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": 2.1275462500969975,
+      "grad_norm": 2.1274919115921556,
       "learning_rate": 1.7421602787456448e-07,
       "logits/chosen": -11.9375,
       "logits/rejected": -10.5,
@@ -25,89 +25,89 @@
     },
     {
       "epoch": 0.010471204188481676,
-      "grad_norm": 1.8697049707182876,
+      "grad_norm": 1.8761961911537475,
       "learning_rate": 1.7421602787456445e-06,
-      "logits/chosen": -11.1875,
+      "logits/chosen": -11.125,
       "logits/rejected": -10.125,
       "logps/chosen": -420.0,
       "logps/rejected": -924.0,
-      "loss": 0.6906,
-      "rewards/accuracies": 0.2013888955116272,
-      "rewards/chosen": 2.849102020263672e-05,
-      "rewards/margins": 0.002899169921875,
-      "rewards/rejected": -0.00286865234375,
+      "loss": 0.6917,
+      "rewards/accuracies": 0.1875,
+      "rewards/chosen": 0.00016307830810546875,
+      "rewards/margins": 0.0013275146484375,
+      "rewards/rejected": -0.0011749267578125,
       "step": 10
     },
     {
       "epoch": 0.020942408376963352,
-      "grad_norm": 2.1248894129001417,
+      "grad_norm": 2.117702639237484,
       "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": 0.4749999940395355,
-      "rewards/chosen": -0.00180816650390625,
-      "rewards/margins": 0.019775390625,
-      "rewards/rejected": -0.021484375,
+      "loss": 0.6802,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.00140380859375,
+      "rewards/margins": 0.021484375,
+      "rewards/rejected": -0.02294921875,
       "step": 20
     },
     {
       "epoch": 0.031413612565445025,
-      "grad_norm": 1.9242791997645654,
+      "grad_norm": 1.91801162090969,
       "learning_rate": 5.226480836236934e-06,
       "logits/chosen": -11.8125,
       "logits/rejected": -11.0,
       "logps/chosen": -418.0,
       "logps/rejected": -796.0,
-      "loss": 0.6298,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -0.042236328125,
+      "loss": 0.6292,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -0.04248046875,
       "rewards/margins": 0.1357421875,
       "rewards/rejected": -0.177734375,
       "step": 30
     },
     {
       "epoch": 0.041884816753926704,
-      "grad_norm": 1.9220197300285,
+      "grad_norm": 1.92613026819293,
       "learning_rate": 6.968641114982578e-06,
       "logits/chosen": -12.25,
       "logits/rejected": -11.4375,
       "logps/chosen": -444.0,
       "logps/rejected": -896.0,
-      "loss": 0.5072,
+      "loss": 0.5074,
       "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -0.32421875,
+      "rewards/chosen": -0.326171875,
       "rewards/margins": 0.6171875,
       "rewards/rejected": -0.94140625,
       "step": 40
     },
     {
       "epoch": 0.05235602094240838,
-      "grad_norm": 1.7529382056270744,
+      "grad_norm": 1.7469579593410194,
       "learning_rate": 8.710801393728225e-06,
       "logits/chosen": -12.0625,
       "logits/rejected": -11.4375,
       "logps/chosen": -482.0,
       "logps/rejected": -1200.0,
-      "loss": 0.4422,
-      "rewards/accuracies": 0.8125,
+      "loss": 0.4426,
+      "rewards/accuracies": 0.8187500238418579,
       "rewards/chosen": -0.734375,
       "rewards/margins": 1.9921875,
-      "rewards/rejected": -2.734375,
+      "rewards/rejected": -2.71875,
       "step": 50
     },
     {
       "epoch": 0.06282722513089005,
-      "grad_norm": 1.6322985161599142,
+      "grad_norm": 1.6806249665662836,
       "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": 0.800000011920929,
+      "logps/rejected": -1020.0,
+      "loss": 0.3991,
+      "rewards/accuracies": 0.8062499761581421,
       "rewards/chosen": -0.7421875,
       "rewards/margins": 1.9921875,
       "rewards/rejected": -2.734375,
@@ -115,4269 +115,4269 @@
     },
     {
       "epoch": 0.07329842931937172,
-      "grad_norm": 2.7143959232959345,
+      "grad_norm": 2.788793675504094,
       "learning_rate": 1.2195121951219513e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -12.875,
+      "logits/chosen": -13.25,
+      "logits/rejected": -12.8125,
       "logps/chosen": -540.0,
       "logps/rejected": -1168.0,
-      "loss": 0.3692,
+      "loss": 0.37,
       "rewards/accuracies": 0.831250011920929,
-      "rewards/chosen": -1.1640625,
+      "rewards/chosen": -1.171875,
       "rewards/margins": 2.625,
       "rewards/rejected": -3.796875,
       "step": 70
     },
     {
       "epoch": 0.08376963350785341,
-      "grad_norm": 2.543170624015684,
+      "grad_norm": 2.0624237578187463,
       "learning_rate": 1.3937282229965156e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -12.9375,
-      "logps/chosen": -528.0,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -12.8125,
+      "logps/chosen": -532.0,
       "logps/rejected": -1384.0,
-      "loss": 0.3663,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -1.4609375,
+      "loss": 0.368,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": -1.5,
       "rewards/margins": 3.390625,
-      "rewards/rejected": -4.84375,
+      "rewards/rejected": -4.90625,
       "step": 80
     },
     {
       "epoch": 0.09424083769633508,
-      "grad_norm": 2.8457615014225035,
+      "grad_norm": 3.07047948223815,
       "learning_rate": 1.56794425087108e-05,
-      "logits/chosen": -11.625,
-      "logits/rejected": -11.25,
-      "logps/chosen": -506.0,
+      "logits/chosen": -11.4375,
+      "logits/rejected": -11.0625,
+      "logps/chosen": -516.0,
       "logps/rejected": -1128.0,
-      "loss": 0.366,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -1.546875,
-      "rewards/margins": 2.9375,
-      "rewards/rejected": -4.46875,
+      "loss": 0.3716,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -1.6328125,
+      "rewards/margins": 2.859375,
+      "rewards/rejected": -4.5,
       "step": 90
     },
     {
       "epoch": 0.10471204188481675,
-      "grad_norm": 1.8743608039660566,
+      "grad_norm": 2.217976090994368,
       "learning_rate": 1.742160278745645e-05,
-      "logits/chosen": -12.4375,
-      "logits/rejected": -11.875,
-      "logps/chosen": -692.0,
-      "logps/rejected": -1544.0,
-      "loss": 0.3669,
+      "logits/chosen": -12.875,
+      "logits/rejected": -12.1875,
+      "logps/chosen": -696.0,
+      "logps/rejected": -1552.0,
+      "loss": 0.3737,
       "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -2.328125,
-      "rewards/margins": 3.6875,
-      "rewards/rejected": -6.0,
+      "rewards/chosen": -2.34375,
+      "rewards/margins": 3.703125,
+      "rewards/rejected": -6.03125,
       "step": 100
     },
     {
       "epoch": 0.10471204188481675,
-      "eval_logits/chosen": -13.25,
-      "eval_logits/rejected": -13.0,
-      "eval_logps/chosen": -460.0,
-      "eval_logps/rejected": -464.0,
-      "eval_loss": 0.6719531416893005,
-      "eval_rewards/accuracies": 0.59765625,
-      "eval_rewards/chosen": -1.421875,
+      "eval_logits/chosen": -13.875,
+      "eval_logits/rejected": -13.625,
+      "eval_logps/chosen": -524.0,
+      "eval_logps/rejected": -528.0,
+      "eval_loss": 0.6770937442779541,
+      "eval_rewards/accuracies": 0.591796875,
+      "eval_rewards/chosen": -2.046875,
       "eval_rewards/margins": 0.330078125,
-      "eval_rewards/rejected": -1.75,
-      "eval_runtime": 48.9206,
-      "eval_samples_per_second": 40.883,
-      "eval_steps_per_second": 0.654,
+      "eval_rewards/rejected": -2.390625,
+      "eval_runtime": 48.422,
+      "eval_samples_per_second": 41.304,
+      "eval_steps_per_second": 0.661,
       "step": 100
     },
     {
       "epoch": 0.11518324607329843,
-      "grad_norm": 2.315798940414535,
+      "grad_norm": 2.284211373940121,
       "learning_rate": 1.9163763066202093e-05,
-      "logits/chosen": -12.875,
-      "logits/rejected": -12.0,
-      "logps/chosen": -548.0,
-      "logps/rejected": -1472.0,
-      "loss": 0.3263,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -1.390625,
-      "rewards/margins": 3.90625,
-      "rewards/rejected": -5.28125,
+      "logits/chosen": -13.375,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -576.0,
+      "logps/rejected": -1536.0,
+      "loss": 0.3283,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -1.671875,
+      "rewards/margins": 4.1875,
+      "rewards/rejected": -5.84375,
       "step": 110
     },
     {
       "epoch": 0.1256544502617801,
-      "grad_norm": 2.653279171062117,
+      "grad_norm": 2.29689992675173,
       "learning_rate": 2.0905923344947736e-05,
-      "logits/chosen": -12.25,
-      "logits/rejected": -11.8125,
-      "logps/chosen": -584.0,
-      "logps/rejected": -1408.0,
-      "loss": 0.3369,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -2.203125,
-      "rewards/margins": 3.875,
-      "rewards/rejected": -6.0625,
+      "logits/chosen": -11.4375,
+      "logits/rejected": -11.0,
+      "logps/chosen": -580.0,
+      "logps/rejected": -1424.0,
+      "loss": 0.3215,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": -2.171875,
+      "rewards/margins": 4.0625,
+      "rewards/rejected": -6.25,
       "step": 120
     },
     {
       "epoch": 0.13612565445026178,
-      "grad_norm": 2.5556065873692706,
+      "grad_norm": 2.979297642091298,
       "learning_rate": 2.264808362369338e-05,
-      "logits/chosen": -12.625,
-      "logits/rejected": -12.25,
-      "logps/chosen": -588.0,
-      "logps/rejected": -1328.0,
-      "loss": 0.3761,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -1.875,
-      "rewards/margins": 3.3125,
-      "rewards/rejected": -5.1875,
+      "logits/chosen": -11.375,
+      "logits/rejected": -11.125,
+      "logps/chosen": -648.0,
+      "logps/rejected": -1400.0,
+      "loss": 0.3632,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -2.484375,
+      "rewards/margins": 3.421875,
+      "rewards/rejected": -5.90625,
       "step": 130
     },
     {
       "epoch": 0.14659685863874344,
-      "grad_norm": 2.9144153102819326,
+      "grad_norm": 3.321228788580682,
       "learning_rate": 2.4390243902439026e-05,
-      "logits/chosen": -13.875,
-      "logits/rejected": -13.125,
-      "logps/chosen": -712.0,
-      "logps/rejected": -1416.0,
-      "loss": 0.3208,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -2.75,
-      "rewards/margins": 3.3125,
-      "rewards/rejected": -6.0625,
+      "logits/chosen": -14.125,
+      "logits/rejected": -13.25,
+      "logps/chosen": -880.0,
+      "logps/rejected": -1624.0,
+      "loss": 0.3428,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.4375,
+      "rewards/margins": 3.640625,
+      "rewards/rejected": -8.125,
       "step": 140
     },
     {
       "epoch": 0.15706806282722513,
-      "grad_norm": 2.9500656576292004,
+      "grad_norm": 2.1347203904622587,
       "learning_rate": 2.6132404181184672e-05,
-      "logits/chosen": -13.0,
-      "logits/rejected": -12.0625,
-      "logps/chosen": -704.0,
-      "logps/rejected": -1472.0,
-      "loss": 0.3588,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -2.828125,
-      "rewards/margins": 3.25,
-      "rewards/rejected": -6.09375,
+      "logits/chosen": -13.375,
+      "logits/rejected": -12.5,
+      "logps/chosen": -736.0,
+      "logps/rejected": -1544.0,
+      "loss": 0.3365,
+      "rewards/accuracies": 0.8687499761581421,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 3.609375,
+      "rewards/rejected": -6.78125,
       "step": 150
     },
     {
       "epoch": 0.16753926701570682,
-      "grad_norm": 2.804750467744131,
+      "grad_norm": 2.501697982774878,
       "learning_rate": 2.7874564459930312e-05,
-      "logits/chosen": -13.375,
-      "logits/rejected": -12.625,
-      "logps/chosen": -616.0,
-      "logps/rejected": -1408.0,
-      "loss": 0.2879,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -2.4375,
-      "rewards/margins": 3.65625,
-      "rewards/rejected": -6.09375,
+      "logits/chosen": -11.4375,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -620.0,
+      "logps/rejected": -1392.0,
+      "loss": 0.2819,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -2.484375,
+      "rewards/margins": 3.5,
+      "rewards/rejected": -6.0,
       "step": 160
     },
     {
       "epoch": 0.17801047120418848,
-      "grad_norm": 1.8038755273950402,
+      "grad_norm": 1.6036018374433882,
       "learning_rate": 2.9616724738675962e-05,
-      "logits/chosen": -13.0625,
-      "logits/rejected": -12.4375,
-      "logps/chosen": -656.0,
-      "logps/rejected": -1544.0,
-      "loss": 0.2837,
+      "logits/chosen": -10.75,
+      "logits/rejected": -10.375,
+      "logps/chosen": -776.0,
+      "logps/rejected": -1720.0,
+      "loss": 0.2949,
       "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -2.3125,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -6.71875,
+      "rewards/chosen": -3.5,
+      "rewards/margins": 4.96875,
+      "rewards/rejected": -8.4375,
       "step": 170
     },
     {
       "epoch": 0.18848167539267016,
-      "grad_norm": 2.316519664295035,
+      "grad_norm": 2.1895434955135946,
       "learning_rate": 3.13588850174216e-05,
-      "logits/chosen": -11.625,
-      "logits/rejected": -10.8125,
-      "logps/chosen": -680.0,
-      "logps/rejected": -1568.0,
-      "loss": 0.2819,
+      "logits/chosen": -10.625,
+      "logits/rejected": -9.75,
+      "logps/chosen": -720.0,
+      "logps/rejected": -1600.0,
+      "loss": 0.2792,
       "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -2.640625,
+      "rewards/chosen": -3.015625,
       "rewards/margins": 4.0625,
-      "rewards/rejected": -6.6875,
+      "rewards/rejected": -7.09375,
       "step": 180
     },
     {
       "epoch": 0.19895287958115182,
-      "grad_norm": 2.8472399625504266,
+      "grad_norm": 6.958897663601635,
       "learning_rate": 3.310104529616725e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -732.0,
-      "logps/rejected": -1704.0,
-      "loss": 0.2973,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -8.1875,
+      "logits/chosen": -10.0,
+      "logits/rejected": -9.0,
+      "logps/chosen": -808.0,
+      "logps/rejected": -1856.0,
+      "loss": 0.3128,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": -4.28125,
+      "rewards/margins": 5.4375,
+      "rewards/rejected": -9.6875,
       "step": 190
     },
     {
       "epoch": 0.2094240837696335,
-      "grad_norm": 1.7672203960274466,
+      "grad_norm": 1.6059161352711844,
       "learning_rate": 3.48432055749129e-05,
-      "logits/chosen": -9.625,
-      "logits/rejected": -9.125,
-      "logps/chosen": -744.0,
-      "logps/rejected": -1552.0,
-      "loss": 0.3019,
-      "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": -2.984375,
-      "rewards/margins": 3.828125,
-      "rewards/rejected": -6.8125,
+      "logits/chosen": -10.875,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -740.0,
+      "logps/rejected": -1600.0,
+      "loss": 0.2761,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 4.34375,
+      "rewards/rejected": -7.28125,
       "step": 200
     },
     {
       "epoch": 0.2094240837696335,
-      "eval_logits/chosen": -8.875,
-      "eval_logits/rejected": -8.25,
-      "eval_logps/chosen": -536.0,
-      "eval_logps/rejected": -544.0,
-      "eval_loss": 0.7078593969345093,
-      "eval_rewards/accuracies": 0.580078125,
-      "eval_rewards/chosen": -2.1875,
-      "eval_rewards/margins": 0.34765625,
-      "eval_rewards/rejected": -2.546875,
-      "eval_runtime": 46.6323,
-      "eval_samples_per_second": 42.889,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -9.3125,
+      "eval_logits/rejected": -8.5625,
+      "eval_logps/chosen": -636.0,
+      "eval_logps/rejected": -664.0,
+      "eval_loss": 0.7596250176429749,
+      "eval_rewards/accuracies": 0.603515625,
+      "eval_rewards/chosen": -3.171875,
+      "eval_rewards/margins": 0.5703125,
+      "eval_rewards/rejected": -3.734375,
+      "eval_runtime": 46.6026,
+      "eval_samples_per_second": 42.916,
+      "eval_steps_per_second": 0.687,
       "step": 200
     },
     {
       "epoch": 0.2198952879581152,
-      "grad_norm": 2.1809321712075005,
+      "grad_norm": 2.092165221145791,
       "learning_rate": 3.6585365853658535e-05,
-      "logits/chosen": -9.3125,
-      "logits/rejected": -8.5,
-      "logps/chosen": -624.0,
-      "logps/rejected": -1400.0,
-      "loss": 0.2949,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -2.703125,
-      "rewards/margins": 4.125,
-      "rewards/rejected": -6.8125,
+      "logits/chosen": -9.125,
+      "logits/rejected": -8.25,
+      "logps/chosen": -820.0,
+      "logps/rejected": -1640.0,
+      "loss": 0.3174,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -4.65625,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -9.1875,
       "step": 210
     },
     {
       "epoch": 0.23036649214659685,
-      "grad_norm": 1.6800260409220835,
+      "grad_norm": 1.9634303716026147,
       "learning_rate": 3.8327526132404185e-05,
-      "logits/chosen": -10.125,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -660.0,
-      "logps/rejected": -1552.0,
-      "loss": 0.3146,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -2.5625,
-      "rewards/margins": 4.25,
-      "rewards/rejected": -6.8125,
+      "logits/chosen": -14.0,
+      "logits/rejected": -12.5625,
+      "logps/chosen": -648.0,
+      "logps/rejected": -1488.0,
+      "loss": 0.2931,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -2.421875,
+      "rewards/margins": 3.78125,
+      "rewards/rejected": -6.1875,
       "step": 220
     },
     {
       "epoch": 0.24083769633507854,
-      "grad_norm": 2.6506543950315447,
+      "grad_norm": 2.264622962603772,
       "learning_rate": 4.006968641114983e-05,
-      "logits/chosen": -9.25,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -724.0,
-      "logps/rejected": -1528.0,
-      "loss": 0.325,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 4.3125,
-      "rewards/rejected": -7.78125,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -13.25,
+      "logps/chosen": -664.0,
+      "logps/rejected": -1560.0,
+      "loss": 0.3025,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 5.15625,
+      "rewards/rejected": -8.0,
       "step": 230
     },
     {
       "epoch": 0.2513089005235602,
-      "grad_norm": 3.791021346335991,
+      "grad_norm": 4.160509603964285,
       "learning_rate": 4.181184668989547e-05,
-      "logits/chosen": -8.625,
-      "logits/rejected": -7.71875,
-      "logps/chosen": -692.0,
-      "logps/rejected": -1520.0,
-      "loss": 0.3408,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -7.78125,
+      "logits/chosen": -10.9375,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -772.0,
+      "logps/rejected": -1664.0,
+      "loss": 0.3232,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.09375,
+      "rewards/margins": 5.15625,
+      "rewards/rejected": -9.25,
       "step": 240
     },
     {
       "epoch": 0.2617801047120419,
-      "grad_norm": 3.0936409725422607,
+      "grad_norm": 2.396670830451119,
       "learning_rate": 4.3554006968641115e-05,
-      "logits/chosen": -7.6875,
-      "logits/rejected": -6.65625,
-      "logps/chosen": -848.0,
-      "logps/rejected": -1768.0,
-      "loss": 0.2631,
+      "logits/chosen": -11.875,
+      "logits/rejected": -10.5,
+      "logps/chosen": -696.0,
+      "logps/rejected": -1464.0,
+      "loss": 0.2585,
       "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -4.125,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -8.9375,
+      "rewards/chosen": -2.609375,
+      "rewards/margins": 3.3125,
+      "rewards/rejected": -5.90625,
       "step": 250
     },
     {
       "epoch": 0.27225130890052357,
-      "grad_norm": 3.2318222788565056,
+      "grad_norm": 2.4339900347367696,
       "learning_rate": 4.529616724738676e-05,
-      "logits/chosen": -9.1875,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -648.0,
-      "logps/rejected": -1504.0,
-      "loss": 0.2664,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -2.859375,
-      "rewards/margins": 3.984375,
-      "rewards/rejected": -6.84375,
+      "logits/chosen": -10.3125,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -744.0,
+      "logps/rejected": -1744.0,
+      "loss": 0.3056,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -3.859375,
+      "rewards/margins": 5.4375,
+      "rewards/rejected": -9.3125,
       "step": 260
     },
     {
       "epoch": 0.28272251308900526,
-      "grad_norm": 2.35441279627441,
+      "grad_norm": 2.201729172934641,
       "learning_rate": 4.703832752613241e-05,
-      "logits/chosen": -9.5,
-      "logits/rejected": -8.3125,
-      "logps/chosen": -676.0,
-      "logps/rejected": -1480.0,
-      "loss": 0.3401,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -3.078125,
-      "rewards/margins": 4.0625,
-      "rewards/rejected": -7.125,
+      "logits/chosen": -12.375,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -608.0,
+      "logps/rejected": -1344.0,
+      "loss": 0.3345,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -2.40625,
+      "rewards/margins": 3.375,
+      "rewards/rejected": -5.78125,
       "step": 270
     },
     {
       "epoch": 0.2931937172774869,
-      "grad_norm": 2.6751995234861465,
+      "grad_norm": 2.0585804618522965,
       "learning_rate": 4.878048780487805e-05,
-      "logits/chosen": -9.1875,
-      "logits/rejected": -8.375,
-      "logps/chosen": -668.0,
-      "logps/rejected": -1400.0,
-      "loss": 0.292,
-      "rewards/accuracies": 0.8500000238418579,
-      "rewards/chosen": -3.109375,
-      "rewards/margins": 3.59375,
-      "rewards/rejected": -6.71875,
+      "logits/chosen": -12.25,
+      "logits/rejected": -11.0625,
+      "logps/chosen": -708.0,
+      "logps/rejected": -1504.0,
+      "loss": 0.2978,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -3.5,
+      "rewards/margins": 4.25,
+      "rewards/rejected": -7.75,
       "step": 280
     },
     {
       "epoch": 0.3036649214659686,
-      "grad_norm": 4.540243010272999,
+      "grad_norm": 2.6266548360648163,
       "learning_rate": 4.999983293481418e-05,
-      "logits/chosen": -9.9375,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -812.0,
-      "logps/rejected": -1848.0,
-      "loss": 0.2545,
+      "logits/chosen": -12.5,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -784.0,
+      "logps/rejected": -1824.0,
+      "loss": 0.2483,
       "rewards/accuracies": 0.90625,
-      "rewards/chosen": -4.375,
-      "rewards/margins": 5.625,
-      "rewards/rejected": -10.0,
+      "rewards/chosen": -4.0625,
+      "rewards/margins": 5.71875,
+      "rewards/rejected": -9.75,
       "step": 290
     },
     {
       "epoch": 0.31413612565445026,
-      "grad_norm": 4.643664734387955,
+      "grad_norm": 2.7725700587125144,
       "learning_rate": 4.9996862949181595e-05,
-      "logits/chosen": -10.125,
-      "logits/rejected": -9.0,
-      "logps/chosen": -932.0,
-      "logps/rejected": -2192.0,
-      "loss": 0.2872,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -5.40625,
-      "rewards/margins": 8.5,
-      "rewards/rejected": -13.9375,
+      "logits/chosen": -10.375,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -884.0,
+      "logps/rejected": -1928.0,
+      "loss": 0.2447,
+      "rewards/accuracies": 0.8687499761581421,
+      "rewards/chosen": -4.90625,
+      "rewards/margins": 6.34375,
+      "rewards/rejected": -11.25,
       "step": 300
     },
     {
       "epoch": 0.31413612565445026,
-      "eval_logits/chosen": -10.8125,
-      "eval_logits/rejected": -10.125,
-      "eval_logps/chosen": -776.0,
-      "eval_logps/rejected": -800.0,
-      "eval_loss": 0.9193124771118164,
-      "eval_rewards/accuracies": 0.55078125,
-      "eval_rewards/chosen": -4.59375,
-      "eval_rewards/margins": 0.51953125,
-      "eval_rewards/rejected": -5.125,
-      "eval_runtime": 46.6166,
-      "eval_samples_per_second": 42.903,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -13.75,
+      "eval_logits/rejected": -13.0625,
+      "eval_logps/chosen": -700.0,
+      "eval_logps/rejected": -736.0,
+      "eval_loss": 0.8172187209129333,
+      "eval_rewards/accuracies": 0.607421875,
+      "eval_rewards/chosen": -3.8125,
+      "eval_rewards/margins": 0.6484375,
+      "eval_rewards/rejected": -4.46875,
+      "eval_runtime": 46.6127,
+      "eval_samples_per_second": 42.907,
+      "eval_steps_per_second": 0.687,
       "step": 300
     },
     {
       "epoch": 0.32460732984293195,
-      "grad_norm": 1.9069050099873621,
+      "grad_norm": 1.8291147629473097,
       "learning_rate": 4.9990180911526896e-05,
-      "logits/chosen": -12.75,
-      "logits/rejected": -11.6875,
-      "logps/chosen": -676.0,
-      "logps/rejected": -1496.0,
-      "loss": 0.2955,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 3.921875,
-      "rewards/rejected": -7.0625,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -12.875,
+      "logps/chosen": -744.0,
+      "logps/rejected": -1784.0,
+      "loss": 0.2748,
+      "rewards/accuracies": 0.8687499761581421,
+      "rewards/chosen": -3.828125,
+      "rewards/margins": 6.09375,
+      "rewards/rejected": -9.9375,
       "step": 310
     },
     {
       "epoch": 0.33507853403141363,
-      "grad_norm": 2.8897603596216284,
+      "grad_norm": 2.255312138573243,
       "learning_rate": 4.997978781413745e-05,
-      "logits/chosen": -10.25,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -996.0,
-      "logps/rejected": -1792.0,
-      "loss": 0.3121,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -5.65625,
-      "rewards/margins": 4.75,
-      "rewards/rejected": -10.4375,
+      "logits/chosen": -16.25,
+      "logits/rejected": -14.8125,
+      "logps/chosen": -932.0,
+      "logps/rejected": -1896.0,
+      "loss": 0.3318,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -5.0,
+      "rewards/margins": 6.4375,
+      "rewards/rejected": -11.4375,
       "step": 320
     },
     {
       "epoch": 0.34554973821989526,
-      "grad_norm": 4.637353580315091,
+      "grad_norm": 5.131793213938642,
       "learning_rate": 4.996568520039565e-05,
-      "logits/chosen": -11.5625,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -748.0,
-      "logps/rejected": -1792.0,
-      "loss": 0.285,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 5.3125,
-      "rewards/rejected": -8.8125,
+      "logits/chosen": -16.75,
+      "logits/rejected": -15.875,
+      "logps/chosen": -696.0,
+      "logps/rejected": -1560.0,
+      "loss": 0.2915,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 3.5625,
+      "rewards/rejected": -6.53125,
       "step": 330
     },
     {
       "epoch": 0.35602094240837695,
-      "grad_norm": 2.663489292151173,
+      "grad_norm": 2.01028175874777,
       "learning_rate": 4.994787516454973e-05,
-      "logits/chosen": -10.4375,
-      "logits/rejected": -8.9375,
-      "logps/chosen": -772.0,
-      "logps/rejected": -1496.0,
-      "loss": 0.2705,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -3.84375,
-      "rewards/margins": 3.75,
-      "rewards/rejected": -7.59375,
+      "logits/chosen": -15.0,
+      "logits/rejected": -13.3125,
+      "logps/chosen": -724.0,
+      "logps/rejected": -1432.0,
+      "loss": 0.2737,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -3.328125,
+      "rewards/margins": 3.609375,
+      "rewards/rejected": -6.9375,
       "step": 340
     },
     {
       "epoch": 0.36649214659685864,
-      "grad_norm": 2.8404515833950272,
+      "grad_norm": 3.9718215017644227,
       "learning_rate": 4.992636035140273e-05,
-      "logits/chosen": -10.6875,
-      "logits/rejected": -8.5625,
-      "logps/chosen": -724.0,
-      "logps/rejected": -1704.0,
-      "loss": 0.2436,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -3.6875,
-      "rewards/margins": 5.25,
-      "rewards/rejected": -8.9375,
+      "logits/chosen": -14.625,
+      "logits/rejected": -12.75,
+      "logps/chosen": -748.0,
+      "logps/rejected": -1888.0,
+      "loss": 0.2645,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -3.9375,
+      "rewards/margins": 6.875,
+      "rewards/rejected": -10.8125,
       "step": 350
     },
     {
       "epoch": 0.3769633507853403,
-      "grad_norm": 2.442461469516304,
+      "grad_norm": 2.4128992355737555,
       "learning_rate": 4.99011439559198e-05,
-      "logits/chosen": -8.125,
-      "logits/rejected": -6.21875,
-      "logps/chosen": -768.0,
-      "logps/rejected": -1648.0,
-      "loss": 0.2625,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -3.78125,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -8.5,
+      "logits/chosen": -12.0,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -784.0,
+      "logps/rejected": -1792.0,
+      "loss": 0.2588,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -3.9375,
+      "rewards/margins": 6.0,
+      "rewards/rejected": -9.9375,
       "step": 360
     },
     {
