
+      "logits/chosen": -7.03125,
+      "logits/rejected": -3.796875,
+      "logps/chosen": -1320.0,
+      "logps/rejected": -3120.0,
+      "loss": 0.006,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -9.125,
       "rewards/margins": 14.3125,
-      "rewards/rejected": -23.0,
+      "rewards/rejected": -23.5,
       "step": 1990
     },
     {
       "epoch": 2.094240837696335,
-      "grad_norm": 0.04948871584936732,
+      "grad_norm": 0.030162034545145527,
       "learning_rate": 1.264980513434752e-05,
-      "logits/chosen": -5.3125,
-      "logits/rejected": -1.953125,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -3392.0,
-      "loss": 0.0026,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.375,
-      "rewards/margins": 15.6875,
-      "rewards/rejected": -24.0,
+      "logits/chosen": -6.53125,
+      "logits/rejected": -3.078125,
+      "logps/chosen": -1304.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0037,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -8.9375,
+      "rewards/margins": 15.3125,
+      "rewards/rejected": -24.25,
       "step": 2000
     },
     {
       "epoch": 2.094240837696335,
-      "eval_logits/chosen": -3.015625,
-      "eval_logits/rejected": -1.09375,
-      "eval_logps/chosen": -1376.0,
-      "eval_logps/rejected": -1528.0,
-      "eval_loss": 1.4025624990463257,
-      "eval_rewards/accuracies": 0.595703125,
-      "eval_rewards/chosen": -10.5625,
-      "eval_rewards/margins": 1.7890625,
-      "eval_rewards/rejected": -12.375,
-      "eval_runtime": 46.6131,
-      "eval_samples_per_second": 42.906,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -4.53125,
+      "eval_logits/rejected": -2.59375,
+      "eval_logps/chosen": -1432.0,
+      "eval_logps/rejected": -1600.0,
+      "eval_loss": 1.3353281021118164,
+      "eval_rewards/accuracies": 0.61328125,
+      "eval_rewards/chosen": -11.125,
+      "eval_rewards/margins": 1.9765625,
+      "eval_rewards/rejected": -13.125,
+      "eval_runtime": 58.2105,
+      "eval_samples_per_second": 34.358,
+      "eval_steps_per_second": 0.55,
       "step": 2000
     },
     {
       "epoch": 2.1047120418848166,
-      "grad_norm": 0.1368090430556953,
+      "grad_norm": 0.05161554859820146,
       "learning_rate": 1.2385844948367324e-05,
-      "logits/chosen": -4.5,
-      "logits/rejected": -2.328125,
-      "logps/chosen": -1208.0,
+      "logits/chosen": -5.96875,
+      "logits/rejected": -3.421875,
+      "logps/chosen": -1256.0,
       "logps/rejected": -2992.0,
-      "loss": 0.0086,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 13.6875,
+      "loss": 0.0094,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -9.0,
+      "rewards/margins": 13.3125,
       "rewards/rejected": -22.25,
       "step": 2010
     },
     {
       "epoch": 2.115183246073298,
-      "grad_norm": 0.1387461358352612,
+      "grad_norm": 0.06246283096476197,
       "learning_rate": 1.2123757973432114e-05,
-      "logits/chosen": -5.09375,
-      "logits/rejected": -1.796875,
-      "logps/chosen": -1312.0,
+      "logits/chosen": -6.375,
+      "logits/rejected": -3.015625,
+      "logps/chosen": -1344.0,
       "logps/rejected": -3088.0,
-      "loss": 0.009,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 13.875,
-      "rewards/rejected": -22.875,
+      "loss": 0.0079,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -9.375,
+      "rewards/margins": 13.5625,
+      "rewards/rejected": -23.0,
       "step": 2020
     },
     {
       "epoch": 2.1256544502617802,
-      "grad_norm": 0.19711086590307134,
+      "grad_norm": 0.08679177112840931,
       "learning_rate": 1.1863583129645106e-05,
-      "logits/chosen": -4.875,
-      "logits/rejected": -2.328125,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -3184.0,
-      "loss": 0.0119,
+      "logits/chosen": -6.0625,
+      "logits/rejected": -3.3125,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0114,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 14.6875,
-      "rewards/rejected": -23.5,
+      "rewards/chosen": -9.8125,
+      "rewards/margins": 14.875,
+      "rewards/rejected": -24.75,
       "step": 2030
     },
     {
       "epoch": 2.136125654450262,
-      "grad_norm": 0.2547246144898433,
+      "grad_norm": 0.3492045896175071,
       "learning_rate": 1.1605359053156607e-05,
-      "logits/chosen": -4.875,
-      "logits/rejected": -1.8125,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0086,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 15.25,
-      "rewards/rejected": -23.75,
+      "logits/chosen": -6.3125,
+      "logits/rejected": -2.984375,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0075,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -9.625,
+      "rewards/margins": 15.75,
+      "rewards/rejected": -25.375,
       "step": 2040
     },
     {
       "epoch": 2.1465968586387434,
-      "grad_norm": 0.179542142301364,
+      "grad_norm": 1.1211714877612753,
       "learning_rate": 1.1349124090426644e-05,
-      "logits/chosen": -5.03125,
-      "logits/rejected": -1.9765625,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -3040.0,
-      "loss": 0.0072,
+      "logits/chosen": -6.0,
+      "logits/rejected": -2.890625,
+      "logps/chosen": -1376.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.0068,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 14.0625,
-      "rewards/rejected": -22.125,
+      "rewards/chosen": -9.8125,
+      "rewards/margins": 15.0625,
+      "rewards/rejected": -24.875,
       "step": 2050
     },
     {
       "epoch": 2.157068062827225,
-      "grad_norm": 0.10698304849662435,
+      "grad_norm": 0.07547649911123329,
       "learning_rate": 1.1094916292530406e-05,
-      "logits/chosen": -4.6875,
-      "logits/rejected": -2.078125,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -2992.0,
-      "loss": 0.0072,
+      "logits/chosen": -5.90625,
+      "logits/rejected": -3.046875,
+      "logps/chosen": -1408.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0063,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 13.4375,
-      "rewards/rejected": -21.75,
+      "rewards/chosen": -9.875,
+      "rewards/margins": 14.6875,
+      "rewards/rejected": -24.625,
       "step": 2060
     },
     {
       "epoch": 2.167539267015707,
-      "grad_norm": 0.41561573801959933,
+      "grad_norm": 0.12516540800032933,
       "learning_rate": 1.0842773409507623e-05,
-      "logits/chosen": -4.65625,
-      "logits/rejected": -1.75,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -3024.0,
-      "loss": 0.0101,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.0,
-      "rewards/margins": 14.375,
-      "rewards/rejected": -22.375,
+      "logits/chosen": -6.0,
+      "logits/rejected": -2.9375,
+      "logps/chosen": -1296.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.0078,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -9.375,
+      "rewards/margins": 15.1875,
+      "rewards/rejected": -24.5,
       "step": 2070
     },
     {
       "epoch": 2.1780104712041886,
-      "grad_norm": 0.02359351706617315,
+      "grad_norm": 0.10558918925859301,
       "learning_rate": 1.0592732884756754e-05,
-      "logits/chosen": -5.65625,
-      "logits/rejected": -2.390625,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -3104.0,
-      "loss": 0.0061,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 14.5,
-      "rewards/rejected": -22.75,
+      "logits/chosen": -6.8125,
+      "logits/rejected": -3.25,
+      "logps/chosen": -1384.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.0068,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -9.8125,
+      "rewards/margins": 15.5,
+      "rewards/rejected": -25.25,
       "step": 2080
     },
     {
       "epoch": 2.18848167539267,
-      "grad_norm": 0.0676757184486897,
+      "grad_norm": 0.28384041943054616,
       "learning_rate": 1.0344831849474507e-05,
-      "logits/chosen": -4.875,
-      "logits/rejected": -1.296875,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.0055,
+      "logits/chosen": -6.0625,
+      "logits/rejected": -2.328125,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0067,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 15.375,
-      "rewards/rejected": -24.375,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -26.625,
       "step": 2090
     },
     {
       "epoch": 2.1989528795811517,
-      "grad_norm": 0.27670542371211604,
+      "grad_norm": 0.24305666180756338,
       "learning_rate": 1.0099107117141879e-05,
-      "logits/chosen": -4.59375,
-      "logits/rejected": -1.671875,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0032,
+      "logits/chosen": -6.21875,
+      "logits/rejected": -2.875,
