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": 1.5821770518053708,
+      "grad_norm": 1.5822940028569477,
       "learning_rate": 1.7421602787456448e-07,
       "logits/chosen": -11.375,
       "logits/rejected": -11.6875,
@@ -25,4750 +25,4750 @@
     },
     {
       "epoch": 0.010471204188481676,
-      "grad_norm": 1.5266999098741172,
+      "grad_norm": 1.5263432760554976,
       "learning_rate": 1.7421602787456445e-06,
       "logits/chosen": -11.3125,
       "logits/rejected": -10.875,
       "logps/chosen": -438.0,
       "logps/rejected": -548.0,
-      "loss": 0.6919,
-      "rewards/accuracies": 0.1944444477558136,
-      "rewards/chosen": 0.001007080078125,
-      "rewards/margins": 0.00139617919921875,
-      "rewards/rejected": -0.0003910064697265625,
+      "loss": 0.6921,
+      "rewards/accuracies": 0.1388888955116272,
+      "rewards/chosen": 0.00028228759765625,
+      "rewards/margins": -0.0004253387451171875,
+      "rewards/rejected": 0.000705718994140625,
       "step": 10
     },
     {
       "epoch": 0.020942408376963352,
-      "grad_norm": 1.5006055995386742,
+      "grad_norm": 1.5049783582640077,
       "learning_rate": 3.484320557491289e-06,
       "logits/chosen": -11.125,
       "logits/rejected": -11.0,
       "logps/chosen": -404.0,
       "logps/rejected": -456.0,
-      "loss": 0.6909,
+      "loss": 0.691,
       "rewards/accuracies": 0.2874999940395355,
-      "rewards/chosen": -0.0014190673828125,
-      "rewards/margins": 0.0052490234375,
-      "rewards/rejected": -0.006683349609375,
+      "rewards/chosen": -0.00141143798828125,
+      "rewards/margins": 0.0042724609375,
+      "rewards/rejected": -0.00567626953125,
       "step": 20
     },
     {
       "epoch": 0.031413612565445025,
-      "grad_norm": 1.4994316123519413,
+      "grad_norm": 1.502524155799114,
       "learning_rate": 5.226480836236934e-06,
       "logits/chosen": -11.5,
       "logits/rejected": -11.625,
       "logps/chosen": -422.0,
       "logps/rejected": -466.0,
-      "loss": 0.6871,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": -0.021240234375,
-      "rewards/margins": 0.0081787109375,
-      "rewards/rejected": -0.0294189453125,
+      "loss": 0.6872,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.0201416015625,
+      "rewards/margins": 0.01007080078125,
+      "rewards/rejected": -0.0302734375,
       "step": 30
     },
     {
       "epoch": 0.041884816753926704,
-      "grad_norm": 1.8563360945823957,
+      "grad_norm": 1.855809629028338,
       "learning_rate": 6.968641114982578e-06,
       "logits/chosen": -12.1875,
       "logits/rejected": -11.9375,
       "logps/chosen": -442.0,
       "logps/rejected": -536.0,
-      "loss": 0.6719,
-      "rewards/accuracies": 0.5874999761581421,
+      "loss": 0.6722,
+      "rewards/accuracies": 0.59375,
       "rewards/chosen": -0.125,
-      "rewards/margins": 0.053955078125,
-      "rewards/rejected": -0.1796875,
+      "rewards/margins": 0.054443359375,
+      "rewards/rejected": -0.1787109375,
       "step": 40
     },
     {
       "epoch": 0.05235602094240838,
-      "grad_norm": 2.0897884802891613,
+      "grad_norm": 2.0883651138973085,
       "learning_rate": 8.710801393728225e-06,
       "logits/chosen": -12.25,
       "logits/rejected": -12.1875,
       "logps/chosen": -450.0,
       "logps/rejected": -592.0,
-      "loss": 0.6403,
-      "rewards/accuracies": 0.6499999761581421,
+      "loss": 0.6404,
+      "rewards/accuracies": 0.65625,
       "rewards/chosen": -0.3203125,
-      "rewards/margins": 0.1669921875,
-      "rewards/rejected": -0.486328125,
+      "rewards/margins": 0.1689453125,
+      "rewards/rejected": -0.48828125,
       "step": 50
     },
     {
       "epoch": 0.06282722513089005,
-      "grad_norm": 2.149489116406211,
+      "grad_norm": 2.142726752534785,
       "learning_rate": 1.0452961672473868e-05,
-      "logits/chosen": -11.1875,
+      "logits/chosen": -11.125,
       "logits/rejected": -10.9375,
       "logps/chosen": -396.0,
       "logps/rejected": -644.0,
-      "loss": 0.6031,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -0.4453125,
-      "rewards/margins": 0.369140625,
-      "rewards/rejected": -0.8125,
+      "loss": 0.6024,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.447265625,
+      "rewards/margins": 0.37109375,
+      "rewards/rejected": -0.81640625,
       "step": 60
     },
     {
       "epoch": 0.07329842931937172,
-      "grad_norm": 2.2708959246495724,
+      "grad_norm": 2.270213750063347,
       "learning_rate": 1.2195121951219513e-05,
       "logits/chosen": -11.5,
-      "logits/rejected": -11.3125,
+      "logits/rejected": -11.375,
       "logps/chosen": -516.0,
       "logps/rejected": -628.0,
-      "loss": 0.6005,
-      "rewards/accuracies": 0.65625,
-      "rewards/chosen": -0.8203125,
+      "loss": 0.6009,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -0.82421875,
       "rewards/margins": 0.27734375,
       "rewards/rejected": -1.1015625,
       "step": 70
     },
     {
       "epoch": 0.08376963350785341,
-      "grad_norm": 2.0939493530760336,
+      "grad_norm": 2.0837915060497463,
       "learning_rate": 1.3937282229965156e-05,
       "logits/chosen": -11.875,
-      "logits/rejected": -11.8125,
+      "logits/rejected": -11.875,
       "logps/chosen": -490.0,
       "logps/rejected": -716.0,
       "loss": 0.5789,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -0.96875,
-      "rewards/margins": 0.7578125,
-      "rewards/rejected": -1.7265625,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -0.9609375,
+      "rewards/margins": 0.75390625,
+      "rewards/rejected": -1.7109375,
       "step": 80
     },
     {
       "epoch": 0.09424083769633508,
-      "grad_norm": 2.2347910797854764,
+      "grad_norm": 2.468707954297056,
       "learning_rate": 1.56794425087108e-05,
       "logits/chosen": -11.9375,
       "logits/rejected": -11.9375,
-      "logps/chosen": -496.0,
-      "logps/rejected": -636.0,
-      "loss": 0.5646,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -1.3359375,
-      "rewards/margins": 0.5390625,
-      "rewards/rejected": -1.875,
+      "logps/chosen": -500.0,
+      "logps/rejected": -640.0,
+      "loss": 0.5661,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -1.375,
+      "rewards/margins": 0.52734375,
+      "rewards/rejected": -1.90625,
       "step": 90
     },
     {
       "epoch": 0.10471204188481675,
-      "grad_norm": 2.5627865438545707,
+      "grad_norm": 2.258005630918792,
       "learning_rate": 1.742160278745645e-05,
-      "logits/chosen": -12.125,
-      "logits/rejected": -12.0,
-      "logps/chosen": -644.0,
-      "logps/rejected": -792.0,
-      "loss": 0.5954,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -1.6953125,
-      "rewards/margins": 0.65625,
-      "rewards/rejected": -2.34375,
+      "logits/chosen": -12.5,
+      "logits/rejected": -12.375,
+      "logps/chosen": -636.0,
+      "logps/rejected": -780.0,
+      "loss": 0.5933,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -1.6171875,
+      "rewards/margins": 0.6328125,
+      "rewards/rejected": -2.25,
       "step": 100
     },
     {
       "epoch": 0.10471204188481675,
-      "eval_logits/chosen": -11.0625,
-      "eval_logits/rejected": -10.75,
-      "eval_logps/chosen": -430.0,
-      "eval_logps/rejected": -414.0,
-      "eval_loss": 0.6902187466621399,
-      "eval_rewards/accuracies": 0.57421875,
-      "eval_rewards/chosen": -1.1171875,
-      "eval_rewards/margins": 0.1416015625,
-      "eval_rewards/rejected": -1.2578125,
-      "eval_runtime": 48.7001,
-      "eval_samples_per_second": 41.068,
-      "eval_steps_per_second": 0.657,
+      "eval_logits/chosen": -11.875,
+      "eval_logits/rejected": -11.5625,
+      "eval_logps/chosen": -434.0,
+      "eval_logps/rejected": -420.0,
+      "eval_loss": 0.6756874918937683,
+      "eval_rewards/accuracies": 0.580078125,
+      "eval_rewards/chosen": -1.15625,
+      "eval_rewards/margins": 0.1611328125,
+      "eval_rewards/rejected": -1.3125,
+      "eval_runtime": 60.3355,
+      "eval_samples_per_second": 33.148,
+      "eval_steps_per_second": 0.53,
       "step": 100
     },
     {
       "epoch": 0.11518324607329843,
-      "grad_norm": 2.223183611822178,
+      "grad_norm": 2.2079303218411304,
       "learning_rate": 1.9163763066202093e-05,
-      "logits/chosen": -10.6875,
-      "logits/rejected": -10.625,
-      "logps/chosen": -568.0,
-      "logps/rejected": -752.0,
-      "loss": 0.5744,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -1.4921875,
-      "rewards/margins": 0.546875,
-      "rewards/rejected": -2.046875,
+      "logits/chosen": -10.875,
+      "logits/rejected": -10.75,
+      "logps/chosen": -584.0,
+      "logps/rejected": -768.0,
+      "loss": 0.5801,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -1.65625,
+      "rewards/margins": 0.5390625,
+      "rewards/rejected": -2.203125,
       "step": 110
     },
     {
       "epoch": 0.1256544502617801,
-      "grad_norm": 2.808900473858744,
+      "grad_norm": 2.6035913965128716,
       "learning_rate": 2.0905923344947736e-05,
-      "logits/chosen": -10.75,
-      "logits/rejected": -10.6875,
-      "logps/chosen": -548.0,
-      "logps/rejected": -696.0,
-      "loss": 0.5553,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -1.7578125,
-      "rewards/margins": 0.609375,
-      "rewards/rejected": -2.375,
+      "logits/chosen": -11.625,
+      "logits/rejected": -11.5625,
+      "logps/chosen": -544.0,
+      "logps/rejected": -688.0,
+      "loss": 0.5524,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -1.7265625,
+      "rewards/margins": 0.58984375,
+      "rewards/rejected": -2.3125,
       "step": 120
     },
     {
       "epoch": 0.13612565445026178,
-      "grad_norm": 2.1401708384915685,
+      "grad_norm": 2.4300718793479223,
       "learning_rate": 2.264808362369338e-05,
-      "logits/chosen": -11.875,
-      "logits/rejected": -11.75,
-      "logps/chosen": -592.0,
-      "logps/rejected": -732.0,
-      "loss": 0.6026,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -1.75,
-      "rewards/margins": 0.494140625,
-      "rewards/rejected": -2.25,
+      "logits/chosen": -12.5625,
+      "logits/rejected": -12.375,
+      "logps/chosen": -632.0,
+      "logps/rejected": -772.0,
+      "loss": 0.6124,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -2.15625,
+      "rewards/margins": 0.50390625,
+      "rewards/rejected": -2.65625,
       "step": 130
     },
     {
       "epoch": 0.14659685863874344,
-      "grad_norm": 2.8783893246311014,
+      "grad_norm": 3.128854965648137,
       "learning_rate": 2.4390243902439026e-05,
-      "logits/chosen": -13.9375,
-      "logits/rejected": -13.5,
-      "logps/chosen": -600.0,
+      "logits/chosen": -13.75,
+      "logits/rejected": -13.375,
+      "logps/chosen": -604.0,
       "logps/rejected": -832.0,
-      "loss": 0.5407,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -1.546875,
-      "rewards/margins": 0.640625,
+      "loss": 0.5503,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -1.59375,
+      "rewards/margins": 0.59375,
       "rewards/rejected": -2.1875,
       "step": 140
     },
     {
       "epoch": 0.15706806282722513,
-      "grad_norm": 2.4716755547029585,
+      "grad_norm": 2.415597330432763,
       "learning_rate": 2.6132404181184672e-05,
-      "logits/chosen": -12.75,
-      "logits/rejected": -12.625,
-      "logps/chosen": -644.0,
-      "logps/rejected": -840.0,
-      "loss": 0.5524,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -2.140625,
-      "rewards/margins": 0.8515625,
-      "rewards/rejected": -3.0,
+      "logits/chosen": -12.25,
+      "logits/rejected": -12.125,
+      "logps/chosen": -608.0,
+      "logps/rejected": -788.0,
+      "loss": 0.5602,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.765625,
+      "rewards/margins": 0.71484375,
+      "rewards/rejected": -2.484375,
       "step": 150
     },
     {
       "epoch": 0.16753926701570682,
-      "grad_norm": 2.375047238648288,
+      "grad_norm": 2.616487528249426,
       "learning_rate": 2.7874564459930312e-05,
-      "logits/chosen": -11.0625,
-      "logits/rejected": -10.8125,
-      "logps/chosen": -652.0,
-      "logps/rejected": -832.0,
-      "loss": 0.5316,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -2.53125,
+      "logits/chosen": -11.375,
+      "logits/rejected": -11.1875,
+      "logps/chosen": -620.0,
+      "logps/rejected": -804.0,
+      "loss": 0.5272,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -2.21875,
       "rewards/margins": 0.7578125,
-      "rewards/rejected": -3.28125,
+      "rewards/rejected": -2.984375,
       "step": 160
     },
     {
       "epoch": 0.17801047120418848,
-      "grad_norm": 2.267338597076113,
+      "grad_norm": 2.5105509691756254,
       "learning_rate": 2.9616724738675962e-05,
-      "logits/chosen": -9.375,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -672.0,
-      "logps/rejected": -856.0,
-      "loss": 0.5116,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -2.5,
-      "rewards/margins": 0.76171875,
-      "rewards/rejected": -3.25,
+      "logits/chosen": -7.65625,
+      "logits/rejected": -7.71875,
+      "logps/chosen": -676.0,
+      "logps/rejected": -872.0,
+      "loss": 0.5266,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -2.5625,
+      "rewards/margins": 0.83984375,
+      "rewards/rejected": -3.40625,
       "step": 170
     },
     {
       "epoch": 0.18848167539267016,
-      "grad_norm": 2.1418594545160867,
+      "grad_norm": 2.09773562490284,
       "learning_rate": 3.13588850174216e-05,
-      "logits/chosen": -8.75,
-      "logits/rejected": -8.5625,
-      "logps/chosen": -744.0,
-      "logps/rejected": -900.0,
-      "loss": 0.5466,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -3.015625,
-      "rewards/margins": 0.95703125,
-      "rewards/rejected": -3.984375,
+      "logits/chosen": -6.65625,
+      "logits/rejected": -6.5,
+      "logps/chosen": -704.0,
+      "logps/rejected": -844.0,
+      "loss": 0.5557,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -2.59375,
+      "rewards/margins": 0.83203125,
+      "rewards/rejected": -3.421875,
       "step": 180
     },
     {
       "epoch": 0.19895287958115182,
-      "grad_norm": 2.5148643119046175,
+      "grad_norm": 2.6999928424432,
       "learning_rate": 3.310104529616725e-05,
-      "logits/chosen": -9.75,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -624.0,
-      "logps/rejected": -828.0,
-      "loss": 0.5441,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -2.28125,
-      "rewards/margins": 0.7421875,
-      "rewards/rejected": -3.015625,
+      "logits/chosen": -8.5625,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -644.0,
+      "logps/rejected": -856.0,
+      "loss": 0.5356,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -2.5,
+      "rewards/margins": 0.76953125,
+      "rewards/rejected": -3.28125,
       "step": 190
     },
     {
       "epoch": 0.2094240837696335,
-      "grad_norm": 2.137445807589994,
+      "grad_norm": 2.3215504459940215,
       "learning_rate": 3.48432055749129e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.6875,
-      "logps/chosen": -696.0,
-      "logps/rejected": -868.0,
-      "loss": 0.5396,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -2.34375,
-      "rewards/margins": 0.79296875,
-      "rewards/rejected": -3.125,
+      "logits/chosen": -9.625,
+      "logits/rejected": -9.5,
+      "logps/chosen": -716.0,
+      "logps/rejected": -888.0,
+      "loss": 0.5649,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -2.515625,
+      "rewards/margins": 0.76953125,
+      "rewards/rejected": -3.28125,
       "step": 200
     },
     {
       "epoch": 0.2094240837696335,
-      "eval_logits/chosen": -10.25,
-      "eval_logits/rejected": -9.6875,
-      "eval_logps/chosen": -572.0,
-      "eval_logps/rejected": -576.0,
-      "eval_loss": 0.6927968859672546,
-      "eval_rewards/accuracies": 0.638671875,
-      "eval_rewards/chosen": -2.515625,
-      "eval_rewards/margins": 0.34375,
-      "eval_rewards/rejected": -2.859375,
-      "eval_runtime": 46.5802,
-      "eval_samples_per_second": 42.937,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -10.5625,
+      "eval_logits/rejected": -10.125,
+      "eval_logps/chosen": -532.0,
+      "eval_logps/rejected": -524.0,
+      "eval_loss": 0.6956562399864197,
+      "eval_rewards/accuracies": 0.607421875,
+      "eval_rewards/chosen": -2.125,
+      "eval_rewards/margins": 0.2451171875,
+      "eval_rewards/rejected": -2.375,
+      "eval_runtime": 58.1658,
+      "eval_samples_per_second": 34.384,
+      "eval_steps_per_second": 0.55,
       "step": 200
     },
     {
       "epoch": 0.2198952879581152,
-      "grad_norm": 2.785740952773544,
+      "grad_norm": 2.542423368401395,
       "learning_rate": 3.6585365853658535e-05,
-      "logits/chosen": -10.75,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -628.0,
-      "logps/rejected": -752.0,
-      "loss": 0.5427,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -2.59375,
-      "rewards/margins": 0.57421875,
-      "rewards/rejected": -3.171875,
+      "logits/chosen": -12.375,
+      "logits/rejected": -12.375,