       "epoch": 0.387434554973822,
-      "grad_norm": 1.7986176057059122,
+      "grad_norm": 2.118756996190423,
       "learning_rate": 4.987222972275369e-05,
-      "logits/chosen": -7.9375,
-      "logits/rejected": -6.46875,
-      "logps/chosen": -812.0,
-      "logps/rejected": -1656.0,
-      "loss": 0.2676,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -4.125,
-      "rewards/margins": 4.625,
-      "rewards/rejected": -8.75,
+      "logits/chosen": -10.0,
+      "logits/rejected": -7.90625,
+      "logps/chosen": -784.0,
+      "logps/rejected": -1576.0,
+      "loss": 0.2723,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -3.875,
+      "rewards/margins": 4.125,
+      "rewards/rejected": -8.0,
       "step": 370
     },
     {
       "epoch": 0.39790575916230364,
-      "grad_norm": 1.6246194971347911,
+      "grad_norm": 1.9489314905522945,
       "learning_rate": 4.983962194568872e-05,
-      "logits/chosen": -7.5625,
-      "logits/rejected": -6.53125,
-      "logps/chosen": -816.0,
-      "logps/rejected": -1744.0,
-      "loss": 0.2976,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -4.1875,
-      "rewards/margins": 4.65625,
-      "rewards/rejected": -8.8125,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -10.125,
+      "logps/chosen": -776.0,
+      "logps/rejected": -1672.0,
+      "loss": 0.2858,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -3.796875,
+      "rewards/margins": 4.34375,
+      "rewards/rejected": -8.125,
       "step": 380
     },
     {
       "epoch": 0.4083769633507853,
-      "grad_norm": 2.0666873815131934,
+      "grad_norm": 1.8569794906322932,
       "learning_rate": 4.9803325467003084e-05,
-      "logits/chosen": -8.5,
-      "logits/rejected": -6.9375,
-      "logps/chosen": -788.0,
-      "logps/rejected": -1808.0,
-      "loss": 0.2366,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -3.875,
-      "rewards/margins": 5.40625,
-      "rewards/rejected": -9.3125,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -876.0,
+      "logps/rejected": -2024.0,
+      "loss": 0.25,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.75,
+      "rewards/margins": 6.6875,
+      "rewards/rejected": -11.4375,
       "step": 390
     },
     {
       "epoch": 0.418848167539267,
-      "grad_norm": 3.528131817056219,
+      "grad_norm": 2.281078674924464,
       "learning_rate": 4.9763345676749816e-05,
-      "logits/chosen": -8.625,
-      "logits/rejected": -8.125,
-      "logps/chosen": -848.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.2766,
-      "rewards/accuracies": 0.8500000238418579,
-      "rewards/chosen": -4.71875,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -9.5,
+      "logits/chosen": -10.875,
+      "logits/rejected": -9.25,
+      "logps/chosen": -844.0,
+      "logps/rejected": -1848.0,
+      "loss": 0.2872,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 5.46875,
+      "rewards/rejected": -10.125,
       "step": 400
     },
     {
       "epoch": 0.418848167539267,
-      "eval_logits/chosen": -8.8125,
-      "eval_logits/rejected": -8.0625,
-      "eval_logps/chosen": -692.0,
-      "eval_logps/rejected": -716.0,
-      "eval_loss": 0.7222187519073486,
-      "eval_rewards/accuracies": 0.607421875,
-      "eval_rewards/chosen": -3.75,
-      "eval_rewards/margins": 0.515625,
-      "eval_rewards/rejected": -4.25,
-      "eval_runtime": 46.5996,
-      "eval_samples_per_second": 42.919,
+      "eval_logits/chosen": -11.4375,
+      "eval_logits/rejected": -10.5625,
+      "eval_logps/chosen": -672.0,
+      "eval_logps/rejected": -692.0,
+      "eval_loss": 0.7366093993186951,
+      "eval_rewards/accuracies": 0.59765625,
+      "eval_rewards/chosen": -3.546875,
+      "eval_rewards/margins": 0.470703125,
+      "eval_rewards/rejected": -4.03125,
+      "eval_runtime": 46.6052,
+      "eval_samples_per_second": 42.914,
       "eval_steps_per_second": 0.687,
       "step": 400
     },
     {
       "epoch": 0.4293193717277487,
-      "grad_norm": 3.0903633191587647,
+      "grad_norm": 2.954049490071907,
       "learning_rate": 4.9719688511956374e-05,
-      "logits/chosen": -9.1875,
-      "logits/rejected": -8.375,
-      "logps/chosen": -796.0,
+      "logits/chosen": -8.625,
+      "logits/rejected": -7.21875,
+      "logps/chosen": -848.0,
       "logps/rejected": -1800.0,
-      "loss": 0.3205,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -3.859375,
-      "rewards/margins": 5.3125,
+      "loss": 0.2998,
+      "rewards/accuracies": 0.8687499761581421,
+      "rewards/chosen": -4.40625,
+      "rewards/margins": 4.8125,
       "rewards/rejected": -9.1875,
       "step": 410
     },
     {
       "epoch": 0.4397905759162304,
-      "grad_norm": 2.8698059114942063,
+      "grad_norm": 2.8425893220007667,
       "learning_rate": 4.967236045574294e-05,
-      "logits/chosen": -7.84375,
-      "logits/rejected": -6.625,
-      "logps/chosen": -820.0,
-      "logps/rejected": -1824.0,
-      "loss": 0.2544,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -4.25,
-      "rewards/margins": 4.90625,
-      "rewards/rejected": -9.125,
+      "logits/chosen": -8.0,
+      "logits/rejected": -6.125,
+      "logps/chosen": -904.0,
+      "logps/rejected": -1928.0,
+      "loss": 0.2606,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -5.125,
+      "rewards/margins": 5.0625,
+      "rewards/rejected": -10.1875,
       "step": 420
     },
     {
       "epoch": 0.450261780104712,
-      "grad_norm": 1.9601996389892722,
+      "grad_norm": 1.813680404195898,
       "learning_rate": 4.96213685363597e-05,
-      "logits/chosen": -7.90625,
-      "logits/rejected": -6.5,
-      "logps/chosen": -804.0,
-      "logps/rejected": -1672.0,
-      "loss": 0.2698,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -4.15625,
-      "rewards/margins": 4.4375,
-      "rewards/rejected": -8.625,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -8.75,
+      "logps/chosen": -824.0,
+      "logps/rejected": -1744.0,
+      "loss": 0.2599,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -4.34375,
+      "rewards/margins": 5.0,
+      "rewards/rejected": -9.3125,
       "step": 430
     },
     {
       "epoch": 0.4607329842931937,
-      "grad_norm": 3.1736852472557198,
+      "grad_norm": 2.2452415094810347,
       "learning_rate": 4.956672032614315e-05,
-      "logits/chosen": -9.0625,
-      "logits/rejected": -7.28125,
-      "logps/chosen": -856.0,
-      "logps/rejected": -1848.0,
-      "loss": 0.2677,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -4.5,
-      "rewards/margins": 5.1875,
-      "rewards/rejected": -9.6875,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.9375,
+      "logps/chosen": -840.0,
+      "logps/rejected": -1928.0,
+      "loss": 0.2972,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.34375,
+      "rewards/margins": 6.15625,
+      "rewards/rejected": -10.5,
       "step": 440
     },
     {
       "epoch": 0.4712041884816754,
-      "grad_norm": 5.170330456485999,
+      "grad_norm": 3.0384409444421325,
       "learning_rate": 4.950842394039157e-05,
-      "logits/chosen": -9.3125,
-      "logits/rejected": -7.8125,
-      "logps/chosen": -756.0,
-      "logps/rejected": -1480.0,
-      "loss": 0.2724,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -3.828125,
-      "rewards/margins": 4.0,
-      "rewards/rejected": -7.8125,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -10.25,
+      "logps/chosen": -708.0,
+      "logps/rejected": -1352.0,
+      "loss": 0.2542,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 3.21875,
+      "rewards/rejected": -6.5625,
       "step": 450
     },
     {
       "epoch": 0.4816753926701571,
-      "grad_norm": 1.8287665337113035,
+      "grad_norm": 2.656574526671651,
       "learning_rate": 4.944648803615994e-05,
-      "logits/chosen": -7.0,
-      "logits/rejected": -5.71875,
-      "logps/chosen": -904.0,
-      "logps/rejected": -1872.0,
-      "loss": 0.2228,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -5.0625,
-      "rewards/margins": 5.15625,
-      "rewards/rejected": -10.1875,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -4.90625,
+      "logps/chosen": -1008.0,
+      "logps/rejected": -1984.0,
+      "loss": 0.2627,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -6.125,
+      "rewards/margins": 5.21875,
+      "rewards/rejected": -11.3125,
       "step": 460
     },
     {
       "epoch": 0.49214659685863876,
-      "grad_norm": 2.280133615757228,
+      "grad_norm": 2.844314589201357,
       "learning_rate": 4.938092181097433e-05,
-      "logits/chosen": -9.0,
-      "logits/rejected": -7.5,
-      "logps/chosen": -816.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.2445,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -4.25,
-      "rewards/margins": 5.21875,
-      "rewards/rejected": -9.4375,
+      "logits/chosen": -9.875,
+      "logits/rejected": -8.125,
+      "logps/chosen": -884.0,
+      "logps/rejected": -1912.0,
+      "loss": 0.2482,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 6.09375,
+      "rewards/rejected": -11.0625,
       "step": 470
     },
     {
       "epoch": 0.5026178010471204,
-      "grad_norm": 6.86685884428158,
+      "grad_norm": 4.949291450418834,
       "learning_rate": 4.931173500146607e-05,
-      "logits/chosen": -6.96875,
-      "logits/rejected": -5.65625,
-      "logps/chosen": -916.0,
-      "logps/rejected": -2432.0,
-      "loss": 0.2944,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -5.5625,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -15.3125,
+      "logits/chosen": -9.875,
+      "logits/rejected": -8.25,
+      "logps/chosen": -828.0,
+      "logps/rejected": -2208.0,
+      "loss": 0.2804,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 8.4375,
+      "rewards/rejected": -13.0625,
       "step": 480
     },
     {
       "epoch": 0.5130890052356021,
-      "grad_norm": 2.257312004055489,
+      "grad_norm": 2.119479310842664,
       "learning_rate": 4.923893788192582e-05,
-      "logits/chosen": -7.25,
-      "logits/rejected": -5.59375,
-      "logps/chosen": -784.0,
-      "logps/rejected": -1840.0,
-      "loss": 0.2845,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -3.78125,
-      "rewards/margins": 6.125,
-      "rewards/rejected": -9.9375,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -7.71875,
+      "logps/chosen": -876.0,
+      "logps/rejected": -1880.0,
+      "loss": 0.2911,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -4.65625,
+      "rewards/margins": 5.625,
+      "rewards/rejected": -10.3125,
       "step": 490
     },
     {
       "epoch": 0.5235602094240838,
-      "grad_norm": 2.8247627167135594,
+      "grad_norm": 2.2743642079963764,
       "learning_rate": 4.9162541262777906e-05,
-      "logits/chosen": -7.1875,
-      "logits/rejected": -6.3125,
-      "logps/chosen": -784.0,
-      "logps/rejected": -1624.0,
-      "loss": 0.2443,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -3.8125,
-      "rewards/margins": 4.59375,
-      "rewards/rejected": -8.4375,
+      "logits/chosen": -12.5625,
+      "logits/rejected": -10.75,
+      "logps/chosen": -768.0,
+      "logps/rejected": -1568.0,
+      "loss": 0.255,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": -3.640625,
+      "rewards/margins": 4.125,
+      "rewards/rejected": -7.78125,
       "step": 500
     },
     {
       "epoch": 0.5235602094240838,
-      "eval_logits/chosen": -5.875,
-      "eval_logits/rejected": -4.9375,
-      "eval_logps/chosen": -836.0,
-      "eval_logps/rejected": -896.0,
-      "eval_loss": 0.8614062666893005,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -5.1875,
-      "eval_rewards/margins": 0.890625,
-      "eval_rewards/rejected": -6.09375,
-      "eval_runtime": 46.6182,
-      "eval_samples_per_second": 42.902,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -12.6875,
+      "eval_logits/rejected": -11.4375,
+      "eval_logps/chosen": -808.0,
+      "eval_logps/rejected": -864.0,
+      "eval_loss": 0.8522812724113464,
+      "eval_rewards/accuracies": 0.603515625,
+      "eval_rewards/chosen": -4.875,
+      "eval_rewards/margins": 0.8671875,
+      "eval_rewards/rejected": -5.75,
+      "eval_runtime": 46.5986,
+      "eval_samples_per_second": 42.92,
+      "eval_steps_per_second": 0.687,
       "step": 500
     },
     {
       "epoch": 0.5340314136125655,
-      "grad_norm": 2.49294117389587,
+      "grad_norm": 2.2795528415039494,
       "learning_rate": 4.90825564889749e-05,
-      "logits/chosen": -7.21875,
-      "logits/rejected": -6.25,
-      "logps/chosen": -912.0,
-      "logps/rejected": -1832.0,
-      "loss": 0.238,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -4.875,
-      "rewards/margins": 6.15625,
-      "rewards/rejected": -11.0625,
+      "logits/chosen": -11.8125,
+      "logits/rejected": -9.75,
+      "logps/chosen": -944.0,
+      "logps/rejected": -1816.0,
+      "loss": 0.2587,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.1875,
+      "rewards/margins": 5.71875,
+      "rewards/rejected": -10.9375,
       "step": 510
     },
     {
       "epoch": 0.5445026178010471,
-      "grad_norm": 2.6920737581996477,
+      "grad_norm": 3.245002777118609,
       "learning_rate": 4.899899543831288e-05,
-      "logits/chosen": -8.25,
-      "logits/rejected": -6.9375,
-      "logps/chosen": -884.0,
-      "logps/rejected": -1928.0,
-      "loss": 0.262,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -4.53125,
-      "rewards/margins": 6.15625,
-      "rewards/rejected": -10.6875,
+      "logits/chosen": -11.625,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -856.0,
+      "logps/rejected": -1832.0,
+      "loss": 0.2778,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -4.25,
+      "rewards/margins": 5.4375,
+      "rewards/rejected": -9.6875,
       "step": 520
     },
     {
       "epoch": 0.5549738219895288,
-      "grad_norm": 4.210756068109314,
+      "grad_norm": 2.4455315382451883,
       "learning_rate": 4.891187051966765e-05,
-      "logits/chosen": -9.3125,
-      "logits/rejected": -8.875,
-      "logps/chosen": -876.0,
+      "logits/chosen": -11.75,
+      "logits/rejected": -9.75,
+      "logps/chosen": -872.0,
       "logps/rejected": -1744.0,
-      "loss": 0.275,
-      "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": -4.6875,
+      "loss": 0.2498,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.625,
       "rewards/margins": 5.0625,
-      "rewards/rejected": -9.75,
+      "rewards/rejected": -9.6875,
       "step": 530
     },
     {
       "epoch": 0.5654450261780105,
-      "grad_norm": 1.4151851777766988,
+      "grad_norm": 1.7182403383689901,
       "learning_rate": 4.882119467115191e-05,
-      "logits/chosen": -10.125,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -772.0,
-      "logps/rejected": -1640.0,
-      "loss": 0.2332,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -3.78125,
-      "rewards/margins": 5.1875,
-      "rewards/rejected": -9.0,
+      "logits/chosen": -12.25,
+      "logits/rejected": -10.125,
+      "logps/chosen": -864.0,
+      "logps/rejected": -1760.0,
+      "loss": 0.2469,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 5.46875,
+      "rewards/rejected": -10.1875,
       "step": 540
     },
     {
       "epoch": 0.5759162303664922,
-      "grad_norm": 2.307741398618567,
+      "grad_norm": 2.1188086901406114,
       "learning_rate": 4.872698135819402e-05,
-      "logits/chosen": -10.0625,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -712.0,
+      "logits/chosen": -12.5625,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -752.0,
       "logps/rejected": -1776.0,
-      "loss": 0.2519,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 5.15625,
+      "loss": 0.2614,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -3.75,
+      "rewards/margins": 4.8125,
       "rewards/rejected": -8.5625,
       "step": 550
     },
     {
       "epoch": 0.5863874345549738,
-      "grad_norm": 2.917195501002264,
+      "grad_norm": 3.0264849663995665,
       "learning_rate": 4.862924457153831e-05,
-      "logits/chosen": -10.25,
-      "logits/rejected": -8.6875,
-      "logps/chosen": -756.0,
-      "logps/rejected": -2048.0,
-      "loss": 0.2446,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -4.25,
-      "rewards/margins": 6.8125,
-      "rewards/rejected": -11.0625,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -10.125,
+      "logps/chosen": -820.0,
+      "logps/rejected": -2112.0,
+      "loss": 0.2378,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 6.96875,
+      "rewards/rejected": -11.8125,
       "step": 560
     },
     {
       "epoch": 0.5968586387434555,
-      "grad_norm": 1.8536654285962237,
+      "grad_norm": 2.5154684516361425,
       "learning_rate": 4.8527998825167484e-05,
-      "logits/chosen": -9.125,
-      "logits/rejected": -8.625,
-      "logps/chosen": -852.0,
-      "logps/rejected": -1904.0,
-      "loss": 0.2057,
+      "logits/chosen": -13.125,
+      "logits/rejected": -11.375,
+      "logps/chosen": -824.0,
+      "logps/rejected": -1888.0,
+      "loss": 0.2264,
       "rewards/accuracies": 0.9375,
-      "rewards/chosen": -4.75,
-      "rewards/margins": 6.6875,
-      "rewards/rejected": -11.4375,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 6.75,
+      "rewards/rejected": -11.25,
       "step": 570
     },
     {
       "epoch": 0.6073298429319371,
-      "grad_norm": 2.1430058730626005,
+      "grad_norm": 3.020749850131992,
       "learning_rate": 4.842325915414728e-05,
-      "logits/chosen": -8.6875,
-      "logits/rejected": -7.5,
-      "logps/chosen": -784.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.2025,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -4.28125,
-      "rewards/margins": 6.1875,
-      "rewards/rejected": -10.4375,
+      "logits/chosen": -13.25,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -752.0,
+      "logps/rejected": -1840.0,
+      "loss": 0.2143,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -3.9375,
+      "rewards/margins": 7.0625,
+      "rewards/rejected": -11.0,
       "step": 580
     },
     {
       "epoch": 0.6178010471204188,
-      "grad_norm": 3.3725131207605217,
+      "grad_norm": 3.232728574756352,
       "learning_rate": 4.831504111239373e-05,
-      "logits/chosen": -8.125,
-      "logits/rejected": -6.625,
-      "logps/chosen": -1016.0,
-      "logps/rejected": -2240.0,
-      "loss": 0.2608,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -6.03125,
-      "rewards/margins": 8.0625,
-      "rewards/rejected": -14.125,
+      "logits/chosen": -10.0625,
+      "logits/rejected": -7.1875,
+      "logps/chosen": -908.0,
+      "logps/rejected": -2080.0,
+      "loss": 0.2425,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 7.59375,
+      "rewards/rejected": -12.5625,
       "step": 590
     },
     {
       "epoch": 0.6282722513089005,
-      "grad_norm": 2.703636338811159,
+      "grad_norm": 2.7867546888157273,
       "learning_rate": 4.8203360770363425e-05,
-      "logits/chosen": -7.03125,
-      "logits/rejected": -5.9375,
-      "logps/chosen": -856.0,
-      "logps/rejected": -1880.0,
-      "loss": 0.2505,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -4.59375,
-      "rewards/margins": 5.8125,
-      "rewards/rejected": -10.375,
+      "logits/chosen": -11.5,
+      "logits/rejected": -9.125,
+      "logps/chosen": -908.0,
+      "logps/rejected": -2016.0,
+      "loss": 0.2543,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 6.65625,
+      "rewards/rejected": -11.75,
       "step": 600
     },
     {
       "epoch": 0.6282722513089005,
-      "eval_logits/chosen": -5.65625,
-      "eval_logits/rejected": -4.59375,
-      "eval_logps/chosen": -768.0,
-      "eval_logps/rejected": -808.0,
-      "eval_loss": 0.8266406059265137,
-      "eval_rewards/accuracies": 0.595703125,
-      "eval_rewards/chosen": -4.5,
-      "eval_rewards/margins": 0.671875,
-      "eval_rewards/rejected": -5.1875,
-      "eval_runtime": 46.58,
-      "eval_samples_per_second": 42.937,
+      "eval_logits/chosen": -14.1875,
+      "eval_logits/rejected": -13.5,
+      "eval_logps/chosen": -784.0,
+      "eval_logps/rejected": -820.0,
+      "eval_loss": 0.8722500205039978,
+      "eval_rewards/accuracies": 0.572265625,
+      "eval_rewards/chosen": -4.65625,
+      "eval_rewards/margins": 0.6328125,
+      "eval_rewards/rejected": -5.28125,
+      "eval_runtime": 46.6119,
+      "eval_samples_per_second": 42.907,
       "eval_steps_per_second": 0.687,
       "step": 600
     },
     {
       "epoch": 0.6387434554973822,
-      "grad_norm": 2.1405484822326093,
+      "grad_norm": 1.7065476709252634,
       "learning_rate": 4.808823471266702e-05,
-      "logits/chosen": -6.46875,
-      "logits/rejected": -5.25,
-      "logps/chosen": -864.0,
-      "logps/rejected": -1688.0,
-      "loss": 0.2514,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -4.75,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -12.6875,
+      "logps/chosen": -820.0,
+      "logps/rejected": -1640.0,
+      "loss": 0.2314,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -4.3125,
       "rewards/margins": 4.625,
-      "rewards/rejected": -9.375,
+      "rewards/rejected": -8.9375,
       "step": 610
     },
     {
       "epoch": 0.6492146596858639,
-      "grad_norm": 2.1911792149945524,
+      "grad_norm": 2.9261475105570094,
       "learning_rate": 4.796968003560639e-05,
-      "logits/chosen": -8.3125,
-      "logits/rejected": -6.9375,
-      "logps/chosen": -908.0,
-      "logps/rejected": -2208.0,
-      "loss": 0.2321,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -5.125,
-      "rewards/margins": 7.25,
-      "rewards/rejected": -12.375,
+      "logits/chosen": -14.0625,
+      "logits/rejected": -11.9375,
+      "logps/chosen": -896.0,
+      "logps/rejected": -2064.0,
+      "loss": 0.2673,
+      "rewards/accuracies": 0.831250011920929,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 6.0625,
+      "rewards/rejected": -11.0625,
       "step": 620
     },
     {
       "epoch": 0.6596858638743456,
-      "grad_norm": 1.9195604835194255,
+      "grad_norm": 1.80447198021574,
       "learning_rate": 4.7847714344635864e-05,
-      "logits/chosen": -10.625,
-      "logits/rejected": -8.75,
-      "logps/chosen": -792.0,
-      "logps/rejected": -1888.0,
-      "loss": 0.2284,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -4.125,
-      "rewards/margins": 6.34375,
-      "rewards/rejected": -10.5,
-      "step": 630
-    },
-    {
+      "logits/chosen": -15.625,
+      "logits/rejected": -12.9375,
+      "logps/chosen": -800.0,
+      "logps/rejected": -1864.0,
+      "loss": 0.2081,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.1875,
+      "rewards/margins": 6.03125,
+      "rewards/rejected": -10.25,
+      "step": 630
+    },
+    {
       "epoch": 0.6701570680628273,
-      "grad_norm": 2.1526624132925316,
+      "grad_norm": 2.9402021152942726,
       "learning_rate": 4.772235575174776e-05,
-      "logits/chosen": -9.625,
-      "logits/rejected": -7.96875,
-      "logps/chosen": -828.0,
-      "logps/rejected": -1584.0,
-      "loss": 0.2554,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -4.09375,
-      "rewards/margins": 4.125,
-      "rewards/rejected": -8.1875,
+      "logits/chosen": -12.9375,
+      "logits/rejected": -10.75,
+      "logps/chosen": -936.0,
+      "logps/rejected": -1896.0,
+      "loss": 0.2583,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -5.1875,
+      "rewards/margins": 6.125,
+      "rewards/rejected": -11.3125,
       "step": 640
     },
     {
       "epoch": 0.680628272251309,
-      "grad_norm": 3.2110530244642965,
+      "grad_norm": 3.0188304906092873,
       "learning_rate": 4.7593622872782754e-05,
-      "logits/chosen": -6.5625,
-      "logits/rejected": -5.78125,
-      "logps/chosen": -916.0,
-      "logps/rejected": -1936.0,
-      "loss": 0.2423,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -5.5625,
-      "rewards/margins": 5.71875,
-      "rewards/rejected": -11.25,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -10.375,
+      "logps/chosen": -920.0,
+      "logps/rejected": -2008.0,
+      "loss": 0.2538,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -5.59375,
+      "rewards/margins": 6.375,
+      "rewards/rejected": -11.9375,
       "step": 650
     },
     {
       "epoch": 0.6910994764397905,
-      "grad_norm": 2.1612514541592605,
+      "grad_norm": 2.5163706786456435,
       "learning_rate": 4.746153482466542e-05,
-      "logits/chosen": -9.125,
-      "logits/rejected": -7.21875,
-      "logps/chosen": -860.0,
-      "logps/rejected": -1880.0,
-      "loss": 0.2479,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -4.8125,
-      "rewards/margins": 6.53125,
-      "rewards/rejected": -11.375,
+      "logits/chosen": -14.25,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -816.0,
+      "logps/rejected": -1776.0,
+      "loss": 0.2342,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -4.375,
+      "rewards/margins": 5.96875,
+      "rewards/rejected": -10.375,
       "step": 660
     },
     {
       "epoch": 0.7015706806282722,
-      "grad_norm": 2.319732553729245,
+      "grad_norm": 2.9362190122225846,
       "learning_rate": 4.732611122256537e-05,
-      "logits/chosen": -9.125,
-      "logits/rejected": -7.875,
-      "logps/chosen": -960.0,
-      "logps/rejected": -1952.0,
-      "loss": 0.2214,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -5.25,
-      "rewards/margins": 5.75,
-      "rewards/rejected": -11.0,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -984.0,
+      "logps/rejected": -2128.0,
+      "loss": 0.2129,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -5.5,
+      "rewards/margins": 7.34375,
+      "rewards/rejected": -12.875,
       "step": 670
     },
     {
       "epoch": 0.7120418848167539,
-      "grad_norm": 2.8650454562616314,
+      "grad_norm": 1.7108064109620083,
       "learning_rate": 4.718737217698435e-05,
-      "logits/chosen": -8.375,
-      "logits/rejected": -7.25,
+      "logits/chosen": -10.875,
+      "logits/rejected": -8.125,
       "logps/chosen": -904.0,
-      "logps/rejected": -2096.0,
-      "loss": 0.2738,
-      "rewards/accuracies": 0.893750011920929,
+      "logps/rejected": -2208.0,
+      "loss": 0.2834,
+      "rewards/accuracies": 0.9312499761581421,
       "rewards/chosen": -5.28125,
-      "rewards/margins": 6.78125,
-      "rewards/rejected": -12.0625,
+      "rewards/margins": 7.90625,
+      "rewards/rejected": -13.1875,
       "step": 680
     },
     {
       "epoch": 0.7225130890052356,
-      "grad_norm": 2.1809813658169643,
+      "grad_norm": 2.09152024205709,
       "learning_rate": 4.704533829076982e-05,
-      "logits/chosen": -7.84375,
-      "logits/rejected": -6.03125,
-      "logps/chosen": -852.0,
-      "logps/rejected": -1704.0,
-      "loss": 0.2271,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -4.5625,
-      "rewards/margins": 5.0,
-      "rewards/rejected": -9.5625,
+      "logits/chosen": -10.1875,
+      "logits/rejected": -6.90625,
+      "logps/chosen": -888.0,
+      "logps/rejected": -1800.0,
+      "loss": 0.2296,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 5.5625,
+      "rewards/rejected": -10.5,
       "step": 690
     },
     {
       "epoch": 0.7329842931937173,
-      "grad_norm": 3.6772673728719907,
+      "grad_norm": 2.3694174119888904,
       "learning_rate": 4.6900030656055455e-05,
-      "logits/chosen": -6.375,
-      "logits/rejected": -4.6875,
-      "logps/chosen": -884.0,
-      "logps/rejected": -1848.0,
-      "loss": 0.2305,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -4.78125,
-      "rewards/margins": 5.125,
-      "rewards/rejected": -9.875,
+      "logits/chosen": -11.125,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -816.0,
+      "logps/rejected": -1808.0,
+      "loss": 0.2317,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -4.09375,
+      "rewards/margins": 5.375,
+      "rewards/rejected": -9.5,
       "step": 700
     },
     {
       "epoch": 0.7329842931937173,
-      "eval_logits/chosen": -5.0625,
-      "eval_logits/rejected": -3.703125,
-      "eval_logps/chosen": -856.0,
-      "eval_logps/rejected": -912.0,
-      "eval_loss": 0.7984218597412109,
-      "eval_rewards/accuracies": 0.62890625,
-      "eval_rewards/chosen": -5.375,
-      "eval_rewards/margins": 0.859375,
-      "eval_rewards/rejected": -6.25,
-      "eval_runtime": 46.6128,
-      "eval_samples_per_second": 42.907,
+      "eval_logits/chosen": -10.8125,
+      "eval_logits/rejected": -9.125,
+      "eval_logps/chosen": -808.0,
+      "eval_logps/rejected": -868.0,
+      "eval_loss": 0.8776249885559082,
+      "eval_rewards/accuracies": 0.603515625,
+      "eval_rewards/chosen": -4.875,
+      "eval_rewards/margins": 0.9140625,
+      "eval_rewards/rejected": -5.8125,