+      "logps/chosen": -1368.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0048,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -25.25,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -26.375,
       "step": 2100
     },
     {
       "epoch": 2.1989528795811517,
-      "eval_logits/chosen": -2.828125,
-      "eval_logits/rejected": -0.875,
-      "eval_logps/chosen": -1424.0,
-      "eval_logps/rejected": -1592.0,
-      "eval_loss": 1.488921880722046,
-      "eval_rewards/accuracies": 0.591796875,
-      "eval_rewards/chosen": -11.0625,
-      "eval_rewards/margins": 1.9296875,
-      "eval_rewards/rejected": -13.0,
-      "eval_runtime": 46.6145,
-      "eval_samples_per_second": 42.905,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -4.71875,
+      "eval_logits/rejected": -2.765625,
+      "eval_logps/chosen": -1536.0,
+      "eval_logps/rejected": -1728.0,
+      "eval_loss": 1.5185468196868896,
+      "eval_rewards/accuracies": 0.599609375,
+      "eval_rewards/chosen": -12.1875,
+      "eval_rewards/margins": 2.203125,
+      "eval_rewards/rejected": -14.375,
+      "eval_runtime": 58.2334,
+      "eval_samples_per_second": 34.345,
+      "eval_steps_per_second": 0.55,
       "step": 2100
     },
     {
       "epoch": 2.2094240837696333,
-      "grad_norm": 0.27002396913138577,
+      "grad_norm": 0.17420038797079168,
       "learning_rate": 9.855595178057334e-06,
-      "logits/chosen": -4.34375,
-      "logits/rejected": -1.671875,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -3632.0,
-      "loss": 0.0085,
+      "logits/chosen": -6.21875,
+      "logits/rejected": -3.09375,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0049,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.25,
-      "rewards/margins": 17.75,
-      "rewards/rejected": -27.0,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -27.875,
       "step": 2110
     },
     {
       "epoch": 2.2198952879581153,
-      "grad_norm": 0.05177117298449773,
+      "grad_norm": 0.04745342669082689,
       "learning_rate": 9.614332193917949e-06,
-      "logits/chosen": -4.84375,
-      "logits/rejected": -2.1875,
-      "logps/chosen": -1416.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0079,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.0,
-      "rewards/margins": 17.25,
-      "rewards/rejected": -27.25,
+      "logits/chosen": -6.71875,
+      "logits/rejected": -3.609375,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0113,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 17.75,
+      "rewards/rejected": -28.25,
       "step": 2120
     },
     {
       "epoch": 2.230366492146597,
-      "grad_norm": 0.02183712295087322,
+      "grad_norm": 0.05990348694372628,
       "learning_rate": 9.375353992449385e-06,
-      "logits/chosen": -5.3125,
-      "logits/rejected": -3.203125,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -3328.0,
-      "loss": 0.0034,
+      "logits/chosen": -7.0625,
+      "logits/rejected": -4.5,
+      "logps/chosen": -1384.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0026,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 15.9375,
-      "rewards/rejected": -25.0,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 15.75,
+      "rewards/rejected": -25.875,
       "step": 2130
     },
     {
       "epoch": 2.2408376963350785,
-      "grad_norm": 0.0673008613707324,
+      "grad_norm": 0.11992303357906868,
       "learning_rate": 9.138696062085442e-06,
-      "logits/chosen": -5.1875,
-      "logits/rejected": -2.21875,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.015,
+      "logits/chosen": -6.96875,
+      "logits/rejected": -3.625,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.01,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 15.9375,
-      "rewards/rejected": -24.875,
+      "rewards/chosen": -9.4375,
+      "rewards/margins": 15.6875,
+      "rewards/rejected": -25.125,
       "step": 2140
     },
     {
       "epoch": 2.25130890052356,
-      "grad_norm": 0.285542696761716,
+      "grad_norm": 0.044232129092162946,
       "learning_rate": 8.904393546698007e-06,
-      "logits/chosen": -5.4375,
-      "logits/rejected": -2.015625,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -3536.0,
-      "loss": 0.0081,
+      "logits/chosen": -7.5625,
+      "logits/rejected": -3.78125,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -3424.0,
+      "loss": 0.0082,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.6875,
-      "rewards/margins": 17.125,
-      "rewards/rejected": -26.875,
+      "rewards/chosen": -9.375,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -25.75,
       "step": 2150
     },
     {
       "epoch": 2.261780104712042,
-      "grad_norm": 0.047589727556458766,
+      "grad_norm": 0.15982454917317138,
       "learning_rate": 8.672481240378142e-06,
-      "logits/chosen": -4.90625,
-      "logits/rejected": -1.15625,
+      "logits/chosen": -6.6875,
+      "logits/rejected": -2.796875,
       "logps/chosen": -1240.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0086,
-      "rewards/accuracies": 1.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.011,
+      "rewards/accuracies": 0.9937499761581421,
       "rewards/chosen": -8.8125,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -25.375,
+      "rewards/margins": 15.5,
+      "rewards/rejected": -24.25,
       "step": 2160
     },
     {
       "epoch": 2.2722513089005236,
-      "grad_norm": 0.14827991800544177,
+      "grad_norm": 0.45284015614229706,
       "learning_rate": 8.44299358226919e-06,
-      "logits/chosen": -4.875,
-      "logits/rejected": -1.984375,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -3472.0,
-      "loss": 0.0068,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 17.125,
-      "rewards/rejected": -25.75,
+      "logits/chosen": -6.5,
+      "logits/rejected": -3.1875,
+      "logps/chosen": -1304.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0076,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -9.3125,
+      "rewards/margins": 16.25,
+      "rewards/rejected": -25.5,
       "step": 2170
     },
-    {
-      "epoch": 2.282722513089005,
-      "grad_norm": 0.04267229774218262,
-      "learning_rate": 8.215964651452457e-06,
-      "logits/chosen": -4.59375,
-      "logits/rejected": -1.53125,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -3456.0,
-      "loss": 0.0091,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.75,
-      "rewards/margins": 17.0,
-      "rewards/rejected": -25.875,
+    {
+      "epoch": 2.282722513089005,
+      "grad_norm": 0.022628696739177015,
+      "learning_rate": 8.215964651452457e-06,
+      "logits/chosen": -5.71875,
+      "logits/rejected": -2.328125,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0063,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -9.5625,
+      "rewards/margins": 17.875,
+      "rewards/rejected": -27.375,
       "step": 2180
     },
     {
       "epoch": 2.2931937172774868,
-      "grad_norm": 0.020446745069524326,
+      "grad_norm": 0.031768797100968346,
       "learning_rate": 7.991428161886502e-06,
-      "logits/chosen": -4.78125,
-      "logits/rejected": -2.234375,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -3280.0,
-      "loss": 0.0058,
+      "logits/chosen": -5.8125,
+      "logits/rejected": -3.140625,
+      "logps/chosen": -1368.0,
+      "logps/rejected": -3424.0,
+      "loss": 0.0062,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 16.0,
-      "rewards/rejected": -25.0,
+      "rewards/chosen": -9.875,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -26.25,
       "step": 2190
     },
     {
       "epoch": 2.303664921465969,
-      "grad_norm": 1.0365648492082784,
+      "grad_norm": 0.03594300479846103,
       "learning_rate": 7.76941745740061e-06,
-      "logits/chosen": -4.5625,
-      "logits/rejected": -1.890625,
-      "logps/chosen": -1368.0,
-      "logps/rejected": -3456.0,
-      "loss": 0.0059,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -2.71875,
+      "logps/chosen": -1464.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0045,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.5625,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -26.125,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 16.75,
+      "rewards/rejected": -27.25,
       "step": 2200
     },
     {
       "epoch": 2.303664921465969,
-      "eval_logits/chosen": -2.75,
-      "eval_logits/rejected": -0.828125,
-      "eval_logps/chosen": -1440.0,
-      "eval_logps/rejected": -1608.0,
-      "eval_loss": 1.5006250143051147,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -11.25,
-      "eval_rewards/margins": 1.9375,
-      "eval_rewards/rejected": -13.1875,
-      "eval_runtime": 46.6237,
-      "eval_samples_per_second": 42.897,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -3.640625,
+      "eval_logits/rejected": -1.75,