+      "logps/chosen": -604.0,
+      "logps/rejected": -724.0,
+      "loss": 0.5441,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -2.328125,
+      "rewards/margins": 0.56640625,
+      "rewards/rejected": -2.890625,
       "step": 210
     },
     {
       "epoch": 0.23036649214659685,
-      "grad_norm": 3.8466932995796443,
+      "grad_norm": 2.303882402104715,
       "learning_rate": 3.8327526132404185e-05,
-      "logits/chosen": -10.9375,
-      "logits/rejected": -10.5,
-      "logps/chosen": -720.0,
-      "logps/rejected": -860.0,
-      "loss": 0.5256,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -3.0,
-      "rewards/margins": 0.79296875,
-      "rewards/rejected": -3.796875,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -13.25,
+      "logps/chosen": -632.0,
+      "logps/rejected": -760.0,
+      "loss": 0.5556,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -2.109375,
+      "rewards/margins": 0.6796875,
+      "rewards/rejected": -2.78125,
       "step": 220
     },
     {
       "epoch": 0.24083769633507854,
-      "grad_norm": 2.1524756483339713,
+      "grad_norm": 2.198308510966628,
       "learning_rate": 4.006968641114983e-05,
-      "logits/chosen": -12.4375,
-      "logits/rejected": -12.1875,
-      "logps/chosen": -612.0,
-      "logps/rejected": -764.0,
-      "loss": 0.5518,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -2.25,
-      "rewards/margins": 0.6015625,
-      "rewards/rejected": -2.84375,
+      "logits/chosen": -12.8125,
+      "logits/rejected": -12.5625,
+      "logps/chosen": -568.0,
+      "logps/rejected": -732.0,
+      "loss": 0.5381,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.8203125,
+      "rewards/margins": 0.6953125,
+      "rewards/rejected": -2.515625,
       "step": 230
     },
     {
       "epoch": 0.2513089005235602,
-      "grad_norm": 3.0868896382509026,
+      "grad_norm": 2.4145625295758126,
       "learning_rate": 4.181184668989547e-05,
-      "logits/chosen": -11.0625,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -624.0,
-      "logps/rejected": -788.0,
-      "loss": 0.5699,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -2.5,
-      "rewards/margins": 0.8671875,
-      "rewards/rejected": -3.375,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -10.9375,
+      "logps/chosen": -628.0,
+      "logps/rejected": -780.0,
+      "loss": 0.5674,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -2.515625,
+      "rewards/margins": 0.78125,
+      "rewards/rejected": -3.3125,
       "step": 240
     },
     {
       "epoch": 0.2617801047120419,
-      "grad_norm": 3.191782213350334,
+      "grad_norm": 4.528555474910485,
       "learning_rate": 4.3554006968641115e-05,
-      "logits/chosen": -11.5,
-      "logits/rejected": -11.25,
-      "logps/chosen": -716.0,
-      "logps/rejected": -888.0,
-      "loss": 0.5159,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -2.59375,
-      "rewards/margins": 0.90234375,
-      "rewards/rejected": -3.5,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -10.875,
+      "logps/chosen": -768.0,
+      "logps/rejected": -940.0,
+      "loss": 0.5292,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 0.93359375,
+      "rewards/rejected": -4.03125,
       "step": 250
     },
     {
       "epoch": 0.27225130890052357,
-      "grad_norm": 2.275497361821708,
+      "grad_norm": 2.0388725005890898,
       "learning_rate": 4.529616724738676e-05,
-      "logits/chosen": -10.375,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -660.0,
-      "logps/rejected": -900.0,
-      "loss": 0.5176,
+      "logits/chosen": -12.9375,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -668.0,
+      "logps/rejected": -904.0,
+      "loss": 0.5208,
       "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -2.8125,
-      "rewards/margins": 1.0703125,
-      "rewards/rejected": -3.890625,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 1.03125,
+      "rewards/rejected": -3.9375,
       "step": 260
     },
     {
       "epoch": 0.28272251308900526,
-      "grad_norm": 2.4579342535508015,
+      "grad_norm": 2.1688322701940574,
       "learning_rate": 4.703832752613241e-05,
-      "logits/chosen": -11.0625,
-      "logits/rejected": -11.0,
-      "logps/chosen": -652.0,
-      "logps/rejected": -808.0,
-      "loss": 0.5529,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -2.703125,
-      "rewards/margins": 0.8359375,
-      "rewards/rejected": -3.546875,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -14.5,
+      "logps/chosen": -628.0,
+      "logps/rejected": -780.0,
+      "loss": 0.5595,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -2.46875,
+      "rewards/margins": 0.81640625,
+      "rewards/rejected": -3.28125,
       "step": 270
     },
     {
       "epoch": 0.2931937172774869,
-      "grad_norm": 2.781034792090926,
+      "grad_norm": 2.418255259207659,
       "learning_rate": 4.878048780487805e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -668.0,
-      "logps/rejected": -840.0,
-      "loss": 0.5725,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -3.0,
-      "rewards/margins": 0.7109375,
-      "rewards/rejected": -3.71875,
+      "logits/chosen": -14.0,
+      "logits/rejected": -13.625,
+      "logps/chosen": -624.0,
+      "logps/rejected": -776.0,
+      "loss": 0.585,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -2.53125,
+      "rewards/margins": 0.5546875,
+      "rewards/rejected": -3.09375,
       "step": 280
     },
     {
       "epoch": 0.3036649214659686,
-      "grad_norm": 2.34417331460503,
+      "grad_norm": 2.8079307634829487,
       "learning_rate": 4.999983293481418e-05,
-      "logits/chosen": -11.0625,
-      "logits/rejected": -11.0,
-      "logps/chosen": -728.0,
-      "logps/rejected": -940.0,
-      "loss": 0.5819,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 0.91796875,
-      "rewards/rejected": -4.25,
+      "logits/chosen": -11.25,
+      "logits/rejected": -11.125,
+      "logps/chosen": -780.0,
+      "logps/rejected": -996.0,
+      "loss": 0.541,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -3.859375,
+      "rewards/margins": 0.96875,
+      "rewards/rejected": -4.8125,
       "step": 290
     },
     {
       "epoch": 0.31413612565445026,
-      "grad_norm": 2.274618457020013,
+      "grad_norm": 2.543616191003804,
       "learning_rate": 4.9996862949181595e-05,
-      "logits/chosen": -11.4375,
-      "logits/rejected": -11.125,
-      "logps/chosen": -732.0,
-      "logps/rejected": -888.0,
-      "loss": 0.5587,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 0.734375,
-      "rewards/rejected": -3.96875,
+      "logits/chosen": -9.25,
+      "logits/rejected": -9.125,
+      "logps/chosen": -812.0,
+      "logps/rejected": -992.0,
+      "loss": 0.5247,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -4.0,
+      "rewards/margins": 0.97265625,
+      "rewards/rejected": -4.96875,
       "step": 300
     },
     {
       "epoch": 0.31413612565445026,
-      "eval_logits/chosen": -9.4375,
-      "eval_logits/rejected": -8.5625,
-      "eval_logps/chosen": -692.0,
-      "eval_logps/rejected": -704.0,
-      "eval_loss": 0.7013906240463257,
+      "eval_logits/chosen": -8.4375,
+      "eval_logits/rejected": -7.6875,
+      "eval_logps/chosen": -764.0,
+      "eval_logps/rejected": -784.0,
+      "eval_loss": 0.7604531049728394,
       "eval_rewards/accuracies": 0.609375,
-      "eval_rewards/chosen": -3.71875,
-      "eval_rewards/margins": 0.435546875,
-      "eval_rewards/rejected": -4.15625,
-      "eval_runtime": 46.5851,
-      "eval_samples_per_second": 42.932,
-      "eval_steps_per_second": 0.687,
+      "eval_rewards/chosen": -4.46875,
+      "eval_rewards/margins": 0.498046875,
+      "eval_rewards/rejected": -4.96875,
+      "eval_runtime": 57.8889,
+      "eval_samples_per_second": 34.549,
+      "eval_steps_per_second": 0.553,
       "step": 300
     },
     {
       "epoch": 0.32460732984293195,
-      "grad_norm": 2.4381658132899253,
+      "grad_norm": 2.431816975507577,
       "learning_rate": 4.9990180911526896e-05,
-      "logits/chosen": -10.8125,
-      "logits/rejected": -10.375,
-      "logps/chosen": -708.0,
-      "logps/rejected": -880.0,
-      "loss": 0.5585,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -3.328125,
-      "rewards/margins": 0.6171875,
-      "rewards/rejected": -3.9375,
+      "logits/chosen": -10.0,
+      "logits/rejected": -9.625,
+      "logps/chosen": -744.0,
+      "logps/rejected": -920.0,
+      "loss": 0.5513,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -3.703125,
+      "rewards/margins": 0.66015625,
+      "rewards/rejected": -4.375,
       "step": 310
     },
     {
       "epoch": 0.33507853403141363,
-      "grad_norm": 2.7153479107577994,
+      "grad_norm": 3.2773365464765547,
       "learning_rate": 4.997978781413745e-05,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -11.0,
-      "logps/chosen": -772.0,
-      "logps/rejected": -936.0,
-      "loss": 0.5276,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 0.94140625,
-      "rewards/rejected": -4.21875,
+      "logits/chosen": -8.3125,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -924.0,
+      "logps/rejected": -1096.0,
+      "loss": 0.5635,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -4.8125,
+      "rewards/margins": 1.0234375,
+      "rewards/rejected": -5.84375,
       "step": 320
     },
     {
       "epoch": 0.34554973821989526,
-      "grad_norm": 2.936101111347803,
+      "grad_norm": 2.6770410980204726,
       "learning_rate": 4.996568520039565e-05,
-      "logits/chosen": -12.0,
-      "logits/rejected": -11.75,
-      "logps/chosen": -724.0,
-      "logps/rejected": -936.0,
-      "loss": 0.5479,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -3.078125,
-      "rewards/margins": 1.0078125,
-      "rewards/rejected": -4.09375,
+      "logits/chosen": -9.5625,
+      "logits/rejected": -9.5,
+      "logps/chosen": -708.0,
+      "logps/rejected": -908.0,
+      "loss": 0.5425,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 0.88671875,
+      "rewards/rejected": -3.796875,
       "step": 330
     },
     {
       "epoch": 0.35602094240837695,
-      "grad_norm": 2.710036665403822,
+      "grad_norm": 2.6863112505344175,
       "learning_rate": 4.994787516454973e-05,
-      "logits/chosen": -11.5625,
-      "logits/rejected": -11.125,
-      "logps/chosen": -756.0,
-      "logps/rejected": -916.0,
-      "loss": 0.5253,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 0.92578125,
-      "rewards/rejected": -4.4375,
+      "logits/chosen": -10.4375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -692.0,
+      "logps/rejected": -844.0,
+      "loss": 0.5293,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -2.859375,
+      "rewards/margins": 0.8671875,
+      "rewards/rejected": -3.734375,
       "step": 340
     },
     {
       "epoch": 0.36649214659685864,
-      "grad_norm": 3.508631393073106,
+      "grad_norm": 5.36458802855477,
       "learning_rate": 4.992636035140273e-05,
-      "logits/chosen": -11.375,
-      "logits/rejected": -11.125,
-      "logps/chosen": -680.0,
-      "logps/rejected": -904.0,
-      "loss": 0.5289,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -3.0625,
-      "rewards/margins": 0.9765625,
-      "rewards/rejected": -4.03125,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -11.1875,
+      "logps/chosen": -700.0,
+      "logps/rejected": -920.0,
+      "loss": 0.5206,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 0.9453125,
+      "rewards/rejected": -4.1875,
       "step": 350
     },
     {
       "epoch": 0.3769633507853403,
-      "grad_norm": 3.6772006188904025,
+      "grad_norm": 2.8913015402395947,
       "learning_rate": 4.99011439559198e-05,
-      "logits/chosen": -12.1875,
-      "logits/rejected": -11.5625,
-      "logps/chosen": -796.0,
-      "logps/rejected": -1056.0,
-      "loss": 0.5549,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -3.890625,
-      "rewards/margins": 1.265625,
-      "rewards/rejected": -5.15625,
+      "logits/chosen": -11.25,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -704.0,
+      "logps/rejected": -948.0,
+      "loss": 0.5507,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 1.109375,
+      "rewards/rejected": -4.0625,
       "step": 360
     },
     {
       "epoch": 0.387434554973822,
-      "grad_norm": 2.9271341064184746,
+      "grad_norm": 2.6206693322700545,
       "learning_rate": 4.987222972275369e-05,
-      "logits/chosen": -9.6875,
-      "logits/rejected": -9.125,
-      "logps/chosen": -988.0,
-      "logps/rejected": -1184.0,
-      "loss": 0.5249,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -5.78125,
-      "rewards/margins": 1.1484375,
-      "rewards/rejected": -6.90625,
+      "logits/chosen": -10.875,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -796.0,
+      "logps/rejected": -980.0,
+      "loss": 0.5233,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -3.875,
+      "rewards/margins": 1.015625,
+      "rewards/rejected": -4.875,
       "step": 370
     },
     {
       "epoch": 0.39790575916230364,
-      "grad_norm": 2.0754386224489756,
+      "grad_norm": 2.1946337136524456,
       "learning_rate": 4.983962194568872e-05,
-      "logits/chosen": -14.3125,
-      "logits/rejected": -13.75,
-      "logps/chosen": -808.0,
-      "logps/rejected": -1064.0,
-      "loss": 0.5346,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -3.890625,
-      "rewards/margins": 0.984375,
-      "rewards/rejected": -4.875,
+      "logits/chosen": -13.125,
+      "logits/rejected": -12.625,
+      "logps/chosen": -780.0,
+      "logps/rejected": -1056.0,
+      "loss": 0.5386,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -3.609375,
+      "rewards/margins": 1.15625,
+      "rewards/rejected": -4.78125,
       "step": 380
     },
     {
       "epoch": 0.4083769633507853,
-      "grad_norm": 2.6859015854368185,
+      "grad_norm": 2.7360610102221865,
       "learning_rate": 4.9803325467003084e-05,
-      "logits/chosen": -12.875,
-      "logits/rejected": -12.0625,
-      "logps/chosen": -856.0,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -12.375,
+      "logps/chosen": -852.0,
       "logps/rejected": -1120.0,
-      "loss": 0.4896,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -4.28125,
-      "rewards/margins": 1.265625,
-      "rewards/rejected": -5.53125,
+      "loss": 0.4907,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -4.25,
+      "rewards/margins": 1.3515625,
+      "rewards/rejected": -5.59375,
       "step": 390
     },
     {
       "epoch": 0.418848167539267,
-      "grad_norm": 2.4931447444186134,
+      "grad_norm": 2.6213367488023116,
       "learning_rate": 4.9763345676749816e-05,
-      "logits/chosen": -10.6875,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -908.0,
-      "logps/rejected": -1200.0,
-      "loss": 0.5067,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -5.0625,
-      "rewards/margins": 1.4921875,
-      "rewards/rejected": -6.5625,
+      "logits/chosen": -12.625,
+      "logits/rejected": -12.125,
+      "logps/chosen": -864.0,
+      "logps/rejected": -1168.0,
+      "loss": 0.5128,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -4.625,
+      "rewards/margins": 1.6171875,
+      "rewards/rejected": -6.21875,
       "step": 400
     },
     {
       "epoch": 0.418848167539267,
-      "eval_logits/chosen": -14.9375,
-      "eval_logits/rejected": -14.0,
-      "eval_logps/chosen": -708.0,
-      "eval_logps/rejected": -724.0,
-      "eval_loss": 0.786859393119812,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -3.875,
-      "eval_rewards/margins": 0.49609375,
-      "eval_rewards/rejected": -4.375,
-      "eval_runtime": 46.5924,
-      "eval_samples_per_second": 42.925,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -13.875,
+      "eval_logits/rejected": -13.25,
+      "eval_logps/chosen": -688.0,
+      "eval_logps/rejected": -712.0,
+      "eval_loss": 0.7886562347412109,
+      "eval_rewards/accuracies": 0.59765625,
+      "eval_rewards/chosen": -3.71875,
+      "eval_rewards/margins": 0.51953125,
+      "eval_rewards/rejected": -4.25,
+      "eval_runtime": 57.876,
+      "eval_samples_per_second": 34.557,
+      "eval_steps_per_second": 0.553,
       "step": 400
     },
     {
       "epoch": 0.4293193717277487,
-      "grad_norm": 3.2594470061421648,
+      "grad_norm": 3.128016104116983,
       "learning_rate": 4.9719688511956374e-05,
-      "logits/chosen": -15.0,
-      "logits/rejected": -14.125,
-      "logps/chosen": -824.0,
-      "logps/rejected": -1080.0,
-      "loss": 0.5185,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -4.0,
-      "rewards/margins": 1.1484375,
-      "rewards/rejected": -5.15625,
+      "logits/chosen": -13.25,
+      "logits/rejected": -12.5,
+      "logps/chosen": -832.0,
+      "logps/rejected": -1120.0,
+      "loss": 0.5221,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -4.125,
+      "rewards/margins": 1.4453125,
+      "rewards/rejected": -5.5625,
       "step": 410
     },
     {
       "epoch": 0.4397905759162304,
-      "grad_norm": 2.8223871004392698,
+      "grad_norm": 2.6953602633646527,
       "learning_rate": 4.967236045574294e-05,