+      "eval_runtime": 46.6115,
+      "eval_samples_per_second": 42.908,
       "eval_steps_per_second": 0.687,
       "step": 700
     },
     {
       "epoch": 0.743455497382199,
-      "grad_norm": 2.684184405429759,
+      "grad_norm": 2.953154017404595,
       "learning_rate": 4.6751470851128906e-05,
-      "logits/chosen": -7.125,
-      "logits/rejected": -5.1875,
-      "logps/chosen": -912.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.2273,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -4.875,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -9.3125,
+      "logits/chosen": -11.625,
+      "logits/rejected": -8.5,
+      "logps/chosen": -896.0,
+      "logps/rejected": -1800.0,
+      "loss": 0.2317,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.75,
+      "rewards/margins": 5.4375,
+      "rewards/rejected": -10.1875,
       "step": 710
     },
     {
       "epoch": 0.7539267015706806,
-      "grad_norm": 2.5803299727904596,
+      "grad_norm": 3.012877005030324,
       "learning_rate": 4.6599680937227454e-05,
-      "logits/chosen": -8.1875,
-      "logits/rejected": -6.09375,
-      "logps/chosen": -924.0,
-      "logps/rejected": -1928.0,
-      "loss": 0.2453,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -5.1875,
+      "logits/chosen": -10.1875,
+      "logits/rejected": -7.375,
+      "logps/chosen": -948.0,
+      "logps/rejected": -1952.0,
+      "loss": 0.2454,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -5.4375,
       "rewards/margins": 6.34375,
-      "rewards/rejected": -11.5,
+      "rewards/rejected": -11.8125,
       "step": 720
     },
     {
       "epoch": 0.7643979057591623,
-      "grad_norm": 1.973234878183987,
+      "grad_norm": 1.9765990495154775,
       "learning_rate": 4.644468345526183e-05,
-      "logits/chosen": -7.84375,
-      "logits/rejected": -6.25,
-      "logps/chosen": -924.0,
-      "logps/rejected": -2032.0,
-      "loss": 0.2359,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -5.25,
-      "rewards/margins": 6.3125,
-      "rewards/rejected": -11.5625,
+      "logits/chosen": -9.8125,
+      "logits/rejected": -7.75,
+      "logps/chosen": -968.0,
+      "logps/rejected": -1960.0,
+      "loss": 0.2333,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -5.6875,
+      "rewards/margins": 5.15625,
+      "rewards/rejected": -10.8125,
       "step": 730
     },
     {
       "epoch": 0.774869109947644,
-      "grad_norm": 3.184944108645906,
+      "grad_norm": 3.1755719420337876,
       "learning_rate": 4.628650142246894e-05,
-      "logits/chosen": -8.0625,
-      "logits/rejected": -5.4375,
-      "logps/chosen": -852.0,
-      "logps/rejected": -1816.0,
-      "loss": 0.2242,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -4.75,
-      "rewards/margins": 5.65625,
-      "rewards/rejected": -10.375,
+      "logits/chosen": -15.0625,
+      "logits/rejected": -11.5625,
+      "logps/chosen": -812.0,
+      "logps/rejected": -1736.0,
+      "loss": 0.2362,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -4.3125,
+      "rewards/margins": 5.34375,
+      "rewards/rejected": -9.625,
       "step": 740
     },
     {
       "epoch": 0.7853403141361257,
-      "grad_norm": 2.9671627362179063,
+      "grad_norm": 2.7478609859946435,
       "learning_rate": 4.6125158328993765e-05,
-      "logits/chosen": -7.15625,
-      "logits/rejected": -5.34375,
-      "logps/chosen": -884.0,
-      "logps/rejected": -1792.0,
-      "loss": 0.2183,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -5.125,
-      "rewards/margins": 5.125,
-      "rewards/rejected": -10.25,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -13.0,
+      "logps/chosen": -824.0,
+      "logps/rejected": -1760.0,
+      "loss": 0.2301,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 5.46875,
+      "rewards/rejected": -9.9375,
       "step": 750
     },
     {
       "epoch": 0.7958115183246073,
-      "grad_norm": 2.925036306888871,
+      "grad_norm": 2.545328717376369,
       "learning_rate": 4.5960678134401064e-05,
-      "logits/chosen": -5.34375,
-      "logits/rejected": -4.15625,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -2112.0,
-      "loss": 0.2328,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -928.0,
+      "logps/rejected": -1944.0,
+      "loss": 0.2204,
       "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -7.03125,
+      "rewards/chosen": -5.3125,
       "rewards/margins": 5.9375,
-      "rewards/rejected": -12.9375,
+      "rewards/rejected": -11.25,
       "step": 760
     },
     {
       "epoch": 0.806282722513089,
-      "grad_norm": 1.6050891982444613,
+      "grad_norm": 1.6452854469283442,
       "learning_rate": 4.579308526411734e-05,
-      "logits/chosen": -4.9375,
-      "logits/rejected": -4.125,
-      "logps/chosen": -1120.0,
-      "logps/rejected": -2320.0,
-      "loss": 0.2116,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.28125,
-      "rewards/margins": 7.5625,
-      "rewards/rejected": -14.875,
+      "logits/chosen": -12.3125,
+      "logits/rejected": -10.0,
+      "logps/chosen": -980.0,
+      "logps/rejected": -2160.0,
+      "loss": 0.1995,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.90625,
+      "rewards/margins": 7.40625,
+      "rewards/rejected": -13.375,
       "step": 770
     },
     {
       "epoch": 0.8167539267015707,
-      "grad_norm": 2.896647431051217,
+      "grad_norm": 3.685751479648318,
       "learning_rate": 4.562240460580367e-05,
-      "logits/chosen": -7.875,
-      "logits/rejected": -6.59375,
-      "logps/chosen": -904.0,
-      "logps/rejected": -2016.0,
-      "loss": 0.248,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -5.125,
-      "rewards/margins": 6.84375,
-      "rewards/rejected": -11.9375,
+      "logits/chosen": -12.4375,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -936.0,
+      "logps/rejected": -2128.0,
+      "loss": 0.2195,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -5.40625,
+      "rewards/margins": 7.65625,
+      "rewards/rejected": -13.0625,
       "step": 780
     },
     {
       "epoch": 0.8272251308900523,
-      "grad_norm": 2.0315563409671484,
+      "grad_norm": 1.8264921795282367,
       "learning_rate": 4.544866150565985e-05,
-      "logits/chosen": -6.8125,
-      "logits/rejected": -5.25,
-      "logps/chosen": -940.0,
-      "logps/rejected": -2256.0,
-      "loss": 0.1937,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -5.53125,
-      "rewards/margins": 8.3125,
-      "rewards/rejected": -13.875,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -8.75,
+      "logps/chosen": -980.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.2115,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -5.9375,
+      "rewards/margins": 7.5,
+      "rewards/rejected": -13.4375,
       "step": 790
     },
     {
       "epoch": 0.837696335078534,
-      "grad_norm": 3.3727134250930093,
+      "grad_norm": 3.000761336327027,
       "learning_rate": 4.527188176466051e-05,
-      "logits/chosen": -6.3125,
-      "logits/rejected": -5.21875,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -2096.0,
-      "loss": 0.2384,
-      "rewards/accuracies": 0.8500000238418579,
-      "rewards/chosen": -6.25,
-      "rewards/margins": 6.75,
-      "rewards/rejected": -13.0,
+      "logits/chosen": -10.8125,
+      "logits/rejected": -8.875,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.2231,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -7.21875,
+      "rewards/margins": 7.0625,
+      "rewards/rejected": -14.3125,
       "step": 800
     },
     {
       "epoch": 0.837696335078534,
-      "eval_logits/chosen": -5.03125,
-      "eval_logits/rejected": -3.828125,
-      "eval_logps/chosen": -904.0,
-      "eval_logps/rejected": -952.0,
-      "eval_loss": 0.9506406188011169,
+      "eval_logits/chosen": -11.5625,
+      "eval_logits/rejected": -10.3125,
+      "eval_logps/chosen": -1088.0,
+      "eval_logps/rejected": -1168.0,
+      "eval_loss": 1.0686718225479126,
       "eval_rewards/accuracies": 0.572265625,
-      "eval_rewards/chosen": -5.875,
-      "eval_rewards/margins": 0.7578125,
-      "eval_rewards/rejected": -6.625,
-      "eval_runtime": 46.6151,
-      "eval_samples_per_second": 42.905,
+      "eval_rewards/chosen": -7.71875,
+      "eval_rewards/margins": 1.0859375,
+      "eval_rewards/rejected": -8.8125,
+      "eval_runtime": 46.6191,
+      "eval_samples_per_second": 42.901,
       "eval_steps_per_second": 0.686,
       "step": 800
     },
     {
       "epoch": 0.8481675392670157,
-      "grad_norm": 1.6793599967067072,
+      "grad_norm": 1.5463844277417869,
       "learning_rate": 4.509209163472357e-05,
-      "logits/chosen": -6.4375,
-      "logits/rejected": -4.5,
-      "logps/chosen": -940.0,
-      "logps/rejected": -2040.0,
-      "loss": 0.2,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -5.375,
-      "rewards/margins": 6.4375,
-      "rewards/rejected": -11.8125,
+      "logits/chosen": -13.125,
+      "logits/rejected": -10.25,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -2352.0,
+      "loss": 0.216,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -6.4375,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -14.9375,
       "step": 810
     },
     {
       "epoch": 0.8586387434554974,
-      "grad_norm": 1.5665736565076624,
+      "grad_norm": 1.1646839134739728,
       "learning_rate": 4.4909317814811896e-05,
-      "logits/chosen": -5.96875,
-      "logits/rejected": -4.75,
-      "logps/chosen": -880.0,
-      "logps/rejected": -1808.0,
-      "loss": 0.2383,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -4.78125,
-      "rewards/margins": 5.0625,
-      "rewards/rejected": -9.875,
+      "logits/chosen": -14.375,
+      "logits/rejected": -12.3125,
+      "logps/chosen": -908.0,
+      "logps/rejected": -1904.0,
+      "loss": 0.2471,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.0625,
+      "rewards/margins": 5.75,
+      "rewards/rejected": -10.8125,
       "step": 820
     },
     {
       "epoch": 0.8691099476439791,
-      "grad_norm": 1.5528906021745623,
+      "grad_norm": 1.5957822712832495,
       "learning_rate": 4.472358744696845e-05,
-      "logits/chosen": -6.09375,
-      "logits/rejected": -5.28125,
-      "logps/chosen": -840.0,
-      "logps/rejected": -1840.0,
-      "loss": 0.2294,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -4.46875,
-      "rewards/margins": 5.5,
-      "rewards/rejected": -10.0,
+      "logits/chosen": -12.4375,
+      "logits/rejected": -9.875,
+      "logps/chosen": -936.0,
+      "logps/rejected": -2048.0,
+      "loss": 0.2144,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -5.46875,
+      "rewards/margins": 6.65625,
+      "rewards/rejected": -12.125,
       "step": 830
     },
     {
       "epoch": 0.8795811518324608,
-      "grad_norm": 3.2124955111761935,
+      "grad_norm": 4.477675782724629,
       "learning_rate": 4.4534928112285645e-05,
-      "logits/chosen": -4.96875,
-      "logits/rejected": -4.1875,
-      "logps/chosen": -976.0,
-      "logps/rejected": -2096.0,
-      "loss": 0.2193,
+      "logits/chosen": -10.25,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -968.0,
+      "logps/rejected": -2176.0,
+      "loss": 0.2187,
       "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -5.9375,
-      "rewards/margins": 6.625,
-      "rewards/rejected": -12.5625,
+      "rewards/chosen": -5.875,
+      "rewards/margins": 7.59375,
+      "rewards/rejected": -13.4375,
       "step": 840
     },
     {
       "epoch": 0.8900523560209425,
-      "grad_norm": 2.5743189308951258,
+      "grad_norm": 2.7387801165370127,
       "learning_rate": 4.43433678268096e-05,
-      "logits/chosen": -6.0,
-      "logits/rejected": -4.8125,
-      "logps/chosen": -924.0,
-      "logps/rejected": -1856.0,
-      "loss": 0.2253,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -5.40625,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -9.125,
+      "logps/chosen": -884.0,
+      "logps/rejected": -1816.0,
+      "loss": 0.2411,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.96875,
       "rewards/margins": 5.78125,
-      "rewards/rejected": -11.1875,
+      "rewards/rejected": -10.75,
       "step": 850
     },
     {
       "epoch": 0.900523560209424,
-      "grad_norm": 2.596921876416307,
+      "grad_norm": 1.7889477835253804,
       "learning_rate": 4.414893503737968e-05,
-      "logits/chosen": -5.3125,
-      "logits/rejected": -4.78125,
-      "logps/chosen": -816.0,
+      "logits/chosen": -10.375,
+      "logits/rejected": -8.5625,
+      "logps/chosen": -844.0,
       "logps/rejected": -1712.0,
-      "loss": 0.2077,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -4.65625,
-      "rewards/margins": 5.125,
+      "loss": 0.199,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 4.84375,
       "rewards/rejected": -9.8125,
       "step": 860
     },
     {
       "epoch": 0.9109947643979057,
-      "grad_norm": 4.05989437793278,
+      "grad_norm": 4.065691865366964,
       "learning_rate": 4.395165861740414e-05,
-      "logits/chosen": -5.71875,
-      "logits/rejected": -4.53125,
-      "logps/chosen": -932.0,
-      "logps/rejected": -2192.0,
-      "loss": 0.1959,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -5.6875,
-      "rewards/margins": 7.5625,
-      "rewards/rejected": -13.25,
+      "logits/chosen": -9.1875,
+      "logits/rejected": -7.28125,
+      "logps/chosen": -964.0,
+      "logps/rejected": -2176.0,
+      "loss": 0.1954,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -6.0,
+      "rewards/margins": 7.0625,
+      "rewards/rejected": -13.0625,
       "step": 870
     },
     {
       "epoch": 0.9214659685863874,
-      "grad_norm": 4.296152810742322,
+      "grad_norm": 3.404031756463821,
       "learning_rate": 4.375156786257241e-05,
-      "logits/chosen": -6.75,
-      "logits/rejected": -5.53125,
-      "logps/chosen": -988.0,
-      "logps/rejected": -2160.0,
-      "loss": 0.2703,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -5.6875,
-      "rewards/margins": 7.46875,
-      "rewards/rejected": -13.125,
+      "logits/chosen": -12.5,
+      "logits/rejected": -10.375,
+      "logps/chosen": -956.0,
+      "logps/rejected": -2080.0,
+      "loss": 0.2858,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -5.375,
+      "rewards/margins": 6.90625,
+      "rewards/rejected": -12.25,
       "step": 880
     },
     {
       "epoch": 0.9319371727748691,
-      "grad_norm": 2.0127605987283803,
+      "grad_norm": 2.293009839247518,
       "learning_rate": 4.354869248650466e-05,
-      "logits/chosen": -6.25,
-      "logits/rejected": -4.125,
-      "logps/chosen": -868.0,
-      "logps/rejected": -1864.0,
-      "loss": 0.2136,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -11.0625,
+      "logps/chosen": -800.0,
+      "logps/rejected": -1704.0,
+      "loss": 0.2289,
       "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -4.6875,
-      "rewards/margins": 6.03125,
-      "rewards/rejected": -10.6875,
+      "rewards/chosen": -4.0,
+      "rewards/margins": 5.0625,
+      "rewards/rejected": -9.0625,
       "step": 890
     },
     {
       "epoch": 0.9424083769633508,
-      "grad_norm": 3.6357700023632677,
+      "grad_norm": 4.060897071047721,
       "learning_rate": 4.334306261633931e-05,
-      "logits/chosen": -4.9375,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -968.0,
-      "logps/rejected": -2080.0,
-      "loss": 0.2003,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -5.5,
-      "rewards/margins": 6.46875,
-      "rewards/rejected": -12.0,
+      "logits/chosen": -9.75,
+      "logits/rejected": -7.96875,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -2112.0,
+      "loss": 0.2171,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": -6.34375,
+      "rewards/margins": 5.96875,
+      "rewards/rejected": -12.3125,
       "step": 900
     },
     {
       "epoch": 0.9424083769633508,
-      "eval_logits/chosen": -3.75,
-      "eval_logits/rejected": -2.5,
-      "eval_logps/chosen": -1000.0,
-      "eval_logps/rejected": -1072.0,
-      "eval_loss": 0.9552500247955322,
-      "eval_rewards/accuracies": 0.59375,
-      "eval_rewards/chosen": -6.84375,
-      "eval_rewards/margins": 0.98828125,
-      "eval_rewards/rejected": -7.8125,
-      "eval_runtime": 46.5943,
-      "eval_samples_per_second": 42.924,
+      "eval_logits/chosen": -11.0,
+      "eval_logits/rejected": -10.125,
+      "eval_logps/chosen": -1096.0,
+      "eval_logps/rejected": -1144.0,
+      "eval_loss": 0.9578906297683716,
+      "eval_rewards/accuracies": 0.576171875,
+      "eval_rewards/chosen": -7.75,
+      "eval_rewards/margins": 0.83984375,
+      "eval_rewards/rejected": -8.5625,
+      "eval_runtime": 46.5879,
+      "eval_samples_per_second": 42.93,
       "eval_steps_per_second": 0.687,
       "step": 900
     },
     {
       "epoch": 0.9528795811518325,
-      "grad_norm": 2.1170374300546557,
+      "grad_norm": 2.775509170834493,
       "learning_rate": 4.313470878825911e-05,
-      "logits/chosen": -6.0625,
-      "logits/rejected": -5.0625,
-      "logps/chosen": -920.0,
-      "logps/rejected": -2416.0,
-      "loss": 0.2258,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -5.25,
-      "rewards/margins": 8.3125,
-      "rewards/rejected": -13.5625,
+      "logits/chosen": -14.0,
+      "logits/rejected": -11.875,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.2324,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -6.59375,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -15.1875,
       "step": 910
     },
     {
       "epoch": 0.9633507853403142,
-      "grad_norm": 6.374006416114005,
+      "grad_norm": 1.1660722743410934,
       "learning_rate": 4.292366194295652e-05,
-      "logits/chosen": -6.09375,
-      "logits/rejected": -4.1875,
+      "logits/chosen": -15.25,
+      "logits/rejected": -12.875,
       "logps/chosen": -904.0,
-      "logps/rejected": -1800.0,
-      "loss": 0.2298,
-      "rewards/accuracies": 0.8812500238418579,
+      "logps/rejected": -1784.0,
+      "loss": 0.2407,
+      "rewards/accuracies": 0.90625,
       "rewards/chosen": -5.0625,
-      "rewards/margins": 5.5,
-      "rewards/rejected": -10.5625,
+      "rewards/margins": 5.40625,
+      "rewards/rejected": -10.5,
       "step": 920
     },
     {
       "epoch": 0.9738219895287958,
-      "grad_norm": 1.63070059688184,
+      "grad_norm": 2.7148276522021457,
       "learning_rate": 4.2709953421038964e-05,
-      "logits/chosen": -6.0,
-      "logits/rejected": -4.34375,
-      "logps/chosen": -864.0,
-      "logps/rejected": -1672.0,
-      "loss": 0.2533,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -4.875,
-      "rewards/margins": 4.34375,
-      "rewards/rejected": -9.1875,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -11.5,
+      "logps/chosen": -844.0,
+      "logps/rejected": -1680.0,
+      "loss": 0.2466,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -4.65625,
+      "rewards/margins": 4.625,
+      "rewards/rejected": -9.3125,
       "step": 930
     },
     {
       "epoch": 0.9842931937172775,
-      "grad_norm": 2.1867665555397138,
+      "grad_norm": 2.608326600501226,
       "learning_rate": 4.249361495837478e-05,
-      "logits/chosen": -5.3125,
-      "logits/rejected": -3.796875,
-      "logps/chosen": -900.0,
-      "logps/rejected": -1920.0,
-      "loss": 0.2421,
-      "rewards/accuracies": 0.9375,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -10.375,
+      "logps/chosen": -904.0,
+      "logps/rejected": -1944.0,
+      "loss": 0.2296,
+      "rewards/accuracies": 0.9437500238418579,
       "rewards/chosen": -5.28125,
-      "rewards/margins": 5.0625,
-      "rewards/rejected": -10.3125,
+      "rewards/margins": 5.25,
+      "rewards/rejected": -10.5625,
       "step": 940
     },
     {
       "epoch": 0.9947643979057592,
-      "grad_norm": 3.4453174636741464,
+      "grad_norm": 3.574321407899499,
       "learning_rate": 4.227467868138035e-05,
-      "logits/chosen": -4.5,
-      "logits/rejected": -3.1875,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -2096.0,
-      "loss": 0.231,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -6.78125,
-      "rewards/margins": 5.8125,
-      "rewards/rejected": -12.625,
+      "logits/chosen": -12.625,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -984.0,
+      "logps/rejected": -1976.0,
+      "loss": 0.2455,
+      "rewards/accuracies": 0.8687499761581421,
+      "rewards/chosen": -5.8125,
+      "rewards/margins": 5.59375,
+      "rewards/rejected": -11.375,
       "step": 950
     },
     {
       "epoch": 1.0052356020942408,
-      "grad_norm": 0.5801358434812638,
+      "grad_norm": 0.8120073865539842,
       "learning_rate": 4.205317710224938e-05,
-      "logits/chosen": -7.0,
-      "logits/rejected": -4.9375,
-      "logps/chosen": -856.0,
-      "logps/rejected": -2096.0,
-      "loss": 0.1373,
+      "logits/chosen": -14.0,
+      "logits/rejected": -11.0,
+      "logps/chosen": -832.0,
+      "logps/rejected": -1992.0,
+      "loss": 0.1564,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -4.5,
-      "rewards/margins": 6.5625,
-      "rewards/rejected": -11.0625,
+      "rewards/chosen": -4.25,
+      "rewards/margins": 5.8125,
+      "rewards/rejected": -10.0625,
       "step": 960
     },
     {
       "epoch": 1.0157068062827226,
-      "grad_norm": 0.8344711365521521,
+      "grad_norm": 1.116999681413669,
       "learning_rate": 4.1829143114124733e-05,
-      "logits/chosen": -6.0625,
-      "logits/rejected": -5.0,
-      "logps/chosen": -920.0,
-      "logps/rejected": -2096.0,
-      "loss": 0.0694,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -5.3125,
-      "rewards/margins": 8.25,
-      "rewards/rejected": -13.5,
+      "logits/chosen": -9.75,
+      "logits/rejected": -7.4375,
+      "logps/chosen": -956.0,
+      "logps/rejected": -2112.0,
+      "loss": 0.0795,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -5.65625,
+      "rewards/margins": 8.0625,
+      "rewards/rejected": -13.75,
       "step": 970
     },
     {
       "epoch": 1.0261780104712042,
-      "grad_norm": 0.5047145856475589,
+      "grad_norm": 0.8705791955218226,
       "learning_rate": 4.160260998621387e-05,
-      "logits/chosen": -5.5625,
-      "logits/rejected": -4.09375,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -2928.0,
-      "loss": 0.0731,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -8.75,
-      "rewards/margins": 12.75,
-      "rewards/rejected": -21.5,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -6.3125,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.0738,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.9375,
+      "rewards/margins": 13.0625,
+      "rewards/rejected": -21.0,
       "step": 980
     },
     {
       "epoch": 1.036649214659686,
-      "grad_norm": 0.7294891916569339,
+      "grad_norm": 1.3900842737593766,
       "learning_rate": 4.137361135884824e-05,
-      "logits/chosen": -7.0,
-      "logits/rejected": -5.1875,
-      "logps/chosen": -948.0,
-      "logps/rejected": -2384.0,
-      "loss": 0.0723,
+      "logits/chosen": -9.6875,
+      "logits/rejected": -7.40625,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0642,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -5.53125,
-      "rewards/margins": 10.0625,
-      "rewards/rejected": -15.5625,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -19.5,
       "step": 990
     },
     {
       "epoch": 1.0471204188481675,
-      "grad_norm": 1.122114769422544,
+      "grad_norm": 3.528808247487733,
       "learning_rate": 4.114218123848777e-05,
-      "logits/chosen": -5.3125,
-      "logits/rejected": -4.1875,
-      "logps/chosen": -976.0,
-      "logps/rejected": -2544.0,
-      "loss": 0.0478,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -6.03125,
-      "rewards/margins": 10.6875,
-      "rewards/rejected": -16.75,
+      "logits/chosen": -8.75,
+      "logits/rejected": -7.0,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.0657,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.8125,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -20.125,
       "step": 1000
     },
     {
       "epoch": 1.0471204188481675,
-      "eval_logits/chosen": -3.515625,
-      "eval_logits/rejected": -1.9453125,
-      "eval_logps/chosen": -1136.0,
-      "eval_logps/rejected": -1224.0,
-      "eval_loss": 1.203281283378601,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -8.1875,
-      "eval_rewards/margins": 1.1640625,
-      "eval_rewards/rejected": -9.3125,
-      "eval_runtime": 46.6073,
-      "eval_samples_per_second": 42.912,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -9.625,
+      "eval_logits/rejected": -8.3125,
+      "eval_logps/chosen": -1048.0,
+      "eval_logps/rejected": -1128.0,
+      "eval_loss": 1.0862656831741333,
+      "eval_rewards/accuracies": 0.583984375,
+      "eval_rewards/chosen": -7.28125,
+      "eval_rewards/margins": 1.1015625,
+      "eval_rewards/rejected": -8.375,
+      "eval_runtime": 46.6164,
+      "eval_samples_per_second": 42.903,
+      "eval_steps_per_second": 0.686,
       "step": 1000
     },
     {
       "epoch": 1.057591623036649,
-      "grad_norm": 1.3565258512363132,
+      "grad_norm": 2.3253620247534794,
       "learning_rate": 4.0908353992670824e-05,
-      "logits/chosen": -5.9375,
-      "logits/rejected": -4.21875,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -2464.0,
-      "loss": 0.0703,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.90625,
-      "rewards/margins": 10.375,
-      "rewards/rejected": -17.25,
+      "logits/chosen": -11.8125,
+      "logits/rejected": -9.75,
+      "logps/chosen": -920.0,
+      "logps/rejected": -2160.0,
+      "loss": 0.0872,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -5.4375,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -14.25,
       "step": 1010
     },
     {
       "epoch": 1.068062827225131,
-      "grad_norm": 0.8179574193749647,
+      "grad_norm": 1.2508641577154647,
       "learning_rate": 4.067216434491059e-05,
-      "logits/chosen": -5.3125,
-      "logits/rejected": -4.1875,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0506,
+      "logits/chosen": -11.8125,
+      "logits/rejected": -10.125,
+      "logps/chosen": -960.0,
+      "logps/rejected": -2320.0,
+      "loss": 0.0552,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.0,
-      "rewards/margins": 10.375,
-      "rewards/rejected": -17.375,
+      "rewards/chosen": -5.75,
+      "rewards/margins": 10.0,
+      "rewards/rejected": -15.75,
       "step": 1020
     },
     {
       "epoch": 1.0785340314136125,
-      "grad_norm": 2.538307237295152,
+      "grad_norm": 2.7976841018357286,
       "learning_rate": 4.043364736953866e-05,
-      "logits/chosen": -5.40625,
-      "logits/rejected": -3.921875,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0796,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.625,
-      "rewards/margins": 10.3125,
-      "rewards/rejected": -16.875,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -2912.0,
+      "loss": 0.0782,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.0625,
+      "rewards/margins": 12.8125,
+      "rewards/rejected": -19.875,
       "step": 1030
     },
     {
       "epoch": 1.0890052356020943,
-      "grad_norm": 0.865427667812597,
+      "grad_norm": 1.2999490291614135,
       "learning_rate": 4.019283848649643e-05,
-      "logits/chosen": -7.8125,
-      "logits/rejected": -5.71875,
-      "logps/chosen": -936.0,
-      "logps/rejected": -2064.0,
-      "loss": 0.0774,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -5.375,
-      "rewards/margins": 7.46875,
-      "rewards/rejected": -12.875,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -10.875,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.05,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -6.9375,
+      "rewards/margins": 10.5,
+      "rewards/rejected": -17.375,
       "step": 1040
     },
     {
       "epoch": 1.0994764397905759,
-      "grad_norm": 0.9839953228754271,
+      "grad_norm": 0.4403691325289348,
       "learning_rate": 3.9949773456075215e-05,
-      "logits/chosen": -5.8125,
-      "logits/rejected": -4.46875,
-      "logps/chosen": -932.0,
-      "logps/rejected": -2288.0,
-      "loss": 0.0531,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -5.5625,
-      "rewards/margins": 8.5625,
-      "rewards/rejected": -14.125,
+      "logits/chosen": -12.375,
+      "logits/rejected": -9.875,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0383,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -6.84375,
+      "rewards/margins": 12.5,
+      "rewards/rejected": -19.375,
       "step": 1050
     },
     {
       "epoch": 1.1099476439790577,
-      "grad_norm": 0.7388746904950071,
+      "grad_norm": 1.6867700184256351,