+      "eval_logps/chosen": -1560.0,
+      "eval_logps/rejected": -1752.0,
+      "eval_loss": 1.5012187957763672,
+      "eval_rewards/accuracies": 0.61328125,
+      "eval_rewards/chosen": -12.4375,
+      "eval_rewards/margins": 2.1875,
+      "eval_rewards/rejected": -14.625,
+      "eval_runtime": 58.2346,
+      "eval_samples_per_second": 34.344,
+      "eval_steps_per_second": 0.55,
       "step": 2200
     },
     {
       "epoch": 2.3141361256544504,
-      "grad_norm": 0.13886793530747335,
+      "grad_norm": 0.21753669317617128,
       "learning_rate": 7.549965506743123e-06,
-      "logits/chosen": -4.34375,
-      "logits/rejected": -1.953125,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0068,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 15.75,
-      "rewards/rejected": -25.125,
+      "logits/chosen": -5.09375,
+      "logits/rejected": -2.5,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.005,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 15.8125,
+      "rewards/rejected": -26.375,
       "step": 2210
     },
     {
       "epoch": 2.324607329842932,
-      "grad_norm": 0.13917563590010362,
+      "grad_norm": 0.1498474460725861,
       "learning_rate": 7.333104898685631e-06,
-      "logits/chosen": -5.0625,
-      "logits/rejected": -1.9765625,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.007,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 15.9375,
-      "rewards/rejected": -24.75,
+      "logits/chosen": -5.5,
+      "logits/rejected": -2.53125,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0079,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 15.75,
+      "rewards/rejected": -26.25,
       "step": 2220
     },
     {
       "epoch": 2.3350785340314135,
-      "grad_norm": 0.09253989684841216,
+      "grad_norm": 0.11589356564895749,
       "learning_rate": 7.118867837183518e-06,
-      "logits/chosen": -5.59375,
-      "logits/rejected": -2.28125,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -3472.0,
-      "loss": 0.0031,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -2.90625,
+      "logps/chosen": -1480.0,
+      "logps/rejected": -3568.0,
+      "loss": 0.0035,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.1875,
-      "rewards/margins": 16.75,
-      "rewards/rejected": -26.0,
+      "rewards/chosen": -10.375,
+      "rewards/margins": 16.625,
+      "rewards/rejected": -27.0,
       "step": 2230
     },
     {
       "epoch": 2.345549738219895,
-      "grad_norm": 0.03573139159702915,
+      "grad_norm": 0.06955519899084536,
       "learning_rate": 6.9072861365936045e-06,
-      "logits/chosen": -5.53125,
-      "logits/rejected": -2.546875,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -3536.0,
-      "loss": 0.0051,
+      "logits/chosen": -6.375,
+      "logits/rejected": -3.203125,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0046,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 16.375,
-      "rewards/rejected": -25.875,
+      "rewards/chosen": -10.5,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -26.625,
       "step": 2240
     },
     {
       "epoch": 2.356020942408377,
-      "grad_norm": 0.11637892692086682,
+      "grad_norm": 0.056582284491731014,
       "learning_rate": 6.6983912169497005e-06,
-      "logits/chosen": -4.625,
-      "logits/rejected": -1.828125,
-      "logps/chosen": -1384.0,
-      "logps/rejected": -3328.0,
-      "loss": 0.0065,
+      "logits/chosen": -5.6875,
+      "logits/rejected": -2.625,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0052,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 15.9375,
-      "rewards/rejected": -25.5,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 15.8125,
+      "rewards/rejected": -26.25,
       "step": 2250
     },
     {
       "epoch": 2.3664921465968587,
-      "grad_norm": 0.17169378603914692,
+      "grad_norm": 0.4624729866522076,
       "learning_rate": 6.49221409929677e-06,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -1.3828125,
-      "logps/chosen": -1288.0,
+      "logits/chosen": -4.4375,
+      "logits/rejected": -1.953125,
+      "logps/chosen": -1368.0,
       "logps/rejected": -3344.0,
-      "loss": 0.0055,
+      "loss": 0.0044,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.25,
-      "rewards/margins": 16.5,
+      "rewards/chosen": -10.0625,
+      "rewards/margins": 15.75,
       "rewards/rejected": -25.75,
       "step": 2260
     },
     {
       "epoch": 2.3769633507853403,
-      "grad_norm": 0.08079639623702707,
+      "grad_norm": 0.22314276350577783,
       "learning_rate": 6.2887854010842075e-06,
-      "logits/chosen": -3.5625,
-      "logits/rejected": -1.1328125,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -3440.0,
-      "loss": 0.0067,
+      "logits/chosen": -4.875,
+      "logits/rejected": -2.015625,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0066,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.9375,
-      "rewards/margins": 16.625,
-      "rewards/rejected": -26.5,
+      "rewards/chosen": -10.5,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -26.625,
       "step": 2270
     },
     {
       "epoch": 2.387434554973822,
-      "grad_norm": 0.06043175650586457,
+      "grad_norm": 0.04252838829986181,
       "learning_rate": 6.0881353316191385e-06,
-      "logits/chosen": -4.84375,
-      "logits/rejected": -1.65625,
-      "logps/chosen": -1352.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0074,
+      "logits/chosen": -5.875,
+      "logits/rejected": -2.484375,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0067,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 15.9375,
-      "rewards/rejected": -25.25,
+      "rewards/chosen": -10.375,
+      "rewards/margins": 16.0,
+      "rewards/rejected": -26.375,
       "step": 2280
     },
     {
       "epoch": 2.3979057591623034,
-      "grad_norm": 1.9324250612468867,
+      "grad_norm": 0.5435298056705987,
       "learning_rate": 5.890293687580381e-06,
-      "logits/chosen": -3.953125,
-      "logits/rejected": -1.1953125,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -3504.0,
-      "loss": 0.0061,
+      "logits/chosen": -5.15625,
+      "logits/rejected": -2.1875,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.004,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.0625,
-      "rewards/margins": 16.875,
-      "rewards/rejected": -27.0,
+      "rewards/chosen": -10.875,
+      "rewards/margins": 17.5,
+      "rewards/rejected": -28.375,
       "step": 2290
     },
     {
       "epoch": 2.4083769633507854,
-      "grad_norm": 0.08486267064754144,
+      "grad_norm": 0.06032056933497945,
       "learning_rate": 5.695289848593532e-06,
-      "logits/chosen": -4.78125,
-      "logits/rejected": -2.078125,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -3424.0,
-      "loss": 0.008,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -9.125,
-      "rewards/margins": 17.375,
-      "rewards/rejected": -26.5,
+      "logits/chosen": -5.9375,
+      "logits/rejected": -3.09375,
+      "logps/chosen": -1376.0,
+      "logps/rejected": -3552.0,
+      "loss": 0.0108,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 17.75,
+      "rewards/rejected": -27.75,
       "step": 2300
     },
     {
       "epoch": 2.4083769633507854,
-      "eval_logits/chosen": -2.40625,
-      "eval_logits/rejected": -0.49609375,
-      "eval_logps/chosen": -1472.0,
-      "eval_logps/rejected": -1640.0,
-      "eval_loss": 1.5363750457763672,
-      "eval_rewards/accuracies": 0.6015625,
-      "eval_rewards/chosen": -11.5,
-      "eval_rewards/margins": 1.96875,
-      "eval_rewards/rejected": -13.5,
-      "eval_runtime": 46.626,
-      "eval_samples_per_second": 42.894,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -3.75,
+      "eval_logits/rejected": -1.8046875,
+      "eval_logps/chosen": -1552.0,
+      "eval_logps/rejected": -1744.0,
+      "eval_loss": 1.528093695640564,
+      "eval_rewards/accuracies": 0.607421875,
+      "eval_rewards/chosen": -12.3125,
+      "eval_rewards/margins": 2.234375,
+      "eval_rewards/rejected": -14.5625,
+      "eval_runtime": 58.2503,
+      "eval_samples_per_second": 34.335,
+      "eval_steps_per_second": 0.549,
       "step": 2300
     },
     {
       "epoch": 2.418848167539267,
-      "grad_norm": 0.5801819817373337,
+      "grad_norm": 0.047743682776225836,
       "learning_rate": 5.50315277286812e-06,
-      "logits/chosen": -4.65625,
-      "logits/rejected": -1.765625,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -3488.0,
+      "logits/chosen": -5.90625,
+      "logits/rejected": -2.546875,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -3648.0,
       "loss": 0.0068,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 17.625,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 18.25,