-      "logits/chosen": -14.6875,
-      "logits/rejected": -14.0,
-      "logps/chosen": -844.0,
-      "logps/rejected": -1056.0,
-      "loss": 0.4966,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -13.25,
+      "logps/chosen": -752.0,
+      "logps/rejected": -956.0,
+      "loss": 0.5205,
       "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -4.3125,
-      "rewards/margins": 1.109375,
-      "rewards/rejected": -5.4375,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 1.0625,
+      "rewards/rejected": -4.4375,
       "step": 420
     },
     {
       "epoch": 0.450261780104712,
-      "grad_norm": 2.013537296666935,
+      "grad_norm": 2.271679575918279,
       "learning_rate": 4.96213685363597e-05,
-      "logits/chosen": -15.3125,
-      "logits/rejected": -14.875,
-      "logps/chosen": -788.0,
-      "logps/rejected": -964.0,
-      "loss": 0.5546,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -3.875,
-      "rewards/margins": 1.015625,
-      "rewards/rejected": -4.90625,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -13.375,
+      "logps/chosen": -768.0,
+      "logps/rejected": -944.0,
+      "loss": 0.5488,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -3.671875,
+      "rewards/margins": 1.0078125,
+      "rewards/rejected": -4.6875,
       "step": 430
     },
     {
       "epoch": 0.4607329842931937,
-      "grad_norm": 3.3253977907838985,
+      "grad_norm": 8.269373918948345,
       "learning_rate": 4.956672032614315e-05,
-      "logits/chosen": -15.5,
-      "logits/rejected": -15.0625,
-      "logps/chosen": -736.0,
-      "logps/rejected": -928.0,
-      "loss": 0.5176,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -3.234375,
-      "rewards/margins": 0.8671875,
-      "rewards/rejected": -4.09375,
+      "logits/chosen": -13.8125,
+      "logits/rejected": -13.4375,
+      "logps/chosen": -772.0,
+      "logps/rejected": -964.0,
+      "loss": 0.5197,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -3.59375,
+      "rewards/margins": 0.875,
+      "rewards/rejected": -4.46875,
       "step": 440
     },
     {
       "epoch": 0.4712041884816754,
-      "grad_norm": 2.5916430921680718,
+      "grad_norm": 2.805706609172665,
       "learning_rate": 4.950842394039157e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -15.625,
-      "logps/chosen": -772.0,
-      "logps/rejected": -972.0,
-      "loss": 0.5184,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -3.875,
-      "rewards/margins": 0.99609375,
-      "rewards/rejected": -4.875,
+      "logits/chosen": -14.0,
+      "logits/rejected": -13.5,
+      "logps/chosen": -840.0,
+      "logps/rejected": -1040.0,
+      "loss": 0.5281,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 1.0546875,
+      "rewards/rejected": -5.5625,
       "step": 450
     },
     {
       "epoch": 0.4816753926701571,
-      "grad_norm": 2.2880461037901316,
+      "grad_norm": 2.4723842970880088,
       "learning_rate": 4.944648803615994e-05,
-      "logits/chosen": -14.625,
-      "logits/rejected": -13.9375,
-      "logps/chosen": -864.0,
-      "logps/rejected": -1056.0,
-      "loss": 0.5358,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -4.46875,
-      "rewards/margins": 1.140625,
-      "rewards/rejected": -5.59375,
+      "logits/chosen": -16.25,
+      "logits/rejected": -15.9375,
+      "logps/chosen": -732.0,
+      "logps/rejected": -916.0,
+      "loss": 0.5342,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 1.0546875,
+      "rewards/rejected": -4.21875,
       "step": 460
     },
     {
       "epoch": 0.49214659685863876,
-      "grad_norm": 3.7177339388307566,
+      "grad_norm": 2.5900979652415,
       "learning_rate": 4.938092181097433e-05,
-      "logits/chosen": -11.5,
-      "logits/rejected": -10.8125,
-      "logps/chosen": -868.0,
-      "logps/rejected": -1064.0,
-      "loss": 0.4827,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -4.65625,
-      "rewards/margins": 1.2109375,
-      "rewards/rejected": -5.875,
+      "logits/chosen": -14.8125,
+      "logits/rejected": -14.5625,
+      "logps/chosen": -864.0,
+      "logps/rejected": -1048.0,
+      "loss": 0.4896,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -4.625,
+      "rewards/margins": 1.1015625,
+      "rewards/rejected": -5.71875,
       "step": 470
     },
     {
       "epoch": 0.5026178010471204,
-      "grad_norm": 3.408297840927194,
+      "grad_norm": 4.721575699261044,
       "learning_rate": 4.931173500146607e-05,
-      "logits/chosen": -10.4375,
-      "logits/rejected": -10.125,
-      "logps/chosen": -880.0,
-      "logps/rejected": -1152.0,
-      "loss": 0.5364,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -4.9375,
-      "rewards/margins": 1.4609375,
-      "rewards/rejected": -6.40625,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -13.125,
+      "logps/chosen": -812.0,
+      "logps/rejected": -1080.0,
+      "loss": 0.5083,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -4.25,
+      "rewards/margins": 1.453125,
+      "rewards/rejected": -5.6875,
       "step": 480
     },
     {
       "epoch": 0.5130890052356021,
-      "grad_norm": 3.119454176443585,
+      "grad_norm": 2.6531690886330836,
       "learning_rate": 4.923893788192582e-05,
-      "logits/chosen": -12.0625,
-      "logits/rejected": -11.4375,
-      "logps/chosen": -784.0,
-      "logps/rejected": -964.0,
-      "loss": 0.502,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -3.640625,
-      "rewards/margins": 1.1171875,
-      "rewards/rejected": -4.75,
+      "logits/chosen": -12.5,
+      "logits/rejected": -11.75,
+      "logps/chosen": -816.0,
+      "logps/rejected": -1004.0,
+      "loss": 0.5168,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -3.953125,
+      "rewards/margins": 1.203125,
+      "rewards/rejected": -5.15625,
       "step": 490
     },
     {
       "epoch": 0.5235602094240838,
-      "grad_norm": 3.3625359520874136,
+      "grad_norm": 3.0530164639159447,
       "learning_rate": 4.9162541262777906e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -816.0,
-      "logps/rejected": -1032.0,
-      "loss": 0.525,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -3.9375,
-      "rewards/margins": 1.0390625,
-      "rewards/rejected": -4.96875,
+      "logits/chosen": -14.0,
+      "logits/rejected": -13.4375,
+      "logps/chosen": -776.0,
+      "logps/rejected": -988.0,
+      "loss": 0.5048,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 1.0234375,
+      "rewards/rejected": -4.5625,
       "step": 500
     },
     {
       "epoch": 0.5235602094240838,
-      "eval_logits/chosen": -8.25,
-      "eval_logits/rejected": -6.84375,
-      "eval_logps/chosen": -780.0,
-      "eval_logps/rejected": -824.0,
-      "eval_loss": 0.849984347820282,
-      "eval_rewards/accuracies": 0.609375,
-      "eval_rewards/chosen": -4.625,
-      "eval_rewards/margins": 0.7109375,
-      "eval_rewards/rejected": -5.34375,
-      "eval_runtime": 46.5938,
-      "eval_samples_per_second": 42.924,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -11.8125,
+      "eval_logits/rejected": -10.6875,
+      "eval_logps/chosen": -744.0,
+      "eval_logps/rejected": -772.0,
+      "eval_loss": 0.7559531331062317,
+      "eval_rewards/accuracies": 0.630859375,
+      "eval_rewards/chosen": -4.25,
+      "eval_rewards/margins": 0.60546875,
+      "eval_rewards/rejected": -4.84375,
+      "eval_runtime": 57.8902,
+      "eval_samples_per_second": 34.548,
+      "eval_steps_per_second": 0.553,
       "step": 500
     },
     {
       "epoch": 0.5340314136125655,
-      "grad_norm": 2.381702575123,
+      "grad_norm": 2.8686473768759666,
       "learning_rate": 4.90825564889749e-05,
-      "logits/chosen": -9.875,
-      "logits/rejected": -9.625,
-      "logps/chosen": -832.0,
-      "logps/rejected": -1020.0,
-      "loss": 0.5124,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -3.921875,
-      "rewards/margins": 1.125,
-      "rewards/rejected": -5.0625,
+      "logits/chosen": -11.25,
+      "logits/rejected": -10.9375,
+      "logps/chosen": -844.0,
+      "logps/rejected": -1024.0,
+      "loss": 0.5099,
+      "rewards/accuracies": 0.793749988079071,
+      "rewards/chosen": -4.0625,
+      "rewards/margins": 1.03125,
+      "rewards/rejected": -5.09375,
       "step": 510
     },
     {
       "epoch": 0.5445026178010471,
-      "grad_norm": 2.4218657069950584,
+      "grad_norm": 2.9172452077419475,
       "learning_rate": 4.899899543831288e-05,
-      "logits/chosen": -8.9375,
-      "logits/rejected": -8.25,
-      "logps/chosen": -948.0,
-      "logps/rejected": -1168.0,
-      "loss": 0.4881,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -5.03125,
-      "rewards/margins": 1.203125,
-      "rewards/rejected": -6.21875,
+      "logits/chosen": -11.6875,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -816.0,
+      "logps/rejected": -1048.0,
+      "loss": 0.4807,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -3.734375,
+      "rewards/margins": 1.3203125,
+      "rewards/rejected": -5.0625,
       "step": 520
     },
     {
       "epoch": 0.5549738219895288,
-      "grad_norm": 3.0828850263323684,
+      "grad_norm": 3.8808187060714467,
       "learning_rate": 4.891187051966765e-05,
-      "logits/chosen": -10.125,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -884.0,
-      "logps/rejected": -1016.0,
-      "loss": 0.5212,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -4.53125,
-      "rewards/margins": 0.859375,
-      "rewards/rejected": -5.375,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -12.875,
+      "logps/chosen": -808.0,
+      "logps/rejected": -956.0,
+      "loss": 0.5162,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -3.765625,
+      "rewards/margins": 1.0546875,
+      "rewards/rejected": -4.8125,
       "step": 530
     },
     {
       "epoch": 0.5654450261780105,
-      "grad_norm": 3.275667584563022,
+      "grad_norm": 3.6737750350820284,
       "learning_rate": 4.882119467115191e-05,
-      "logits/chosen": -11.875,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -736.0,
-      "logps/rejected": -964.0,
-      "loss": 0.5403,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 1.0390625,
-      "rewards/rejected": -4.40625,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -13.3125,
+      "logps/chosen": -900.0,
+      "logps/rejected": -1128.0,
+      "loss": 0.5531,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -5.0,
+      "rewards/margins": 1.03125,
+      "rewards/rejected": -6.03125,
       "step": 540
     },
     {
       "epoch": 0.5759162303664922,
-      "grad_norm": 2.697007031543629,
+      "grad_norm": 2.6830161938866994,
       "learning_rate": 4.872698135819402e-05,
-      "logits/chosen": -11.1875,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -808.0,
-      "logps/rejected": -1040.0,
-      "loss": 0.5273,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -3.90625,
-      "rewards/margins": 1.0546875,
-      "rewards/rejected": -4.96875,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -14.0,
+      "logps/chosen": -844.0,
+      "logps/rejected": -1072.0,
+      "loss": 0.5238,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -4.3125,
+      "rewards/margins": 0.921875,
+      "rewards/rejected": -5.21875,
       "step": 550
     },
     {
       "epoch": 0.5863874345549738,
-      "grad_norm": 3.279372549268849,
+      "grad_norm": 3.4941946648351356,
       "learning_rate": 4.862924457153831e-05,
-      "logits/chosen": -9.1875,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -852.0,
-      "logps/rejected": -1104.0,
-      "loss": 0.4897,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -5.09375,
-      "rewards/margins": 1.1953125,
-      "rewards/rejected": -6.28125,
+      "logits/chosen": -14.0625,
+      "logits/rejected": -13.1875,
+      "logps/chosen": -744.0,
+      "logps/rejected": -1004.0,
+      "loss": 0.4856,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -4.0,
+      "rewards/margins": 1.2890625,
+      "rewards/rejected": -5.28125,
       "step": 560
     },
     {
       "epoch": 0.5968586387434555,
-      "grad_norm": 2.255169684786711,
+      "grad_norm": 2.493423532661518,
       "learning_rate": 4.8527998825167484e-05,
-      "logits/chosen": -8.0625,
-      "logits/rejected": -8.0,
-      "logps/chosen": -1000.0,
-      "logps/rejected": -1224.0,
-      "loss": 0.4724,
+      "logits/chosen": -13.375,
+      "logits/rejected": -12.875,
+      "logps/chosen": -832.0,
+      "logps/rejected": -1048.0,
+      "loss": 0.4793,
       "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -6.15625,
-      "rewards/margins": 1.421875,
-      "rewards/rejected": -7.59375,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 1.3671875,
+      "rewards/rejected": -5.875,
       "step": 570
     },
     {
       "epoch": 0.6073298429319371,
-      "grad_norm": 2.4132308024639038,
+      "grad_norm": 3.469573192238688,
       "learning_rate": 4.842325915414728e-05,
-      "logits/chosen": -10.375,
-      "logits/rejected": -10.0,
-      "logps/chosen": -856.0,
-      "logps/rejected": -1136.0,
-      "loss": 0.5036,
+      "logits/chosen": -15.0625,
+      "logits/rejected": -14.3125,
+      "logps/chosen": -768.0,
+      "logps/rejected": -1048.0,
+      "loss": 0.5095,
       "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -4.8125,
+      "rewards/chosen": -3.953125,
       "rewards/margins": 1.5859375,
-      "rewards/rejected": -6.40625,
+      "rewards/rejected": -5.53125,
       "step": 580
     },
     {
       "epoch": 0.6178010471204188,
-      "grad_norm": 3.2730968506575446,
+      "grad_norm": 3.7877896440572343,
       "learning_rate": 4.831504111239373e-05,
-      "logits/chosen": -11.5,
-      "logits/rejected": -10.875,
-      "logps/chosen": -972.0,
-      "logps/rejected": -1256.0,
-      "loss": 0.4879,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -5.53125,
-      "rewards/margins": 1.609375,
-      "rewards/rejected": -7.125,
+      "logits/chosen": -13.625,
+      "logits/rejected": -12.6875,
+      "logps/chosen": -916.0,
+      "logps/rejected": -1200.0,
+      "loss": 0.4926,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 1.5859375,
+      "rewards/rejected": -6.5625,
       "step": 590
     },
     {
       "epoch": 0.6282722513089005,
-      "grad_norm": 2.5844217364618784,
+      "grad_norm": 2.7632660925150576,
       "learning_rate": 4.8203360770363425e-05,
-      "logits/chosen": -10.8125,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -1320.0,
-      "loss": 0.5173,
+      "logits/chosen": -14.75,
+      "logits/rejected": -13.625,
+      "logps/chosen": -908.0,
+      "logps/rejected": -1160.0,
+      "loss": 0.4935,
       "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -6.59375,
-      "rewards/margins": 1.515625,
-      "rewards/rejected": -8.125,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 1.6484375,
+      "rewards/rejected": -6.53125,
       "step": 600
     },
     {
       "epoch": 0.6282722513089005,
-      "eval_logits/chosen": -10.0,
-      "eval_logits/rejected": -8.875,
-      "eval_logps/chosen": -944.0,
-      "eval_logps/rejected": -960.0,
-      "eval_loss": 0.7292031049728394,
-      "eval_rewards/accuracies": 0.572265625,
-      "eval_rewards/chosen": -6.25,
-      "eval_rewards/margins": 0.46875,
-      "eval_rewards/rejected": -6.71875,
-      "eval_runtime": 46.6021,
-      "eval_samples_per_second": 42.917,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -14.875,
+      "eval_logits/rejected": -14.1875,
+      "eval_logps/chosen": -784.0,
+      "eval_logps/rejected": -800.0,
+      "eval_loss": 0.7499843835830688,
+      "eval_rewards/accuracies": 0.578125,
+      "eval_rewards/chosen": -4.65625,
+      "eval_rewards/margins": 0.447265625,
+      "eval_rewards/rejected": -5.09375,
+      "eval_runtime": 57.8679,
+      "eval_samples_per_second": 34.561,
+      "eval_steps_per_second": 0.553,
       "step": 600
     },
     {
       "epoch": 0.6387434554973822,
-      "grad_norm": 2.8607736256068312,
+      "grad_norm": 2.7964885899745466,
       "learning_rate": 4.808823471266702e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.375,
-      "logps/chosen": -948.0,
-      "logps/rejected": -1168.0,
-      "loss": 0.4909,
+      "logits/chosen": -15.125,
+      "logits/rejected": -14.4375,
+      "logps/chosen": -864.0,
+      "logps/rejected": -1088.0,
+      "loss": 0.5046,
       "rewards/accuracies": 0.75,
-      "rewards/chosen": -5.46875,
-      "rewards/margins": 1.203125,
-      "rewards/rejected": -6.65625,
+      "rewards/chosen": -4.59375,
+      "rewards/margins": 1.3046875,
+      "rewards/rejected": -5.875,
       "step": 610
     },
     {
       "epoch": 0.6492146596858639,
-      "grad_norm": 2.5521045206372035,
+      "grad_norm": 2.2217185425883508,
       "learning_rate": 4.796968003560639e-05,
-      "logits/chosen": -12.625,
-      "logits/rejected": -11.75,
-      "logps/chosen": -984.0,
-      "logps/rejected": -1280.0,
-      "loss": 0.5128,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -5.78125,
-      "rewards/margins": 1.4921875,
-      "rewards/rejected": -7.28125,
+      "logits/chosen": -16.375,
+      "logits/rejected": -15.75,