       "learning_rate": 3.970448837360585e-05,
-      "logits/chosen": -5.84375,
-      "logits/rejected": -4.65625,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -2384.0,
-      "loss": 0.0599,
+      "logits/chosen": -12.25,
+      "logits/rejected": -9.75,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0784,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.75,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -15.625,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 10.6875,
+      "rewards/rejected": -17.875,
       "step": 1060
     },
     {
       "epoch": 1.1204188481675392,
-      "grad_norm": 1.2592928093021656,
+      "grad_norm": 1.577069303118089,
       "learning_rate": 3.945701966409846e-05,
-      "logits/chosen": -6.1875,
-      "logits/rejected": -4.78125,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0472,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.65625,
-      "rewards/margins": 11.6875,
-      "rewards/rejected": -18.375,
+      "logits/chosen": -10.5625,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -1016.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0552,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -6.34375,
+      "rewards/margins": 11.625,
+      "rewards/rejected": -18.0,
       "step": 1070
     },
     {
       "epoch": 1.130890052356021,
-      "grad_norm": 0.47446092068918994,
+      "grad_norm": 1.5263376333706904,
       "learning_rate": 3.9207404076833374e-05,
-      "logits/chosen": -4.59375,
-      "logits/rejected": -2.59375,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -2992.0,
-      "loss": 0.0545,
+      "logits/chosen": -8.6875,
+      "logits/rejected": -6.3125,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0588,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 12.9375,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -7.625,
+      "rewards/margins": 11.4375,
+      "rewards/rejected": -19.0,
       "step": 1080
     },
     {
       "epoch": 1.1413612565445026,
-      "grad_norm": 1.0610806677877183,
+      "grad_norm": 3.825667424031301,
       "learning_rate": 3.895567867990379e-05,
-      "logits/chosen": -4.71875,
-      "logits/rejected": -2.859375,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.0439,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 13.5,
-      "rewards/rejected": -21.875,
+      "logits/chosen": -8.875,
+      "logits/rejected": -6.71875,
+      "logps/chosen": -1408.0,
+      "logps/rejected": -3360.0,
+      "loss": 0.0634,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -9.5625,
+      "rewards/margins": 14.4375,
+      "rewards/rejected": -24.0,
       "step": 1090
     },
     {
       "epoch": 1.1518324607329844,
-      "grad_norm": 1.3629624576031214,
+      "grad_norm": 1.492788000938283,
       "learning_rate": 3.870188085471114e-05,
-      "logits/chosen": -4.21875,
-      "logits/rejected": -3.03125,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0626,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.875,
-      "rewards/margins": 10.6875,
-      "rewards/rejected": -18.5,
+      "logits/chosen": -10.375,
+      "logits/rejected": -8.5,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.0641,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.53125,
+      "rewards/margins": 10.625,
+      "rewards/rejected": -18.125,
       "step": 1100
     },
     {
       "epoch": 1.1518324607329844,
-      "eval_logits/chosen": -3.203125,
-      "eval_logits/rejected": -1.578125,
-      "eval_logps/chosen": -1136.0,
-      "eval_logps/rejected": -1256.0,
-      "eval_loss": 1.1789531707763672,
-      "eval_rewards/accuracies": 0.591796875,
-      "eval_rewards/chosen": -8.1875,
-      "eval_rewards/margins": 1.515625,
-      "eval_rewards/rejected": -9.6875,
-      "eval_runtime": 46.6149,
-      "eval_samples_per_second": 42.905,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -10.125,
+      "eval_logits/rejected": -8.9375,
+      "eval_logps/chosen": -1120.0,
+      "eval_logps/rejected": -1208.0,
+      "eval_loss": 1.139718770980835,
+      "eval_rewards/accuracies": 0.59765625,
+      "eval_rewards/chosen": -8.0,
+      "eval_rewards/margins": 1.171875,
+      "eval_rewards/rejected": -9.1875,
+      "eval_runtime": 46.6106,
+      "eval_samples_per_second": 42.909,
+      "eval_steps_per_second": 0.687,
       "step": 1100
     },
     {
       "epoch": 1.162303664921466,
-      "grad_norm": 0.5191893255195431,
+      "grad_norm": 0.8068549149786886,
       "learning_rate": 3.844604829041395e-05,
-      "logits/chosen": -4.34375,
-      "logits/rejected": -2.828125,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0573,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.40625,
-      "rewards/margins": 10.875,
-      "rewards/rejected": -18.25,
+      "logits/chosen": -10.0,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.072,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 10.3125,
+      "rewards/rejected": -18.5,
       "step": 1110
     },
     {
       "epoch": 1.1727748691099475,
-      "grad_norm": 1.5414984051419345,
+      "grad_norm": 1.9684803773121133,
       "learning_rate": 3.818821897833095e-05,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -2.28125,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -2512.0,
-      "loss": 0.082,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -1020.0,
+      "logps/rejected": -2256.0,
+      "loss": 0.0763,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.5,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -17.375,
+      "rewards/chosen": -6.125,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -14.6875,
       "step": 1120
     },
     {
       "epoch": 1.1832460732984293,
-      "grad_norm": 2.7460207514056347,
+      "grad_norm": 1.2337422293242988,
       "learning_rate": 3.792843120629935e-05,
-      "logits/chosen": -5.46875,
-      "logits/rejected": -3.421875,
-      "logps/chosen": -924.0,
-      "logps/rejected": -2336.0,
-      "loss": 0.074,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -5.5,
-      "rewards/margins": 9.625,
-      "rewards/rejected": -15.125,
+      "logits/chosen": -11.75,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -928.0,
+      "logps/rejected": -2400.0,
+      "loss": 0.0619,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -5.53125,
+      "rewards/margins": 10.25,
+      "rewards/rejected": -15.75,
       "step": 1130
     },
     {
       "epoch": 1.193717277486911,
-      "grad_norm": 1.5837523275699386,
+      "grad_norm": 0.44040185265128595,
       "learning_rate": 3.766672355298907e-05,
-      "logits/chosen": -5.4375,
-      "logits/rejected": -3.46875,
-      "logps/chosen": -996.0,
-      "logps/rejected": -2448.0,
-      "loss": 0.0563,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.3125,
-      "rewards/margins": 10.0,
-      "rewards/rejected": -16.375,
+      "logits/chosen": -8.875,
+      "logits/rejected": -6.5625,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -2992.0,
+      "loss": 0.0493,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 13.1875,
+      "rewards/rejected": -21.625,
       "step": 1140
     },
     {
       "epoch": 1.2041884816753927,
-      "grad_norm": 3.1926846097170127,
+      "grad_norm": 4.1861605782168665,
       "learning_rate": 3.740313488217373e-05,
-      "logits/chosen": -4.25,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.0794,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 10.625,
-      "rewards/rejected": -19.625,
+      "logits/chosen": -9.5625,
+      "logits/rejected": -7.78125,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0744,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.71875,
+      "rewards/margins": 10.25,
+      "rewards/rejected": -18.0,
       "step": 1150
     },
     {
       "epoch": 1.2146596858638743,
-      "grad_norm": 1.8770780235182813,
+      "grad_norm": 2.4687719547221745,
       "learning_rate": 3.7137704336959464e-05,
-      "logits/chosen": -3.71875,
-      "logits/rejected": -2.796875,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0615,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 9.9375,
-      "rewards/rejected": -18.25,
+      "logits/chosen": -10.125,
+      "logits/rejected": -7.84375,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0515,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -6.75,
+      "rewards/margins": 10.0625,
+      "rewards/rejected": -16.875,
       "step": 1160
     },
     {
       "epoch": 1.225130890052356,
-      "grad_norm": 1.7849574178807202,
+      "grad_norm": 3.7305132651327266,
       "learning_rate": 3.687047133397201e-05,
-      "logits/chosen": -5.34375,
-      "logits/rejected": -3.71875,
-      "logps/chosen": -1004.0,
-      "logps/rejected": -2416.0,
-      "loss": 0.0549,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -6.25,
-      "rewards/margins": 10.125,
-      "rewards/rejected": -16.375,
+      "logits/chosen": -9.875,
+      "logits/rejected": -7.28125,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0651,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.09375,
+      "rewards/margins": 12.0625,
+      "rewards/rejected": -19.125,
       "step": 1170
     },
     {
       "epoch": 1.2356020942408377,
-      "grad_norm": 0.628673401888252,
+      "grad_norm": 1.084927985402219,
       "learning_rate": 3.6601475557503415e-05,
-      "logits/chosen": -5.78125,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -920.0,
+      "logits/chosen": -9.5,
+      "logits/rejected": -6.6875,
+      "logps/chosen": -976.0,
       "logps/rejected": -2352.0,
-      "loss": 0.0674,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -5.25,
-      "rewards/margins": 10.0625,
-      "rewards/rejected": -15.3125,
+      "loss": 0.0743,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -5.75,
+      "rewards/margins": 9.5625,
+      "rewards/rejected": -15.375,
       "step": 1180
     },
     {
       "epoch": 1.2460732984293195,
-      "grad_norm": 1.7915966337452056,
+      "grad_norm": 2.436507455374378,
       "learning_rate": 3.633075695361882e-05,
-      "logits/chosen": -3.25,
-      "logits/rejected": -2.296875,
-      "logps/chosen": -984.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0589,
+      "logits/chosen": -9.0625,
+      "logits/rejected": -6.75,
+      "logps/chosen": -912.0,
+      "logps/rejected": -2240.0,
+      "loss": 0.0694,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.09375,
-      "rewards/margins": 10.5625,
-      "rewards/rejected": -16.625,
+      "rewards/chosen": -5.34375,
+      "rewards/margins": 8.75,
+      "rewards/rejected": -14.0625,
       "step": 1190
     },
     {
       "epoch": 1.256544502617801,
-      "grad_norm": 0.9976838229178505,
+      "grad_norm": 0.8589614748038182,
       "learning_rate": 3.6058355724224477e-05,
-      "logits/chosen": -3.296875,
-      "logits/rejected": -2.3125,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.0518,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.9375,
-      "rewards/margins": 11.375,
-      "rewards/rejected": -18.375,
+      "logits/chosen": -9.1875,
+      "logits/rejected": -6.84375,
+      "logps/chosen": -1020.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.0464,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -6.3125,
+      "rewards/margins": 11.0,
+      "rewards/rejected": -17.375,
       "step": 1200
     },
     {
       "epoch": 1.256544502617801,
-      "eval_logits/chosen": -1.8515625,
-      "eval_logits/rejected": -0.271484375,
-      "eval_logps/chosen": -1144.0,
-      "eval_logps/rejected": -1240.0,
-      "eval_loss": 1.1557968854904175,
+      "eval_logits/chosen": -7.96875,
+      "eval_logits/rejected": -6.34375,
+      "eval_logps/chosen": -1152.0,
+      "eval_logps/rejected": -1248.0,
+      "eval_loss": 1.2448437213897705,
       "eval_rewards/accuracies": 0.6015625,
       "eval_rewards/chosen": -8.3125,
-      "eval_rewards/margins": 1.203125,
-      "eval_rewards/rejected": -9.5,
-      "eval_runtime": 46.6228,
-      "eval_samples_per_second": 42.897,
-      "eval_steps_per_second": 0.686,
+      "eval_rewards/margins": 1.2734375,
+      "eval_rewards/rejected": -9.5625,
+      "eval_runtime": 46.5859,
+      "eval_samples_per_second": 42.931,
+      "eval_steps_per_second": 0.687,
       "step": 1200
     },
     {
       "epoch": 1.2670157068062826,
-      "grad_norm": 2.049283323729441,
+      "grad_norm": 3.4049623834713,
       "learning_rate": 3.5784312321097744e-05,
-      "logits/chosen": -4.15625,
-      "logits/rejected": -2.890625,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0521,
+      "logits/chosen": -8.375,
+      "logits/rejected": -5.78125,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.068,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.625,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -18.125,
+      "rewards/chosen": -7.46875,
+      "rewards/margins": 12.0625,
+      "rewards/rejected": -19.5,
       "step": 1210
     },
     {
       "epoch": 1.2774869109947644,
-      "grad_norm": 1.2257077245239623,
+      "grad_norm": 1.0232609734216676,
       "learning_rate": 3.550866743987994e-05,
-      "logits/chosen": -4.9375,
-      "logits/rejected": -3.15625,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0566,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.3125,
-      "rewards/margins": 11.8125,
-      "rewards/rejected": -19.125,
+      "logits/chosen": -9.5,
+      "logits/rejected": -6.28125,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.0684,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -6.375,
+      "rewards/margins": 9.75,
+      "rewards/rejected": -16.125,
       "step": 1220
     },
     {
       "epoch": 1.2879581151832462,
-      "grad_norm": 1.2783697577342665,
+      "grad_norm": 1.0889713148807807,
       "learning_rate": 3.5231462014033024e-05,
-      "logits/chosen": -4.28125,
-      "logits/rejected": -3.375,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -2848.0,
-      "loss": 0.0375,
+      "logits/chosen": -8.3125,
+      "logits/rejected": -5.9375,
+      "logps/chosen": -948.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0344,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.75,
-      "rewards/margins": 13.125,
-      "rewards/rejected": -19.875,
+      "rewards/chosen": -5.75,
+      "rewards/margins": 10.125,
+      "rewards/rejected": -15.875,
       "step": 1230
     },
     {
       "epoch": 1.2984293193717278,
-      "grad_norm": 6.8690084882183315,
+      "grad_norm": 1.7169142178411654,
       "learning_rate": 3.495273720876095e-05,
-      "logits/chosen": -3.8125,
-      "logits/rejected": -3.078125,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0525,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.65625,
-      "rewards/margins": 10.75,
-      "rewards/rejected": -18.375,
+      "logits/chosen": -4.78125,
+      "logits/rejected": -3.28125,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -2576.0,
+      "loss": 0.04,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.9375,
+      "rewards/margins": 10.0625,
+      "rewards/rejected": -18.0,
       "step": 1240
     },
     {
       "epoch": 1.3089005235602094,
-      "grad_norm": 1.1979662996718428,
+      "grad_norm": 0.9011108162832775,
       "learning_rate": 3.4672534414896616e-05,
-      "logits/chosen": -5.6875,
-      "logits/rejected": -3.4375,
-      "logps/chosen": -1112.0,
+      "logits/chosen": -7.25,
+      "logits/rejected": -4.46875,
+      "logps/chosen": -1192.0,
       "logps/rejected": -2736.0,
-      "loss": 0.0601,
+      "loss": 0.0672,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.03125,
-      "rewards/margins": 12.0,
+      "rewards/chosen": -7.84375,
+      "rewards/margins": 11.25,
       "rewards/rejected": -19.0,
       "step": 1250
     },
     {
       "epoch": 1.3193717277486912,
-      "grad_norm": 2.6109733674067113,
+      "grad_norm": 5.8581541700435835,
       "learning_rate": 3.439089524275526e-05,
-      "logits/chosen": -5.1875,
-      "logits/rejected": -3.46875,
-      "logps/chosen": -948.0,
-      "logps/rejected": -2544.0,
-      "loss": 0.0743,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -5.78125,
-      "rewards/margins": 11.1875,
-      "rewards/rejected": -17.0,
+      "logits/chosen": -8.125,
+      "logits/rejected": -5.78125,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0698,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 11.125,
+      "rewards/rejected": -17.875,
       "step": 1260
     },
     {
       "epoch": 1.3298429319371727,
-      "grad_norm": 0.8802384159756188,
+      "grad_norm": 1.8164353562464701,
       "learning_rate": 3.410786151595533e-05,
-      "logits/chosen": -4.84375,
-      "logits/rejected": -2.984375,
-      "logps/chosen": -948.0,
-      "logps/rejected": -2304.0,
-      "loss": 0.0708,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -5.75,
-      "rewards/margins": 9.5,
-      "rewards/rejected": -15.1875,
+      "logits/chosen": -8.875,
+      "logits/rejected": -6.53125,
+      "logps/chosen": -1024.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0743,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -6.46875,
+      "rewards/margins": 10.125,
+      "rewards/rejected": -16.625,
       "step": 1270
     },
     {
       "epoch": 1.3403141361256545,
-      "grad_norm": 0.34044334661294434,
+      "grad_norm": 0.841060887058342,
       "learning_rate": 3.382347526520765e-05,
-      "logits/chosen": -4.5,
-      "logits/rejected": -2.71875,
-      "logps/chosen": -928.0,
-      "logps/rejected": -2400.0,
-      "loss": 0.0603,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -5.375,
-      "rewards/margins": 10.5,
-      "rewards/rejected": -15.875,
+      "logits/chosen": -10.375,
+      "logits/rejected": -7.71875,
+      "logps/chosen": -920.0,
+      "logps/rejected": -2304.0,
+      "loss": 0.0642,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -5.28125,
+      "rewards/margins": 9.625,
+      "rewards/rejected": -14.9375,
       "step": 1280
     },
     {
       "epoch": 1.350785340314136,
-      "grad_norm": 2.7041329234638596,
+      "grad_norm": 3.0437567297358714,
       "learning_rate": 3.353777872207381e-05,
-      "logits/chosen": -3.765625,
-      "logits/rejected": -2.6875,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0576,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.46875,
-      "rewards/margins": 11.75,
-      "rewards/rejected": -18.25,
+      "logits/chosen": -10.125,
+      "logits/rejected": -7.40625,
+      "logps/chosen": -988.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0654,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -5.96875,
+      "rewards/margins": 10.0625,
+      "rewards/rejected": -16.0,
       "step": 1290
     },
     {
       "epoch": 1.3612565445026177,
-      "grad_norm": 1.2778437203112551,
+      "grad_norm": 1.7464317779987542,
       "learning_rate": 3.325081431269475e-05,
-      "logits/chosen": -2.1875,
-      "logits/rejected": -1.1328125,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0627,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.84375,
-      "rewards/margins": 12.4375,
-      "rewards/rejected": -19.25,
+      "logits/chosen": -9.5,
+      "logits/rejected": -6.8125,
+      "logps/chosen": -984.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.0455,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -6.25,
+      "rewards/margins": 10.125,
+      "rewards/rejected": -16.375,
       "step": 1300
     },
     {
       "epoch": 1.3612565445026177,
-      "eval_logits/chosen": -2.453125,
-      "eval_logits/rejected": -0.94140625,
-      "eval_logps/chosen": -1120.0,
+      "eval_logits/chosen": -10.0625,
+      "eval_logits/rejected": -8.5,
+      "eval_logps/chosen": -1136.0,
       "eval_logps/rejected": -1232.0,
-      "eval_loss": 1.2759687900543213,
-      "eval_rewards/accuracies": 0.591796875,
-      "eval_rewards/chosen": -8.0625,
-      "eval_rewards/margins": 1.3671875,
+      "eval_loss": 1.236828088760376,
+      "eval_rewards/accuracies": 0.5859375,
+      "eval_rewards/chosen": -8.1875,
+      "eval_rewards/margins": 1.25,
       "eval_rewards/rejected": -9.4375,
-      "eval_runtime": 46.6136,
-      "eval_samples_per_second": 42.906,
+      "eval_runtime": 46.6296,
+      "eval_samples_per_second": 42.891,
       "eval_steps_per_second": 0.686,
       "step": 1300
     },
     {
       "epoch": 1.3717277486910995,
-      "grad_norm": 1.7316435668351948,
+      "grad_norm": 2.766342619732496,
       "learning_rate": 3.296262465149046e-05,
-      "logits/chosen": -4.59375,
-      "logits/rejected": -2.96875,
+      "logits/chosen": -11.0625,
+      "logits/rejected": -7.75,
       "logps/chosen": -1048.0,
-      "logps/rejected": -2816.0,
-      "loss": 0.0606,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -6.46875,
-      "rewards/margins": 12.8125,
-      "rewards/rejected": -19.25,
+      "logps/rejected": -2672.0,
+      "loss": 0.049,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -6.4375,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -17.875,
       "step": 1310
     },
     {
       "epoch": 1.3821989528795813,
-      "grad_norm": 0.7592957805468828,
+      "grad_norm": 0.6044969857270757,
       "learning_rate": 3.2673252534831687e-05,
-      "logits/chosen": -4.25,
-      "logits/rejected": -2.484375,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -2544.0,
-      "loss": 0.0517,
+      "logits/chosen": -9.125,
+      "logits/rejected": -6.28125,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0424,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.5,
-      "rewards/margins": 11.5,
-      "rewards/rejected": -18.0,
+      "rewards/chosen": -7.0,
+      "rewards/margins": 11.6875,
+      "rewards/rejected": -18.625,
       "step": 1320
     },
     {
       "epoch": 1.3926701570680629,
-      "grad_norm": 2.2333545618443296,
+      "grad_norm": 1.4103616083861292,
       "learning_rate": 3.2382740934684696e-05,
-      "logits/chosen": -2.4375,
-      "logits/rejected": -1.6484375,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.103,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.8125,
-      "rewards/margins": 10.75,
-      "rewards/rejected": -17.5,
+      "logits/chosen": -7.125,
+      "logits/rejected": -5.0625,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0916,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.53125,
+      "rewards/margins": 11.0625,
+      "rewards/rejected": -18.625,
       "step": 1330
     },
     {
       "epoch": 1.4031413612565444,
-      "grad_norm": 0.9907823349526859,
+      "grad_norm": 1.6786342156766705,
       "learning_rate": 3.209113299222982e-05,
-      "logits/chosen": -0.94140625,
-      "logits/rejected": -1.0390625,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0606,
+      "logits/chosen": -6.53125,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.0628,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.15625,
-      "rewards/margins": 9.9375,
-      "rewards/rejected": -17.125,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 11.3125,
+      "rewards/rejected": -18.875,
       "step": 1340
     },
     {
       "epoch": 1.4136125654450262,
-      "grad_norm": 1.1484308352772061,
+      "grad_norm": 2.5064463597510613,
       "learning_rate": 3.179847201145505e-05,
-      "logits/chosen": -2.875,
-      "logits/rejected": -1.1953125,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -2512.0,
-      "loss": 0.0585,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.25,
-      "rewards/margins": 10.25,
-      "rewards/rejected": -16.5,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -5.59375,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -2576.0,
+      "loss": 0.0691,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 10.3125,
+      "rewards/rejected": -17.125,
       "step": 1350
     },
     {
       "epoch": 1.4240837696335078,
-      "grad_norm": 1.061818844234854,
+      "grad_norm": 3.0971403940241897,
       "learning_rate": 3.150480145272528e-05,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -1.6796875,
+      "logits/chosen": -9.8125,
+      "logits/rejected": -6.0,
       "logps/chosen": -1088.0,
-      "logps/rejected": -2400.0,
-      "loss": 0.0529,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.9375,
-      "rewards/margins": 9.5,
-      "rewards/rejected": -16.375,
+      "logps/rejected": -2368.0,
+      "loss": 0.0621,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -6.875,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -16.0,
       "step": 1360
     },
     {
       "epoch": 1.4345549738219896,
-      "grad_norm": 1.1519231535293004,
+      "grad_norm": 1.1618091783528295,
       "learning_rate": 3.121016492632848e-05,
-      "logits/chosen": -3.75,
-      "logits/rejected": -2.046875,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0745,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.96875,
-      "rewards/margins": 11.25,
-      "rewards/rejected": -18.25,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -6.125,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0481,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 10.375,
+      "rewards/rejected": -17.625,
       "step": 1370
     },
     {
       "epoch": 1.4450261780104712,
-      "grad_norm": 1.2966080282758348,
+      "grad_norm": 1.7790637191676755,
       "learning_rate": 3.091460618599951e-05,
-      "logits/chosen": -2.6875,
-      "logits/rejected": -1.515625,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0429,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.15625,
-      "rewards/margins": 11.25,
-      "rewards/rejected": -18.375,
-      "step": 1380
-    },
+      "logits/chosen": -9.25,
+      "logits/rejected": -5.9375,
+      "logps/chosen": -1080.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.061,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 10.0625,
+      "rewards/rejected": -16.875,
+      "step": 1380
+    },
     {
       "epoch": 1.4554973821989527,
-      "grad_norm": 0.6534081902638929,
+      "grad_norm": 3.34587047631057,
       "learning_rate": 3.061816912242265e-05,
-      "logits/chosen": -2.296875,
-      "logits/rejected": -0.71875,
-      "logps/chosen": -1152.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.0528,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.5,
-      "rewards/margins": 12.125,
-      "rewards/rejected": -19.625,
+      "logits/chosen": -10.5,
+      "logits/rejected": -6.96875,
+      "logps/chosen": -1004.0,
+      "logps/rejected": -2304.0,
+      "loss": 0.0654,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -6.0625,
+      "rewards/margins": 9.25,
+      "rewards/rejected": -15.3125,
       "step": 1390
     },
     {
       "epoch": 1.4659685863874345,
-      "grad_norm": 0.3432173943911499,
+      "grad_norm": 0.4199748826732826,
       "learning_rate": 3.0320897756713785e-05,
-      "logits/chosen": -2.078125,
-      "logits/rejected": -1.15625,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -2976.0,
-      "loss": 0.067,
+      "logits/chosen": -9.625,
+      "logits/rejected": -7.375,
+      "logps/chosen": -1024.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0614,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.5,
-      "rewards/margins": 12.75,
-      "rewards/rejected": -19.25,
+      "rewards/chosen": -6.25,
+      "rewards/margins": 11.1875,
+      "rewards/rejected": -17.375,
       "step": 1400
     },
     {
       "epoch": 1.4659685863874345,
-      "eval_logits/chosen": -0.43359375,
-      "eval_logits/rejected": 1.1875,
-      "eval_logps/chosen": -1080.0,
-      "eval_logps/rejected": -1184.0,
-      "eval_loss": 1.1144062280654907,
-      "eval_rewards/accuracies": 0.611328125,
-      "eval_rewards/chosen": -7.625,
-      "eval_rewards/margins": 1.3515625,
-      "eval_rewards/rejected": -9.0,
-      "eval_runtime": 46.6149,
-      "eval_samples_per_second": 42.905,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -9.375,
+      "eval_logits/rejected": -7.875,
+      "eval_logps/chosen": -1120.0,
+      "eval_logps/rejected": -1216.0,
+      "eval_loss": 1.2304843664169312,
+      "eval_rewards/accuracies": 0.599609375,
+      "eval_rewards/chosen": -8.0,
+      "eval_rewards/margins": 1.2578125,
+      "eval_rewards/rejected": -9.3125,
+      "eval_runtime": 46.613,
+      "eval_samples_per_second": 42.907,
+      "eval_steps_per_second": 0.687,
       "step": 1400
     },
     {
       "epoch": 1.4764397905759163,
-      "grad_norm": 7.636536164805229,
+      "grad_norm": 1.4807719388534133,
       "learning_rate": 3.002283623388332e-05,
-      "logits/chosen": -1.796875,
-      "logits/rejected": -0.53125,
-      "logps/chosen": -1012.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0585,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.53125,
-      "rewards/margins": 10.8125,