+      "rewards/rejected": -28.5,
       "step": 2310
     },
     {
       "epoch": 2.4293193717277486,
-      "grad_norm": 0.3223977413301785,
+      "grad_norm": 0.981406400313708,
       "learning_rate": 5.313910992897281e-06,
-      "logits/chosen": -4.34375,
-      "logits/rejected": -0.9765625,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.007,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -1.8515625,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -3488.0,
+      "loss": 0.0072,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 16.75,
-      "rewards/rejected": -25.625,
+      "rewards/chosen": -9.5,
+      "rewards/margins": 17.5,
+      "rewards/rejected": -27.0,
       "step": 2320
     },
     {
       "epoch": 2.4397905759162306,
-      "grad_norm": 0.17698923519397072,
+      "grad_norm": 0.11402287415800015,
       "learning_rate": 5.127592611220658e-06,
-      "logits/chosen": -4.5625,
-      "logits/rejected": -1.9453125,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -3600.0,
-      "loss": 0.0123,
+      "logits/chosen": -5.28125,
+      "logits/rejected": -2.53125,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0119,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 17.625,
-      "rewards/rejected": -26.625,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -28.375,
       "step": 2330
     },
     {
       "epoch": 2.450261780104712,
-      "grad_norm": 0.10899371132231304,
+      "grad_norm": 0.022229325334366966,
       "learning_rate": 4.944225296251159e-06,
-      "logits/chosen": -4.5,
-      "logits/rejected": -2.234375,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -3184.0,
-      "loss": 0.0056,
+      "logits/chosen": -5.28125,
+      "logits/rejected": -2.921875,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.0051,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.4375,
+      "rewards/chosen": -9.9375,
       "rewards/margins": 15.875,
-      "rewards/rejected": -24.25,
+      "rewards/rejected": -25.875,
       "step": 2340
     },
     {
       "epoch": 2.4607329842931938,
-      "grad_norm": 0.04494918375681616,
+      "grad_norm": 0.5182942950473136,
       "learning_rate": 4.763836278166172e-06,
-      "logits/chosen": -4.65625,
-      "logits/rejected": -1.5859375,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -3104.0,
-      "loss": 0.0048,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 15.375,
-      "rewards/rejected": -24.25,
+      "logits/chosen": -5.71875,
+      "logits/rejected": -2.40625,
+      "logps/chosen": -1368.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0064,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -9.875,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -26.0,
       "step": 2350
     },
     {
       "epoch": 2.4712041884816753,
-      "grad_norm": 0.04484649040155957,
+      "grad_norm": 0.22043053440020735,
       "learning_rate": 4.586452344863876e-06,
-      "logits/chosen": -5.3125,
-      "logits/rejected": -2.046875,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -3360.0,
+      "logits/chosen": -6.28125,
+      "logits/rejected": -2.765625,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -3456.0,
       "loss": 0.0067,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.875,
+      "rewards/chosen": -9.6875,
       "rewards/margins": 16.625,
-      "rewards/rejected": -25.5,
+      "rewards/rejected": -26.375,
       "step": 2360
     },
     {
       "epoch": 2.481675392670157,
-      "grad_norm": 0.222496723089707,
+      "grad_norm": 0.2079645800005788,
       "learning_rate": 4.412099837985192e-06,
-      "logits/chosen": -4.78125,
-      "logits/rejected": -1.5078125,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.0052,
+      "logits/chosen": -5.78125,
+      "logits/rejected": -2.390625,
+      "logps/chosen": -1384.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.0054,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 15.875,
-      "rewards/rejected": -24.875,
+      "rewards/chosen": -9.8125,
+      "rewards/margins": 16.0,
+      "rewards/rejected": -25.875,
       "step": 2370
     },
     {
       "epoch": 2.492146596858639,
-      "grad_norm": 0.15125130177276486,
+      "grad_norm": 0.15902843765497746,
       "learning_rate": 4.240804649002089e-06,
-      "logits/chosen": -5.0,
-      "logits/rejected": -1.578125,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -3488.0,
-      "loss": 0.0066,
+      "logits/chosen": -6.0,
+      "logits/rejected": -2.3125,
+      "logps/chosen": -1320.0,
+      "logps/rejected": -3600.0,
+      "loss": 0.0087,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 17.5,
-      "rewards/rejected": -26.375,
+      "rewards/chosen": -9.625,
+      "rewards/margins": 17.75,
+      "rewards/rejected": -27.5,
       "step": 2380
     },
     {
       "epoch": 2.5026178010471205,
-      "grad_norm": 1.2186103963794295,
+      "grad_norm": 0.08933160446153633,
       "learning_rate": 4.072592215372598e-06,
-      "logits/chosen": -5.25,
-      "logits/rejected": -2.4375,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -3488.0,
-      "loss": 0.0049,
+      "logits/chosen": -6.1875,
+      "logits/rejected": -3.125,
+      "logps/chosen": -1464.0,
+      "logps/rejected": -3568.0,
+      "loss": 0.0044,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.6875,
-      "rewards/margins": 17.0,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 17.125,
+      "rewards/rejected": -27.5,
       "step": 2390
     },
     {
       "epoch": 2.513089005235602,
-      "grad_norm": 0.36379749049724197,
+      "grad_norm": 0.43907224800917544,
       "learning_rate": 3.907487516763389e-06,
-      "logits/chosen": -4.625,
-      "logits/rejected": -1.2578125,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0059,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -25.5,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -2.078125,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0056,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.125,
+      "rewards/margins": 15.9375,
+      "rewards/rejected": -26.125,
       "step": 2400
     },
     {
       "epoch": 2.513089005235602,
-      "eval_logits/chosen": -2.609375,
-      "eval_logits/rejected": -0.65625,
-      "eval_logps/chosen": -1464.0,
-      "eval_logps/rejected": -1640.0,
-      "eval_loss": 1.571062445640564,
-      "eval_rewards/accuracies": 0.58984375,
-      "eval_rewards/chosen": -11.5,
-      "eval_rewards/margins": 2.046875,
-      "eval_rewards/rejected": -13.5,
-      "eval_runtime": 46.6281,
-      "eval_samples_per_second": 42.893,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -3.625,
+      "eval_logits/rejected": -1.6796875,
+      "eval_logps/chosen": -1528.0,
+      "eval_logps/rejected": -1720.0,
+      "eval_loss": 1.5154374837875366,
+      "eval_rewards/accuracies": 0.607421875,
+      "eval_rewards/chosen": -12.125,
+      "eval_rewards/margins": 2.21875,
+      "eval_rewards/rejected": -14.3125,
+      "eval_runtime": 58.2112,
+      "eval_samples_per_second": 34.358,
+      "eval_steps_per_second": 0.55,
       "step": 2400
     },
     {
-      "epoch": 2.5235602094240837,
-      "grad_norm": 0.051040842479329024,
-      "learning_rate": 3.745515071340286e-06,
-      "logits/chosen": -4.5625,
-      "logits/rejected": -1.21875,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -3712.0,
-      "loss": 0.0054,
+      "epoch": 2.5235602094240837,
+      "grad_norm": 0.036255823472614475,
+      "learning_rate": 3.745515071340286e-06,
+      "logits/chosen": -5.65625,
+      "logits/rejected": -2.265625,
+      "logps/chosen": -1480.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0056,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.0,
-      "rewards/margins": 18.375,
-      "rewards/rejected": -28.375,
+      "rewards/chosen": -10.75,
+      "rewards/margins": 18.5,
+      "rewards/rejected": -29.25,
       "step": 2410
     },
     {
       "epoch": 2.5340314136125652,
-      "grad_norm": 0.13125040037229135,
+      "grad_norm": 0.05945710029695591,
       "learning_rate": 3.586698932127236e-06,
-      "logits/chosen": -5.0,
-      "logits/rejected": -1.5546875,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -3456.0,
-      "loss": 0.0027,
+      "logits/chosen": -6.1875,
+      "logits/rejected": -2.484375,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -3488.0,
+      "loss": 0.0018,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.1875,
-      "rewards/margins": 17.0,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -9.4375,
+      "rewards/margins": 17.125,
+      "rewards/rejected": -26.625,
       "step": 2420
     },
     {
       "epoch": 2.5445026178010473,
-      "grad_norm": 0.19185786771714228,
+      "grad_norm": 0.06364564878118428,