+      "logps/chosen": -892.0,
+      "logps/rejected": -1176.0,
+      "loss": 0.5293,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -4.84375,
+      "rewards/margins": 1.375,
+      "rewards/rejected": -6.21875,
       "step": 620
     },
     {
       "epoch": 0.6596858638743456,
-      "grad_norm": 2.100242351111436,
+      "grad_norm": 2.09077116869423,
       "learning_rate": 4.7847714344635864e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -15.4375,
-      "logps/chosen": -828.0,
-      "logps/rejected": -1004.0,
-      "loss": 0.5078,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -4.40625,
-      "rewards/margins": 1.1015625,
-      "rewards/rejected": -5.5,
+      "logits/chosen": -17.625,
+      "logits/rejected": -17.125,
+      "logps/chosen": -788.0,
+      "logps/rejected": -964.0,
+      "loss": 0.5002,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -3.96875,
+      "rewards/margins": 1.1171875,
+      "rewards/rejected": -5.09375,
       "step": 630
     },
     {
       "epoch": 0.6701570680628273,
-      "grad_norm": 2.88389896572159,
+      "grad_norm": 3.469256715668109,
       "learning_rate": 4.772235575174776e-05,
-      "logits/chosen": -14.6875,
-      "logits/rejected": -13.8125,
-      "logps/chosen": -996.0,
-      "logps/rejected": -1296.0,
-      "loss": 0.4789,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -5.59375,
-      "rewards/margins": 1.6171875,
-      "rewards/rejected": -7.21875,
+      "logits/chosen": -16.125,
+      "logits/rejected": -15.3125,
+      "logps/chosen": -956.0,
+      "logps/rejected": -1240.0,
+      "loss": 0.4807,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -5.1875,
+      "rewards/margins": 1.4375,
+      "rewards/rejected": -6.625,
       "step": 640
     },
     {
       "epoch": 0.680628272251309,
-      "grad_norm": 2.5649835223172537,
+      "grad_norm": 2.426539705044726,
       "learning_rate": 4.7593622872782754e-05,
-      "logits/chosen": -11.625,
-      "logits/rejected": -10.875,
-      "logps/chosen": -1008.0,
-      "logps/rejected": -1224.0,
-      "loss": 0.5015,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -6.28125,
-      "rewards/margins": 1.046875,
-      "rewards/rejected": -7.3125,
+      "logits/chosen": -14.375,
+      "logits/rejected": -13.75,
+      "logps/chosen": -880.0,
+      "logps/rejected": -1080.0,
+      "loss": 0.5056,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 0.92578125,
+      "rewards/rejected": -5.875,
       "step": 650
     },
     {
       "epoch": 0.6910994764397905,
-      "grad_norm": 2.2912862379158616,
+      "grad_norm": 1.9379816129169367,
       "learning_rate": 4.746153482466542e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -10.25,
-      "logps/chosen": -1012.0,
-      "logps/rejected": -1184.0,
-      "loss": 0.4881,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -6.15625,
-      "rewards/margins": 1.1171875,
-      "rewards/rejected": -7.28125,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -14.0,
+      "logps/chosen": -828.0,
+      "logps/rejected": -976.0,
+      "loss": 0.4996,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -4.3125,
+      "rewards/margins": 0.875,
+      "rewards/rejected": -5.1875,
       "step": 660
     },
     {
       "epoch": 0.7015706806282722,
-      "grad_norm": 2.2914271293332886,
+      "grad_norm": 2.2872104225745344,
       "learning_rate": 4.732611122256537e-05,
-      "logits/chosen": -12.0,
-      "logits/rejected": -11.1875,
-      "logps/chosen": -996.0,
-      "logps/rejected": -1200.0,
-      "loss": 0.4787,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -5.65625,
-      "rewards/margins": 1.3125,
-      "rewards/rejected": -6.96875,
+      "logits/chosen": -15.0,
+      "logits/rejected": -14.25,
+      "logps/chosen": -852.0,
+      "logps/rejected": -1040.0,
+      "loss": 0.4638,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -4.1875,
+      "rewards/margins": 1.203125,
+      "rewards/rejected": -5.40625,
       "step": 670
     },
     {
       "epoch": 0.7120418848167539,
-      "grad_norm": 2.4054534560605676,
+      "grad_norm": 2.434553671885649,
       "learning_rate": 4.718737217698435e-05,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -10.6875,
-      "logps/chosen": -876.0,
-      "logps/rejected": -1072.0,
-      "loss": 0.5335,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -4.96875,
-      "rewards/margins": 0.99609375,
-      "rewards/rejected": -5.96875,
-      "step": 680
+      "logits/chosen": -14.75,
+      "logits/rejected": -13.8125,
+      "logps/chosen": -800.0,
+      "logps/rejected": -996.0,
+      "loss": 0.534,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -4.21875,
+      "rewards/margins": 1.0390625,
+      "rewards/rejected": -5.25,
+      "step": 680
     },
     {
       "epoch": 0.7225130890052356,
-      "grad_norm": 2.815491143179486,
+      "grad_norm": 2.6969988740315793,
       "learning_rate": 4.704533829076982e-05,
-      "logits/chosen": -11.125,
-      "logits/rejected": -10.0,
-      "logps/chosen": -840.0,
-      "logps/rejected": -1184.0,
-      "loss": 0.463,
-      "rewards/accuracies": 0.831250011920929,
-      "rewards/chosen": -4.40625,
-      "rewards/margins": 1.671875,
-      "rewards/rejected": -6.0625,
+      "logits/chosen": -15.125,
+      "logits/rejected": -14.0,
+      "logps/chosen": -848.0,
+      "logps/rejected": -1192.0,
+      "loss": 0.4577,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -4.46875,
+      "rewards/margins": 1.6796875,
+      "rewards/rejected": -6.125,
       "step": 690
     },
     {
       "epoch": 0.7329842931937173,
-      "grad_norm": 2.8274069934382697,
+      "grad_norm": 2.6930106582605227,
       "learning_rate": 4.6900030656055455e-05,
-      "logits/chosen": -9.375,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -840.0,
-      "logps/rejected": -1048.0,
-      "loss": 0.4944,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -4.25,
-      "rewards/margins": 1.1484375,
-      "rewards/rejected": -5.375,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -13.6875,
+      "logps/chosen": -900.0,
+      "logps/rejected": -1104.0,
+      "loss": 0.4879,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -4.8125,
+      "rewards/margins": 1.1328125,
+      "rewards/rejected": -5.96875,
       "step": 700
     },
     {
       "epoch": 0.7329842931937173,
-      "eval_logits/chosen": -8.0,
-      "eval_logits/rejected": -6.40625,
-      "eval_logps/chosen": -832.0,
+      "eval_logits/chosen": -13.8125,
+      "eval_logits/rejected": -12.5,
+      "eval_logps/chosen": -828.0,
       "eval_logps/rejected": -868.0,
-      "eval_loss": 0.7880781292915344,
-      "eval_rewards/accuracies": 0.603515625,
-      "eval_rewards/chosen": -5.15625,
-      "eval_rewards/margins": 0.64453125,
+      "eval_loss": 0.7732343673706055,
+      "eval_rewards/accuracies": 0.623046875,
+      "eval_rewards/chosen": -5.09375,
+      "eval_rewards/margins": 0.6796875,
       "eval_rewards/rejected": -5.78125,
-      "eval_runtime": 46.5807,
-      "eval_samples_per_second": 42.936,
-      "eval_steps_per_second": 0.687,
+      "eval_runtime": 57.9869,
+      "eval_samples_per_second": 34.491,
+      "eval_steps_per_second": 0.552,
       "step": 700
     },
     {
       "epoch": 0.743455497382199,
-      "grad_norm": 2.9362926685321957,
+      "grad_norm": 3.7380948524511517,
       "learning_rate": 4.6751470851128906e-05,
-      "logits/chosen": -10.25,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -872.0,
-      "logps/rejected": -1096.0,
-      "loss": 0.4684,
+      "logits/chosen": -13.625,
+      "logits/rejected": -12.375,
+      "logps/chosen": -892.0,
+      "logps/rejected": -1120.0,
+      "loss": 0.4739,
       "rewards/accuracies": 0.75,
-      "rewards/chosen": -4.375,
-      "rewards/margins": 1.3359375,
-      "rewards/rejected": -5.71875,
+      "rewards/chosen": -4.5625,
+      "rewards/margins": 1.3515625,
+      "rewards/rejected": -5.90625,
       "step": 710
     },
     {
       "epoch": 0.7539267015706806,
-      "grad_norm": 2.302381436322528,
+      "grad_norm": 2.520017019528966,
       "learning_rate": 4.6599680937227454e-05,
-      "logits/chosen": -10.5625,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -856.0,
-      "logps/rejected": -1080.0,
-      "loss": 0.4686,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -4.3125,
-      "rewards/margins": 1.4140625,
-      "rewards/rejected": -5.71875,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -12.125,
+      "logps/chosen": -924.0,
+      "logps/rejected": -1144.0,
+      "loss": 0.4781,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 1.4453125,
+      "rewards/rejected": -6.40625,
       "step": 720
     },
     {
       "epoch": 0.7643979057591623,
-      "grad_norm": 2.8863256594220235,
+      "grad_norm": 2.8266851785758216,
       "learning_rate": 4.644468345526183e-05,
-      "logits/chosen": -9.875,
-      "logits/rejected": -9.25,
-      "logps/chosen": -952.0,
-      "logps/rejected": -1272.0,
-      "loss": 0.507,
+      "logits/chosen": -14.25,
+      "logits/rejected": -13.625,
+      "logps/chosen": -884.0,
+      "logps/rejected": -1184.0,
+      "loss": 0.5196,
       "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -5.40625,
-      "rewards/margins": 1.6796875,
-      "rewards/rejected": -7.0625,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 1.453125,
+      "rewards/rejected": -6.1875,
       "step": 730
     },
     {
       "epoch": 0.774869109947644,
-      "grad_norm": 2.6253706594055104,
+      "grad_norm": 3.238588562409636,
       "learning_rate": 4.628650142246894e-05,
-      "logits/chosen": -9.5,
-      "logits/rejected": -8.875,
-      "logps/chosen": -884.0,
-      "logps/rejected": -1112.0,
-      "loss": 0.5276,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -4.96875,
-      "rewards/margins": 1.34375,
-      "rewards/rejected": -6.3125,
+      "logits/chosen": -14.125,
+      "logits/rejected": -13.4375,
+      "logps/chosen": -852.0,
+      "logps/rejected": -1080.0,
+      "loss": 0.5057,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 1.265625,
+      "rewards/rejected": -5.9375,
       "step": 740
     },
     {
       "epoch": 0.7853403141361257,
-      "grad_norm": 2.6943779399491388,
+      "grad_norm": 2.6770722832147,
       "learning_rate": 4.6125158328993765e-05,
-      "logits/chosen": -11.0625,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -792.0,
-      "logps/rejected": -1012.0,
-      "loss": 0.4943,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -4.03125,
-      "rewards/margins": 0.87890625,
-      "rewards/rejected": -4.90625,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -12.375,
+      "logps/chosen": -872.0,
+      "logps/rejected": -1096.0,
+      "loss": 0.4977,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -4.84375,
+      "rewards/margins": 0.9140625,
+      "rewards/rejected": -5.75,
       "step": 750
     },
     {
       "epoch": 0.7958115183246073,
-      "grad_norm": 2.607331422722883,
+      "grad_norm": 2.395883433797317,
       "learning_rate": 4.5960678134401064e-05,
-      "logits/chosen": -10.9375,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -868.0,
-      "logps/rejected": -1136.0,
-      "loss": 0.4627,
-      "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -4.53125,
-      "rewards/margins": 1.4609375,
-      "rewards/rejected": -6.0,
+      "logits/chosen": -14.375,
+      "logits/rejected": -13.375,
+      "logps/chosen": -904.0,
+      "logps/rejected": -1176.0,
+      "loss": 0.4721,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 1.46875,
+      "rewards/rejected": -6.34375,
       "step": 760
     },
     {
       "epoch": 0.806282722513089,
-      "grad_norm": 3.452760958521426,
+      "grad_norm": 3.730763590305456,
       "learning_rate": 4.579308526411734e-05,
-      "logits/chosen": -7.875,
-      "logits/rejected": -7.28125,
-      "logps/chosen": -956.0,
-      "logps/rejected": -1168.0,
-      "loss": 0.4596,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -5.5,
-      "rewards/margins": 1.4296875,
-      "rewards/rejected": -6.9375,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -13.4375,
+      "logps/chosen": -980.0,
+      "logps/rejected": -1208.0,
+      "loss": 0.4523,
+      "rewards/accuracies": 0.8187500238418579,
+      "rewards/chosen": -5.75,
+      "rewards/margins": 1.546875,
+      "rewards/rejected": -7.28125,
       "step": 770
     },
     {
       "epoch": 0.8167539267015707,
-      "grad_norm": 3.816721506686251,
+      "grad_norm": 4.231115314464132,
       "learning_rate": 4.562240460580367e-05,
-      "logits/chosen": -9.8125,
-      "logits/rejected": -9.0,
-      "logps/chosen": -852.0,
-      "logps/rejected": -1096.0,
-      "loss": 0.4573,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -4.5,
-      "rewards/margins": 1.3515625,
-      "rewards/rejected": -5.84375,
+      "logits/chosen": -14.75,
+      "logits/rejected": -13.8125,
+      "logps/chosen": -940.0,
+      "logps/rejected": -1200.0,
+      "loss": 0.4347,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -5.40625,
+      "rewards/margins": 1.484375,
+      "rewards/rejected": -6.90625,
       "step": 780
     },
     {
       "epoch": 0.8272251308900523,
-      "grad_norm": 2.8447243093799233,
+      "grad_norm": 2.5056967898721236,
       "learning_rate": 4.544866150565985e-05,
-      "logits/chosen": -9.0625,
-      "logits/rejected": -8.125,
-      "logps/chosen": -1000.0,
+      "logits/chosen": -14.625,
+      "logits/rejected": -14.0625,
+      "logps/chosen": -996.0,
       "logps/rejected": -1312.0,
-      "loss": 0.5113,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -5.875,
-      "rewards/margins": 1.71875,
-      "rewards/rejected": -7.5625,
+      "loss": 0.5046,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -5.8125,
+      "rewards/margins": 1.7578125,
+      "rewards/rejected": -7.59375,
       "step": 790
     },
     {
       "epoch": 0.837696335078534,
-      "grad_norm": 2.626819607914269,
+      "grad_norm": 3.1135407306796976,
       "learning_rate": 4.527188176466051e-05,
-      "logits/chosen": -9.6875,
-      "logits/rejected": -9.5,
-      "logps/chosen": -972.0,
-      "logps/rejected": -1184.0,
-      "loss": 0.5113,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -5.40625,
-      "rewards/margins": 1.21875,
-      "rewards/rejected": -6.625,
+      "logits/chosen": -14.625,
+      "logits/rejected": -14.375,
+      "logps/chosen": -904.0,
+      "logps/rejected": -1112.0,
+      "loss": 0.4911,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 1.1875,
+      "rewards/rejected": -5.90625,
       "step": 800
     },
     {
       "epoch": 0.837696335078534,
-      "eval_logits/chosen": -10.8125,
-      "eval_logits/rejected": -9.4375,
-      "eval_logps/chosen": -796.0,
-      "eval_logps/rejected": -824.0,
-      "eval_loss": 0.7106093764305115,
-      "eval_rewards/accuracies": 0.611328125,
-      "eval_rewards/chosen": -4.78125,
-      "eval_rewards/margins": 0.55859375,
-      "eval_rewards/rejected": -5.34375,
-      "eval_runtime": 46.5838,
-      "eval_samples_per_second": 42.933,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -14.25,
+      "eval_logits/rejected": -13.375,
+      "eval_logps/chosen": -816.0,
+      "eval_logps/rejected": -852.0,
+      "eval_loss": 0.7705937623977661,
+      "eval_rewards/accuracies": 0.625,
+      "eval_rewards/chosen": -5.0,
+      "eval_rewards/margins": 0.640625,
+      "eval_rewards/rejected": -5.625,
+      "eval_runtime": 57.8708,
+      "eval_samples_per_second": 34.56,
+      "eval_steps_per_second": 0.553,
       "step": 800
     },
     {
       "epoch": 0.8481675392670157,
-      "grad_norm": 2.361204175031886,
+      "grad_norm": 2.566096647042146,
       "learning_rate": 4.509209163472357e-05,
-      "logits/chosen": -11.625,
-      "logits/rejected": -10.625,
-      "logps/chosen": -916.0,
-      "logps/rejected": -1240.0,
-      "loss": 0.4621,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -4.84375,
-      "rewards/margins": 1.6171875,
-      "rewards/rejected": -6.4375,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -13.6875,
+      "logps/chosen": -976.0,
+      "logps/rejected": -1312.0,
+      "loss": 0.4737,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -5.40625,
+      "rewards/margins": 1.6953125,
+      "rewards/rejected": -7.125,
       "step": 810
     },
     {
       "epoch": 0.8586387434554974,
-      "grad_norm": 1.9857028579014593,
+      "grad_norm": 1.9963865717848797,
       "learning_rate": 4.4909317814811896e-05,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -10.5,
+      "logits/chosen": -14.125,
+      "logits/rejected": -13.5,
       "logps/chosen": -912.0,
-      "logps/rejected": -1200.0,
-      "loss": 0.4702,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -4.96875,
-      "rewards/margins": 1.6640625,
-      "rewards/rejected": -6.65625,
+      "logps/rejected": -1184.0,
+      "loss": 0.4942,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -5.0,
+      "rewards/margins": 1.4921875,
+      "rewards/rejected": -6.46875,
       "step": 820
     },
     {
       "epoch": 0.8691099476439791,