-      "rewards/rejected": -17.375,
+      "logits/chosen": -9.9375,
+      "logits/rejected": -7.1875,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -2576.0,
+      "loss": 0.0584,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -7.3125,
+      "rewards/margins": 10.9375,
+      "rewards/rejected": -18.25,
       "step": 1410
     },
     {
       "epoch": 1.486910994764398,
-      "grad_norm": 3.2729847436812145,
+      "grad_norm": 2.074451914761993,
       "learning_rate": 2.9724028816280507e-05,
-      "logits/chosen": -2.40625,
-      "logits/rejected": -0.76171875,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0494,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.9375,
-      "rewards/margins": 11.8125,
-      "rewards/rejected": -18.75,
+      "logits/chosen": -10.625,
+      "logits/rejected": -7.28125,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0499,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.375,
+      "rewards/margins": 11.9375,
+      "rewards/rejected": -19.25,
       "step": 1420
     },
     {
       "epoch": 1.4973821989528795,
-      "grad_norm": 1.4631253724517745,
+      "grad_norm": 0.4198785413974353,
       "learning_rate": 2.9424519877020524e-05,
-      "logits/chosen": -4.34375,
-      "logits/rejected": -1.8046875,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.0491,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -6.71875,
-      "rewards/margins": 14.9375,
-      "rewards/rejected": -21.625,
+      "logits/chosen": -12.625,
+      "logits/rejected": -9.25,
+      "logps/chosen": -1020.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.0657,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -5.96875,
+      "rewards/margins": 13.0,
+      "rewards/rejected": -19.0,
       "step": 1430
     },
     {
       "epoch": 1.5078534031413613,
-      "grad_norm": 0.3904054682401217,
+      "grad_norm": 0.4454534168949265,
       "learning_rate": 2.912435389339504e-05,
-      "logits/chosen": -3.125,
-      "logits/rejected": -0.328125,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0485,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.5625,
-      "rewards/margins": 13.1875,
-      "rewards/rejected": -19.75,
+      "logits/chosen": -12.125,
+      "logits/rejected": -7.96875,
+      "logps/chosen": -988.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0498,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -6.0625,
+      "rewards/margins": 11.6875,
+      "rewards/rejected": -17.75,
       "step": 1440
     },
     {
       "epoch": 1.518324607329843,
-      "grad_norm": 1.5916925034702496,
+      "grad_norm": 1.7116400585113292,
       "learning_rate": 2.8823575440267226e-05,
-      "logits/chosen": -2.875,
-      "logits/rejected": -2.125,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0529,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.84375,
-      "rewards/margins": 11.6875,
-      "rewards/rejected": -18.5,
+      "logits/chosen": -10.875,
+      "logits/rejected": -7.90625,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0572,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.21875,
+      "rewards/margins": 10.75,
+      "rewards/rejected": -18.0,
       "step": 1450
     },
     {
       "epoch": 1.5287958115183247,
-      "grad_norm": 0.8752828757385837,
+      "grad_norm": 0.6559486647957037,
       "learning_rate": 2.8522229183452465e-05,
-      "logits/chosen": -3.171875,
-      "logits/rejected": -1.2734375,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0918,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -6.8125,
-      "rewards/margins": 11.375,
-      "rewards/rejected": -18.125,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -7.9375,
+      "logps/chosen": -1064.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.0821,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -6.5625,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -16.375,
       "step": 1460
     },
     {
       "epoch": 1.5392670157068062,
-      "grad_norm": 1.9474563693873936,
+      "grad_norm": 1.761828277619678,
       "learning_rate": 2.8220359873085373e-05,
-      "logits/chosen": -2.875,
-      "logits/rejected": -1.3203125,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.053,
+      "logits/chosen": -9.4375,
+      "logits/rejected": -6.5,
+      "logps/chosen": -1032.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0615,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.84375,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -18.25,
+      "rewards/chosen": -6.3125,
+      "rewards/margins": 10.0,
+      "rewards/rejected": -16.375,
       "step": 1470
     },
     {
       "epoch": 1.5497382198952878,
-      "grad_norm": 1.3098604504961686,
+      "grad_norm": 1.163077107914401,
       "learning_rate": 2.7918012336974388e-05,
-      "logits/chosen": -2.109375,
-      "logits/rejected": -1.6953125,
-      "logps/chosen": -984.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0585,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.03125,
-      "rewards/margins": 11.6875,
-      "rewards/rejected": -17.75,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -8.5,
+      "logps/chosen": -972.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.1619,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -5.90625,
+      "rewards/margins": 10.125,
+      "rewards/rejected": -16.0,
       "step": 1480
     },
     {
       "epoch": 1.5602094240837696,
-      "grad_norm": 0.8023320906608231,
+      "grad_norm": 1.369328165726986,
       "learning_rate": 2.7615231473944815e-05,
-      "logits/chosen": -1.6484375,
-      "logits/rejected": -0.251953125,
-      "logps/chosen": -1008.0,
-      "logps/rejected": -2384.0,
-      "loss": 0.0577,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -6.3125,
-      "rewards/margins": 10.1875,
-      "rewards/rejected": -16.5,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -9.75,
+      "logps/chosen": -956.0,
+      "logps/rejected": -2240.0,
+      "loss": 0.0479,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -5.8125,
+      "rewards/margins": 9.25,
+      "rewards/rejected": -15.0625,
       "step": 1490
     },
     {
       "epoch": 1.5706806282722514,
-      "grad_norm": 1.2073362083247476,
+      "grad_norm": 1.288888724297925,
       "learning_rate": 2.7312062247171327e-05,
-      "logits/chosen": -2.15625,
-      "logits/rejected": -1.1875,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.057,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.21875,
-      "rewards/margins": 12.0625,
-      "rewards/rejected": -19.25,
+      "logits/chosen": -13.0,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -992.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0573,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -6.3125,
+      "rewards/margins": 10.625,
+      "rewards/rejected": -17.0,
       "step": 1500
     },
     {
       "epoch": 1.5706806282722514,
-      "eval_logits/chosen": -0.026611328125,
-      "eval_logits/rejected": 1.4453125,
-      "eval_logps/chosen": -1200.0,
-      "eval_logps/rejected": -1312.0,
-      "eval_loss": 1.23842191696167,
-      "eval_rewards/accuracies": 0.578125,
-      "eval_rewards/chosen": -8.8125,
-      "eval_rewards/margins": 1.4453125,
-      "eval_rewards/rejected": -10.25,
-      "eval_runtime": 46.616,
-      "eval_samples_per_second": 42.904,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -12.375,
+      "eval_logits/rejected": -11.0625,
+      "eval_logps/chosen": -1104.0,
+      "eval_logps/rejected": -1184.0,
+      "eval_loss": 1.217046856880188,
+      "eval_rewards/accuracies": 0.572265625,
+      "eval_rewards/chosen": -7.8125,
+      "eval_rewards/margins": 1.1640625,
+      "eval_rewards/rejected": -9.0,
+      "eval_runtime": 46.6082,
+      "eval_samples_per_second": 42.911,
+      "eval_steps_per_second": 0.687,
       "step": 1500
     },
     {
       "epoch": 1.581151832460733,
-      "grad_norm": 0.5234266411993196,
+      "grad_norm": 0.6434094923068135,
       "learning_rate": 2.7008549677500876e-05,
-      "logits/chosen": -2.921875,
-      "logits/rejected": -0.99609375,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -2880.0,
-      "loss": 0.0652,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.53125,
-      "rewards/margins": 13.0,
-      "rewards/rejected": -20.5,
+      "logits/chosen": -12.3125,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0628,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -7.15625,
+      "rewards/margins": 10.5,
+      "rewards/rejected": -17.75,
       "step": 1510
     },
     {
       "epoch": 1.5916230366492146,
-      "grad_norm": 0.40511633202977493,
+      "grad_norm": 0.4728421952780738,
       "learning_rate": 2.670473883676709e-05,
-      "logits/chosen": -2.140625,
-      "logits/rejected": -0.8515625,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.0504,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.125,
-      "rewards/margins": 12.4375,
-      "rewards/rejected": -19.5,
+      "logits/chosen": -11.0,
+      "logits/rejected": -7.375,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0588,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 10.4375,
+      "rewards/rejected": -17.25,
       "step": 1520
     },
     {
       "epoch": 1.6020942408376964,
-      "grad_norm": 1.214533818826857,
+      "grad_norm": 1.8936808173044952,
       "learning_rate": 2.6400674841097074e-05,
-      "logits/chosen": -3.109375,
-      "logits/rejected": -1.5390625,
-      "logps/chosen": -1152.0,
-      "logps/rejected": -3072.0,
-      "loss": 0.039,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.46875,
-      "rewards/margins": 13.8125,
-      "rewards/rejected": -21.25,
+      "logits/chosen": -9.6875,
+      "logits/rejected": -6.25,
+      "logps/chosen": -1020.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.0467,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -6.125,
+      "rewards/margins": 10.0625,
+      "rewards/rejected": -16.125,
       "step": 1530
     },
     {
       "epoch": 1.6125654450261782,
-      "grad_norm": 2.8719465691480783,
+      "grad_norm": 1.096355938526378,
       "learning_rate": 2.609640284421158e-05,
-      "logits/chosen": -3.09375,
-      "logits/rejected": -0.984375,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.063,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 14.4375,
-      "rewards/rejected": -22.5,
+      "logits/chosen": -9.8125,
+      "logits/rejected": -5.53125,
+      "logps/chosen": -996.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0568,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -5.9375,
+      "rewards/margins": 11.25,
+      "rewards/rejected": -17.25,
       "step": 1540
     },
     {
       "epoch": 1.6230366492146597,
-      "grad_norm": 0.3877520690652679,
+      "grad_norm": 0.7607456182475383,
       "learning_rate": 2.5791968030719726e-05,
-      "logits/chosen": -3.0625,
-      "logits/rejected": -1.3125,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0465,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -6.625,
-      "rewards/margins": 11.3125,
-      "rewards/rejected": -17.875,
+      "logits/chosen": -10.625,
+      "logits/rejected": -7.125,
+      "logps/chosen": -1012.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.0519,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -5.9375,
+      "rewards/margins": 10.1875,
+      "rewards/rejected": -16.125,
       "step": 1550
     },
     {
       "epoch": 1.6335078534031413,
-      "grad_norm": 2.326679393513896,
+      "grad_norm": 1.5979255107403063,
       "learning_rate": 2.548741560940902e-05,
-      "logits/chosen": -2.890625,
-      "logits/rejected": -1.4140625,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.0591,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.53125,
-      "rewards/margins": 11.9375,
-      "rewards/rejected": -18.5,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -8.375,
+      "logps/chosen": -992.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0533,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -6.09375,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -17.625,
       "step": 1560
     },
     {
       "epoch": 1.6439790575916229,
-      "grad_norm": 1.0820043163274844,
+      "grad_norm": 1.1899919000601975,
       "learning_rate": 2.5182790806531787e-05,
-      "logits/chosen": -2.515625,
-      "logits/rejected": -1.5,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0679,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -6.59375,
-      "rewards/margins": 11.5625,
-      "rewards/rejected": -18.125,
+      "logits/chosen": -12.375,
+      "logits/rejected": -9.625,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0795,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.34375,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -17.75,
       "step": 1570
     },
     {
       "epoch": 1.6544502617801047,
-      "grad_norm": 0.5282337856296464,
+      "grad_norm": 1.4097926726421348,
       "learning_rate": 2.4878138859089074e-05,
-      "logits/chosen": -2.40625,
-      "logits/rejected": -0.58984375,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0472,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -10.125,
+      "logps/chosen": -988.0,
+      "logps/rejected": -2384.0,
+      "loss": 0.0518,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.78125,
-      "rewards/margins": 10.8125,
-      "rewards/rejected": -17.625,
+      "rewards/chosen": -5.9375,
+      "rewards/margins": 10.5,
+      "rewards/rejected": -16.375,
       "step": 1580
     },
     {
       "epoch": 1.6649214659685865,
-      "grad_norm": 2.089003017045723,
+      "grad_norm": 1.6584712255862577,
       "learning_rate": 2.4573505008112924e-05,
-      "logits/chosen": -2.484375,
-      "logits/rejected": -0.427734375,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0458,
+      "logits/chosen": -11.75,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0547,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.4375,
-      "rewards/margins": 10.5625,
-      "rewards/rejected": -18.0,
+      "rewards/chosen": -7.125,
+      "rewards/margins": 10.75,
+      "rewards/rejected": -17.875,
       "step": 1590
     },
     {
       "epoch": 1.675392670157068,
-      "grad_norm": 1.0365161984996114,
+      "grad_norm": 0.7106220768599325,
       "learning_rate": 2.4268934491948032e-05,
-      "logits/chosen": -2.546875,
-      "logits/rejected": -1.0390625,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -2448.0,
-      "loss": 0.0549,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.9375,
-      "rewards/margins": 9.875,
-      "rewards/rejected": -16.75,
+      "logits/chosen": -9.875,
+      "logits/rejected": -6.78125,
+      "logps/chosen": -1160.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0561,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 10.6875,
+      "rewards/rejected": -18.25,
       "step": 1600
     },
     {
       "epoch": 1.675392670157068,
-      "eval_logits/chosen": -0.046630859375,
-      "eval_logits/rejected": 1.4921875,
-      "eval_logps/chosen": -1104.0,
-      "eval_logps/rejected": -1208.0,
-      "eval_loss": 1.103906273841858,
-      "eval_rewards/accuracies": 0.6015625,
-      "eval_rewards/chosen": -7.875,
-      "eval_rewards/margins": 1.3046875,
-      "eval_rewards/rejected": -9.1875,
-      "eval_runtime": 46.6227,
-      "eval_samples_per_second": 42.898,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -8.75,
+      "eval_logits/rejected": -6.90625,
+      "eval_logps/chosen": -1240.0,
+      "eval_logps/rejected": -1336.0,
+      "eval_loss": 1.17954683303833,
+      "eval_rewards/accuracies": 0.609375,
+      "eval_rewards/chosen": -9.1875,
+      "eval_rewards/margins": 1.265625,
+      "eval_rewards/rejected": -10.5,
+      "eval_runtime": 46.6012,
+      "eval_samples_per_second": 42.917,
+      "eval_steps_per_second": 0.687,
       "step": 1600
     },
     {
       "epoch": 1.6858638743455496,
-      "grad_norm": 1.8284734995452323,
+      "grad_norm": 1.3129214235173163,
       "learning_rate": 2.396447253953385e-05,
-      "logits/chosen": -3.40625,
-      "logits/rejected": -1.953125,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0392,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.71875,
-      "rewards/margins": 11.5,
-      "rewards/rejected": -18.25,
+      "logits/chosen": -10.5,
+      "logits/rejected": -6.9375,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.0485,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.0,
+      "rewards/margins": 11.75,
+      "rewards/rejected": -18.75,
       "step": 1610
     },
     {
       "epoch": 1.6963350785340314,
-      "grad_norm": 2.2162918981496844,
+      "grad_norm": 1.535951012755932,
       "learning_rate": 2.3660164363687998e-05,
-      "logits/chosen": -2.265625,
-      "logits/rejected": -0.875,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -2976.0,
-      "loss": 0.057,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.9375,
-      "rewards/margins": 11.9375,
-      "rewards/rejected": -19.875,
+      "logits/chosen": -9.875,
+      "logits/rejected": -6.21875,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0512,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -6.875,
+      "rewards/margins": 11.1875,
+      "rewards/rejected": -18.0,
       "step": 1620
     },
     {
       "epoch": 1.7068062827225132,
-      "grad_norm": 1.4470096330344102,
+      "grad_norm": 1.7742187383840784,
       "learning_rate": 2.33560551543922e-05,
-      "logits/chosen": -2.953125,
-      "logits/rejected": -1.640625,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -2512.0,
-      "loss": 0.0438,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -5.65625,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.0416,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.09375,
-      "rewards/margins": 10.625,
-      "rewards/rejected": -17.75,
+      "rewards/chosen": -6.96875,
+      "rewards/margins": 10.375,
+      "rewards/rejected": -17.375,
       "step": 1630
     },
     {
       "epoch": 1.7172774869109948,
-      "grad_norm": 0.7559415078675955,
+      "grad_norm": 0.7177587645715898,
       "learning_rate": 2.3052190072081493e-05,
-      "logits/chosen": -2.4375,
-      "logits/rejected": -0.96875,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0683,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.78125,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -19.125,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -5.65625,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.0436,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -7.5625,
+      "rewards/margins": 12.75,
+      "rewards/rejected": -20.25,
       "step": 1640
     },
     {
       "epoch": 1.7277486910994764,
-      "grad_norm": 0.7402912564828471,
+      "grad_norm": 0.9644507296710271,
       "learning_rate": 2.2748614240937864e-05,
-      "logits/chosen": -1.859375,
-      "logits/rejected": -0.765625,
-      "logps/chosen": -964.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.0478,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -5.9375,
-      "rewards/margins": 12.4375,
-      "rewards/rejected": -18.375,
+      "logits/chosen": -7.75,
+      "logits/rejected": -4.40625,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -2992.0,
+      "loss": 0.0606,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 13.375,
+      "rewards/rejected": -20.625,
       "step": 1650
     },
     {
       "epoch": 1.738219895287958,
-      "grad_norm": 0.6918685496114385,
+      "grad_norm": 0.9995270770969238,
       "learning_rate": 2.2445372742189334e-05,
-      "logits/chosen": -2.40625,
-      "logits/rejected": -1.1640625,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -2848.0,
-      "loss": 0.0493,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.28125,
-      "rewards/margins": 12.5625,
-      "rewards/rejected": -18.875,
+      "logits/chosen": -7.6875,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.0462,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -6.96875,
+      "rewards/margins": 12.9375,
+      "rewards/rejected": -19.875,
       "step": 1660
     },
     {
       "epoch": 1.7486910994764397,
-      "grad_norm": 2.875754840014667,
+      "grad_norm": 1.28111481267349,
       "learning_rate": 2.214251060741528e-05,
-      "logits/chosen": -2.515625,
-      "logits/rejected": -0.9296875,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -2816.0,
-      "loss": 0.057,
+      "logits/chosen": -9.1875,
+      "logits/rejected": -5.75,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -2576.0,
+      "loss": 0.0503,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.125,
-      "rewards/margins": 12.625,
-      "rewards/rejected": -19.75,
+      "rewards/chosen": -6.4375,
+      "rewards/margins": 10.8125,
+      "rewards/rejected": -17.25,
       "step": 1670
     },
     {
       "epoch": 1.7591623036649215,
-      "grad_norm": 1.5732696410460978,
+      "grad_norm": 2.0406873749415277,
       "learning_rate": 2.1840072811859302e-05,
-      "logits/chosen": -2.359375,
-      "logits/rejected": -1.34375,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -2816.0,
-      "loss": 0.0466,
+      "logits/chosen": -9.1875,
+      "logits/rejected": -6.40625,
+      "logps/chosen": -1024.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0602,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.625,
-      "rewards/margins": 11.625,
-      "rewards/rejected": -18.25,
+      "rewards/chosen": -6.15625,
+      "rewards/margins": 10.875,
+      "rewards/rejected": -17.0,
       "step": 1680
     },
     {
       "epoch": 1.7696335078534031,
-      "grad_norm": 0.9904017418624957,
+      "grad_norm": 1.0654978021241128,
       "learning_rate": 2.1538104267750287e-05,
-      "logits/chosen": -1.609375,
-      "logits/rejected": -1.1328125,
-      "logps/chosen": -1004.0,
-      "logps/rejected": -2544.0,
-      "loss": 0.0471,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -6.34375,
-      "rewards/margins": 10.5625,
-      "rewards/rejected": -16.875,
+      "logits/chosen": -8.375,
+      "logits/rejected": -6.21875,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0395,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -6.65625,
+      "rewards/margins": 10.75,
+      "rewards/rejected": -17.375,
       "step": 1690
     },
     {
       "epoch": 1.7801047120418847,
-      "grad_norm": 1.2971262546850524,
+      "grad_norm": 1.2862285865685201,
       "learning_rate": 2.1236649817632952e-05,
-      "logits/chosen": -1.21875,
-      "logits/rejected": -0.75,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.065,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.8125,
-      "rewards/margins": 11.3125,
-      "rewards/rejected": -18.125,
+      "logits/chosen": -8.875,
+      "logits/rejected": -6.46875,
+      "logps/chosen": -1080.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.0735,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -19.875,
       "step": 1700
     },
     {
       "epoch": 1.7801047120418847,
-      "eval_logits/chosen": -0.0018157958984375,
-      "eval_logits/rejected": 1.5390625,
-      "eval_logps/chosen": -1136.0,
-      "eval_logps/rejected": -1272.0,
-      "eval_loss": 1.212499976158142,
+      "eval_logits/chosen": -8.8125,
+      "eval_logits/rejected": -7.09375,
+      "eval_logps/chosen": -1224.0,
+      "eval_logps/rejected": -1352.0,
+      "eval_loss": 1.3455936908721924,
       "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -8.1875,
-      "eval_rewards/margins": 1.6015625,
-      "eval_rewards/rejected": -9.8125,
-      "eval_runtime": 46.6229,
-      "eval_samples_per_second": 42.897,
-      "eval_steps_per_second": 0.686,
+      "eval_rewards/chosen": -9.0625,
+      "eval_rewards/margins": 1.5859375,
+      "eval_rewards/rejected": -10.625,
+      "eval_runtime": 46.6003,
+      "eval_samples_per_second": 42.918,
+      "eval_steps_per_second": 0.687,
       "step": 1700
     },
     {
       "epoch": 1.7905759162303665,
-      "grad_norm": 0.9202076211650413,
+      "grad_norm": 2.0663901288558573,
       "learning_rate": 2.093575422770872e-05,
-      "logits/chosen": -1.1328125,
-      "logits/rejected": -0.03125,
-      "logps/chosen": -1004.0,
-      "logps/rejected": -2400.0,
-      "loss": 0.043,
+      "logits/chosen": -9.625,
+      "logits/rejected": -6.34375,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0463,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.40625,
-      "rewards/margins": 10.375,
-      "rewards/rejected": -16.75,
+      "rewards/chosen": -6.9375,
+      "rewards/margins": 11.0625,
+      "rewards/rejected": -18.0,
       "step": 1710
     },
     {
       "epoch": 1.8010471204188483,
-      "grad_norm": 4.402270604716534,
+      "grad_norm": 0.6245438569510241,
       "learning_rate": 2.0635462181187827e-05,
-      "logits/chosen": -2.796875,
-      "logits/rejected": -1.625,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -2992.0,
-      "loss": 0.056,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.90625,
-      "rewards/margins": 13.375,
-      "rewards/rejected": -20.25,
+      "logits/chosen": -11.0625,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -1080.0,
+      "logps/rejected": -2976.0,
+      "loss": 0.0559,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 13.3125,
+      "rewards/rejected": -20.125,
       "step": 1720
     },
-    {
-      "epoch": 1.8115183246073299,
-      "grad_norm": 1.461185659614161,
-      "learning_rate": 2.0335818271653917e-05,
-      "logits/chosen": -2.140625,
-      "logits/rejected": -1.03125,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.058,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -6.53125,
-      "rewards/margins": 12.0,
-      "rewards/rejected": -18.5,
+    {
+      "epoch": 1.8115183246073299,
+      "grad_norm": 1.1654328508335727,
+      "learning_rate": 2.0335818271653917e-05,
+      "logits/chosen": -10.625,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -1000.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0536,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -5.96875,
+      "rewards/margins": 11.3125,
+      "rewards/rejected": -17.25,
       "step": 1730
     },
     {
       "epoch": 1.8219895287958114,
-      "grad_norm": 0.8613325312714369,
+      "grad_norm": 1.045223972173717,
       "learning_rate": 2.0036866996441815e-05,
-      "logits/chosen": -1.1484375,
-      "logits/rejected": 0.021728515625,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -2880.0,
-      "loss": 0.0291,
+      "logits/chosen": -10.8125,
+      "logits/rejected": -6.75,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0386,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 12.6875,
-      "rewards/rejected": -20.375,
+      "rewards/chosen": -6.40625,
+      "rewards/margins": 11.8125,
+      "rewards/rejected": -18.125,
       "step": 1740
     },
     {
       "epoch": 1.8324607329842932,
-      "grad_norm": 2.4697234715161454,
+      "grad_norm": 0.7328327341918704,
       "learning_rate": 1.9738652750029677e-05,
-      "logits/chosen": -1.328125,
-      "logits/rejected": 0.10400390625,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.0403,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.875,
-      "rewards/margins": 12.8125,
-      "rewards/rejected": -20.75,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -6.09375,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0388,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.3125,
+      "rewards/margins": 12.4375,
+      "rewards/rejected": -19.75,
       "step": 1750
     },
     {
       "epoch": 1.8429319371727748,
-      "grad_norm": 3.770933475705704,
+      "grad_norm": 2.658228238252288,
       "learning_rate": 1.944121981744631e-05,
-      "logits/chosen": -2.53125,
-      "logits/rejected": -0.27734375,
-      "logps/chosen": -1216.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0463,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.65625,
-      "rewards/margins": 12.5,
-      "rewards/rejected": -20.125,
+      "logits/chosen": -9.75,
+      "logits/rejected": -5.84375,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.0453,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -21.125,
       "step": 1760
     },
     {
       "epoch": 1.8534031413612566,
-      "grad_norm": 1.247200149561182,
+      "grad_norm": 1.3248844128535708,
       "learning_rate": 1.9144612367694887e-05,
-      "logits/chosen": -2.421875,
-      "logits/rejected": -0.87890625,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0422,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.75,
-      "rewards/margins": 11.8125,
-      "rewards/rejected": -18.625,
+      "logits/chosen": -9.375,