       "learning_rate": 3.4310626834344744e-06,
-      "logits/chosen": -5.0625,
-      "logits/rejected": -2.46875,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -3312.0,
-      "loss": 0.0066,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -3.234375,
+      "logps/chosen": -1384.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 16.25,
-      "rewards/rejected": -25.25,
+      "rewards/chosen": -9.75,
+      "rewards/margins": 15.875,
+      "rewards/rejected": -25.625,
       "step": 2430
     },
     {
       "epoch": 2.554973821989529,
-      "grad_norm": 0.25035121094515306,
+      "grad_norm": 0.03046845929222081,
       "learning_rate": 3.2786294373562343e-06,
-      "logits/chosen": -4.5625,
-      "logits/rejected": -1.640625,
-      "logps/chosen": -1336.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0033,
+      "logits/chosen": -5.59375,
+      "logits/rejected": -2.546875,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0025,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.3125,
-      "rewards/margins": 16.625,
-      "rewards/rejected": -26.0,
+      "rewards/chosen": -9.875,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -26.375,
       "step": 2440
     },
     {
       "epoch": 2.5654450261780104,
-      "grad_norm": 0.3335287321166872,
+      "grad_norm": 0.23662880767529607,
       "learning_rate": 3.1294218303385526e-06,
-      "logits/chosen": -4.75,
-      "logits/rejected": -1.6640625,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -3424.0,
-      "loss": 0.0042,
+      "logits/chosen": -5.84375,
+      "logits/rejected": -2.5625,
+      "logps/chosen": -1456.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0056,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -26.5,
       "step": 2450
     },
     {
       "epoch": 2.5759162303664924,
-      "grad_norm": 0.0933514781917977,
+      "grad_norm": 0.4245248898686133,
       "learning_rate": 2.9834620198177483e-06,
-      "logits/chosen": -4.5,
-      "logits/rejected": -1.4140625,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0109,
+      "logits/chosen": -5.65625,
+      "logits/rejected": -2.5,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0106,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.8125,
-      "rewards/margins": 18.75,
-      "rewards/rejected": -28.5,
+      "rewards/chosen": -10.5,
+      "rewards/margins": 18.25,
+      "rewards/rejected": -28.875,
       "step": 2460
     },
     {
       "epoch": 2.5863874345549736,
-      "grad_norm": 0.03620863602796955,
+      "grad_norm": 0.02128443838390143,
       "learning_rate": 2.840771680930068e-06,
-      "logits/chosen": -4.78125,
-      "logits/rejected": -1.921875,
-      "logps/chosen": -1440.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0054,
+      "logits/chosen": -5.625,
+      "logits/rejected": -2.65625,
+      "logps/chosen": -1528.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0037,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.25,
-      "rewards/margins": 17.875,
-      "rewards/rejected": -28.0,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 18.25,
+      "rewards/rejected": -29.375,
       "step": 2470
     },
     {
       "epoch": 2.5968586387434556,
-      "grad_norm": 0.16312912647748445,
+      "grad_norm": 0.21694697586510897,
       "learning_rate": 2.701372003292868e-06,
-      "logits/chosen": -3.90625,
-      "logits/rejected": -1.5625,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -3552.0,
-      "loss": 0.0091,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.1875,
+      "logits/chosen": -4.96875,
+      "logits/rejected": -2.0625,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0082,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -9.75,
       "rewards/margins": 17.875,
-      "rewards/rejected": -27.125,
+      "rewards/rejected": -27.625,
       "step": 2480
     },
     {
       "epoch": 2.607329842931937,
-      "grad_norm": 0.14646917165101753,
+      "grad_norm": 0.04113141586634902,
       "learning_rate": 2.56528368785795e-06,
-      "logits/chosen": -4.875,
-      "logits/rejected": -1.7109375,
-      "logps/chosen": -1328.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.002,
+      "logits/chosen": -5.5625,
+      "logits/rejected": -2.28125,
+      "logps/chosen": -1408.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.0022,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 15.5625,
-      "rewards/rejected": -24.625,
+      "rewards/chosen": -9.9375,
+      "rewards/margins": 15.375,
+      "rewards/rejected": -25.25,
       "step": 2490
     },
     {
       "epoch": 2.6178010471204187,
-      "grad_norm": 0.044582155081154894,
+      "grad_norm": 0.11998014061126595,
       "learning_rate": 2.432526943837471e-06,
-      "logits/chosen": -4.5,
-      "logits/rejected": -0.76953125,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -3552.0,
-      "loss": 0.0063,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 18.5,
-      "rewards/rejected": -27.625,
+      "logits/chosen": -5.34375,
+      "logits/rejected": -1.484375,
+      "logps/chosen": -1352.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0051,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -9.8125,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -29.125,
       "step": 2500
     },
     {
       "epoch": 2.6178010471204187,
-      "eval_logits/chosen": -2.578125,
-      "eval_logits/rejected": -0.63671875,
-      "eval_logps/chosen": -1456.0,
-      "eval_logps/rejected": -1632.0,
-      "eval_loss": 1.5533593893051147,
-      "eval_rewards/accuracies": 0.587890625,
-      "eval_rewards/chosen": -11.375,
-      "eval_rewards/margins": 2.046875,
-      "eval_rewards/rejected": -13.4375,
-      "eval_runtime": 46.6233,
-      "eval_samples_per_second": 42.897,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -3.453125,
+      "eval_logits/rejected": -1.5234375,
+      "eval_logps/chosen": -1536.0,
+      "eval_logps/rejected": -1728.0,
+      "eval_loss": 1.5115312337875366,
+      "eval_rewards/accuracies": 0.603515625,
+      "eval_rewards/chosen": -12.1875,
+      "eval_rewards/margins": 2.21875,
+      "eval_rewards/rejected": -14.4375,
+      "eval_runtime": 58.2666,
+      "eval_samples_per_second": 34.325,
+      "eval_steps_per_second": 0.549,
       "step": 2500
     },
     {
       "epoch": 2.6282722513089007,
-      "grad_norm": 0.02185774831434233,
+      "grad_norm": 0.0330025415512247,
       "learning_rate": 2.3031214857028446e-06,
-      "logits/chosen": -4.59375,
-      "logits/rejected": -1.625,
-      "logps/chosen": -1328.0,
-      "logps/rejected": -3600.0,
-      "loss": 0.0053,
+      "logits/chosen": -5.4375,
+      "logits/rejected": -2.3125,
+      "logps/chosen": -1376.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0068,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.125,
-      "rewards/margins": 18.375,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -9.625,
+      "rewards/margins": 18.75,
+      "rewards/rejected": -28.375,
       "step": 2510
     },
     {
       "epoch": 2.6387434554973823,
-      "grad_norm": 0.18127490750064293,
+      "grad_norm": 0.025703459842164194,
       "learning_rate": 2.177086530257139e-06,
-      "logits/chosen": -4.65625,
-      "logits/rejected": -1.8671875,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -3136.0,
-      "loss": 0.0081,
+      "logits/chosen": -5.4375,
+      "logits/rejected": -2.546875,
+      "logps/chosen": -1400.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0082,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.1875,
-      "rewards/margins": 14.6875,
-      "rewards/rejected": -23.875,
+      "rewards/chosen": -10.125,
+      "rewards/margins": 14.8125,
+      "rewards/rejected": -25.0,
       "step": 2520
     },
     {
       "epoch": 2.649214659685864,
-      "grad_norm": 0.10286487460392181,
+      "grad_norm": 0.09154398451068889,
       "learning_rate": 2.054440793781354e-06,
-      "logits/chosen": -5.0625,
-      "logits/rejected": -1.65625,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0085,
+      "logits/chosen": -5.84375,
+      "logits/rejected": -2.46875,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0074,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 16.375,
-      "rewards/rejected": -25.25,
+      "rewards/chosen": -10.0625,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -26.125,
       "step": 2530
     },
     {
       "epoch": 2.6596858638743455,
-      "grad_norm": 0.01957039918016148,
+      "grad_norm": 0.13915429744639582,
       "learning_rate": 1.9352024892550587e-06,
-      "logits/chosen": -4.4375,
-      "logits/rejected": -1.6796875,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -3392.0,
-      "loss": 0.006,
+      "logits/chosen": -5.34375,