-      "grad_norm": 2.548305522140382,
+      "grad_norm": 2.3828326362349954,
       "learning_rate": 4.472358744696845e-05,
-      "logits/chosen": -11.875,
-      "logits/rejected": -10.875,
-      "logps/chosen": -884.0,
+      "logits/chosen": -14.5,
+      "logits/rejected": -13.6875,
+      "logps/chosen": -888.0,
       "logps/rejected": -1104.0,
-      "loss": 0.4807,
+      "loss": 0.4997,
       "rewards/accuracies": 0.75,
-      "rewards/chosen": -4.8125,
-      "rewards/margins": 1.4921875,
-      "rewards/rejected": -6.3125,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 1.3828125,
+      "rewards/rejected": -6.25,
       "step": 830
     },
     {
       "epoch": 0.8795811518324608,
-      "grad_norm": 2.753026085927344,
+      "grad_norm": 2.678470041891397,
       "learning_rate": 4.4534928112285645e-05,
-      "logits/chosen": -10.9375,
-      "logits/rejected": -9.75,
-      "logps/chosen": -964.0,
-      "logps/rejected": -1216.0,
-      "loss": 0.4749,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -5.4375,
-      "rewards/margins": 1.46875,
-      "rewards/rejected": -6.90625,
+      "logits/chosen": -13.375,
+      "logits/rejected": -12.125,
+      "logps/chosen": -904.0,
+      "logps/rejected": -1152.0,
+      "loss": 0.4888,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -4.84375,
+      "rewards/margins": 1.4140625,
+      "rewards/rejected": -6.28125,
       "step": 840
     },
     {
       "epoch": 0.8900523560209425,
-      "grad_norm": 2.233938187435334,
+      "grad_norm": 2.182200584644557,
       "learning_rate": 4.43433678268096e-05,
-      "logits/chosen": -11.4375,
-      "logits/rejected": -10.375,
-      "logps/chosen": -912.0,
-      "logps/rejected": -1128.0,
-      "loss": 0.4686,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -5.1875,
-      "rewards/margins": 1.265625,
-      "rewards/rejected": -6.4375,
+      "logits/chosen": -12.3125,
+      "logits/rejected": -11.3125,
+      "logps/chosen": -896.0,
+      "logps/rejected": -1136.0,
+      "loss": 0.447,
+      "rewards/accuracies": 0.793749988079071,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 1.4765625,
+      "rewards/rejected": -6.5,
       "step": 850
     },
     {
       "epoch": 0.900523560209424,
-      "grad_norm": 2.5826262691515955,
+      "grad_norm": 2.5147799206327837,
       "learning_rate": 4.414893503737968e-05,
-      "logits/chosen": -9.25,
-      "logits/rejected": -7.90625,
-      "logps/chosen": -940.0,
-      "logps/rejected": -1128.0,
-      "loss": 0.5042,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -5.625,
-      "rewards/margins": 1.2265625,
-      "rewards/rejected": -6.84375,
+      "logits/chosen": -13.125,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -880.0,
+      "logps/rejected": -1080.0,
+      "loss": 0.4933,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 1.3125,
+      "rewards/rejected": -6.34375,
       "step": 860
     },
     {
       "epoch": 0.9109947643979057,
-      "grad_norm": 3.1050627151549555,
+      "grad_norm": 3.5427794163334405,
       "learning_rate": 4.395165861740414e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.125,
-      "logps/chosen": -860.0,
-      "logps/rejected": -1120.0,
-      "loss": 0.4484,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -4.84375,
-      "rewards/margins": 1.5234375,
-      "rewards/rejected": -6.34375,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -12.625,
+      "logps/chosen": -932.0,
+      "logps/rejected": -1208.0,
+      "loss": 0.4532,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -5.53125,
+      "rewards/margins": 1.71875,
+      "rewards/rejected": -7.25,
       "step": 870
     },
     {
       "epoch": 0.9214659685863874,
-      "grad_norm": 2.214330309632272,
+      "grad_norm": 3.1749624924134725,
       "learning_rate": 4.375156786257241e-05,
-      "logits/chosen": -12.625,
-      "logits/rejected": -12.0,
-      "logps/chosen": -872.0,
-      "logps/rejected": -1056.0,
-      "loss": 0.4964,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -4.28125,
-      "rewards/margins": 1.0234375,
-      "rewards/rejected": -5.3125,
+      "logits/chosen": -13.375,
+      "logits/rejected": -13.0,
+      "logps/chosen": -948.0,
+      "logps/rejected": -1144.0,
+      "loss": 0.5063,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 1.15625,
+      "rewards/rejected": -6.1875,
       "step": 880
     },
     {
       "epoch": 0.9319371727748691,
-      "grad_norm": 2.3984833683683155,
+      "grad_norm": 2.763480600786539,
       "learning_rate": 4.354869248650466e-05,
-      "logits/chosen": -12.9375,
-      "logits/rejected": -12.3125,
-      "logps/chosen": -820.0,
-      "logps/rejected": -984.0,
-      "loss": 0.475,
-      "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -4.09375,
-      "rewards/margins": 1.0703125,
-      "rewards/rejected": -5.15625,
+      "logits/chosen": -13.25,
+      "logits/rejected": -12.75,
+      "logps/chosen": -864.0,
+      "logps/rejected": -1020.0,
+      "loss": 0.4915,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 0.984375,
+      "rewards/rejected": -5.5,
       "step": 890
     },
     {
       "epoch": 0.9424083769633508,
-      "grad_norm": 3.801034037687458,
+      "grad_norm": 3.7067905737488944,
       "learning_rate": 4.334306261633931e-05,
-      "logits/chosen": -10.25,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -1288.0,
-      "loss": 0.4589,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -6.25,
-      "rewards/margins": 1.3984375,
-      "rewards/rejected": -7.65625,
+      "logits/chosen": -11.5,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1264.0,
+      "loss": 0.4586,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -6.09375,
+      "rewards/margins": 1.3125,
+      "rewards/rejected": -7.40625,
       "step": 900
     },
     {
       "epoch": 0.9424083769633508,
-      "eval_logits/chosen": -8.125,
-      "eval_logits/rejected": -6.1875,
-      "eval_logps/chosen": -1064.0,
-      "eval_logps/rejected": -1112.0,
-      "eval_loss": 0.8806718587875366,
-      "eval_rewards/accuracies": 0.609375,
-      "eval_rewards/chosen": -7.4375,
-      "eval_rewards/margins": 0.765625,
-      "eval_rewards/rejected": -8.1875,
-      "eval_runtime": 46.5789,
-      "eval_samples_per_second": 42.938,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -10.6875,
+      "eval_logits/rejected": -9.0,
+      "eval_logps/chosen": -1072.0,
+      "eval_logps/rejected": -1120.0,
+      "eval_loss": 0.9273124933242798,
+      "eval_rewards/accuracies": 0.611328125,
+      "eval_rewards/chosen": -7.53125,
+      "eval_rewards/margins": 0.77734375,
+      "eval_rewards/rejected": -8.3125,
+      "eval_runtime": 57.8643,
+      "eval_samples_per_second": 34.564,
+      "eval_steps_per_second": 0.553,
       "step": 900
     },
     {
       "epoch": 0.9528795811518325,
-      "grad_norm": 2.9904209983798546,
+      "grad_norm": 3.290140051290833,
       "learning_rate": 4.313470878825911e-05,
-      "logits/chosen": -10.8125,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -980.0,
-      "logps/rejected": -1296.0,
-      "loss": 0.4786,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -5.75,
-      "rewards/margins": 1.6328125,
-      "rewards/rejected": -7.375,
+      "logits/chosen": -12.625,
+      "logits/rejected": -11.5625,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1384.0,
+      "loss": 0.4607,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -6.34375,
+      "rewards/margins": 1.9453125,
+      "rewards/rejected": -8.25,
       "step": 910
     },
     {
       "epoch": 0.9633507853403142,
-      "grad_norm": 2.048544130700443,
+      "grad_norm": 2.1638760848578578,
       "learning_rate": 4.292366194295652e-05,
-      "logits/chosen": -11.875,
-      "logits/rejected": -10.8125,
-      "logps/chosen": -908.0,
-      "logps/rejected": -1112.0,
-      "loss": 0.473,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -4.875,
-      "rewards/margins": 1.515625,
-      "rewards/rejected": -6.375,
+      "logits/chosen": -12.5625,
+      "logits/rejected": -11.6875,
+      "logps/chosen": -972.0,
+      "logps/rejected": -1192.0,
+      "loss": 0.4799,
+      "rewards/accuracies": 0.793749988079071,
+      "rewards/chosen": -5.5,
+      "rewards/margins": 1.671875,
+      "rewards/rejected": -7.1875,
       "step": 920
     },
     {
       "epoch": 0.9738219895287958,
-      "grad_norm": 2.670494441144423,
+      "grad_norm": 2.7150100210030392,
       "learning_rate": 4.2709953421038964e-05,
-      "logits/chosen": -12.8125,
-      "logits/rejected": -11.8125,
-      "logps/chosen": -844.0,
-      "logps/rejected": -1072.0,
-      "loss": 0.4564,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -4.59375,
-      "rewards/margins": 1.2890625,
-      "rewards/rejected": -5.875,
+      "logits/chosen": -13.25,
+      "logits/rejected": -12.375,
+      "logps/chosen": -884.0,
+      "logps/rejected": -1104.0,
+      "loss": 0.4733,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 1.2421875,
+      "rewards/rejected": -6.21875,
       "step": 930
     },
     {
       "epoch": 0.9842931937172775,
-      "grad_norm": 2.3412165901378437,
+      "grad_norm": 3.0380334866879015,
       "learning_rate": 4.249361495837478e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -12.25,
-      "logps/chosen": -824.0,
-      "logps/rejected": -1056.0,
-      "loss": 0.5025,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -4.25,
-      "rewards/margins": 1.34375,
-      "rewards/rejected": -5.59375,
+      "logits/chosen": -12.9375,
+      "logits/rejected": -12.1875,
+      "logps/chosen": -872.0,
+      "logps/rejected": -1096.0,
+      "loss": 0.4976,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 1.3125,
+      "rewards/rejected": -6.03125,
       "step": 940
     },
     {
       "epoch": 0.9947643979057592,
-      "grad_norm": 2.8654359850365436,
+      "grad_norm": 2.858014230955159,
       "learning_rate": 4.227467868138035e-05,
-      "logits/chosen": -10.25,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -916.0,
-      "logps/rejected": -1200.0,
-      "loss": 0.4668,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -5.0,
-      "rewards/margins": 1.546875,
-      "rewards/rejected": -6.5625,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -896.0,
+      "logps/rejected": -1168.0,
+      "loss": 0.4683,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -4.8125,
+      "rewards/margins": 1.4140625,
+      "rewards/rejected": -6.21875,
       "step": 950
     },
     {
       "epoch": 1.0052356020942408,
-      "grad_norm": 1.899384396327492,
+      "grad_norm": 2.0625686308908215,
       "learning_rate": 4.205317710224938e-05,
-      "logits/chosen": -10.5625,
-      "logits/rejected": -9.25,
-      "logps/chosen": -884.0,
-      "logps/rejected": -1208.0,
-      "loss": 0.3006,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -4.59375,
-      "rewards/margins": 2.484375,
-      "rewards/rejected": -7.0625,
+      "logits/chosen": -13.75,
+      "logits/rejected": -13.25,
+      "logps/chosen": -880.0,
+      "logps/rejected": -1184.0,
+      "loss": 0.3164,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -4.53125,
+      "rewards/margins": 2.28125,
+      "rewards/rejected": -6.8125,
       "step": 960
     },
     {
       "epoch": 1.0157068062827226,
-      "grad_norm": 0.527601993020238,
+      "grad_norm": 0.7861371586876117,
       "learning_rate": 4.1829143114124733e-05,
-      "logits/chosen": -9.6875,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -1012.0,
-      "logps/rejected": -1520.0,
-      "loss": 0.1607,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -6.0,
-      "rewards/margins": 3.890625,
-      "rewards/rejected": -9.875,
+      "logits/chosen": -14.0,
+      "logits/rejected": -12.75,
+      "logps/chosen": -1064.0,
+      "logps/rejected": -1576.0,
+      "loss": 0.1645,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.5625,
+      "rewards/margins": 3.859375,
+      "rewards/rejected": -10.4375,
       "step": 970
     },
     {
       "epoch": 1.0261780104712042,
-      "grad_norm": 0.9330992482939198,
+      "grad_norm": 1.0136909886768888,
       "learning_rate": 4.160260998621387e-05,
-      "logits/chosen": -10.8125,
-      "logits/rejected": -9.0,
-      "logps/chosen": -1012.0,
-      "logps/rejected": -1552.0,
-      "loss": 0.1368,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -5.59375,
-      "rewards/margins": 4.3125,
-      "rewards/rejected": -9.9375,
+      "logits/chosen": -14.125,
+      "logits/rejected": -12.6875,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -1656.0,
+      "loss": 0.1389,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.59375,
+      "rewards/margins": 4.375,
+      "rewards/rejected": -11.0,
       "step": 980
     },
     {
       "epoch": 1.036649214659686,
-      "grad_norm": 1.3225241089012347,
+      "grad_norm": 1.1721629449164703,
       "learning_rate": 4.137361135884824e-05,
-      "logits/chosen": -7.96875,
-      "logits/rejected": -6.71875,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -1664.0,
-      "loss": 0.1318,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.96875,
-      "rewards/margins": 4.25,
-      "rewards/rejected": -11.25,
+      "logits/chosen": -10.875,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -1720.0,
+      "loss": 0.1345,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -7.6875,
+      "rewards/margins": 4.125,
+      "rewards/rejected": -11.8125,
       "step": 990
     },
     {
       "epoch": 1.0471204188481675,
-      "grad_norm": 2.091545662265001,
+      "grad_norm": 2.9482010259196465,
       "learning_rate": 4.114218123848777e-05,
-      "logits/chosen": -7.125,
-      "logits/rejected": -5.8125,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -1704.0,
-      "loss": 0.1368,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.53125,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -11.9375,
+      "logits/chosen": -10.4375,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1712.0,
+      "loss": 0.1423,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.5625,
+      "rewards/margins": 4.4375,
+      "rewards/rejected": -12.0,
       "step": 1000
     },
     {
       "epoch": 1.0471204188481675,
-      "eval_logits/chosen": -5.375,
-      "eval_logits/rejected": -3.453125,
-      "eval_logps/chosen": -1160.0,
-      "eval_logps/rejected": -1232.0,
-      "eval_loss": 1.1006406545639038,
+      "eval_logits/chosen": -9.125,
+      "eval_logits/rejected": -7.0625,
+      "eval_logps/chosen": -1192.0,
+      "eval_logps/rejected": -1256.0,
+      "eval_loss": 1.1068437099456787,
       "eval_rewards/accuracies": 0.587890625,
-      "eval_rewards/chosen": -8.375,
-      "eval_rewards/margins": 1.0546875,
-      "eval_rewards/rejected": -9.4375,
-      "eval_runtime": 46.5592,
-      "eval_samples_per_second": 42.956,
-      "eval_steps_per_second": 0.687,
+      "eval_rewards/chosen": -8.75,
+      "eval_rewards/margins": 0.9609375,
+      "eval_rewards/rejected": -9.6875,
+      "eval_runtime": 57.877,
+      "eval_samples_per_second": 34.556,
+      "eval_steps_per_second": 0.553,
       "step": 1000
     },
     {
       "epoch": 1.057591623036649,
-      "grad_norm": 2.4010039681166724,
+      "grad_norm": 2.1332446359165673,
       "learning_rate": 4.0908353992670824e-05,
-      "logits/chosen": -7.25,
-      "logits/rejected": -5.9375,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -1624.0,
-      "loss": 0.1445,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -9.875,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -1632.0,
+      "loss": 0.1466,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -6.3125,
-      "rewards/margins": 4.9375,
-      "rewards/rejected": -11.25,
+      "rewards/chosen": -6.78125,
+      "rewards/margins": 4.59375,
+      "rewards/rejected": -11.375,
       "step": 1010
     },
     {
       "epoch": 1.068062827225131,
-      "grad_norm": 2.693821873827417,
+      "grad_norm": 3.849793690439324,
       "learning_rate": 4.067216434491059e-05,
-      "logits/chosen": -6.6875,
-      "logits/rejected": -5.34375,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -1592.0,
-      "loss": 0.1363,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -6.3125,
-      "rewards/margins": 4.4375,
-      "rewards/rejected": -10.75,
+      "logits/chosen": -9.0,
+      "logits/rejected": -7.5,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -1728.0,
+      "loss": 0.1444,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -7.8125,
+      "rewards/margins": 4.28125,
+      "rewards/rejected": -12.0625,
       "step": 1020
     },
     {
       "epoch": 1.0785340314136125,
-      "grad_norm": 1.1632939310956245,
+      "grad_norm": 1.345579129636241,
       "learning_rate": 4.043364736953866e-05,
-      "logits/chosen": -6.65625,
-      "logits/rejected": -5.3125,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -1696.0,
-      "loss": 0.1431,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -6.875,
-      "rewards/margins": 4.4375,
-      "rewards/rejected": -11.3125,
+      "logits/chosen": -9.4375,
+      "logits/rejected": -7.71875,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -1768.0,