+      "logits/rejected": -6.09375,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.0403,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -7.65625,
+      "rewards/margins": 12.125,
+      "rewards/rejected": -19.75,
       "step": 1770
     },
     {
       "epoch": 1.8638743455497382,
-      "grad_norm": 0.6643836340344415,
+      "grad_norm": 0.6237436976652655,
       "learning_rate": 1.8848874447193803e-05,
-      "logits/chosen": -1.9765625,
-      "logits/rejected": -0.41015625,
-      "logps/chosen": -1008.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.047,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.3125,
-      "rewards/margins": 10.625,
-      "rewards/rejected": -17.0,
+      "logits/chosen": -9.375,
+      "logits/rejected": -6.21875,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.0511,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 11.3125,
+      "rewards/rejected": -18.625,
       "step": 1780
     },
     {
       "epoch": 1.8743455497382198,
-      "grad_norm": 0.7465320733051719,
+      "grad_norm": 1.5787673649037564,
       "learning_rate": 1.855404997323571e-05,
-      "logits/chosen": -2.609375,
-      "logits/rejected": -1.484375,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0307,
+      "logits/chosen": -10.375,
+      "logits/rejected": -7.125,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0365,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.03125,
-      "rewards/margins": 12.125,
-      "rewards/rejected": -19.125,
+      "rewards/chosen": -7.21875,
+      "rewards/margins": 11.625,
+      "rewards/rejected": -18.875,
       "step": 1790
     },
     {
       "epoch": 1.8848167539267016,
-      "grad_norm": 1.4489706454435407,
+      "grad_norm": 1.5703097455816861,
       "learning_rate": 1.82601827274658e-05,
-      "logits/chosen": -1.5859375,
-      "logits/rejected": -0.8828125,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0477,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.21875,
-      "rewards/margins": 11.3125,
-      "rewards/rejected": -18.5,
+      "logits/chosen": -9.75,
+      "logits/rejected": -6.90625,
+      "logps/chosen": -1080.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0646,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.09375,
+      "rewards/margins": 10.75,
+      "rewards/rejected": -17.875,
       "step": 1800
     },
     {
       "epoch": 1.8848167539267016,
-      "eval_logits/chosen": 0.55078125,
-      "eval_logits/rejected": 2.046875,
-      "eval_logps/chosen": -1160.0,
-      "eval_logps/rejected": -1288.0,
-      "eval_loss": 1.2241874933242798,
-      "eval_rewards/accuracies": 0.603515625,
-      "eval_rewards/chosen": -8.4375,
-      "eval_rewards/margins": 1.546875,
-      "eval_rewards/rejected": -10.0,
-      "eval_runtime": 46.6281,
-      "eval_samples_per_second": 42.893,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -7.875,
+      "eval_logits/rejected": -6.34375,
+      "eval_logps/chosen": -1144.0,
+      "eval_logps/rejected": -1248.0,
+      "eval_loss": 1.1507188081741333,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -8.25,
+      "eval_rewards/margins": 1.34375,
+      "eval_rewards/rejected": -9.625,
+      "eval_runtime": 46.6037,
+      "eval_samples_per_second": 42.915,
+      "eval_steps_per_second": 0.687,
       "step": 1800
     },
     {
       "epoch": 1.8952879581151834,
-      "grad_norm": 3.995058535055145,
+      "grad_norm": 3.1947574871879185,
       "learning_rate": 1.7967316349380213e-05,
-      "logits/chosen": -1.4375,
-      "logits/rejected": -0.4375,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.049,
+      "logits/chosen": -8.625,
+      "logits/rejected": -6.21875,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0556,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.03125,
-      "rewards/margins": 11.6875,
-      "rewards/rejected": -18.75,
+      "rewards/chosen": -7.09375,
+      "rewards/margins": 10.5625,
+      "rewards/rejected": -17.625,
       "step": 1810
     },
     {
       "epoch": 1.905759162303665,
-      "grad_norm": 0.6222829554230952,
+      "grad_norm": 1.4464928611882828,
       "learning_rate": 1.7675494329845516e-05,
-      "logits/chosen": -1.1953125,
-      "logits/rejected": -0.1298828125,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0253,
+      "logits/chosen": -8.375,
+      "logits/rejected": -5.96875,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0312,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.125,
-      "rewards/margins": 11.625,
-      "rewards/rejected": -18.75,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 10.5,
+      "rewards/rejected": -17.75,
       "step": 1820
     },
     {
       "epoch": 1.9162303664921465,
-      "grad_norm": 0.2993435161426848,
+      "grad_norm": 0.3104040938044856,
       "learning_rate": 1.738476000464026e-05,
-      "logits/chosen": -0.58203125,
-      "logits/rejected": 0.1845703125,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.057,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -6.96875,
+      "logits/chosen": -8.0625,
+      "logits/rejected": -5.40625,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0394,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.34375,
       "rewards/margins": 12.0625,
-      "rewards/rejected": -19.0,
+      "rewards/rejected": -19.375,
       "step": 1830
     },
     {
       "epoch": 1.9267015706806283,
-      "grad_norm": 2.240978076045901,
+      "grad_norm": 1.156828951510642,
       "learning_rate": 1.7095156548019646e-05,
-      "logits/chosen": -1.6171875,
-      "logits/rejected": -0.6640625,
-      "logps/chosen": -1024.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0539,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.53125,
-      "rewards/margins": 11.625,
-      "rewards/rejected": -18.125,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -6.4375,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.05,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -6.75,
+      "rewards/margins": 11.125,
+      "rewards/rejected": -17.875,
       "step": 1840
     },
     {
       "epoch": 1.93717277486911,
-      "grad_norm": 1.6812515877884868,
+      "grad_norm": 0.9397629325279463,
       "learning_rate": 1.6806726966304064e-05,
-      "logits/chosen": -1.9140625,
-      "logits/rejected": 0.10009765625,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0484,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -6.125,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.0486,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.875,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -19.125,
+      "rewards/chosen": -6.90625,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -18.25,
       "step": 1850
     },
     {
       "epoch": 1.9476439790575917,
-      "grad_norm": 0.9500942126190716,
+      "grad_norm": 1.6562507277902645,
       "learning_rate": 1.6519514091492625e-05,
-      "logits/chosen": -1.5703125,
-      "logits/rejected": -0.68359375,
+      "logits/chosen": -9.4375,
+      "logits/rejected": -6.53125,
       "logps/chosen": -1064.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.035,
-      "rewards/accuracies": 0.9750000238418579,
+      "logps/rejected": -2576.0,
+      "loss": 0.04,
+      "rewards/accuracies": 0.981249988079071,
       "rewards/chosen": -7.0,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -18.5,
+      "rewards/margins": 11.25,
+      "rewards/rejected": -18.25,
       "step": 1860
     },
     {
       "epoch": 1.9581151832460733,
-      "grad_norm": 1.4518396251728989,
+      "grad_norm": 2.2687784818468457,
       "learning_rate": 1.6233560574902627e-05,
-      "logits/chosen": -1.2109375,
-      "logits/rejected": -0.2001953125,
-      "logps/chosen": -1024.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0726,
+      "logits/chosen": -9.4375,
+      "logits/rejected": -6.375,
+      "logps/chosen": -1008.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.0711,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.78125,
-      "rewards/margins": 12.4375,
-      "rewards/rejected": -19.25,
+      "rewards/chosen": -6.59375,
+      "rewards/margins": 11.875,
+      "rewards/rejected": -18.5,
       "step": 1870
     },
     {
       "epoch": 1.9685863874345548,
-      "grad_norm": 1.0601077231803497,
+      "grad_norm": 0.9616997966462891,
       "learning_rate": 1.5948908880835753e-05,
-      "logits/chosen": -1.1171875,
-      "logits/rejected": 0.4453125,
-      "logps/chosen": -1016.0,
+      "logits/chosen": -10.0,
+      "logits/rejected": -6.40625,
+      "logps/chosen": -1032.0,
       "logps/rejected": -2272.0,
-      "loss": 0.0463,
+      "loss": 0.0412,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.28125,
-      "rewards/margins": 10.25,
+      "rewards/chosen": -6.5,
+      "rewards/margins": 10.0625,
       "rewards/rejected": -16.5,
       "step": 1880
     },
     {
       "epoch": 1.9790575916230366,
-      "grad_norm": 0.687596452086469,
+      "grad_norm": 0.250839414492072,
       "learning_rate": 1.5665601280272124e-05,
-      "logits/chosen": -1.84375,
-      "logits/rejected": 0.146484375,
-      "logps/chosen": -1008.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0488,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.25,
-      "rewards/margins": 11.5,
+      "logits/chosen": -10.0,
+      "logits/rejected": -6.84375,
+      "logps/chosen": -1024.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0381,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -6.40625,
+      "rewards/margins": 11.25,
       "rewards/rejected": -17.75,
       "step": 1890
     },
     {
       "epoch": 1.9895287958115184,
-      "grad_norm": 2.2229011721601655,
+      "grad_norm": 1.4387060810597163,
       "learning_rate": 1.538367984459298e-05,
-      "logits/chosen": -1.109375,
-      "logits/rejected": -0.484375,
-      "logps/chosen": -1000.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0232,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.375,
-      "rewards/margins": 11.5625,
-      "rewards/rejected": -18.0,
+      "logits/chosen": -9.375,
+      "logits/rejected": -6.4375,
+      "logps/chosen": -1024.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0256,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -6.625,
+      "rewards/margins": 11.625,
+      "rewards/rejected": -18.25,
       "step": 1900
     },
     {
       "epoch": 1.9895287958115184,
-      "eval_logits/chosen": 0.41796875,
-      "eval_logits/rejected": 1.9296875,
-      "eval_logps/chosen": -1128.0,
-      "eval_logps/rejected": -1256.0,
-      "eval_loss": 1.1593905687332153,
-      "eval_rewards/accuracies": 0.615234375,
-      "eval_rewards/chosen": -8.125,
-      "eval_rewards/margins": 1.59375,
-      "eval_rewards/rejected": -9.6875,
-      "eval_runtime": 46.6333,
-      "eval_samples_per_second": 42.888,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -8.5,
+      "eval_logits/rejected": -6.90625,
+      "eval_logps/chosen": -1184.0,
+      "eval_logps/rejected": -1304.0,
+      "eval_loss": 1.2246718406677246,
+      "eval_rewards/accuracies": 0.59765625,
+      "eval_rewards/chosen": -8.625,
+      "eval_rewards/margins": 1.4921875,
+      "eval_rewards/rejected": -10.125,
+      "eval_runtime": 46.6018,
+      "eval_samples_per_second": 42.917,
+      "eval_steps_per_second": 0.687,
       "step": 1900
     },
     {
       "epoch": 2.0,
-      "grad_norm": 1.5618971886372357,
+      "grad_norm": 0.6912634059116118,
       "learning_rate": 1.5103186439333133e-05,
-      "logits/chosen": -0.56640625,
-      "logits/rejected": 0.3125,
-      "logps/chosen": -1048.0,
+      "logits/chosen": -9.125,
+      "logits/rejected": -5.8125,
+      "logps/chosen": -1056.0,
       "logps/rejected": -2736.0,
-      "loss": 0.0779,
+      "loss": 0.06,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.71875,
-      "rewards/margins": 12.4375,
-      "rewards/rejected": -19.125,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 12.1875,
+      "rewards/rejected": -19.0,
       "step": 1910
     },
     {
       "epoch": 2.0104712041884816,
-      "grad_norm": 0.46511859107465076,
+      "grad_norm": 0.26432028752253783,
       "learning_rate": 1.4824162717963829e-05,
-      "logits/chosen": -1.5078125,
-      "logits/rejected": 0.58203125,
-      "logps/chosen": -1016.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.008,
+      "logits/chosen": -9.1875,
+      "logits/rejected": -5.1875,
+      "logps/chosen": -1064.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.0082,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -6.3125,
-      "rewards/margins": 12.375,
-      "rewards/rejected": -18.625,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 12.4375,
+      "rewards/rejected": -19.25,
       "step": 1920
     },
     {
       "epoch": 2.020942408376963,
-      "grad_norm": 0.12155853048961496,
+      "grad_norm": 0.031229439814562852,
       "learning_rate": 1.4546650115707192e-05,
-      "logits/chosen": -1.4609375,
-      "logits/rejected": -0.53515625,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -3024.0,
-      "loss": 0.0064,
+      "logits/chosen": -8.625,
+      "logits/rejected": -5.125,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.0052,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -6.96875,
-      "rewards/margins": 14.25,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -7.21875,
+      "rewards/margins": 15.25,
+      "rewards/rejected": -22.5,
       "step": 1930
     },
     {
       "epoch": 2.031413612565445,
-      "grad_norm": 0.16902628329398756,
+      "grad_norm": 0.12039724059248401,
       "learning_rate": 1.4270689843383111e-05,
-      "logits/chosen": -1.625,
-      "logits/rejected": -0.75,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -2944.0,
-      "loss": 0.01,
+      "logits/chosen": -8.125,
+      "logits/rejected": -5.09375,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -3008.0,
+      "loss": 0.0112,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -6.875,
-      "rewards/margins": 14.0625,
-      "rewards/rejected": -20.875,
+      "rewards/chosen": -7.125,
+      "rewards/margins": 14.5,
+      "rewards/rejected": -21.625,
       "step": 1940
     },
     {
       "epoch": 2.0418848167539267,
-      "grad_norm": 0.19359889705674835,
+      "grad_norm": 0.06177957270344228,
       "learning_rate": 1.3996322881289347e-05,
-      "logits/chosen": -2.21875,
-      "logits/rejected": -0.02392578125,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.0104,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -2928.0,
+      "loss": 0.0127,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.25,
-      "rewards/margins": 13.25,
-      "rewards/rejected": -20.5,
+      "rewards/chosen": -7.625,
+      "rewards/margins": 13.875,
+      "rewards/rejected": -21.5,
       "step": 1950
     },
     {
       "epoch": 2.0523560209424083,
-      "grad_norm": 0.19057618012643068,
+      "grad_norm": 0.16831676713737106,
       "learning_rate": 1.3723589973115963e-05,
-      "logits/chosen": -1.171875,
-      "logits/rejected": -0.52734375,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -2976.0,
-      "loss": 0.0052,
+      "logits/chosen": -9.0,
+      "logits/rejected": -5.6875,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.0067,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.15625,
-      "rewards/margins": 14.375,
-      "rewards/rejected": -21.5,
+      "rewards/chosen": -7.5625,
+      "rewards/margins": 14.5625,
+      "rewards/rejected": -22.125,
       "step": 1960
     },
     {
       "epoch": 2.06282722513089,
-      "grad_norm": 1.3037147193317367,
+      "grad_norm": 0.15697895980153734,
       "learning_rate": 1.3452531619894837e-05,
-      "logits/chosen": -1.3125,
-      "logits/rejected": 0.0908203125,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -3056.0,
-      "loss": 0.0074,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -5.15625,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.0067,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.1875,
-      "rewards/margins": 14.5625,
-      "rewards/rejected": -21.75,
+      "rewards/chosen": -7.34375,
+      "rewards/margins": 15.3125,
+      "rewards/rejected": -22.625,
       "step": 1970
     },
     {
       "epoch": 2.073298429319372,
-      "grad_norm": 0.10615792601074006,
+      "grad_norm": 0.2111192197901564,
       "learning_rate": 1.3183188073985245e-05,
-      "logits/chosen": -1.734375,
-      "logits/rejected": -0.2119140625,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -3104.0,
-      "loss": 0.0088,
+      "logits/chosen": -8.875,
+      "logits/rejected": -5.0,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.0073,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.46875,
-      "rewards/margins": 14.9375,
-      "rewards/rejected": -22.375,
+      "rewards/chosen": -7.6875,
+      "rewards/margins": 15.0625,
+      "rewards/rejected": -22.75,
       "step": 1980
     },
     {
       "epoch": 2.0837696335078535,
-      "grad_norm": 0.10237947798484894,
+      "grad_norm": 0.09564817029414235,
       "learning_rate": 1.2915599333096351e-05,
-      "logits/chosen": -2.21875,
-      "logits/rejected": -0.78125,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -2976.0,
-      "loss": 0.0076,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.59375,
-      "rewards/margins": 14.5,
-      "rewards/rejected": -22.0,
+      "logits/chosen": -9.5,
+      "logits/rejected": -5.84375,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -3008.0,
+      "loss": 0.0082,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -7.6875,
+      "rewards/margins": 14.75,
+      "rewards/rejected": -22.5,
       "step": 1990
     },
     {
       "epoch": 2.094240837696335,
-      "grad_norm": 0.09847848295760914,
+      "grad_norm": 0.03985411600367326,
       "learning_rate": 1.264980513434752e-05,
-      "logits/chosen": -1.9375,
-      "logits/rejected": 0.035400390625,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0025,
+      "logits/chosen": -8.625,
+      "logits/rejected": -4.875,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.34375,
-      "rewards/margins": 15.875,
-      "rewards/rejected": -23.25,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 17.25,
+      "rewards/rejected": -24.75,
       "step": 2000
     },
     {
       "epoch": 2.094240837696335,
-      "eval_logits/chosen": 0.56640625,
-      "eval_logits/rejected": 2.09375,
-      "eval_logps/chosen": -1232.0,
-      "eval_logps/rejected": -1392.0,
-      "eval_loss": 1.2468827962875366,
-      "eval_rewards/accuracies": 0.603515625,
-      "eval_rewards/chosen": -9.1875,
-      "eval_rewards/margins": 1.84375,
-      "eval_rewards/rejected": -11.0,
-      "eval_runtime": 46.6294,
-      "eval_samples_per_second": 42.891,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -7.3125,
+      "eval_logits/rejected": -5.46875,
+      "eval_logps/chosen": -1312.0,
+      "eval_logps/rejected": -1456.0,
+      "eval_loss": 1.4274219274520874,
+      "eval_rewards/accuracies": 0.595703125,
+      "eval_rewards/chosen": -9.9375,
+      "eval_rewards/margins": 1.75,
+      "eval_rewards/rejected": -11.6875,
+      "eval_runtime": 46.6125,
+      "eval_samples_per_second": 42.907,
+      "eval_steps_per_second": 0.687,
       "step": 2000
     },
     {
       "epoch": 2.1047120418848166,
-      "grad_norm": 0.3205380744657223,
+      "grad_norm": 0.11186365184774691,
       "learning_rate": 1.2385844948367324e-05,
-      "logits/chosen": -0.93359375,
-      "logits/rejected": -0.1669921875,
-      "logps/chosen": -1120.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.0072,
+      "logits/chosen": -7.90625,
+      "logits/rejected": -4.84375,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -3056.0,
+      "loss": 0.0069,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 13.75,
-      "rewards/rejected": -21.375,
+      "rewards/chosen": -8.125,
+      "rewards/margins": 14.875,
+      "rewards/rejected": -23.0,
       "step": 2010
     },
     {
       "epoch": 2.115183246073298,
-      "grad_norm": 0.06949735502180925,
+      "grad_norm": 0.10219217293169225,
       "learning_rate": 1.2123757973432114e-05,
-      "logits/chosen": -1.8359375,
-      "logits/rejected": 0.08203125,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -2992.0,
-      "loss": 0.0068,
+      "logits/chosen": -8.25,
+      "logits/rejected": -4.46875,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -3120.0,
+      "loss": 0.0096,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.875,
-      "rewards/margins": 14.1875,
-      "rewards/rejected": -22.125,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 15.0,
+      "rewards/rejected": -23.25,
       "step": 2020
     },
     {
       "epoch": 2.1256544502617802,
-      "grad_norm": 0.18365978545496986,
+      "grad_norm": 0.5361201603547617,
       "learning_rate": 1.1863583129645106e-05,
-      "logits/chosen": -1.7578125,
-      "logits/rejected": -0.62890625,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.0187,
+      "logits/chosen": -8.25,
+      "logits/rejected": -4.65625,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0126,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.90625,
-      "rewards/margins": 15.4375,
-      "rewards/rejected": -23.375,
+      "rewards/chosen": -8.5625,
+      "rewards/margins": 15.9375,
+      "rewards/rejected": -24.5,
       "step": 2030
     },
     {
       "epoch": 2.136125654450262,
-      "grad_norm": 0.9038835692276321,
+      "grad_norm": 1.5807754444566973,
       "learning_rate": 1.1605359053156607e-05,
-      "logits/chosen": -1.8671875,
-      "logits/rejected": -0.2041015625,
-      "logps/chosen": -1152.0,
-      "logps/rejected": -3136.0,
-      "loss": 0.008,
+      "logits/chosen": -7.6875,
+      "logits/rejected": -4.09375,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -3328.0,
+      "loss": 0.0075,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.46875,
-      "rewards/margins": 15.5625,
-      "rewards/rejected": -23.0,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -24.875,
       "step": 2040
     },
     {
       "epoch": 2.1465968586387434,
-      "grad_norm": 0.15257220691317425,
+      "grad_norm": 0.13657445035187418,
       "learning_rate": 1.1349124090426644e-05,
-      "logits/chosen": -1.5390625,
-      "logits/rejected": 0.05810546875,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -3024.0,
-      "loss": 0.008,
+      "logits/chosen": -7.40625,
+      "logits/rejected": -4.25,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.011,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.34375,
-      "rewards/margins": 14.6875,
-      "rewards/rejected": -22.0,
+      "rewards/chosen": -7.96875,
+      "rewards/margins": 15.9375,
+      "rewards/rejected": -23.875,
       "step": 2050
     },
     {
       "epoch": 2.157068062827225,
-      "grad_norm": 0.03157020896990601,
+      "grad_norm": 0.09917222465378253,
       "learning_rate": 1.1094916292530406e-05,
-      "logits/chosen": -0.8671875,
-      "logits/rejected": 0.248046875,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -3008.0,
-      "loss": 0.0079,
+      "logits/chosen": -7.5625,
+      "logits/rejected": -4.3125,
+      "logps/chosen": -1264.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.0083,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.71875,
-      "rewards/margins": 14.125,
-      "rewards/rejected": -21.875,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 14.875,
+      "rewards/rejected": -23.375,
       "step": 2060
     },
     {
       "epoch": 2.167539267015707,
-      "grad_norm": 1.1629995990490773,
+      "grad_norm": 0.1486045066404541,
       "learning_rate": 1.0842773409507623e-05,
-      "logits/chosen": -0.890625,
-      "logits/rejected": 0.154296875,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -3008.0,
-      "loss": 0.0094,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.5,
-      "rewards/margins": 14.75,
-      "rewards/rejected": -22.25,
+      "logits/chosen": -7.84375,
+      "logits/rejected": -4.25,
+      "logps/chosen": -1160.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.011,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -8.0,
+      "rewards/margins": 15.8125,
+      "rewards/rejected": -23.75,
       "step": 2070
     },
     {
       "epoch": 2.1780104712041886,
-      "grad_norm": 0.04191315826034855,
+      "grad_norm": 0.6242097445378998,
       "learning_rate": 1.0592732884756754e-05,
-      "logits/chosen": -1.7578125,
-      "logits/rejected": -0.35546875,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -3072.0,
-      "loss": 0.006,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -4.875,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.0068,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.78125,
-      "rewards/margins": 14.75,
-      "rewards/rejected": -22.5,
+      "rewards/chosen": -7.84375,
+      "rewards/margins": 15.4375,
+      "rewards/rejected": -23.25,
       "step": 2080
     },
     {
       "epoch": 2.18848167539267,
-      "grad_norm": 0.06909945461863397,
+      "grad_norm": 0.11165960413934461,
       "learning_rate": 1.0344831849474507e-05,
-      "logits/chosen": -1.1328125,
-      "logits/rejected": 0.671875,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.0078,
+      "logits/chosen": -8.625,
+      "logits/rejected": -4.3125,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -3264.0,
+      "loss": 0.008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.5625,
-      "rewards/margins": 15.6875,
-      "rewards/rejected": -24.25,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -24.625,
       "step": 2090
     },
     {
       "epoch": 2.1989528795811517,
-      "grad_norm": 0.22937476503632495,
+      "grad_norm": 0.4368344855572855,
       "learning_rate": 1.0099107117141879e-05,
-      "logits/chosen": -0.73828125,
-      "logits/rejected": 0.46875,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.0064,
+      "logits/chosen": -8.6875,
+      "logits/rejected": -4.875,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0035,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 16.25,
-      "rewards/rejected": -24.625,
+      "rewards/chosen": -8.0,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -24.5,
       "step": 2100
     },
     {
       "epoch": 2.1989528795811517,
-      "eval_logits/chosen": 0.83203125,
-      "eval_logits/rejected": 2.328125,
-      "eval_logps/chosen": -1336.0,
-      "eval_logps/rejected": -1504.0,
-      "eval_loss": 1.3712031841278076,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -10.1875,
-      "eval_rewards/margins": 1.984375,
-      "eval_rewards/rejected": -12.1875,
-      "eval_runtime": 46.6255,
-      "eval_samples_per_second": 42.895,
+      "eval_logits/chosen": -7.5625,
+      "eval_logits/rejected": -5.625,
+      "eval_logps/chosen": -1328.0,
+      "eval_logps/rejected": -1488.0,
+      "eval_loss": 1.512109398841858,
+      "eval_rewards/accuracies": 0.6015625,
+      "eval_rewards/chosen": -10.125,
+      "eval_rewards/margins": 1.84375,
+      "eval_rewards/rejected": -12.0,
+      "eval_runtime": 46.615,
+      "eval_samples_per_second": 42.905,
       "eval_steps_per_second": 0.686,
       "step": 2100
     },
     {
       "epoch": 2.2094240837696333,
-      "grad_norm": 0.2401066099778482,
+      "grad_norm": 0.21650482671260762,
       "learning_rate": 9.855595178057334e-06,
-      "logits/chosen": -1.21875,
-      "logits/rejected": 0.11279296875,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -3600.0,
-      "loss": 0.0039,
+      "logits/chosen": -7.78125,
+      "logits/rejected": -4.375,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0043,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 17.75,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -8.5625,
+      "rewards/margins": 18.375,
+      "rewards/rejected": -26.875,
       "step": 2110
     },
     {
       "epoch": 2.2198952879581153,
-      "grad_norm": 0.03064895673055832,
+      "grad_norm": 0.05469422243421969,
       "learning_rate": 9.614332193917949e-06,
-      "logits/chosen": -0.94140625,
-      "logits/rejected": -0.1865234375,
-      "logps/chosen": -1360.0,
+      "logits/chosen": -8.25,
+      "logits/rejected": -4.28125,
+      "logps/chosen": -1296.0,
       "logps/rejected": -3696.0,
-      "loss": 0.0134,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.4375,
-      "rewards/margins": 18.0,
+      "loss": 0.0099,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -8.875,
+      "rewards/margins": 18.625,
       "rewards/rejected": -27.5,
       "step": 2120
     },
     {
       "epoch": 2.230366492146597,
-      "grad_norm": 1.9217835733149424,
+      "grad_norm": 0.20712838629585972,
       "learning_rate": 9.375353992449385e-06,
-      "logits/chosen": -1.15625,