+      "logits/rejected": -2.484375,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -3472.0,
+      "loss": 0.0064,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.25,
-      "rewards/margins": 16.875,
-      "rewards/rejected": -26.125,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 16.625,
+      "rewards/rejected": -26.875,
       "step": 2540
     },
     {
       "epoch": 2.670157068062827,
-      "grad_norm": 0.12195627385958806,
+      "grad_norm": 0.12667227514338306,
       "learning_rate": 1.8193893236517379e-06,
-      "logits/chosen": -4.6875,
-      "logits/rejected": -1.578125,
-      "logps/chosen": -1336.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0081,
+      "logits/chosen": -5.625,
+      "logits/rejected": -2.375,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0071,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.3125,
-      "rewards/margins": 19.375,
-      "rewards/rejected": -28.75,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 19.5,
+      "rewards/rejected": -29.75,
       "step": 2550
     },
     {
       "epoch": 2.680628272251309,
-      "grad_norm": 0.10346872451146909,
+      "grad_norm": 0.4002033246834211,
       "learning_rate": 1.7070184953092956e-06,
-      "logits/chosen": -4.5,
-      "logits/rejected": -0.640625,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -3392.0,
-      "loss": 0.0074,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.5625,
-      "rewards/margins": 17.75,
-      "rewards/rejected": -26.25,
+      "logits/chosen": -5.4375,
+      "logits/rejected": -1.453125,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -3424.0,
+      "loss": 0.0111,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -9.5,
+      "rewards/margins": 17.125,
+      "rewards/rejected": -26.625,
       "step": 2560
     },
     {
       "epoch": 2.6910994764397906,
-      "grad_norm": 0.100589110474654,
+      "grad_norm": 0.17408255248950816,
       "learning_rate": 1.5981066913760918e-06,
-      "logits/chosen": -3.84375,
-      "logits/rejected": -0.6171875,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -3280.0,
+      "logits/chosen": -5.0625,
+      "logits/rejected": -1.546875,
+      "logps/chosen": -1296.0,
+      "logps/rejected": -3264.0,
       "loss": 0.0039,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.75,
-      "rewards/margins": 16.875,
-      "rewards/rejected": -25.625,
+      "rewards/chosen": -9.3125,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -25.5,
       "step": 2570
     },
     {
       "epoch": 2.701570680628272,
-      "grad_norm": 0.06080814485687144,
+      "grad_norm": 0.10294543826619756,
       "learning_rate": 1.492670085332884e-06,
-      "logits/chosen": -4.90625,
-      "logits/rejected": -1.5234375,
-      "logps/chosen": -1352.0,
-      "logps/rejected": -3632.0,
-      "loss": 0.0081,
+      "logits/chosen": -6.03125,
+      "logits/rejected": -2.4375,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.4375,
-      "rewards/margins": 18.375,
-      "rewards/rejected": -27.875,
+      "rewards/chosen": -10.125,
+      "rewards/margins": 17.875,
+      "rewards/rejected": -28.0,
       "step": 2580
     },
     {
       "epoch": 2.712041884816754,
-      "grad_norm": 0.019934056806947008,
+      "grad_norm": 0.08942773490351473,
       "learning_rate": 1.3907243345910787e-06,
-      "logits/chosen": -4.71875,
-      "logits/rejected": -1.640625,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -3360.0,
-      "loss": 0.0033,
+      "logits/chosen": -5.59375,
+      "logits/rejected": -2.453125,
+      "logps/chosen": -1384.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0032,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -25.375,
+      "rewards/chosen": -9.5625,
+      "rewards/margins": 15.875,
+      "rewards/rejected": -25.5,
       "step": 2590
     },
     {
       "epoch": 2.7225130890052354,
-      "grad_norm": 0.1338158916746071,
+      "grad_norm": 0.05614154557132533,
       "learning_rate": 1.292284578167538e-06,
-      "logits/chosen": -4.34375,
-      "logits/rejected": -1.765625,
-      "logps/chosen": -1272.0,
+      "logits/chosen": -5.375,
+      "logits/rejected": -2.546875,
+      "logps/chosen": -1312.0,
       "logps/rejected": -3520.0,
-      "loss": 0.0075,
+      "loss": 0.0041,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 17.625,
+      "rewards/chosen": -9.3125,
+      "rewards/margins": 17.125,
       "rewards/rejected": -26.5,
       "step": 2600
     },
     {
       "epoch": 2.7225130890052354,
-      "eval_logits/chosen": -2.53125,
-      "eval_logits/rejected": -0.5859375,
-      "eval_logps/chosen": -1472.0,
-      "eval_logps/rejected": -1648.0,
-      "eval_loss": 1.5769844055175781,
-      "eval_rewards/accuracies": 0.5859375,
-      "eval_rewards/chosen": -11.5625,
-      "eval_rewards/margins": 2.078125,
-      "eval_rewards/rejected": -13.625,
-      "eval_runtime": 46.6296,
-      "eval_samples_per_second": 42.891,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -3.75,
+      "eval_logits/rejected": -1.8046875,
+      "eval_logps/chosen": -1504.0,
+      "eval_logps/rejected": -1696.0,
+      "eval_loss": 1.4846405982971191,
+      "eval_rewards/accuracies": 0.59375,
+      "eval_rewards/chosen": -11.8125,
+      "eval_rewards/margins": 2.203125,
+      "eval_rewards/rejected": -14.0625,
+      "eval_runtime": 58.2948,
+      "eval_samples_per_second": 34.308,
+      "eval_steps_per_second": 0.549,
       "step": 2600
     },
     {
       "epoch": 2.7329842931937174,
-      "grad_norm": 0.05164935236054819,
+      "grad_norm": 0.23820188482605056,
       "learning_rate": 1.1973654344364925e-06,
-      "logits/chosen": -4.40625,
-      "logits/rejected": -1.8671875,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -3552.0,
-      "loss": 0.0103,
+      "logits/chosen": -5.71875,
+      "logits/rejected": -2.875,
+      "logps/chosen": -1328.0,
+      "logps/rejected": -3568.0,
+      "loss": 0.0055,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 17.375,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -9.25,
+      "rewards/margins": 17.25,
+      "rewards/rejected": -26.5,
       "step": 2610
     },
     {
       "epoch": 2.743455497382199,
-      "grad_norm": 0.039308878248630816,
+      "grad_norm": 0.030655914196202026,
       "learning_rate": 1.1059809989586583e-06,
-      "logits/chosen": -4.34375,
-      "logits/rejected": -1.046875,
-      "logps/chosen": -1328.0,
-      "logps/rejected": -3488.0,
-      "loss": 0.0047,
+      "logits/chosen": -5.59375,
+      "logits/rejected": -2.0625,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0033,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.3125,
-      "rewards/margins": 17.375,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -9.5,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -26.375,
       "step": 2620
     },
     {
       "epoch": 2.7539267015706805,
-      "grad_norm": 0.018018274035608725,
+      "grad_norm": 0.06982458689884719,
       "learning_rate": 1.018144842388044e-06,
-      "logits/chosen": -4.15625,
-      "logits/rejected": -1.171875,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -3648.0,
-      "loss": 0.0074,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.1875,
-      "rewards/margins": 18.375,
-      "rewards/rejected": -27.625,
+      "logits/chosen": -5.25,
+      "logits/rejected": -2.125,
+      "logps/chosen": -1384.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.005,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -9.5625,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -27.25,
       "step": 2630
     },
-    {
-      "epoch": 2.7643979057591626,
-      "grad_norm": 0.03959159611971814,
-      "learning_rate": 9.338700084567109e-07,
-      "logits/chosen": -4.46875,
-      "logits/rejected": -2.3125,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -3552.0,
-      "loss": 0.0048,
+    {
+      "epoch": 2.7643979057591626,
+      "grad_norm": 0.02892761533764223,
+      "learning_rate": 9.338700084567109e-07,
+      "logits/chosen": -5.46875,
+      "logits/rejected": -3.03125,
+      "logps/chosen": -1368.0,
+      "logps/rejected": -3520.0,
+      "loss": 0.0038,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.1875,
-      "rewards/margins": 17.5,
-      "rewards/rejected": -26.625,
+      "rewards/chosen": -9.875,
+      "rewards/margins": 16.625,
+      "rewards/rejected": -26.5,
       "step": 2640
     },
     {
       "epoch": 2.774869109947644,
-      "grad_norm": 0.24405498457885635,
+      "grad_norm": 0.04416508285965663,
       "learning_rate": 8.531690120377606e-07,
-      "logits/chosen": -4.71875,
-      "logits/rejected": -1.75,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -3600.0,
-      "loss": 0.0055,