+      "loss": 0.1385,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.71875,
+      "rewards/margins": 4.28125,
+      "rewards/rejected": -12.0,
       "step": 1030
     },
     {
       "epoch": 1.0890052356020943,
-      "grad_norm": 2.419462562336923,
+      "grad_norm": 1.7967136520644447,
       "learning_rate": 4.019283848649643e-05,
-      "logits/chosen": -8.5,
-      "logits/rejected": -7.125,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -1560.0,
-      "loss": 0.1655,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -6.5625,
-      "rewards/margins": 3.890625,
-      "rewards/rejected": -10.4375,
+      "logits/chosen": -12.375,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1608.0,
+      "loss": 0.1566,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.875,
+      "rewards/margins": 4.03125,
+      "rewards/rejected": -10.9375,
       "step": 1040
     },
     {
       "epoch": 1.0994764397905759,
-      "grad_norm": 1.3362076022796896,
+      "grad_norm": 1.5322431515177632,
       "learning_rate": 3.9949773456075215e-05,
-      "logits/chosen": -7.53125,
-      "logits/rejected": -6.625,
-      "logps/chosen": -988.0,
-      "logps/rejected": -1512.0,
-      "loss": 0.1303,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.125,
-      "rewards/margins": 4.09375,
-      "rewards/rejected": -10.1875,
+      "logits/chosen": -10.6875,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -1568.0,
+      "loss": 0.1355,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -6.78125,
+      "rewards/margins": 4.0,
+      "rewards/rejected": -10.8125,
       "step": 1050
     },
     {
       "epoch": 1.1099476439790577,
-      "grad_norm": 1.2610354298980535,
+      "grad_norm": 1.7505039291752302,
       "learning_rate": 3.970448837360585e-05,
-      "logits/chosen": -8.3125,
-      "logits/rejected": -6.625,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -1720.0,
-      "loss": 0.1308,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -7.5,
-      "rewards/margins": 4.1875,
-      "rewards/rejected": -11.6875,
+      "logits/chosen": -9.4375,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -1800.0,
+      "loss": 0.1437,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -8.75,
+      "rewards/margins": 3.78125,
+      "rewards/rejected": -12.5,
       "step": 1060
     },
     {
       "epoch": 1.1204188481675392,
-      "grad_norm": 1.3678416681954628,
+      "grad_norm": 1.375735744092172,
       "learning_rate": 3.945701966409846e-05,
-      "logits/chosen": -7.65625,
-      "logits/rejected": -6.28125,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.1338,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -9.375,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -1704.0,
+      "loss": 0.1198,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -6.96875,
-      "rewards/margins": 4.875,
+      "rewards/chosen": -7.0,
+      "rewards/margins": 4.84375,
       "rewards/rejected": -11.8125,
       "step": 1070
     },
     {
       "epoch": 1.130890052356021,
-      "grad_norm": 1.5162831364279024,
+      "grad_norm": 4.371293926049364,
       "learning_rate": 3.9207404076833374e-05,
-      "logits/chosen": -8.3125,
-      "logits/rejected": -6.75,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -1664.0,
-      "loss": 0.1449,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.21875,
-      "rewards/margins": 4.125,
-      "rewards/rejected": -11.375,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -10.0,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -1704.0,
+      "loss": 0.1441,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -7.40625,
+      "rewards/margins": 4.28125,
+      "rewards/rejected": -11.6875,
       "step": 1080
     },
     {
       "epoch": 1.1413612565445026,
-      "grad_norm": 2.271707183617078,
+      "grad_norm": 1.6577635225919498,
       "learning_rate": 3.895567867990379e-05,
-      "logits/chosen": -7.75,
-      "logits/rejected": -5.84375,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -1864.0,
-      "loss": 0.1396,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.4375,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -13.1875,
+      "logits/chosen": -13.625,
+      "logits/rejected": -11.5625,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -1720.0,
+      "loss": 0.152,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -7.125,
+      "rewards/margins": 4.625,
+      "rewards/rejected": -11.75,
       "step": 1090
     },
     {
       "epoch": 1.1518324607329844,
-      "grad_norm": 1.6871139983069965,
+      "grad_norm": 1.7957341503824324,
       "learning_rate": 3.870188085471114e-05,
-      "logits/chosen": -7.28125,
-      "logits/rejected": -5.6875,
-      "logps/chosen": -1216.0,
-      "logps/rejected": -1728.0,
-      "loss": 0.138,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.96875,
-      "rewards/margins": 4.5625,
-      "rewards/rejected": -12.5625,
+      "logits/chosen": -13.625,
+      "logits/rejected": -12.25,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -1536.0,
+      "loss": 0.1457,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.5,
+      "rewards/margins": 4.1875,
+      "rewards/rejected": -10.6875,
       "step": 1100
     },
     {
       "epoch": 1.1518324607329844,
-      "eval_logits/chosen": -5.96875,
-      "eval_logits/rejected": -4.09375,
-      "eval_logps/chosen": -1160.0,
-      "eval_logps/rejected": -1224.0,
-      "eval_loss": 1.028640627861023,
-      "eval_rewards/accuracies": 0.59765625,
-      "eval_rewards/chosen": -8.375,
-      "eval_rewards/margins": 0.953125,
-      "eval_rewards/rejected": -9.375,
-      "eval_runtime": 46.5679,
-      "eval_samples_per_second": 42.948,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -12.375,
+      "eval_logits/rejected": -10.75,
+      "eval_logps/chosen": -1128.0,
+      "eval_logps/rejected": -1192.0,
+      "eval_loss": 1.1010937690734863,
+      "eval_rewards/accuracies": 0.580078125,
+      "eval_rewards/chosen": -8.125,
+      "eval_rewards/margins": 0.9140625,
+      "eval_rewards/rejected": -9.0625,
+      "eval_runtime": 57.8778,
+      "eval_samples_per_second": 34.556,
+      "eval_steps_per_second": 0.553,
       "step": 1100
     },
     {
       "epoch": 1.162303664921466,
-      "grad_norm": 2.6258548839528424,
+      "grad_norm": 2.355058006268161,
       "learning_rate": 3.844604829041395e-05,
-      "logits/chosen": -6.75,
-      "logits/rejected": -4.9375,
-      "logps/chosen": -1020.0,
-      "logps/rejected": -1512.0,
-      "loss": 0.1377,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -6.46875,
-      "rewards/margins": 4.0625,
-      "rewards/rejected": -10.5625,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -11.375,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1568.0,
+      "loss": 0.1331,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -6.75,
+      "rewards/margins": 4.40625,
+      "rewards/rejected": -11.125,
       "step": 1110
     },
     {
       "epoch": 1.1727748691099475,
-      "grad_norm": 3.512283842722846,
+      "grad_norm": 2.5281025366813092,
       "learning_rate": 3.818821897833095e-05,
-      "logits/chosen": -4.78125,
-      "logits/rejected": -3.890625,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -1704.0,
-      "loss": 0.1507,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.40625,
+      "logits/chosen": -11.4375,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1696.0,
+      "loss": 0.1551,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -7.3125,
       "rewards/margins": 4.5625,
-      "rewards/rejected": -11.9375,
+      "rewards/rejected": -11.875,
       "step": 1120
     },
     {
       "epoch": 1.1832460732984293,
-      "grad_norm": 2.547504293958598,
+      "grad_norm": 2.7570795741590888,
       "learning_rate": 3.792843120629935e-05,
-      "logits/chosen": -7.03125,
-      "logits/rejected": -5.15625,
-      "logps/chosen": -988.0,
-      "logps/rejected": -1520.0,
-      "loss": 0.1614,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -5.875,
-      "rewards/margins": 4.21875,
-      "rewards/rejected": -10.125,
+      "logits/chosen": -12.5,
+      "logits/rejected": -10.25,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1616.0,
+      "loss": 0.1459,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.5,
+      "rewards/margins": 4.625,
+      "rewards/rejected": -11.125,
       "step": 1130
     },
     {
       "epoch": 1.193717277486911,
-      "grad_norm": 3.684159912507558,
+      "grad_norm": 4.614183309676998,
       "learning_rate": 3.766672355298907e-05,
-      "logits/chosen": -6.4375,
-      "logits/rejected": -5.5625,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -1608.0,
-      "loss": 0.1352,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.90625,
-      "rewards/margins": 4.25,
-      "rewards/rejected": -11.1875,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -11.5625,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -1648.0,
+      "loss": 0.1424,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -7.1875,
+      "rewards/margins": 4.34375,
+      "rewards/rejected": -11.5625,
       "step": 1140
     },
     {
       "epoch": 1.2041884816753927,
-      "grad_norm": 2.819233324788751,
+      "grad_norm": 2.3836252026751144,
       "learning_rate": 3.740313488217373e-05,
-      "logits/chosen": -7.0625,
-      "logits/rejected": -5.53125,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1696.0,
-      "loss": 0.1377,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.09375,
-      "rewards/margins": 4.9375,
-      "rewards/rejected": -12.0625,
+      "logits/chosen": -13.25,
+      "logits/rejected": -11.625,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1760.0,
+      "loss": 0.1327,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -7.65625,
+      "rewards/margins": 5.0625,
+      "rewards/rejected": -12.75,
       "step": 1150
     },
     {
       "epoch": 1.2146596858638743,
-      "grad_norm": 2.6316218075740903,
+      "grad_norm": 1.6611020790313147,
       "learning_rate": 3.7137704336959464e-05,
-      "logits/chosen": -6.21875,
-      "logits/rejected": -4.59375,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -1664.0,
-      "loss": 0.1223,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -6.8125,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -11.5,
+      "logits/chosen": -13.25,
+      "logits/rejected": -11.3125,
+      "logps/chosen": -1160.0,
+      "logps/rejected": -1800.0,
+      "loss": 0.1332,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -7.78125,
+      "rewards/margins": 5.0625,
+      "rewards/rejected": -12.875,
       "step": 1160
     },
     {
       "epoch": 1.225130890052356,
-      "grad_norm": 2.568592697706442,
+      "grad_norm": 2.335807379312117,
       "learning_rate": 3.687047133397201e-05,
-      "logits/chosen": -8.0625,
-      "logits/rejected": -6.0,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -1696.0,
-      "loss": 0.134,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.78125,
-      "rewards/margins": 5.0,
-      "rewards/rejected": -11.8125,
+      "logits/chosen": -12.875,
+      "logits/rejected": -11.0,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -1832.0,
+      "loss": 0.142,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.0625,
+      "rewards/margins": 5.09375,
+      "rewards/rejected": -13.1875,
       "step": 1170
     },
     {
       "epoch": 1.2356020942408377,
-      "grad_norm": 2.127605618222479,
+      "grad_norm": 1.9641604371607848,
       "learning_rate": 3.6601475557503415e-05,
-      "logits/chosen": -6.78125,
-      "logits/rejected": -4.65625,
-      "logps/chosen": -1160.0,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -9.125,
+      "logps/chosen": -1168.0,
       "logps/rejected": -1728.0,
-      "loss": 0.1209,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -7.375,
-      "rewards/margins": 4.59375,
-      "rewards/rejected": -12.0,
+      "loss": 0.1417,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.5,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -12.0625,
       "step": 1180
     },
     {
       "epoch": 1.2460732984293195,
-      "grad_norm": 1.928952092925142,
+      "grad_norm": 1.696464271010807,
       "learning_rate": 3.633075695361882e-05,
-      "logits/chosen": -4.8125,
-      "logits/rejected": -3.765625,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1816.0,
-      "loss": 0.1217,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 5.0625,
-      "rewards/rejected": -13.1875,
-      "step": 1190
-    },
-    {
-      "epoch": 1.256544502617801,
-      "grad_norm": 1.32199700898028,
-      "learning_rate": 3.6058355724224477e-05,
-      "logits/chosen": -5.75,
-      "logits/rejected": -4.21875,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.1376,
+      "logits/chosen": -10.6875,
+      "logits/rejected": -8.375,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1752.0,
+      "loss": 0.1402,
       "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.5,
-      "rewards/margins": 4.53125,
-      "rewards/rejected": -12.0625,
+      "rewards/chosen": -7.6875,
+      "rewards/margins": 4.875,
+      "rewards/rejected": -12.5625,
+      "step": 1190
+    },
+    {
+      "epoch": 1.256544502617801,
+      "grad_norm": 1.8319408688318246,
+      "learning_rate": 3.6058355724224477e-05,
+      "logits/chosen": -10.125,
+      "logits/rejected": -7.9375,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -1648.0,
+      "loss": 0.1344,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -7.0625,
+      "rewards/margins": 4.3125,
+      "rewards/rejected": -11.375,
       "step": 1200
     },
     {
       "epoch": 1.256544502617801,
-      "eval_logits/chosen": -3.078125,
-      "eval_logits/rejected": -1.2265625,
-      "eval_logps/chosen": -1192.0,
-      "eval_logps/rejected": -1264.0,
-      "eval_loss": 1.0961874723434448,
-      "eval_rewards/accuracies": 0.603515625,
-      "eval_rewards/chosen": -8.6875,
-      "eval_rewards/margins": 1.03125,
-      "eval_rewards/rejected": -9.75,
-      "eval_runtime": 46.5696,
-      "eval_samples_per_second": 42.946,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -7.40625,
+      "eval_logits/rejected": -5.34375,
+      "eval_logps/chosen": -1152.0,
+      "eval_logps/rejected": -1224.0,
+      "eval_loss": 1.0088906288146973,
+      "eval_rewards/accuracies": 0.599609375,
+      "eval_rewards/chosen": -8.375,
+      "eval_rewards/margins": 0.98828125,
+      "eval_rewards/rejected": -9.375,
+      "eval_runtime": 57.8629,
+      "eval_samples_per_second": 34.564,
+      "eval_steps_per_second": 0.553,
       "step": 1200
     },
     {
       "epoch": 1.2670157068062826,
-      "grad_norm": 2.735910201834722,
+      "grad_norm": 2.6284631559432587,
       "learning_rate": 3.5784312321097744e-05,
-      "logits/chosen": -4.75,
-      "logits/rejected": -3.515625,
-      "logps/chosen": -1168.0,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -6.9375,
+      "logps/chosen": -1184.0,
       "logps/rejected": -1720.0,
-      "loss": 0.1435,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.4375,
-      "rewards/margins": 4.75,
+      "loss": 0.1348,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.53125,
+      "rewards/margins": 4.625,
       "rewards/rejected": -12.1875,
       "step": 1210
     },
     {
       "epoch": 1.2774869109947644,
-      "grad_norm": 2.7969356617821246,
+      "grad_norm": 3.5081549493883033,
       "learning_rate": 3.550866743987994e-05,
-      "logits/chosen": -4.5,
-      "logits/rejected": -3.4375,
-      "logps/chosen": -1216.0,
-      "logps/rejected": -1752.0,
-      "loss": 0.1436,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.90625,
-      "rewards/margins": 4.5625,
-      "rewards/rejected": -12.5,
+      "logits/chosen": -8.375,
+      "logits/rejected": -6.75,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -1808.0,
+      "loss": 0.1557,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -13.0625,
       "step": 1220
     },
     {
       "epoch": 1.2879581151832462,
-      "grad_norm": 2.49774269111401,
+      "grad_norm": 1.7247823604605028,
       "learning_rate": 3.5231462014033024e-05,
-      "logits/chosen": -5.96875,
-      "logits/rejected": -4.375,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -1608.0,
-      "loss": 0.1166,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -6.6875,
-      "rewards/margins": 4.78125,
-      "rewards/rejected": -11.5,
+      "logits/chosen": -10.4375,
+      "logits/rejected": -8.5625,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -1648.0,
+      "loss": 0.1321,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -11.875,
       "step": 1230
     },
     {
       "epoch": 1.2984293193717278,
-      "grad_norm": 2.044560967921475,
+      "grad_norm": 1.853588498086977,
       "learning_rate": 3.495273720876095e-05,
-      "logits/chosen": -3.21875,
-      "logits/rejected": -2.21875,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -1840.0,
-      "loss": 0.1223,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.4375,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -13.3125,