-      "logits/rejected": -0.7109375,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -3392.0,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -5.21875,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -3312.0,
       "loss": 0.0057,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.125,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -25.625,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -24.875,
       "step": 2130
     },
     {
       "epoch": 2.2408376963350785,
-      "grad_norm": 0.09716414462954667,
+      "grad_norm": 0.14068457268147141,
       "learning_rate": 9.138696062085442e-06,
-      "logits/chosen": -1.484375,
-      "logits/rejected": -0.169921875,
-      "logps/chosen": -1216.0,
+      "logits/chosen": -8.5625,
+      "logits/rejected": -4.6875,
+      "logps/chosen": -1200.0,
       "logps/rejected": -3120.0,
-      "loss": 0.0133,
+      "loss": 0.0106,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 15.625,
+      "rewards/chosen": -8.125,
+      "rewards/margins": 15.75,
       "rewards/rejected": -23.875,
       "step": 2140
     },
     {
       "epoch": 2.25130890052356,
-      "grad_norm": 0.07289620412915172,
+      "grad_norm": 0.042052645147495724,
       "learning_rate": 8.904393546698007e-06,
-      "logits/chosen": -1.796875,
-      "logits/rejected": -0.1103515625,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -3392.0,
-      "loss": 0.0096,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1224.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0118,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 16.625,
-      "rewards/rejected": -25.5,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -24.25,
       "step": 2150
     },
     {
       "epoch": 2.261780104712042,
-      "grad_norm": 0.2787280454307116,
+      "grad_norm": 0.24459410495305786,
       "learning_rate": 8.672481240378142e-06,
-      "logits/chosen": -1.2421875,
-      "logits/rejected": 0.78515625,
-      "logps/chosen": -1152.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.0136,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.9375,
-      "rewards/margins": 15.9375,
-      "rewards/rejected": -23.875,
+      "logits/chosen": -9.1875,
+      "logits/rejected": -4.625,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -3056.0,
+      "loss": 0.0144,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -7.1875,
+      "rewards/margins": 15.75,
+      "rewards/rejected": -22.875,
       "step": 2160
     },
     {
       "epoch": 2.2722513089005236,
-      "grad_norm": 1.1955058320617347,
+      "grad_norm": 0.13309618093961817,
       "learning_rate": 8.44299358226919e-06,
-      "logits/chosen": -1.3046875,
-      "logits/rejected": -0.0986328125,
+      "logits/chosen": -8.625,
+      "logits/rejected": -5.09375,
       "logps/chosen": -1144.0,
-      "logps/rejected": -3312.0,
-      "loss": 0.008,
-      "rewards/accuracies": 0.9937499761581421,
+      "logps/rejected": -3280.0,
+      "loss": 0.0087,
+      "rewards/accuracies": 1.0,
       "rewards/chosen": -7.71875,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -24.125,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -23.875,
       "step": 2170
     },
     {
       "epoch": 2.282722513089005,
-      "grad_norm": 0.11678887902387922,
+      "grad_norm": 0.05993128322396106,
       "learning_rate": 8.215964651452457e-06,
-      "logits/chosen": -1.1484375,
-      "logits/rejected": 0.2314453125,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -4.5,
       "logps/chosen": -1144.0,
-      "logps/rejected": -3328.0,
-      "loss": 0.0076,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 17.0,
-      "rewards/rejected": -24.625,
+      "logps/rejected": -3296.0,
+      "loss": 0.0072,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.5625,
+      "rewards/margins": 16.625,
+      "rewards/rejected": -24.125,
       "step": 2180
     },
     {
       "epoch": 2.2931937172774868,
-      "grad_norm": 0.057091894748249356,
+      "grad_norm": 0.0674464308808311,
       "learning_rate": 7.991428161886502e-06,
-      "logits/chosen": -1.4765625,
-      "logits/rejected": -0.56640625,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -3088.0,
-      "loss": 0.0052,
+      "logits/chosen": -8.3125,
+      "logits/rejected": -4.96875,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -3136.0,
+      "loss": 0.0054,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.84375,
-      "rewards/margins": 15.25,
-      "rewards/rejected": -23.125,
+      "rewards/chosen": -8.0,
+      "rewards/margins": 15.5625,
+      "rewards/rejected": -23.625,
       "step": 2190
     },
     {
       "epoch": 2.303664921465969,
-      "grad_norm": 0.050141965090892866,
+      "grad_norm": 0.1955699644825176,
       "learning_rate": 7.76941745740061e-06,
-      "logits/chosen": -1.296875,
-      "logits/rejected": -0.296875,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -3280.0,
-      "loss": 0.0068,
+      "logits/chosen": -8.1875,
+      "logits/rejected": -4.71875,
+      "logps/chosen": -1280.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.0073,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.4375,
-      "rewards/margins": 15.8125,
-      "rewards/rejected": -24.25,
+      "rewards/chosen": -8.75,
+      "rewards/margins": 15.6875,
+      "rewards/rejected": -24.5,
       "step": 2200
     },
     {
       "epoch": 2.303664921465969,
-      "eval_logits/chosen": 0.6328125,
-      "eval_logits/rejected": 2.109375,
-      "eval_logps/chosen": -1280.0,
-      "eval_logps/rejected": -1432.0,
-      "eval_loss": 1.2939062118530273,
-      "eval_rewards/accuracies": 0.609375,
-      "eval_rewards/chosen": -9.5625,
-      "eval_rewards/margins": 1.8359375,
-      "eval_rewards/rejected": -11.4375,
-      "eval_runtime": 46.6289,
-      "eval_samples_per_second": 42.892,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -7.25,
+      "eval_logits/rejected": -5.40625,
+      "eval_logps/chosen": -1344.0,
+      "eval_logps/rejected": -1496.0,
+      "eval_loss": 1.4838438034057617,
+      "eval_rewards/accuracies": 0.599609375,
+      "eval_rewards/chosen": -10.25,
+      "eval_rewards/margins": 1.8125,
+      "eval_rewards/rejected": -12.0625,
+      "eval_runtime": 46.6089,
+      "eval_samples_per_second": 42.91,
+      "eval_steps_per_second": 0.687,
       "step": 2200
     },
     {
       "epoch": 2.3141361256544504,
-      "grad_norm": 0.18044026854613746,
+      "grad_norm": 0.7410086131460206,
       "learning_rate": 7.549965506743123e-06,
-      "logits/chosen": -0.87890625,
-      "logits/rejected": 0.041748046875,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -3072.0,
-      "loss": 0.0042,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.4375,
-      "rewards/margins": 14.875,
-      "rewards/rejected": -23.25,
+      "logits/chosen": -7.875,
+      "logits/rejected": -4.8125,
+      "logps/chosen": -1264.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0067,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -8.8125,
+      "rewards/margins": 15.8125,
+      "rewards/rejected": -24.625,
       "step": 2210
     },
     {
       "epoch": 2.324607329842932,
-      "grad_norm": 0.3324428300597743,
+      "grad_norm": 0.16685875442252432,
       "learning_rate": 7.333104898685631e-06,
-      "logits/chosen": -1.7421875,
-      "logits/rejected": -0.466796875,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -3104.0,
-      "loss": 0.0091,
+      "logits/chosen": -7.90625,
+      "logits/rejected": -4.25,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0075,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 15.1875,
-      "rewards/rejected": -23.375,
+      "rewards/chosen": -8.625,
+      "rewards/margins": 15.8125,
+      "rewards/rejected": -24.375,
       "step": 2220
     },
     {
       "epoch": 2.3350785340314135,
-      "grad_norm": 0.3609491450836267,
+      "grad_norm": 0.13288534802125307,
       "learning_rate": 7.118867837183518e-06,
-      "logits/chosen": -2.671875,
-      "logits/rejected": -0.828125,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -3312.0,
-      "loss": 0.0057,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.90625,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -24.5,
+      "logits/chosen": -8.25,
+      "logits/rejected": -4.40625,
+      "logps/chosen": -1288.0,
+      "logps/rejected": -3440.0,
+      "loss": 0.0069,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 17.25,
+      "rewards/rejected": -25.75,
       "step": 2230
     },
     {
       "epoch": 2.345549738219895,
-      "grad_norm": 0.1435608571114668,
+      "grad_norm": 0.043988513442951176,
       "learning_rate": 6.9072861365936045e-06,
-      "logits/chosen": -2.40625,
-      "logits/rejected": -0.8984375,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -3328.0,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -4.96875,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -3520.0,
       "loss": 0.0048,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.0,
-      "rewards/margins": 15.6875,
-      "rewards/rejected": -23.625,
+      "rewards/chosen": -8.9375,
+      "rewards/margins": 16.625,
+      "rewards/rejected": -25.625,
       "step": 2240
     },
     {
       "epoch": 2.356020942408377,
-      "grad_norm": 0.11555069559886802,
+      "grad_norm": 0.09769902807842423,
       "learning_rate": 6.6983912169497005e-06,
-      "logits/chosen": -2.0,
-      "logits/rejected": -0.5078125,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.0055,
+      "logits/chosen": -7.75,
+      "logits/rejected": -4.125,
+      "logps/chosen": -1296.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 15.6875,
-      "rewards/rejected": -23.75,
+      "rewards/chosen": -8.625,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -25.125,
       "step": 2250
     },
     {
       "epoch": 2.3664921465968587,
-      "grad_norm": 0.08657070282327808,
+      "grad_norm": 0.1405892789994459,
       "learning_rate": 6.49221409929677e-06,
-      "logits/chosen": -0.265625,
-      "logits/rejected": 0.427734375,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.0063,
+      "logits/chosen": -7.4375,
+      "logits/rejected": -4.5,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.4375,
-      "rewards/margins": 15.5625,
-      "rewards/rejected": -24.0,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -24.75,
       "step": 2260
     },
     {
       "epoch": 2.3769633507853403,
-      "grad_norm": 0.23221610321507827,
+      "grad_norm": 0.12515873808181377,
       "learning_rate": 6.2887854010842075e-06,
-      "logits/chosen": -0.90625,
-      "logits/rejected": 0.2734375,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -3280.0,
+      "logits/chosen": -7.40625,
+      "logits/rejected": -4.0,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -3296.0,
       "loss": 0.0074,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -8.75,
-      "rewards/margins": 16.125,
-      "rewards/rejected": -24.875,
+      "rewards/chosen": -8.875,
+      "rewards/margins": 16.25,
+      "rewards/rejected": -25.0,
       "step": 2270
     },
     {
       "epoch": 2.387434554973822,
-      "grad_norm": 0.12400593901405332,
+      "grad_norm": 0.0801432267890652,
       "learning_rate": 6.0881353316191385e-06,
-      "logits/chosen": -1.828125,
-      "logits/rejected": -0.0986328125,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -3136.0,
-      "loss": 0.0078,
+      "logits/chosen": -8.0,
+      "logits/rejected": -4.34375,
+      "logps/chosen": -1264.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0076,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 15.6875,
-      "rewards/rejected": -24.0,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 16.25,
+      "rewards/rejected": -24.875,
       "step": 2280
     },
     {
       "epoch": 2.3979057591623034,
-      "grad_norm": 0.17047164961101463,
+      "grad_norm": 0.7777665123815218,
       "learning_rate": 5.890293687580381e-06,
-      "logits/chosen": -1.265625,
-      "logits/rejected": 0.1953125,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -3440.0,
-      "loss": 0.0042,
+      "logits/chosen": -7.0,
+      "logits/rejected": -3.78125,
+      "logps/chosen": -1328.0,
+      "logps/rejected": -3488.0,
+      "loss": 0.0069,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 17.125,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -9.25,
+      "rewards/margins": 17.375,
+      "rewards/rejected": -26.75,
       "step": 2290
     },
     {
       "epoch": 2.4083769633507854,
-      "grad_norm": 0.053385421022568796,
+      "grad_norm": 0.31202559396526053,
       "learning_rate": 5.695289848593532e-06,
-      "logits/chosen": -2.21875,
-      "logits/rejected": -0.81640625,
+      "logits/chosen": -7.71875,
+      "logits/rejected": -4.46875,
       "logps/chosen": -1216.0,
       "logps/rejected": -3392.0,
-      "loss": 0.0106,
+      "loss": 0.0081,
       "rewards/accuracies": 0.9750000238418579,
       "rewards/chosen": -8.4375,
-      "rewards/margins": 17.75,
+      "rewards/margins": 17.625,
       "rewards/rejected": -26.125,
       "step": 2300
     },
     {
       "epoch": 2.4083769633507854,
-      "eval_logits/chosen": 0.75390625,
-      "eval_logits/rejected": 2.234375,
-      "eval_logps/chosen": -1360.0,
-      "eval_logps/rejected": -1528.0,
-      "eval_loss": 1.3934218883514404,
-      "eval_rewards/accuracies": 0.607421875,
-      "eval_rewards/chosen": -10.375,
-      "eval_rewards/margins": 1.9765625,
-      "eval_rewards/rejected": -12.375,
-      "eval_runtime": 46.6241,
-      "eval_samples_per_second": 42.896,
+      "eval_logits/chosen": -6.65625,
+      "eval_logits/rejected": -4.8125,
+      "eval_logps/chosen": -1408.0,
+      "eval_logps/rejected": -1568.0,
+      "eval_loss": 1.5809531211853027,
+      "eval_rewards/accuracies": 0.595703125,
+      "eval_rewards/chosen": -10.875,
+      "eval_rewards/margins": 1.8984375,
+      "eval_rewards/rejected": -12.75,
+      "eval_runtime": 46.627,
+      "eval_samples_per_second": 42.894,
       "eval_steps_per_second": 0.686,
       "step": 2300
     },
     {
       "epoch": 2.418848167539267,
-      "grad_norm": 0.03275377997536498,
+      "grad_norm": 0.040919133364681205,
       "learning_rate": 5.50315277286812e-06,
-      "logits/chosen": -1.53125,
-      "logits/rejected": -0.109375,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -3408.0,
-      "loss": 0.0077,
+      "logits/chosen": -8.375,
+      "logits/rejected": -4.65625,
+      "logps/chosen": -1264.0,
+      "logps/rejected": -3536.0,
+      "loss": 0.0084,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 17.5,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -9.0,
+      "rewards/margins": 18.375,
+      "rewards/rejected": -27.5,
       "step": 2310
     },
     {
       "epoch": 2.4293193717277486,
-      "grad_norm": 0.1679708875285526,
+      "grad_norm": 0.18551828476806592,
       "learning_rate": 5.313910992897281e-06,
-      "logits/chosen": -1.1484375,
-      "logits/rejected": 0.60546875,
+      "logits/chosen": -7.6875,
+      "logits/rejected": -3.671875,
       "logps/chosen": -1176.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0074,
+      "logps/rejected": -3328.0,
+      "loss": 0.0067,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.96875,
-      "rewards/margins": 16.875,
-      "rewards/rejected": -24.75,
+      "rewards/chosen": -7.9375,
+      "rewards/margins": 17.5,
+      "rewards/rejected": -25.5,
       "step": 2320
     },
     {
       "epoch": 2.4397905759162306,
-      "grad_norm": 0.07598059422423965,
+      "grad_norm": 0.07821953557886552,
       "learning_rate": 5.127592611220658e-06,
-      "logits/chosen": -1.2734375,
-      "logits/rejected": -0.1318359375,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -3568.0,
-      "loss": 0.0127,
+      "logits/chosen": -7.59375,
+      "logits/rejected": -4.375,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -3632.0,
+      "loss": 0.0117,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.625,
-      "rewards/margins": 17.625,
-      "rewards/rejected": -26.125,
+      "rewards/chosen": -8.75,
+      "rewards/margins": 18.125,
+      "rewards/rejected": -26.875,
       "step": 2330
     },
     {
       "epoch": 2.450261780104712,
-      "grad_norm": 0.024583685238023404,
+      "grad_norm": 0.13275752330291243,
       "learning_rate": 4.944225296251159e-06,
-      "logits/chosen": -1.015625,
-      "logits/rejected": -0.71875,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -3168.0,
-      "loss": 0.004,
+      "logits/chosen": -8.0,
+      "logits/rejected": -4.90625,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0047,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 15.9375,
-      "rewards/rejected": -24.125,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -24.625,
       "step": 2340
     },
     {
       "epoch": 2.4607329842931938,
-      "grad_norm": 0.37305316467867383,
+      "grad_norm": 0.21445599211267555,
       "learning_rate": 4.763836278166172e-06,
-      "logits/chosen": -1.2734375,
-      "logits/rejected": 0.01348876953125,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -3056.0,
-      "loss": 0.0073,
+      "logits/chosen": -8.1875,
+      "logits/rejected": -4.40625,
+      "logps/chosen": -1224.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.0069,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 15.6875,
-      "rewards/rejected": -23.875,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 16.25,
+      "rewards/rejected": -24.75,
       "step": 2350
     },
     {
       "epoch": 2.4712041884816753,
-      "grad_norm": 0.06892585797297679,
+      "grad_norm": 0.10255454640112936,
       "learning_rate": 4.586452344863876e-06,
-      "logits/chosen": -1.984375,
-      "logits/rejected": -0.49609375,
+      "logits/chosen": -8.75,
+      "logits/rejected": -4.75,
       "logps/chosen": -1216.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0078,
+      "logps/rejected": -3376.0,
+      "loss": 0.0087,
       "rewards/accuracies": 0.9937499761581421,
       "rewards/chosen": -8.25,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -24.75,
+      "rewards/margins": 17.25,
+      "rewards/rejected": -25.625,
       "step": 2360
     },
     {
       "epoch": 2.481675392670157,
-      "grad_norm": 0.09769089343444261,
+      "grad_norm": 0.06384900655914158,
       "learning_rate": 4.412099837985192e-06,
-      "logits/chosen": -1.4296875,
-      "logits/rejected": 0.0244140625,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -3184.0,
-      "loss": 0.0187,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 16.125,
-      "rewards/rejected": -24.25,
+      "logits/chosen": -8.5,
+      "logits/rejected": -4.5,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.006,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -8.375,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -24.75,
       "step": 2370
     },
     {
       "epoch": 2.492146596858639,
-      "grad_norm": 0.31999882128194007,
+      "grad_norm": 0.27509199100364545,
       "learning_rate": 4.240804649002089e-06,
-      "logits/chosen": -1.8203125,
-      "logits/rejected": -0.2421875,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -3376.0,
-      "loss": 0.0085,
+      "logits/chosen": -8.75,
+      "logits/rejected": -4.4375,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -3472.0,
+      "loss": 0.0079,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 17.125,
-      "rewards/rejected": -25.25,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 17.875,
+      "rewards/rejected": -26.125,
       "step": 2380
     },
     {
       "epoch": 2.5026178010471205,
-      "grad_norm": 0.06997334372847906,
+      "grad_norm": 2.246719185758602,
       "learning_rate": 4.072592215372598e-06,
-      "logits/chosen": -2.375,
-      "logits/rejected": -0.96875,
+      "logits/chosen": -8.75,
+      "logits/rejected": -5.15625,
       "logps/chosen": -1328.0,
-      "logps/rejected": -3440.0,
-      "loss": 0.0043,
+      "logps/rejected": -3472.0,
+      "loss": 0.0068,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 17.0,
-      "rewards/rejected": -26.125,
+      "rewards/chosen": -9.0,
+      "rewards/margins": 17.5,
+      "rewards/rejected": -26.5,
       "step": 2390
     },
     {
       "epoch": 2.513089005235602,
-      "grad_norm": 0.03811520754059408,
+      "grad_norm": 0.5837801108086123,
       "learning_rate": 3.907487516763389e-06,
-      "logits/chosen": -1.3515625,
-      "logits/rejected": 0.265625,
+      "logits/chosen": -8.6875,
+      "logits/rejected": -4.28125,
       "logps/chosen": -1208.0,
       "logps/rejected": -3248.0,
       "loss": 0.0074,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 16.375,
-      "rewards/rejected": -24.625,
+      "rewards/chosen": -8.375,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -24.5,
       "step": 2400
     },
     {
       "epoch": 2.513089005235602,
-      "eval_logits/chosen": 0.63671875,
-      "eval_logits/rejected": 2.125,
-      "eval_logps/chosen": -1360.0,
-      "eval_logps/rejected": -1536.0,
-      "eval_loss": 1.4225937128067017,
-      "eval_rewards/accuracies": 0.615234375,
-      "eval_rewards/chosen": -10.4375,
-      "eval_rewards/margins": 2.03125,
-      "eval_rewards/rejected": -12.4375,
-      "eval_runtime": 46.648,
-      "eval_samples_per_second": 42.874,
+      "eval_logits/chosen": -7.21875,
+      "eval_logits/rejected": -5.34375,
+      "eval_logps/chosen": -1400.0,
+      "eval_logps/rejected": -1560.0,
+      "eval_loss": 1.6087656021118164,
+      "eval_rewards/accuracies": 0.591796875,
+      "eval_rewards/chosen": -10.8125,
+      "eval_rewards/margins": 1.9140625,
+      "eval_rewards/rejected": -12.6875,
+      "eval_runtime": 46.6164,
+      "eval_samples_per_second": 42.903,
       "eval_steps_per_second": 0.686,
       "step": 2400
     },
     {
       "epoch": 2.5235602094240837,
-      "grad_norm": 0.08552148883154066,
+      "grad_norm": 0.15456604538654667,
       "learning_rate": 3.745515071340286e-06,
-      "logits/chosen": -1.421875,
-      "logits/rejected": 0.034912109375,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0064,
+      "logits/chosen": -8.6875,
+      "logits/rejected": -4.15625,
+      "logps/chosen": -1328.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0054,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.5625,
-      "rewards/margins": 18.5,
-      "rewards/rejected": -28.125,
+      "rewards/chosen": -9.1875,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -28.375,
       "step": 2410
     },
     {
       "epoch": 2.5340314136125652,
-      "grad_norm": 0.02638913440092958,
+      "grad_norm": 0.07208272634095589,
       "learning_rate": 3.586698932127236e-06,
-      "logits/chosen": -2.203125,
-      "logits/rejected": -0.330078125,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -3424.0,
-      "loss": 0.0033,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -4.65625,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -3440.0,
+      "loss": 0.0023,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 17.875,
-      "rewards/rejected": -26.0,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -26.125,
       "step": 2420
     },
     {
       "epoch": 2.5445026178010473,
-      "grad_norm": 0.9307749344880539,
+      "grad_norm": 0.24229559938765557,
       "learning_rate": 3.4310626834344744e-06,
-      "logits/chosen": -2.203125,
-      "logits/rejected": -0.921875,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -3216.0,
-      "loss": 0.0074,
+      "logits/chosen": -8.25,
+      "logits/rejected": -5.03125,
+      "logps/chosen": -1296.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0091,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 15.75,
-      "rewards/rejected": -24.25,
+      "rewards/chosen": -8.875,
+      "rewards/margins": 16.0,
+      "rewards/rejected": -24.875,
       "step": 2430
     },
     {
       "epoch": 2.554973821989529,
-      "grad_norm": 0.02999989824340152,
+      "grad_norm": 0.05888255624700516,
       "learning_rate": 3.2786294373562343e-06,
-      "logits/chosen": -1.6796875,
-      "logits/rejected": -0.58203125,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0041,
+      "logits/chosen": -8.25,
+      "logits/rejected": -4.1875,
+      "logps/chosen": -1264.0,
+      "logps/rejected": -3328.0,
+      "loss": 0.0062,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 16.375,
-      "rewards/rejected": -24.625,
+      "rewards/chosen": -8.625,
+      "rewards/margins": 17.25,
+      "rewards/rejected": -25.875,
       "step": 2440
     },
     {
       "epoch": 2.5654450261780104,
-      "grad_norm": 0.14902807923492303,
+      "grad_norm": 0.2567072399302382,
       "learning_rate": 3.1294218303385526e-06,
-      "logits/chosen": -1.96875,
-      "logits/rejected": -0.515625,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -3376.0,
-      "loss": 0.0047,
+      "logits/chosen": -8.3125,
+      "logits/rejected": -4.25,
+      "logps/chosen": -1328.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0046,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.75,
-      "rewards/margins": 16.875,
-      "rewards/rejected": -25.625,
+      "rewards/chosen": -9.0,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -26.0,
       "step": 2450
     },
     {
       "epoch": 2.5759162303664924,
-      "grad_norm": 0.0980008182330775,
+      "grad_norm": 0.08849688417884374,
       "learning_rate": 2.9834620198177483e-06,
-      "logits/chosen": -2.25,
-      "logits/rejected": -0.56640625,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -3664.0,
-      "loss": 0.0111,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 19.125,
-      "rewards/rejected": -27.875,
+      "logits/chosen": -8.125,
+      "logits/rejected": -4.1875,
+      "logps/chosen": -1280.0,
+      "logps/rejected": -3632.0,
+      "loss": 0.0095,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -8.875,
+      "rewards/margins": 18.75,
+      "rewards/rejected": -27.625,
       "step": 2460
     },
     {
       "epoch": 2.5863874345549736,
-      "grad_norm": 0.3213169030520241,
+      "grad_norm": 0.08867257292947205,
       "learning_rate": 2.840771680930068e-06,
-      "logits/chosen": -2.125,
-      "logits/rejected": -0.65234375,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -3776.0,
-      "loss": 0.0035,
+      "logits/chosen": -8.375,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1368.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0061,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 18.75,
-      "rewards/rejected": -28.5,
+      "rewards/chosen": -9.5,
+      "rewards/margins": 18.5,
+      "rewards/rejected": -28.0,
       "step": 2470
     },
     {
       "epoch": 2.5968586387434556,
-      "grad_norm": 0.549764744850555,
+      "grad_norm": 0.1824159575032018,
       "learning_rate": 2.701372003292868e-06,
-      "logits/chosen": -0.341796875,
-      "logits/rejected": 0.275390625,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -3568.0,
-      "loss": 0.009,
+      "logits/chosen": -8.0,
+      "logits/rejected": -4.5,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -3488.0,
+      "loss": 0.0101,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.625,
-      "rewards/margins": 18.625,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -8.4375,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -26.5,
       "step": 2480
     },
     {
       "epoch": 2.607329842931937,
-      "grad_norm": 0.13949812258783909,
+      "grad_norm": 0.035544858385887244,
       "learning_rate": 2.56528368785795e-06,
-      "logits/chosen": -1.3046875,
-      "logits/rejected": 0.0225830078125,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.0047,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 15.8125,
-      "rewards/rejected": -24.625,
+      "logits/chosen": -9.0,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -3184.0,
+      "loss": 0.0024,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 15.625,
+      "rewards/rejected": -24.0,
       "step": 2490
     },
     {
       "epoch": 2.6178010471204187,
-      "grad_norm": 0.031757230869942955,
+      "grad_norm": 0.08483714052206529,
       "learning_rate": 2.432526943837471e-06,
-      "logits/chosen": -1.1796875,
-      "logits/rejected": 0.6328125,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -3568.0,
-      "loss": 0.0055,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -4.21875,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0063,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.5625,
-      "rewards/margins": 19.25,
-      "rewards/rejected": -27.75,
+      "rewards/chosen": -8.0625,
+      "rewards/margins": 18.5,
+      "rewards/rejected": -26.625,
       "step": 2500
     },
     {