+      "logits/chosen": -5.5625,
+      "logits/rejected": -2.671875,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.0041,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.5625,
+      "rewards/chosen": -10.1875,
       "rewards/margins": 17.5,
-      "rewards/rejected": -27.125,
+      "rewards/rejected": -27.625,
       "step": 2650
     },
     {
       "epoch": 2.7853403141361257,
-      "grad_norm": 0.2905476869814392,
+      "grad_norm": 0.41846545695405934,
       "learning_rate": 7.760538372868636e-07,
-      "logits/chosen": -4.625,
-      "logits/rejected": -1.78125,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -3472.0,
-      "loss": 0.0102,
+      "logits/chosen": -5.78125,
+      "logits/rejected": -2.6875,
+      "logps/chosen": -1352.0,
+      "logps/rejected": -3504.0,
+      "loss": 0.0079,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.1875,
-      "rewards/margins": 17.25,
-      "rewards/rejected": -26.375,
+      "rewards/chosen": -9.6875,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -26.625,
       "step": 2660
     },
     {
       "epoch": 2.7958115183246073,
-      "grad_norm": 0.23013884049604932,
+      "grad_norm": 0.0955745510159564,
       "learning_rate": 7.025359358626166e-07,
-      "logits/chosen": -5.15625,
-      "logits/rejected": -1.65625,
-      "logps/chosen": -1368.0,
-      "logps/rejected": -3568.0,
-      "loss": 0.0028,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.4375,
+      "logits/chosen": -5.84375,
+      "logits/rejected": -2.5625,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -3632.0,
+      "loss": 0.0037,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.125,
       "rewards/margins": 17.875,
-      "rewards/rejected": -27.375,
+      "rewards/rejected": -28.0,
       "step": 2670
     },
     {
       "epoch": 2.806282722513089,
-      "grad_norm": 0.17371708518176998,
+      "grad_norm": 0.35624164204172204,
       "learning_rate": 6.326262252259463e-07,
-      "logits/chosen": -4.9375,
-      "logits/rejected": -1.546875,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -3088.0,
-      "loss": 0.004,
+      "logits/chosen": -5.90625,
+      "logits/rejected": -2.4375,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -3136.0,
+      "loss": 0.0034,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 15.375,
-      "rewards/rejected": -24.125,
+      "rewards/chosen": -9.4375,
+      "rewards/margins": 15.0625,
+      "rewards/rejected": -24.5,
       "step": 2680
     },
     {
       "epoch": 2.816753926701571,
-      "grad_norm": 0.7324354528835144,
+      "grad_norm": 0.020497933735202346,
       "learning_rate": 5.66335087018871e-07,
-      "logits/chosen": -5.09375,
-      "logits/rejected": -1.3046875,
-      "logps/chosen": -1336.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.0032,
+      "logits/chosen": -6.03125,
+      "logits/rejected": -2.125,
+      "logps/chosen": -1408.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0035,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.25,
-      "rewards/margins": 20.125,
-      "rewards/rejected": -29.25,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -29.875,
       "step": 2690
     },
     {
       "epoch": 2.8272251308900525,
-      "grad_norm": 0.09460492677095046,
+      "grad_norm": 0.13034074403898002,
       "learning_rate": 5.036723655228226e-07,
-      "logits/chosen": -5.0625,
-      "logits/rejected": -1.5703125,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -3376.0,
-      "loss": 0.005,
+      "logits/chosen": -5.875,
+      "logits/rejected": -2.46875,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -3472.0,
+      "loss": 0.0049,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.1875,
-      "rewards/margins": 17.25,
-      "rewards/rejected": -26.5,
+      "rewards/chosen": -9.75,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -27.375,
       "step": 2700
     },
     {
       "epoch": 2.8272251308900525,
-      "eval_logits/chosen": -2.609375,
-      "eval_logits/rejected": -0.65625,
-      "eval_logps/chosen": -1464.0,
-      "eval_logps/rejected": -1640.0,
-      "eval_loss": 1.5728750228881836,
-      "eval_rewards/accuracies": 0.583984375,
-      "eval_rewards/chosen": -11.4375,
-      "eval_rewards/margins": 2.09375,
-      "eval_rewards/rejected": -13.5,
-      "eval_runtime": 46.6088,
-      "eval_samples_per_second": 42.91,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -3.671875,
+      "eval_logits/rejected": -1.7265625,
+      "eval_logps/chosen": -1520.0,
+      "eval_logps/rejected": -1712.0,
+      "eval_loss": 1.5019999742507935,
+      "eval_rewards/accuracies": 0.59765625,
+      "eval_rewards/chosen": -12.0,
+      "eval_rewards/margins": 2.234375,
+      "eval_rewards/rejected": -14.25,
+      "eval_runtime": 58.2243,
+      "eval_samples_per_second": 34.35,
+      "eval_steps_per_second": 0.55,
       "step": 2700
     },
     {
       "epoch": 2.837696335078534,
-      "grad_norm": 0.09565340164614645,
+      "grad_norm": 0.06555522740771393,
       "learning_rate": 4.446473661967432e-07,
-      "logits/chosen": -5.0,
-      "logits/rejected": -2.25,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -3168.0,
-      "loss": 0.0045,
+      "logits/chosen": -5.875,
+      "logits/rejected": -3.0625,
+      "logps/chosen": -1368.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0074,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.125,
-      "rewards/margins": 15.25,
-      "rewards/rejected": -24.375,
+      "rewards/chosen": -9.8125,
+      "rewards/margins": 15.375,
+      "rewards/rejected": -25.25,
       "step": 2710
     },
     {
       "epoch": 2.8481675392670156,
-      "grad_norm": 0.27347025347442405,
+      "grad_norm": 0.050288331111448474,
       "learning_rate": 3.892688542952411e-07,
-      "logits/chosen": -4.3125,
-      "logits/rejected": -1.03125,
-      "logps/chosen": -1232.0,
+      "logits/chosen": -5.21875,
+      "logits/rejected": -1.9140625,
+      "logps/chosen": -1296.0,
       "logps/rejected": -3408.0,
-      "loss": 0.005,
+      "loss": 0.0048,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.5625,
-      "rewards/margins": 17.125,
-      "rewards/rejected": -25.625,
+      "rewards/chosen": -9.1875,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -25.75,
       "step": 2720
     },
     {
       "epoch": 2.858638743455497,
-      "grad_norm": 0.28217522894230995,
+      "grad_norm": 0.3274931540370359,
       "learning_rate": 3.3754505356694e-07,
-      "logits/chosen": -4.71875,
-      "logits/rejected": -1.890625,
-      "logps/chosen": -1328.0,
-      "logps/rejected": -3328.0,
-      "loss": 0.009,
+      "logits/chosen": -5.875,
+      "logits/rejected": -2.859375,
+      "logps/chosen": -1400.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0053,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.25,
+      "rewards/chosen": -10.0,
       "rewards/margins": 16.25,
-      "rewards/rejected": -25.5,
+      "rewards/rejected": -26.25,
       "step": 2730
     },
     {
       "epoch": 2.869109947643979,
-      "grad_norm": 0.038559048353935456,
+      "grad_norm": 0.04602948776482805,
       "learning_rate": 2.8948364503322276e-07,
-      "logits/chosen": -4.53125,
-      "logits/rejected": -1.515625,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -3360.0,
-      "loss": 0.0048,
+      "logits/chosen": -5.75,
+      "logits/rejected": -2.46875,
+      "logps/chosen": -1320.0,
+      "logps/rejected": -3392.0,
+      "loss": 0.0042,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 17.75,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -9.5,
+      "rewards/margins": 17.5,
+      "rewards/rejected": -27.0,
       "step": 2740
     },
     {
       "epoch": 2.8795811518324608,
-      "grad_norm": 0.1670327186497815,
+      "grad_norm": 0.05126710804313756,
       "learning_rate": 2.4509176584762706e-07,
-      "logits/chosen": -4.78125,
-      "logits/rejected": -2.046875,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -3392.0,
-      "loss": 0.0052,
+      "logits/chosen": -5.96875,
+      "logits/rejected": -2.96875,
+      "logps/chosen": -1312.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0037,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.75,
-      "rewards/margins": 16.75,
-      "rewards/rejected": -25.5,
+      "rewards/chosen": -9.3125,
+      "rewards/margins": 16.0,
+      "rewards/rejected": -25.375,
       "step": 2750
     },
     {
       "epoch": 2.8900523560209423,
-      "grad_norm": 0.02448797796061219,
+      "grad_norm": 0.06071311577101956,
       "learning_rate": 2.0437600823595694e-07,
-      "logits/chosen": -4.96875,
-      "logits/rejected": -1.4453125,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -3328.0,
-      "loss": 0.006,