+      "logits/chosen": -10.6875,
+      "logits/rejected": -9.125,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -1712.0,
+      "loss": 0.1278,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -7.5,
+      "rewards/margins": 4.4375,
+      "rewards/rejected": -11.9375,
       "step": 1240
     },
     {
       "epoch": 1.3089005235602094,
-      "grad_norm": 1.8033499695336526,
+      "grad_norm": 2.5716686200695427,
       "learning_rate": 3.4672534414896616e-05,
-      "logits/chosen": -5.65625,
-      "logits/rejected": -4.15625,
-      "logps/chosen": -1120.0,
-      "logps/rejected": -1600.0,
-      "loss": 0.1363,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -1672.0,
+      "loss": 0.1585,
       "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -6.78125,
-      "rewards/margins": 4.21875,
-      "rewards/rejected": -11.0,
+      "rewards/chosen": -7.5,
+      "rewards/margins": 4.25,
+      "rewards/rejected": -11.75,
       "step": 1250
     },
     {
       "epoch": 1.3193717277486912,
-      "grad_norm": 1.9440641331794801,
+      "grad_norm": 1.3328590082465688,
       "learning_rate": 3.439089524275526e-05,
-      "logits/chosen": -4.5625,
-      "logits/rejected": -3.375,
-      "logps/chosen": -1024.0,
-      "logps/rejected": -1528.0,
-      "loss": 0.1313,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -6.375,
-      "rewards/margins": 4.15625,
-      "rewards/rejected": -10.5,
+      "logits/chosen": -11.4375,
+      "logits/rejected": -10.25,
+      "logps/chosen": -1004.0,
+      "logps/rejected": -1456.0,
+      "loss": 0.1454,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.1875,
+      "rewards/margins": 3.640625,
+      "rewards/rejected": -9.8125,
       "step": 1260
     },
     {
       "epoch": 1.3298429319371727,
-      "grad_norm": 2.080518644365482,
+      "grad_norm": 2.845371840442894,
       "learning_rate": 3.410786151595533e-05,
-      "logits/chosen": -4.8125,
-      "logits/rejected": -3.46875,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -1624.0,
+      "logits/chosen": -10.875,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -1552.0,
       "loss": 0.1453,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.375,
-      "rewards/margins": 4.34375,
-      "rewards/rejected": -11.75,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 4.125,
+      "rewards/rejected": -11.0,
       "step": 1270
     },
     {
       "epoch": 1.3403141361256545,
-      "grad_norm": 1.7423708302589305,
+      "grad_norm": 2.4741622719528564,
       "learning_rate": 3.382347526520765e-05,
-      "logits/chosen": -5.8125,
-      "logits/rejected": -4.40625,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1648.0,
-      "loss": 0.1216,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -6.96875,
-      "rewards/margins": 4.15625,
-      "rewards/rejected": -11.125,
+      "logits/chosen": -10.5,
+      "logits/rejected": -8.625,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -1776.0,
+      "loss": 0.1488,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -8.125,
+      "rewards/margins": 4.21875,
+      "rewards/rejected": -12.375,
       "step": 1280
     },
     {
       "epoch": 1.350785340314136,
-      "grad_norm": 1.9662250821522929,
+      "grad_norm": 1.867310708360032,
       "learning_rate": 3.353777872207381e-05,
-      "logits/chosen": -5.09375,
-      "logits/rejected": -4.125,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -1640.0,
-      "loss": 0.145,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -7.28125,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -7.25,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -1704.0,
+      "loss": 0.1335,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -7.875,
       "rewards/margins": 4.375,
-      "rewards/rejected": -11.625,
+      "rewards/rejected": -12.25,
       "step": 1290
     },
     {
       "epoch": 1.3612565445026177,
-      "grad_norm": 1.7089326714778734,
+      "grad_norm": 2.169775650411397,
       "learning_rate": 3.325081431269475e-05,
-      "logits/chosen": -4.9375,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -1688.0,
-      "loss": 0.1434,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -7.5625,
-      "rewards/margins": 4.59375,
-      "rewards/rejected": -12.125,
+      "logits/chosen": -8.375,
+      "logits/rejected": -6.15625,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -1664.0,
+      "loss": 0.1369,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -7.34375,
+      "rewards/margins": 4.5625,
+      "rewards/rejected": -11.875,
       "step": 1300
     },
     {
       "epoch": 1.3612565445026177,
-      "eval_logits/chosen": -5.6875,
-      "eval_logits/rejected": -3.703125,
-      "eval_logps/chosen": -1256.0,
-      "eval_logps/rejected": -1336.0,
-      "eval_loss": 1.1220155954360962,
-      "eval_rewards/accuracies": 0.580078125,
-      "eval_rewards/chosen": -9.375,
-      "eval_rewards/margins": 1.1171875,
-      "eval_rewards/rejected": -10.5,
-      "eval_runtime": 46.5589,
-      "eval_samples_per_second": 42.956,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -7.59375,
+      "eval_logits/rejected": -5.53125,
+      "eval_logps/chosen": -1264.0,
+      "eval_logps/rejected": -1352.0,
+      "eval_loss": 1.054031252861023,
+      "eval_rewards/accuracies": 0.6015625,
+      "eval_rewards/chosen": -9.4375,
+      "eval_rewards/margins": 1.1796875,
+      "eval_rewards/rejected": -10.625,
+      "eval_runtime": 57.94,
+      "eval_samples_per_second": 34.518,
+      "eval_steps_per_second": 0.552,
       "step": 1300
     },
     {
       "epoch": 1.3717277486910995,
-      "grad_norm": 3.5166744612956014,
+      "grad_norm": 1.8481247288761367,
       "learning_rate": 3.296262465149046e-05,
-      "logits/chosen": -7.875,
-      "logits/rejected": -5.8125,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1840.0,
-      "loss": 0.1278,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.65625,
-      "rewards/margins": 5.125,
-      "rewards/rejected": -12.75,
+      "logits/chosen": -9.1875,
+      "logits/rejected": -7.0,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -1816.0,
+      "loss": 0.1359,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.8125,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -12.5625,
       "step": 1310
     },
     {
       "epoch": 1.3821989528795813,
-      "grad_norm": 3.1745359969873754,
+      "grad_norm": 2.5390310940092498,
       "learning_rate": 3.2673252534831687e-05,
-      "logits/chosen": -8.125,
-      "logits/rejected": -6.21875,
-      "logps/chosen": -1152.0,
-      "logps/rejected": -1728.0,
-      "loss": 0.1625,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -7.46875,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -12.25,
+      "logits/chosen": -9.5625,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -1696.0,
+      "loss": 0.1409,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.28125,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -11.9375,
       "step": 1320
     },
     {
       "epoch": 1.3926701570680629,
-      "grad_norm": 1.9166983354994007,
+      "grad_norm": 1.6171442390218278,
       "learning_rate": 3.2382740934684696e-05,
-      "logits/chosen": -7.5625,
-      "logits/rejected": -5.78125,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -1696.0,
-      "loss": 0.1475,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.09375,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -11.5625,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -7.1875,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -1752.0,
+      "loss": 0.152,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -7.78125,
+      "rewards/margins": 4.375,
+      "rewards/rejected": -12.1875,
       "step": 1330
     },
     {
       "epoch": 1.4031413612565444,
-      "grad_norm": 2.3999374742349917,
+      "grad_norm": 2.107246508384085,
       "learning_rate": 3.209113299222982e-05,
-      "logits/chosen": -6.625,
-      "logits/rejected": -5.78125,
-      "logps/chosen": -1040.0,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -1032.0,
       "logps/rejected": -1600.0,
-      "loss": 0.1473,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.875,
-      "rewards/margins": 4.1875,
-      "rewards/rejected": -11.0625,
+      "loss": 0.1452,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 4.28125,
+      "rewards/rejected": -11.125,
       "step": 1340
     },
     {
       "epoch": 1.4136125654450262,
-      "grad_norm": 2.693206966118984,
+      "grad_norm": 2.462715733268011,
       "learning_rate": 3.179847201145505e-05,
-      "logits/chosen": -4.96875,
-      "logits/rejected": -3.8125,
-      "logps/chosen": -1216.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.1225,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.0,
-      "rewards/margins": 4.28125,
-      "rewards/rejected": -12.3125,
+      "logits/chosen": -8.25,
+      "logits/rejected": -6.71875,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1672.0,
+      "loss": 0.1245,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.40625,
+      "rewards/margins": 4.0,
+      "rewards/rejected": -11.375,
       "step": 1350
     },
     {
       "epoch": 1.4240837696335078,
-      "grad_norm": 3.3985074894470784,
+      "grad_norm": 2.2979055637738623,
       "learning_rate": 3.150480145272528e-05,
-      "logits/chosen": -6.3125,
-      "logits/rejected": -5.21875,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -1672.0,
-      "loss": 0.1501,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.8125,
-      "rewards/margins": 3.921875,
-      "rewards/rejected": -11.75,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -7.4375,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -1728.0,
+      "loss": 0.1612,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 4.0625,
+      "rewards/rejected": -12.25,
       "step": 1360
     },
     {
       "epoch": 1.4345549738219896,
-      "grad_norm": 1.1658891436912604,
+      "grad_norm": 1.1522710177742748,
       "learning_rate": 3.121016492632848e-05,
-      "logits/chosen": -8.1875,
-      "logits/rejected": -6.59375,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -1640.0,
-      "loss": 0.1499,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -7.0,
-      "rewards/margins": 4.375,
-      "rewards/rejected": -11.375,
+      "logits/chosen": -9.1875,
+      "logits/rejected": -7.40625,
+      "logps/chosen": -1224.0,
+      "logps/rejected": -1792.0,
+      "loss": 0.1297,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -12.875,
       "step": 1370
     },
     {
       "epoch": 1.4450261780104712,
-      "grad_norm": 2.223333609241383,
+      "grad_norm": 2.284134055131499,
       "learning_rate": 3.091460618599951e-05,
-      "logits/chosen": -7.40625,
-      "logits/rejected": -5.15625,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -1736.0,
-      "loss": 0.1199,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -7.59375,
-      "rewards/margins": 4.46875,
-      "rewards/rejected": -12.0625,
+      "logits/chosen": -8.25,
+      "logits/rejected": -6.21875,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -1824.0,
+      "loss": 0.1278,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -13.0,
       "step": 1380
     },
     {
       "epoch": 1.4554973821989527,
-      "grad_norm": 2.353720366885145,
+      "grad_norm": 3.1287604065097723,
       "learning_rate": 3.061816912242265e-05,
       "logits/chosen": -8.125,
-      "logits/rejected": -6.34375,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -1768.0,
-      "loss": 0.1192,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.46875,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -12.1875,
+      "logits/rejected": -6.75,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -1872.0,
+      "loss": 0.1355,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 5.0,
+      "rewards/rejected": -13.1875,
       "step": 1390
     },
     {
       "epoch": 1.4659685863874345,
-      "grad_norm": 1.4754634082222138,
+      "grad_norm": 1.844598678276241,
       "learning_rate": 3.0320897756713785e-05,
-      "logits/chosen": -8.1875,
-      "logits/rejected": -6.0625,
-      "logps/chosen": -1144.0,
+      "logits/chosen": -9.125,
+      "logits/rejected": -7.4375,
+      "logps/chosen": -1152.0,
       "logps/rejected": -1736.0,
-      "loss": 0.1386,
+      "loss": 0.1225,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.21875,
-      "rewards/margins": 4.65625,
-      "rewards/rejected": -11.875,
+      "rewards/chosen": -7.28125,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -11.8125,
       "step": 1400
     },
     {
       "epoch": 1.4659685863874345,
-      "eval_logits/chosen": -5.46875,
-      "eval_logits/rejected": -3.5,
-      "eval_logps/chosen": -1256.0,
-      "eval_logps/rejected": -1328.0,
-      "eval_loss": 1.0637500286102295,
-      "eval_rewards/accuracies": 0.623046875,
-      "eval_rewards/chosen": -9.4375,
-      "eval_rewards/margins": 0.95703125,
-      "eval_rewards/rejected": -10.375,
-      "eval_runtime": 46.5595,
-      "eval_samples_per_second": 42.956,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -7.375,
+      "eval_logits/rejected": -5.59375,
+      "eval_logps/chosen": -1272.0,
+      "eval_logps/rejected": -1352.0,
+      "eval_loss": 1.1049062013626099,
+      "eval_rewards/accuracies": 0.603515625,
+      "eval_rewards/chosen": -9.5625,
+      "eval_rewards/margins": 1.0859375,
+      "eval_rewards/rejected": -10.625,
+      "eval_runtime": 57.8681,
+      "eval_samples_per_second": 34.561,
+      "eval_steps_per_second": 0.553,
       "step": 1400
     },
     {
       "epoch": 1.4764397905759163,
-      "grad_norm": 1.7743467440627028,
+      "grad_norm": 1.3300041566926881,
       "learning_rate": 3.002283623388332e-05,
-      "logits/chosen": -6.28125,
-      "logits/rejected": -5.15625,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.129,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 4.375,
-      "rewards/rejected": -12.4375,
+      "logits/chosen": -9.625,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -1696.0,
+      "loss": 0.1255,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -7.53125,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -12.25,
       "step": 1410
     },
     {
       "epoch": 1.486910994764398,
-      "grad_norm": 4.423939456254239,
+      "grad_norm": 2.543778404959387,
       "learning_rate": 2.9724028816280507e-05,
-      "logits/chosen": -7.09375,
-      "logits/rejected": -6.03125,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -1728.0,
-      "loss": 0.1256,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 4.09375,
-      "rewards/rejected": -12.1875,
+      "logits/chosen": -9.875,
+      "logits/rejected": -8.75,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1680.0,
+      "loss": 0.1362,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -7.125,
+      "rewards/margins": 4.59375,
+      "rewards/rejected": -11.75,
       "step": 1420
     },
     {
       "epoch": 1.4973821989528795,
-      "grad_norm": 2.898950117797185,
+      "grad_norm": 2.68918424674839,
       "learning_rate": 2.9424519877020524e-05,
-      "logits/chosen": -9.375,
-      "logits/rejected": -7.875,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -1856.0,
-      "loss": 0.1383,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 4.625,
-      "rewards/rejected": -12.75,
+      "logits/chosen": -10.375,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -1776.0,
+      "loss": 0.1416,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.03125,
+      "rewards/margins": 4.84375,
+      "rewards/rejected": -11.875,
       "step": 1430
     },
     {
       "epoch": 1.5078534031413613,
-      "grad_norm": 2.8872519159612233,
+      "grad_norm": 1.4621980209087713,
       "learning_rate": 2.912435389339504e-05,
-      "logits/chosen": -10.5,
-      "logits/rejected": -8.25,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -1632.0,
-      "loss": 0.1387,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.34375,
-      "rewards/margins": 4.34375,
-      "rewards/rejected": -11.6875,
+      "logits/chosen": -10.0,
+      "logits/rejected": -7.625,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1544.0,
+      "loss": 0.1267,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.375,
+      "rewards/margins": 4.40625,
+      "rewards/rejected": -10.8125,
       "step": 1440
     },
     {
       "epoch": 1.518324607329843,
-      "grad_norm": 2.415377354370345,
+      "grad_norm": 3.293110744051904,
       "learning_rate": 2.8823575440267226e-05,
-      "logits/chosen": -9.625,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -1152.0,
-      "logps/rejected": -1704.0,
-      "loss": 0.1361,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.5625,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -12.0,
+      "logits/chosen": -9.125,
+      "logits/rejected": -7.5,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -1664.0,
+      "loss": 0.1428,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -7.34375,
+      "rewards/margins": 4.28125,
+      "rewards/rejected": -11.625,
       "step": 1450
     },
     {
       "epoch": 1.5287958115183247,