       "epoch": 2.6178010471204187,
-      "eval_logits/chosen": 0.62109375,
-      "eval_logits/rejected": 2.109375,
-      "eval_logps/chosen": -1376.0,
-      "eval_logps/rejected": -1552.0,
-      "eval_loss": 1.431890606880188,
-      "eval_rewards/accuracies": 0.615234375,
-      "eval_rewards/chosen": -10.5625,
-      "eval_rewards/margins": 2.0625,
-      "eval_rewards/rejected": -12.625,
-      "eval_runtime": 46.6335,
-      "eval_samples_per_second": 42.888,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -7.4375,
+      "eval_logits/rejected": -5.5625,
+      "eval_logps/chosen": -1368.0,
+      "eval_logps/rejected": -1528.0,
+      "eval_loss": 1.5735156536102295,
+      "eval_rewards/accuracies": 0.59375,
+      "eval_rewards/chosen": -10.5,
+      "eval_rewards/margins": 1.8828125,
+      "eval_rewards/rejected": -12.375,
+      "eval_runtime": 46.596,
+      "eval_samples_per_second": 42.922,
+      "eval_steps_per_second": 0.687,
       "step": 2500
     },
     {
       "epoch": 2.6282722513089007,
-      "grad_norm": 0.05942975961657911,
+      "grad_norm": 0.10473941253531054,
       "learning_rate": 2.3031214857028446e-06,
-      "logits/chosen": -1.296875,
-      "logits/rejected": -0.0703125,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -3552.0,
-      "loss": 0.006,
+      "logits/chosen": -8.375,
+      "logits/rejected": -4.71875,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -3536.0,
+      "loss": 0.0062,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.625,
-      "rewards/margins": 18.25,
-      "rewards/rejected": -26.875,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 18.5,
+      "rewards/rejected": -26.75,
       "step": 2510
     },
     {
       "epoch": 2.6387434554973823,
-      "grad_norm": 0.03572761056998649,
+      "grad_norm": 0.038719725000748204,
       "learning_rate": 2.177086530257139e-06,
-      "logits/chosen": -1.2421875,
-      "logits/rejected": -0.2255859375,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -3104.0,
-      "loss": 0.0096,
+      "logits/chosen": -8.875,
+      "logits/rejected": -5.21875,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -3056.0,
+      "loss": 0.0087,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.8125,
+      "rewards/chosen": -8.25,
       "rewards/margins": 14.875,
-      "rewards/rejected": -23.625,
+      "rewards/rejected": -23.125,
       "step": 2520
     },
     {
       "epoch": 2.649214659685864,
-      "grad_norm": 0.03846660104673547,
+      "grad_norm": 0.08005165239945668,
       "learning_rate": 2.054440793781354e-06,
-      "logits/chosen": -1.7734375,
-      "logits/rejected": -0.236328125,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.0087,
+      "logits/chosen": -9.0625,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0069,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.375,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -24.875,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -24.625,
       "step": 2530
     },
     {
       "epoch": 2.6596858638743455,
-      "grad_norm": 0.374971797012904,
+      "grad_norm": 0.08302626875153249,
       "learning_rate": 1.9352024892550587e-06,
-      "logits/chosen": -1.8125,
-      "logits/rejected": -0.5546875,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -3312.0,
-      "loss": 0.0052,
+      "logits/chosen": -8.375,
+      "logits/rejected": -4.625,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.4375,
-      "rewards/margins": 16.875,
-      "rewards/rejected": -25.375,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 16.625,
+      "rewards/rejected": -24.875,
       "step": 2540
     },
     {
       "epoch": 2.670157068062827,
-      "grad_norm": 0.28650504693157575,
+      "grad_norm": 0.31265992585889363,
       "learning_rate": 1.8193893236517379e-06,
-      "logits/chosen": -2.078125,
-      "logits/rejected": -0.333984375,
+      "logits/chosen": -8.1875,
+      "logits/rejected": -4.4375,
       "logps/chosen": -1264.0,
-      "logps/rejected": -3648.0,
-      "loss": 0.0091,
+      "logps/rejected": -3664.0,
+      "loss": 0.011,
       "rewards/accuracies": 1.0,
       "rewards/chosen": -8.625,
-      "rewards/margins": 19.25,
-      "rewards/rejected": -27.875,
+      "rewards/margins": 19.5,
+      "rewards/rejected": -28.125,
       "step": 2550
     },
     {
       "epoch": 2.680628272251309,
-      "grad_norm": 0.05944442898747252,
+      "grad_norm": 0.22649436434820946,
       "learning_rate": 1.7070184953092956e-06,
-      "logits/chosen": -1.4453125,
-      "logits/rejected": 0.72265625,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.0084,
+      "logits/chosen": -8.375,
+      "logits/rejected": -3.9375,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0065,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.6875,
-      "rewards/margins": 17.125,
-      "rewards/rejected": -24.75,
+      "rewards/chosen": -7.53125,
+      "rewards/margins": 17.75,
+      "rewards/rejected": -25.25,
       "step": 2560
     },
     {
       "epoch": 2.6910994764397906,
-      "grad_norm": 0.03917188411316824,
+      "grad_norm": 0.017969325262499496,
       "learning_rate": 1.5981066913760918e-06,
-      "logits/chosen": -0.71875,
-      "logits/rejected": 0.6875,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -3120.0,
-      "loss": 0.0037,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -4.28125,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -3136.0,
+      "loss": 0.0025,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.9375,
-      "rewards/margins": 16.0,
-      "rewards/rejected": -24.0,
+      "rewards/chosen": -7.5,
+      "rewards/margins": 16.75,
+      "rewards/rejected": -24.25,
       "step": 2570
     },
     {
       "epoch": 2.701570680628272,
-      "grad_norm": 0.28052792478669675,
+      "grad_norm": 0.16158651461730886,
       "learning_rate": 1.492670085332884e-06,
-      "logits/chosen": -2.265625,
-      "logits/rejected": -0.609375,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -3488.0,
-      "loss": 0.0094,
+      "logits/chosen": -8.875,
+      "logits/rejected": -4.3125,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0073,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.4375,
-      "rewards/margins": 18.0,
-      "rewards/rejected": -26.375,
+      "rewards/chosen": -8.6875,
+      "rewards/margins": 18.625,
+      "rewards/rejected": -27.25,
       "step": 2580
     },
     {
       "epoch": 2.712041884816754,
-      "grad_norm": 0.03455889587798861,
+      "grad_norm": 0.1247271225041803,
       "learning_rate": 1.3907243345910787e-06,
-      "logits/chosen": -1.6328125,
-      "logits/rejected": -0.384765625,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0052,
+      "logits/chosen": -9.0,
+      "logits/rejected": -4.6875,
+      "logps/chosen": -1224.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0073,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 15.75,
-      "rewards/rejected": -23.875,
+      "rewards/chosen": -8.0,
+      "rewards/margins": 16.625,
+      "rewards/rejected": -24.625,
       "step": 2590
     },
     {
       "epoch": 2.7225130890052354,
-      "grad_norm": 0.16428575193938777,
+      "grad_norm": 0.13005485897724398,
       "learning_rate": 1.292284578167538e-06,
-      "logits/chosen": -1.5,
-      "logits/rejected": -0.52734375,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0094,
+      "logits/chosen": -8.25,
+      "logits/rejected": -4.6875,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -3424.0,
+      "loss": 0.0058,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.875,
-      "rewards/margins": 16.875,
-      "rewards/rejected": -24.75,
+      "rewards/chosen": -8.0625,
+      "rewards/margins": 17.5,
+      "rewards/rejected": -25.5,
       "step": 2600
     },
     {
       "epoch": 2.7225130890052354,
-      "eval_logits/chosen": 0.43359375,
-      "eval_logits/rejected": 1.9375,
-      "eval_logps/chosen": -1328.0,
-      "eval_logps/rejected": -1504.0,
-      "eval_loss": 1.39829683303833,
-      "eval_rewards/accuracies": 0.615234375,
-      "eval_rewards/chosen": -10.125,
-      "eval_rewards/margins": 2.015625,
-      "eval_rewards/rejected": -12.125,
-      "eval_runtime": 46.6245,
-      "eval_samples_per_second": 42.896,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -7.53125,
+      "eval_logits/rejected": -5.65625,
+      "eval_logps/chosen": -1368.0,
+      "eval_logps/rejected": -1528.0,
+      "eval_loss": 1.5901561975479126,
+      "eval_rewards/accuracies": 0.591796875,
+      "eval_rewards/chosen": -10.5,
+      "eval_rewards/margins": 1.890625,
+      "eval_rewards/rejected": -12.4375,
+      "eval_runtime": 46.6016,
+      "eval_samples_per_second": 42.917,
+      "eval_steps_per_second": 0.687,
       "step": 2600
     },
     {
       "epoch": 2.7329842931937174,
-      "grad_norm": 0.46108105350297496,
+      "grad_norm": 0.19575848647241598,
       "learning_rate": 1.1973654344364925e-06,
-      "logits/chosen": -1.625,
-      "logits/rejected": -0.6640625,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -3456.0,
-      "loss": 0.0066,
+      "logits/chosen": -8.5625,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -3472.0,
+      "loss": 0.0053,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.84375,
-      "rewards/margins": 17.375,
-      "rewards/rejected": -25.25,
+      "rewards/chosen": -7.75,
+      "rewards/margins": 17.75,
+      "rewards/rejected": -25.5,
       "step": 2610
     },
     {
       "epoch": 2.743455497382199,
-      "grad_norm": 0.04133685615953533,
+      "grad_norm": 0.03335682928778873,
       "learning_rate": 1.1059809989586583e-06,
-      "logits/chosen": -1.2421875,
-      "logits/rejected": 0.11669921875,
+      "logits/chosen": -8.6875,
+      "logits/rejected": -4.0,
       "logps/chosen": -1216.0,
-      "logps/rejected": -3312.0,
-      "loss": 0.006,
+      "logps/rejected": -3360.0,
+      "loss": 0.004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 16.75,
-      "rewards/rejected": -25.0,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 17.25,
+      "rewards/rejected": -25.5,
       "step": 2620
     },
     {
       "epoch": 2.7539267015706805,
-      "grad_norm": 0.07145181889846582,
+      "grad_norm": 0.47795721008775993,
       "learning_rate": 1.018144842388044e-06,
-      "logits/chosen": -1.15625,
-      "logits/rejected": 0.2060546875,
+      "logits/chosen": -8.25,
+      "logits/rejected": -4.21875,
       "logps/chosen": -1240.0,
-      "logps/rejected": -3552.0,
-      "loss": 0.0044,
+      "logps/rejected": -3568.0,
+      "loss": 0.005,
       "rewards/accuracies": 1.0,
       "rewards/chosen": -8.125,
       "rewards/margins": 18.625,
@@ -4386,389 +4386,389 @@
     },
     {
       "epoch": 2.7643979057591626,
-      "grad_norm": 0.05575194575845501,
+      "grad_norm": 0.029368571241583586,
       "learning_rate": 9.338700084567109e-07,
-      "logits/chosen": -1.671875,
-      "logits/rejected": -1.015625,
-      "logps/chosen": -1216.0,
-      "logps/rejected": -3472.0,
-      "loss": 0.0037,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -5.0625,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -3520.0,
+      "loss": 0.0056,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 17.625,
-      "rewards/rejected": -26.0,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -26.5,
       "step": 2640
     },
     {
       "epoch": 2.774869109947644,
-      "grad_norm": 0.056817129143399744,
+      "grad_norm": 0.1307078000098744,
       "learning_rate": 8.531690120377606e-07,
-      "logits/chosen": -1.953125,
-      "logits/rejected": -0.46484375,
+      "logits/chosen": -8.375,
+      "logits/rejected": -4.8125,
       "logps/chosen": -1288.0,
-      "logps/rejected": -3536.0,
-      "loss": 0.0036,
+      "logps/rejected": -3568.0,
+      "loss": 0.0041,
       "rewards/accuracies": 1.0,
       "rewards/chosen": -8.6875,
-      "rewards/margins": 17.75,
-      "rewards/rejected": -26.375,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -26.75,
       "step": 2650
     },
     {
       "epoch": 2.7853403141361257,
-      "grad_norm": 0.22406633795816527,
+      "grad_norm": 0.41982950645420475,
       "learning_rate": 7.760538372868636e-07,
-      "logits/chosen": -1.234375,
-      "logits/rejected": -0.251953125,
-      "logps/chosen": -1216.0,
-      "logps/rejected": -3328.0,
-      "loss": 0.0099,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -4.9375,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -3440.0,
+      "loss": 0.0093,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 16.75,
-      "rewards/rejected": -25.0,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -26.125,
       "step": 2660
     },
     {
       "epoch": 2.7958115183246073,
-      "grad_norm": 0.03691563736343768,
+      "grad_norm": 0.832388011162078,
       "learning_rate": 7.025359358626166e-07,
-      "logits/chosen": -1.90625,
-      "logits/rejected": -0.3046875,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -3472.0,
-      "loss": 0.0024,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 18.0,
-      "rewards/rejected": -26.5,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1288.0,
+      "logps/rejected": -3568.0,
+      "loss": 0.0047,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -8.625,
+      "rewards/margins": 18.75,
+      "rewards/rejected": -27.375,
       "step": 2670
     },
     {
       "epoch": 2.806282722513089,
-      "grad_norm": 0.8935751225894819,
+      "grad_norm": 0.3657771100397663,
       "learning_rate": 6.326262252259463e-07,
-      "logits/chosen": -2.09375,
-      "logits/rejected": -0.435546875,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -3008.0,
-      "loss": 0.0076,
+      "logits/chosen": -8.5,
+      "logits/rejected": -4.125,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.0049,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.90625,
-      "rewards/margins": 15.4375,
-      "rewards/rejected": -23.375,
+      "rewards/chosen": -7.96875,
+      "rewards/margins": 15.5625,
+      "rewards/rejected": -23.625,
       "step": 2680
     },
     {
       "epoch": 2.816753926701571,
-      "grad_norm": 0.6039606051444781,
+      "grad_norm": 0.19968583063064413,
       "learning_rate": 5.66335087018871e-07,
-      "logits/chosen": -2.078125,
-      "logits/rejected": -0.2578125,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -3712.0,
-      "loss": 0.0039,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 19.875,
-      "rewards/rejected": -28.0,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -4.28125,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.006,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -28.5,
       "step": 2690
     },
     {
       "epoch": 2.8272251308900525,
-      "grad_norm": 0.22354548184376424,
+      "grad_norm": 0.17743973382330416,
       "learning_rate": 5.036723655228226e-07,
-      "logits/chosen": -1.9453125,
-      "logits/rejected": -0.3671875,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0045,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.125,
+      "logits/chosen": -9.0625,
+      "logits/rejected": -4.8125,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.0071,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -8.4375,
       "rewards/margins": 17.375,
-      "rewards/rejected": -25.375,
+      "rewards/rejected": -25.875,
       "step": 2700
     },
     {
       "epoch": 2.8272251308900525,
-      "eval_logits/chosen": 0.423828125,
-      "eval_logits/rejected": 1.9296875,
-      "eval_logps/chosen": -1320.0,
-      "eval_logps/rejected": -1488.0,
-      "eval_loss": 1.3869374990463257,
-      "eval_rewards/accuracies": 0.61328125,
-      "eval_rewards/chosen": -10.0,
-      "eval_rewards/margins": 2.015625,
-      "eval_rewards/rejected": -12.0,
-      "eval_runtime": 46.6252,
-      "eval_samples_per_second": 42.895,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -7.625,
+      "eval_logits/rejected": -5.71875,
+      "eval_logps/chosen": -1360.0,
+      "eval_logps/rejected": -1512.0,
+      "eval_loss": 1.57798433303833,
+      "eval_rewards/accuracies": 0.58984375,
+      "eval_rewards/chosen": -10.375,
+      "eval_rewards/margins": 1.8828125,
+      "eval_rewards/rejected": -12.25,
+      "eval_runtime": 46.6076,
+      "eval_samples_per_second": 42.911,
+      "eval_steps_per_second": 0.687,
       "step": 2700
     },
     {
       "epoch": 2.837696335078534,
-      "grad_norm": 0.11329810735390831,
+      "grad_norm": 0.41132929021626863,
       "learning_rate": 4.446473661967432e-07,
-      "logits/chosen": -2.171875,
-      "logits/rejected": -0.86328125,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -3024.0,
-      "loss": 0.008,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 15.0,
-      "rewards/rejected": -23.0,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -5.5,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -3072.0,
+      "loss": 0.0025,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 15.1875,
+      "rewards/rejected": -23.5,
       "step": 2710
     },
     {
       "epoch": 2.8481675392670156,
-      "grad_norm": 0.2314113440614339,
+      "grad_norm": 0.22067201941135434,
       "learning_rate": 3.892688542952411e-07,
-      "logits/chosen": -1.09375,
-      "logits/rejected": 0.55078125,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0059,
+      "logits/chosen": -8.5,
+      "logits/rejected": -4.4375,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.006,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 16.625,
-      "rewards/rejected": -24.25,
+      "rewards/chosen": -7.53125,
+      "rewards/margins": 17.125,
+      "rewards/rejected": -24.625,
       "step": 2720
     },
     {
       "epoch": 2.858638743455497,
-      "grad_norm": 0.5963751619341928,
+      "grad_norm": 0.30831390429451977,
       "learning_rate": 3.3754505356694e-07,
-      "logits/chosen": -2.125,
-      "logits/rejected": -0.5703125,
+      "logits/chosen": -8.75,
+      "logits/rejected": -5.15625,
       "logps/chosen": -1232.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0074,
+      "logps/rejected": -3216.0,
+      "loss": 0.0059,
       "rewards/accuracies": 1.0,
       "rewards/chosen": -8.3125,
-      "rewards/margins": 15.9375,
-      "rewards/rejected": -24.25,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -24.375,
       "step": 2730
     },
     {
       "epoch": 2.869109947643979,
-      "grad_norm": 0.01964194172212382,
+      "grad_norm": 0.12552738168155977,
       "learning_rate": 2.8948364503322276e-07,
-      "logits/chosen": -1.5703125,
-      "logits/rejected": -0.3125,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0044,
+      "logits/chosen": -8.875,
+      "logits/rejected": -4.71875,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.0042,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.90625,
-      "rewards/margins": 17.25,
-      "rewards/rejected": -25.125,
+      "rewards/chosen": -8.125,
+      "rewards/margins": 17.375,
+      "rewards/rejected": -25.375,
       "step": 2740
     },
     {
       "epoch": 2.8795811518324608,
-      "grad_norm": 0.0842643294599748,
+      "grad_norm": 0.49387684482289984,
       "learning_rate": 2.4509176584762706e-07,
-      "logits/chosen": -2.015625,
-      "logits/rejected": -0.91015625,
-      "logps/chosen": -1152.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0056,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -5.15625,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -3328.0,
+      "loss": 0.0045,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.75,
-      "rewards/margins": 16.375,
-      "rewards/rejected": -24.125,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 17.25,
+      "rewards/rejected": -24.875,
       "step": 2750
     },
     {
       "epoch": 2.8900523560209423,
-      "grad_norm": 0.0523078118583025,
+      "grad_norm": 0.040042006911317304,
       "learning_rate": 2.0437600823595694e-07,
-      "logits/chosen": -2.1875,
-      "logits/rejected": -0.306640625,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.0058,
+      "logits/chosen": -8.625,
+      "logits/rejected": -4.46875,
+      "logps/chosen": -1224.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0032,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 16.375,
-      "rewards/rejected": -24.5,
+      "rewards/chosen": -7.96875,
+      "rewards/margins": 16.25,
+      "rewards/rejected": -24.25,
       "step": 2760
     },
     {
       "epoch": 2.900523560209424,
-      "grad_norm": 0.03217661633152188,
+      "grad_norm": 0.026370166309928264,
       "learning_rate": 1.6734241851733e-07,
-      "logits/chosen": -2.078125,
-      "logits/rejected": -0.42578125,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -3424.0,
-      "loss": 0.0068,
+      "logits/chosen": -8.6875,
+      "logits/rejected": -4.40625,
+      "logps/chosen": -1296.0,
+      "logps/rejected": -3472.0,
+      "loss": 0.0065,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 17.125,
-      "rewards/rejected": -25.875,
+      "rewards/chosen": -8.625,
+      "rewards/margins": 17.75,
+      "rewards/rejected": -26.25,
       "step": 2770
     },
     {
       "epoch": 2.9109947643979055,
-      "grad_norm": 0.18772737630671385,
+      "grad_norm": 0.2982190839295278,
       "learning_rate": 1.339964962062984e-07,
-      "logits/chosen": -2.390625,
-      "logits/rejected": -1.625,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.005,
+      "logits/chosen": -8.875,
+      "logits/rejected": -5.25,
+      "logps/chosen": -1160.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0045,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 17.125,
-      "rewards/rejected": -25.125,
+      "rewards/chosen": -8.0,
+      "rewards/margins": 18.375,
+      "rewards/rejected": -26.375,
       "step": 2780
     },
     {
       "epoch": 2.9214659685863875,
-      "grad_norm": 0.0806871935356948,
+      "grad_norm": 0.04684861273101893,
       "learning_rate": 1.0434319319617136e-07,
-      "logits/chosen": -1.890625,
-      "logits/rejected": -0.9609375,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -3088.0,
-      "loss": 0.0143,
+      "logits/chosen": -8.0625,
+      "logits/rejected": -5.09375,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0114,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.96875,
-      "rewards/margins": 15.3125,
-      "rewards/rejected": -23.25,
+      "rewards/chosen": -7.90625,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -24.375,
       "step": 2790
     },
     {
       "epoch": 2.931937172774869,
-      "grad_norm": 0.2543799838260187,
+      "grad_norm": 0.24839769790542174,
       "learning_rate": 7.838691302365363e-08,
-      "logits/chosen": -1.734375,
-      "logits/rejected": 0.494140625,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -3136.0,
-      "loss": 0.0065,
+      "logits/chosen": -9.0625,
+      "logits/rejected": -4.21875,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0063,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.6875,
-      "rewards/margins": 16.0,
-      "rewards/rejected": -23.75,
+      "rewards/chosen": -7.5,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -24.5,
       "step": 2800
     },
     {
       "epoch": 2.931937172774869,
-      "eval_logits/chosen": 0.439453125,
-      "eval_logits/rejected": 1.9375,
-      "eval_logps/chosen": -1328.0,
-      "eval_logps/rejected": -1496.0,
-      "eval_loss": 1.3937655687332153,
-      "eval_rewards/accuracies": 0.615234375,
-      "eval_rewards/chosen": -10.0625,
-      "eval_rewards/margins": 2.015625,
-      "eval_rewards/rejected": -12.0625,
-      "eval_runtime": 46.6232,
-      "eval_samples_per_second": 42.897,
+      "eval_logits/chosen": -7.59375,
+      "eval_logits/rejected": -5.6875,
+      "eval_logps/chosen": -1360.0,
+      "eval_logps/rejected": -1520.0,
+      "eval_loss": 1.5814374685287476,
+      "eval_rewards/accuracies": 0.59375,
+      "eval_rewards/chosen": -10.375,
+      "eval_rewards/margins": 1.890625,
+      "eval_rewards/rejected": -12.25,
+      "eval_runtime": 46.617,
+      "eval_samples_per_second": 42.903,
       "eval_steps_per_second": 0.686,
       "step": 2800
     },
     {
       "epoch": 2.9424083769633507,
-      "grad_norm": 0.0953113356122145,
+      "grad_norm": 0.5506909876681924,
       "learning_rate": 5.613151021490459e-08,
-      "logits/chosen": -1.328125,
-      "logits/rejected": 0.333984375,
+      "logits/chosen": -8.625,
+      "logits/rejected": -4.125,
       "logps/chosen": -1096.0,
-      "logps/rejected": -2960.0,
-      "loss": 0.0048,
+      "logps/rejected": -2992.0,
+      "loss": 0.0044,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.65625,
-      "rewards/margins": 15.25,
-      "rewards/rejected": -22.875,
+      "rewards/chosen": -7.625,
+      "rewards/margins": 15.6875,
+      "rewards/rejected": -23.25,
       "step": 2810
     },
     {
       "epoch": 2.9528795811518327,
-      "grad_norm": 0.011985384898715123,
+      "grad_norm": 0.046564986965591984,
       "learning_rate": 3.7580289713159965e-08,
-      "logits/chosen": -1.328125,
-      "logits/rejected": -0.08203125,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -3424.0,
-      "loss": 0.0069,
+      "logits/chosen": -8.75,
+      "logits/rejected": -4.8125,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -3440.0,
+      "loss": 0.007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.1875,
+      "rewards/chosen": -8.3125,
       "rewards/margins": 17.5,
-      "rewards/rejected": -25.625,
+      "rewards/rejected": -25.75,
       "step": 2820
     },
     {
       "epoch": 2.9633507853403143,
-      "grad_norm": 0.12632150406108508,
+      "grad_norm": 0.2978141149867284,
       "learning_rate": 2.2736006387927277e-08,
-      "logits/chosen": -1.9375,
-      "logits/rejected": -0.75390625,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -3520.0,
-      "loss": 0.0049,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -5.15625,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -3488.0,
+      "loss": 0.0048,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 18.125,
-      "rewards/rejected": -26.5,
+      "rewards/chosen": -8.375,
+      "rewards/margins": 17.875,
+      "rewards/rejected": -26.25,
       "step": 2830
     },
     {
       "epoch": 2.973821989528796,
-      "grad_norm": 0.014673652923435353,
+      "grad_norm": 0.047410137463969076,
       "learning_rate": 1.1600864625893604e-08,
-      "logits/chosen": -2.234375,
-      "logits/rejected": -0.546875,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -3504.0,
-      "loss": 0.0047,
+      "logits/chosen": -8.75,
+      "logits/rejected": -4.5,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -3600.0,
+      "loss": 0.0038,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.84375,
-      "rewards/margins": 18.75,
-      "rewards/rejected": -26.625,
+      "rewards/chosen": -8.0,
+      "rewards/margins": 19.5,
+      "rewards/rejected": -27.5,
       "step": 2840
     },
     {
       "epoch": 2.9842931937172774,
-      "grad_norm": 0.05893350407251536,
+      "grad_norm": 0.3085281568027718,
       "learning_rate": 4.176518003573549e-09,
-      "logits/chosen": -2.234375,
-      "logits/rejected": -0.30078125,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -3280.0,
-      "loss": 0.0048,
+      "logits/chosen": -9.0,
+      "logits/rejected": -4.59375,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0079,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -24.75,
+      "rewards/chosen": -7.90625,
+      "rewards/margins": 17.875,
+      "rewards/rejected": -25.75,
       "step": 2850
     },
     {
       "epoch": 2.994764397905759,
-      "grad_norm": 0.13867621353622797,
+      "grad_norm": 0.2634998253912702,
       "learning_rate": 4.640690417528903e-10,
-      "logits/chosen": -1.8203125,
-      "logits/rejected": -0.87890625,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -2976.0,
-      "loss": 0.0091,
+      "logits/chosen": -8.75,
+      "logits/rejected": -5.65625,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -2992.0,
+      "loss": 0.0095,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 14.75,
-      "rewards/rejected": -23.0,
+      "rewards/chosen": -8.125,
+      "rewards/margins": 15.0,
+      "rewards/rejected": -23.125,
       "step": 2860
     },
     {
       "epoch": 3.0,
       "step": 2865,
       "total_flos": 0.0,
-      "train_loss": 0.11531559581726945,
-      "train_runtime": 12304.0145,
-      "train_samples_per_second": 14.902,
-      "train_steps_per_second": 0.233
+      "train_loss": 0.116292925227765,
+      "train_runtime": 12240.7967,
+      "train_samples_per_second": 14.979,
+      "train_steps_per_second": 0.234
     }
   ],
   "logging_steps": 10,