+      "logits/chosen": -5.75,
+      "logits/rejected": -2.234375,
+      "logps/chosen": -1408.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0043,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 16.625,
-      "rewards/rejected": -25.5,
+      "rewards/chosen": -9.8125,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -26.125,
       "step": 2760
     },
     {
       "epoch": 2.900523560209424,
-      "grad_norm": 0.0951971978227471,
+      "grad_norm": 0.01039544456699555,
       "learning_rate": 1.6734241851733e-07,
-      "logits/chosen": -5.125,
-      "logits/rejected": -1.5234375,
-      "logps/chosen": -1384.0,
-      "logps/rejected": -3600.0,
-      "loss": 0.0065,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 18.125,
-      "rewards/rejected": -27.625,
+      "logits/chosen": -5.84375,
+      "logits/rejected": -2.40625,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0044,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -10.375,
+      "rewards/margins": 17.375,
+      "rewards/rejected": -27.75,
       "step": 2770
     },
     {
       "epoch": 2.9109947643979055,
-      "grad_norm": 0.4705009005730346,
+      "grad_norm": 0.19696639262511806,
       "learning_rate": 1.339964962062984e-07,
-      "logits/chosen": -5.125,
-      "logits/rejected": -2.6875,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -3488.0,
-      "loss": 0.007,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 17.5,
-      "rewards/rejected": -26.5,
+      "logits/chosen": -6.0625,
+      "logits/rejected": -3.40625,
+      "logps/chosen": -1320.0,
+      "logps/rejected": -3552.0,
+      "loss": 0.0054,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -9.625,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -27.25,
       "step": 2780
     },
     {
       "epoch": 2.9214659685863875,
-      "grad_norm": 0.06481324830587623,
+      "grad_norm": 0.03206459275183846,
       "learning_rate": 1.0434319319617136e-07,
-      "logits/chosen": -4.71875,
-      "logits/rejected": -2.375,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -3216.0,
-      "loss": 0.0091,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 15.5625,
-      "rewards/rejected": -24.5,
+      "logits/chosen": -5.65625,
+      "logits/rejected": -3.015625,
+      "logps/chosen": -1328.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0055,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -9.5,
+      "rewards/margins": 15.8125,
+      "rewards/rejected": -25.375,
       "step": 2790
     },
     {
       "epoch": 2.931937172774869,
-      "grad_norm": 0.3545201142155869,
+      "grad_norm": 0.2536395750299021,
       "learning_rate": 7.838691302365363e-08,
-      "logits/chosen": -4.65625,
-      "logits/rejected": -0.87109375,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0065,
+      "logits/chosen": -5.78125,
+      "logits/rejected": -1.765625,
+      "logps/chosen": -1328.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.0063,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 16.75,
-      "rewards/rejected": -25.5,
+      "rewards/chosen": -9.25,
+      "rewards/margins": 16.625,
+      "rewards/rejected": -26.0,
       "step": 2800
     },
     {
       "epoch": 2.931937172774869,
-      "eval_logits/chosen": -2.59375,
-      "eval_logits/rejected": -0.6484375,
-      "eval_logps/chosen": -1464.0,
-      "eval_logps/rejected": -1648.0,
-      "eval_loss": 1.5774062871932983,
-      "eval_rewards/accuracies": 0.587890625,
-      "eval_rewards/chosen": -11.5,
-      "eval_rewards/margins": 2.09375,
-      "eval_rewards/rejected": -13.5625,
-      "eval_runtime": 46.5837,
-      "eval_samples_per_second": 42.933,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -3.6875,
+      "eval_logits/rejected": -1.7421875,
+      "eval_logps/chosen": -1520.0,
+      "eval_logps/rejected": -1712.0,
+      "eval_loss": 1.5022187232971191,
+      "eval_rewards/accuracies": 0.599609375,
+      "eval_rewards/chosen": -12.0,
+      "eval_rewards/margins": 2.21875,
+      "eval_rewards/rejected": -14.25,
+      "eval_runtime": 58.2325,
+      "eval_samples_per_second": 34.345,
+      "eval_steps_per_second": 0.55,
       "step": 2800
     },
     {
       "epoch": 2.9424083769633507,
-      "grad_norm": 0.10498352428375558,
+      "grad_norm": 0.12124275017440117,
       "learning_rate": 5.613151021490459e-08,
-      "logits/chosen": -4.3125,
-      "logits/rejected": -0.90234375,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -3104.0,
-      "loss": 0.0018,
+      "logits/chosen": -5.34375,
+      "logits/rejected": -1.8046875,
+      "logps/chosen": -1280.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.0037,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.625,
-      "rewards/margins": 15.75,
-      "rewards/rejected": -24.375,
+      "rewards/chosen": -9.5,
+      "rewards/margins": 15.3125,
+      "rewards/rejected": -24.75,
       "step": 2810
     },
     {
       "epoch": 2.9528795811518327,
-      "grad_norm": 0.11790065909688109,
+      "grad_norm": 0.02558741645440495,
       "learning_rate": 3.7580289713159965e-08,
-      "logits/chosen": -4.53125,
-      "logits/rejected": -1.546875,
-      "logps/chosen": -1336.0,
-      "logps/rejected": -3536.0,
-      "loss": 0.0055,
+      "logits/chosen": -5.53125,
+      "logits/rejected": -2.34375,
+      "logps/chosen": -1408.0,
+      "logps/rejected": -3568.0,
+      "loss": 0.0062,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.25,
-      "rewards/margins": 17.5,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -9.9375,
+      "rewards/margins": 17.25,
+      "rewards/rejected": -27.125,
       "step": 2820
     },
     {
       "epoch": 2.9633507853403143,
-      "grad_norm": 0.1612374300881366,
+      "grad_norm": 0.23011217113283375,
       "learning_rate": 2.2736006387927277e-08,
-      "logits/chosen": -4.625,
-      "logits/rejected": -2.046875,
-      "logps/chosen": -1368.0,
-      "logps/rejected": -3632.0,
-      "loss": 0.0077,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 18.125,
-      "rewards/rejected": -27.75,
+      "logits/chosen": -5.875,
+      "logits/rejected": -2.828125,
+      "logps/chosen": -1400.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 17.5,
+      "rewards/rejected": -27.5,
       "step": 2830
     },
     {
       "epoch": 2.973821989528796,
-      "grad_norm": 0.02612630092448086,
+      "grad_norm": 0.01961946923203549,
       "learning_rate": 1.1600864625893604e-08,
-      "logits/chosen": -5.125,
-      "logits/rejected": -1.6015625,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0044,
+      "logits/chosen": -5.96875,
+      "logits/rejected": -2.5625,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0061,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 19.375,
-      "rewards/rejected": -28.25,
+      "rewards/chosen": -9.5,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -28.625,
       "step": 2840
     },
     {
       "epoch": 2.9842931937172774,
-      "grad_norm": 0.04025595010774536,
+      "grad_norm": 0.037385698898153796,
       "learning_rate": 4.176518003573549e-09,
-      "logits/chosen": -5.125,
-      "logits/rejected": -1.6015625,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -3392.0,
-      "loss": 0.005,
+      "logits/chosen": -5.84375,
+      "logits/rejected": -2.328125,
+      "logps/chosen": -1368.0,
+      "logps/rejected": -3472.0,
+      "loss": 0.0078,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.125,
+      "rewards/chosen": -9.875,
       "rewards/margins": 16.875,
-      "rewards/rejected": -26.0,
+      "rewards/rejected": -26.75,
       "step": 2850
     },
     {
       "epoch": 2.994764397905759,
-      "grad_norm": 0.1817470313509824,
+      "grad_norm": 0.1704030102970896,
       "learning_rate": 4.640690417528903e-10,
-      "logits/chosen": -4.78125,
-      "logits/rejected": -2.359375,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -3136.0,
-      "loss": 0.0089,
+      "logits/chosen": -5.625,
+      "logits/rejected": -3.25,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -3088.0,
+      "loss": 0.0097,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.25,
-      "rewards/margins": 15.375,
-      "rewards/rejected": -24.625,
+      "rewards/chosen": -9.75,
+      "rewards/margins": 14.4375,
+      "rewards/rejected": -24.125,
       "step": 2860
     },
     {
       "epoch": 3.0,
       "step": 2865,
       "total_flos": 0.0,
-      "train_loss": 0.11596642152112088,
-      "train_runtime": 12300.8596,
-      "train_samples_per_second": 14.906,
-      "train_steps_per_second": 0.233
+      "train_loss": 0.11265950318471403,
+      "train_runtime": 12465.6533,
+      "train_samples_per_second": 14.709,
+      "train_steps_per_second": 0.23
     }
   ],
   "logging_steps": 10,