-      "grad_norm": 2.380509713310526,
+      "grad_norm": 2.0633014585494562,
       "learning_rate": 2.8522229183452465e-05,
-      "logits/chosen": -7.1875,
-      "logits/rejected": -5.53125,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -1704.0,
-      "loss": 0.1813,
-      "rewards/accuracies": 0.9375,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -6.78125,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -1728.0,
+      "loss": 0.1811,
+      "rewards/accuracies": 0.9437500238418579,
       "rewards/chosen": -7.875,
-      "rewards/margins": 4.1875,
-      "rewards/rejected": -12.0625,
+      "rewards/margins": 4.40625,
+      "rewards/rejected": -12.3125,
       "step": 1460
     },
     {
       "epoch": 1.5392670157068062,
-      "grad_norm": 2.291791143945341,
+      "grad_norm": 3.129375925356749,
       "learning_rate": 2.8220359873085373e-05,
-      "logits/chosen": -6.59375,
-      "logits/rejected": -5.4375,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.1246,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.40625,
-      "rewards/margins": 4.5625,
-      "rewards/rejected": -12.0,
+      "logits/chosen": -9.1875,
+      "logits/rejected": -7.75,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -1688.0,
+      "loss": 0.1198,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -7.15625,
+      "rewards/margins": 4.625,
+      "rewards/rejected": -11.75,
       "step": 1470
     },
     {
       "epoch": 1.5497382198952878,
-      "grad_norm": 2.2269248522353453,
+      "grad_norm": 2.554500756929723,
       "learning_rate": 2.7918012336974388e-05,
-      "logits/chosen": -6.4375,
-      "logits/rejected": -4.53125,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -1648.0,
-      "loss": 0.1821,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.59375,
-      "rewards/margins": 4.59375,
-      "rewards/rejected": -12.1875,
+      "logits/chosen": -8.75,
+      "logits/rejected": -6.46875,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -1600.0,
+      "loss": 0.1186,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.21875,
+      "rewards/margins": 4.46875,
+      "rewards/rejected": -11.6875,
       "step": 1480
     },
     {
       "epoch": 1.5602094240837696,
-      "grad_norm": 2.532735313429492,
+      "grad_norm": 2.115029618579577,
       "learning_rate": 2.7615231473944815e-05,
-      "logits/chosen": -6.0,
-      "logits/rejected": -4.53125,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -1728.0,
-      "loss": 0.1238,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.84375,
-      "rewards/margins": 4.71875,
-      "rewards/rejected": -12.5625,
+      "logits/chosen": -6.28125,
+      "logits/rejected": -4.875,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -1816.0,
+      "loss": 0.1234,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -8.6875,
+      "rewards/margins": 4.78125,
+      "rewards/rejected": -13.4375,
       "step": 1490
     },
     {
       "epoch": 1.5706806282722514,
-      "grad_norm": 2.601166300855045,
+      "grad_norm": 3.074090667336648,
       "learning_rate": 2.7312062247171327e-05,
-      "logits/chosen": -7.21875,
-      "logits/rejected": -5.375,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -1664.0,
-      "loss": 0.1258,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.40625,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -11.875,
+      "logits/chosen": -5.96875,
+      "logits/rejected": -4.3125,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -1768.0,
+      "loss": 0.1276,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 4.59375,
+      "rewards/rejected": -12.875,
       "step": 1500
     },
     {
       "epoch": 1.5706806282722514,
-      "eval_logits/chosen": -4.5625,
-      "eval_logits/rejected": -2.40625,
-      "eval_logps/chosen": -1368.0,
-      "eval_logps/rejected": -1464.0,
-      "eval_loss": 1.192312479019165,
-      "eval_rewards/accuracies": 0.6015625,
-      "eval_rewards/chosen": -10.5,
-      "eval_rewards/margins": 1.1953125,
-      "eval_rewards/rejected": -11.75,
-      "eval_runtime": 46.542,
-      "eval_samples_per_second": 42.972,
-      "eval_steps_per_second": 0.688,
+      "eval_logits/chosen": -3.0625,
+      "eval_logits/rejected": -1.0625,
+      "eval_logps/chosen": -1416.0,
+      "eval_logps/rejected": -1512.0,
+      "eval_loss": 1.178531289100647,
+      "eval_rewards/accuracies": 0.607421875,
+      "eval_rewards/chosen": -11.0,
+      "eval_rewards/margins": 1.234375,
+      "eval_rewards/rejected": -12.25,
+      "eval_runtime": 57.8706,
+      "eval_samples_per_second": 34.56,
+      "eval_steps_per_second": 0.553,
       "step": 1500
     },
     {
       "epoch": 1.581151832460733,
-      "grad_norm": 1.7225582039170644,
+      "grad_norm": 2.2558625538479906,
       "learning_rate": 2.7008549677500876e-05,
-      "logits/chosen": -7.1875,
-      "logits/rejected": -5.15625,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -1968.0,
-      "loss": 0.1079,
+      "logits/chosen": -7.125,
+      "logits/rejected": -5.1875,
+      "logps/chosen": -1320.0,
+      "logps/rejected": -1896.0,
+      "loss": 0.119,
       "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -9.25,
-      "rewards/margins": 5.09375,
-      "rewards/rejected": -14.3125,
+      "rewards/chosen": -8.875,
+      "rewards/margins": 4.75,
+      "rewards/rejected": -13.625,
       "step": 1510
     },
     {
       "epoch": 1.5916230366492146,
-      "grad_norm": 1.7959707440532422,
+      "grad_norm": 1.5456222505046924,
       "learning_rate": 2.670473883676709e-05,
-      "logits/chosen": -5.34375,
-      "logits/rejected": -3.703125,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -1976.0,
-      "loss": 0.078,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 5.34375,
-      "rewards/rejected": -14.3125,
+      "logits/chosen": -6.625,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1280.0,
+      "logps/rejected": -1912.0,
+      "loss": 0.0874,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -8.6875,
+      "rewards/margins": 5.03125,
+      "rewards/rejected": -13.6875,
       "step": 1520
     },
     {
       "epoch": 1.6020942408376964,
-      "grad_norm": 3.9624068534923462,
+      "grad_norm": 2.008923182976819,
       "learning_rate": 2.6400674841097074e-05,
-      "logits/chosen": -5.90625,
-      "logits/rejected": -3.984375,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -1936.0,
-      "loss": 0.1112,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -9.5625,
-      "rewards/margins": 5.03125,
-      "rewards/rejected": -14.625,
+      "logits/chosen": -5.875,
+      "logits/rejected": -4.03125,
+      "logps/chosen": -1504.0,
+      "logps/rejected": -2064.0,
+      "loss": 0.1212,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -10.8125,
+      "rewards/margins": 5.0625,
+      "rewards/rejected": -15.875,
       "step": 1530
     },
     {
       "epoch": 1.6125654450261782,
-      "grad_norm": 2.220629912318349,
+      "grad_norm": 2.1471027153757536,
       "learning_rate": 2.609640284421158e-05,
-      "logits/chosen": -6.4375,
-      "logits/rejected": -4.125,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -1816.0,
-      "loss": 0.1436,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 4.875,
-      "rewards/rejected": -13.1875,
+      "logits/chosen": -6.625,
+      "logits/rejected": -4.40625,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -1984.0,
+      "loss": 0.163,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -9.8125,
+      "rewards/margins": 5.15625,
+      "rewards/rejected": -14.9375,
       "step": 1540
     },
     {
       "epoch": 1.6230366492146597,
-      "grad_norm": 2.1315596040693467,
+      "grad_norm": 2.2431714153450226,
       "learning_rate": 2.5791968030719726e-05,
-      "logits/chosen": -8.0,
-      "logits/rejected": -6.375,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -1720.0,
-      "loss": 0.1263,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.21875,
-      "rewards/margins": 4.1875,
-      "rewards/rejected": -11.375,
+      "logits/chosen": -8.875,
+      "logits/rejected": -7.09375,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -1848.0,
+      "loss": 0.1131,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -8.0,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -12.6875,
       "step": 1550
     },
     {
       "epoch": 1.6335078534031413,
-      "grad_norm": 1.8039145616733827,
+      "grad_norm": 2.8149678802639815,
       "learning_rate": 2.548741560940902e-05,
-      "logits/chosen": -8.1875,
-      "logits/rejected": -6.71875,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -1720.0,
-      "loss": 0.1292,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -7.125,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -11.5625,
+      "logits/chosen": -8.625,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -1800.0,
+      "loss": 0.1423,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -7.5,
+      "rewards/margins": 4.75,
+      "rewards/rejected": -12.25,
       "step": 1560
     },
     {
       "epoch": 1.6439790575916229,
-      "grad_norm": 2.0292341717593336,
+      "grad_norm": 1.5870551995990518,
       "learning_rate": 2.5182790806531787e-05,
-      "logits/chosen": -5.875,
-      "logits/rejected": -4.40625,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -1848.0,
-      "loss": 0.1485,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.625,
-      "rewards/margins": 4.4375,
-      "rewards/rejected": -13.0625,
+      "logits/chosen": -7.21875,
+      "logits/rejected": -5.96875,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -1832.0,
+      "loss": 0.1425,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -12.9375,
       "step": 1570
     },
     {
       "epoch": 1.6544502617801047,
-      "grad_norm": 1.907816471595626,
+      "grad_norm": 2.4992246770819535,
       "learning_rate": 2.4878138859089074e-05,
-      "logits/chosen": -6.28125,
-      "logits/rejected": -4.59375,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -1792.0,
-      "loss": 0.1233,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.875,
-      "rewards/margins": 4.78125,
-      "rewards/rejected": -12.625,
+      "logits/chosen": -7.65625,
+      "logits/rejected": -6.3125,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -1744.0,
+      "loss": 0.1267,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -7.5625,
+      "rewards/margins": 4.5625,
+      "rewards/rejected": -12.125,
       "step": 1580
     },
     {
       "epoch": 1.6649214659685865,
-      "grad_norm": 1.4674864118156452,
+      "grad_norm": 1.421377187702302,
       "learning_rate": 2.4573505008112924e-05,
-      "logits/chosen": -6.75,
-      "logits/rejected": -5.25,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.1155,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.96875,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -12.5,
+      "logits/chosen": -8.125,
+      "logits/rejected": -6.5625,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -1752.0,
+      "loss": 0.1159,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.84375,
+      "rewards/margins": 4.5625,
+      "rewards/rejected": -12.375,
       "step": 1590
     },
     {
       "epoch": 1.675392670157068,
-      "grad_norm": 2.6409381475702216,
+      "grad_norm": 1.9011977773447022,
       "learning_rate": 2.4268934491948032e-05,
-      "logits/chosen": -7.59375,
-      "logits/rejected": -6.0,
-      "logps/chosen": -1152.0,
-      "logps/rejected": -1672.0,
-      "loss": 0.1269,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.21875,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -11.625,
+      "logits/chosen": -7.84375,
+      "logits/rejected": -6.53125,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -1736.0,
+      "loss": 0.1177,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -7.71875,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -12.25,
       "step": 1600
     },
     {
       "epoch": 1.675392670157068,
-      "eval_logits/chosen": -5.21875,
-      "eval_logits/rejected": -2.84375,
-      "eval_logps/chosen": -1264.0,
-      "eval_logps/rejected": -1352.0,
-      "eval_loss": 1.2008594274520874,
-      "eval_rewards/accuracies": 0.607421875,
-      "eval_rewards/chosen": -9.4375,
-      "eval_rewards/margins": 1.15625,
-      "eval_rewards/rejected": -10.625,
-      "eval_runtime": 46.5585,
-      "eval_samples_per_second": 42.957,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -5.90625,
+      "eval_logits/rejected": -3.859375,
+      "eval_logps/chosen": -1272.0,
+      "eval_logps/rejected": -1368.0,
+      "eval_loss": 1.148593783378601,
+      "eval_rewards/accuracies": 0.609375,
+      "eval_rewards/chosen": -9.5,
+      "eval_rewards/margins": 1.25,
+      "eval_rewards/rejected": -10.75,
+      "eval_runtime": 57.9113,
+      "eval_samples_per_second": 34.536,
+      "eval_steps_per_second": 0.553,
       "step": 1600
     },
     {
       "epoch": 1.6858638743455496,
-      "grad_norm": 2.3095591960316724,
+      "grad_norm": 2.756657922492984,
       "learning_rate": 2.396447253953385e-05,
-      "logits/chosen": -7.96875,
-      "logits/rejected": -5.84375,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.1454,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.6875,
-      "rewards/margins": 4.78125,
-      "rewards/rejected": -12.5,
+      "logits/chosen": -9.25,
+      "logits/rejected": -7.25,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -1752.0,
+      "loss": 0.145,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.5,
+      "rewards/margins": 4.875,
+      "rewards/rejected": -12.375,
       "step": 1610
     },
     {
       "epoch": 1.6963350785340314,
-      "grad_norm": 2.1408802896934787,
+      "grad_norm": 2.171146445919755,
       "learning_rate": 2.3660164363687998e-05,
-      "logits/chosen": -7.03125,
-      "logits/rejected": -5.46875,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -1872.0,
-      "loss": 0.1248,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 4.75,
-      "rewards/rejected": -12.8125,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -7.4375,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -1800.0,
+      "loss": 0.12,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -7.4375,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -12.125,
       "step": 1620
     },
     {
       "epoch": 1.7068062827225132,
-      "grad_norm": 2.704269424888906,
+      "grad_norm": 2.8423712168844553,
       "learning_rate": 2.33560551543922e-05,
-      "logits/chosen": -7.28125,
-      "logits/rejected": -5.6875,
-      "logps/chosen": -1216.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.1359,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 4.375,
-      "rewards/rejected": -12.0,
+      "logits/chosen": -8.75,
+      "logits/rejected": -7.15625,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -1720.0,
+      "loss": 0.1443,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -7.8125,
+      "rewards/margins": 4.28125,
+      "rewards/rejected": -12.0625,
       "step": 1630
     },
     {
       "epoch": 1.7172774869109948,
-      "grad_norm": 2.603411633265208,
+      "grad_norm": 2.443160796033812,
       "learning_rate": 2.3052190072081493e-05,
-      "logits/chosen": -8.5,
-      "logits/rejected": -6.1875,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1592.0,
-      "loss": 0.1185,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.75,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -11.125,
+      "logits/chosen": -9.875,
+      "logits/rejected": -7.875,
+      "logps/chosen": -1160.0,
+      "logps/rejected": -1680.0,
+      "loss": 0.1314,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.375,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -12.0625,
       "step": 1640
     },
     {
       "epoch": 1.7277486910994764,
-      "grad_norm": 1.0759247859926822,
+      "grad_norm": 1.5309581244523511,
       "learning_rate": 2.2748614240937864e-05,
-      "logits/chosen": -7.03125,
-      "logits/rejected": -5.0625,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -1744.0,
-      "loss": 0.1028,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.53125,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -12.25,
+      "logits/chosen": -9.0625,
+      "logits/rejected": -7.03125,
+      "logps/chosen": -1160.0,
+      "logps/rejected": -1760.0,
+      "loss": 0.1064,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.71875,
+      "rewards/margins": 4.625,
+      "rewards/rejected": -12.3125,
       "step": 1650
     },
     {
       "epoch": 1.738219895287958,
-      "grad_norm": 2.3902873401345612,
+      "grad_norm": 2.1189649645983355,
       "learning_rate": 2.2445372742189334e-05,
-      "logits/chosen": -6.21875,
-      "logits/rejected": -4.65625,
+      "logits/chosen": -8.5625,
+      "logits/rejected": -6.90625,
       "logps/chosen": -1336.0,
-      "logps/rejected": -2000.0,
-      "loss": 0.1049,
+      "logps/rejected": -1984.0,
+      "loss": 0.1066,
       "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -9.125,
-      "rewards/margins": 5.3125,
-      "rewards/rejected": -14.4375,
+      "rewards/chosen": -9.1875,
+      "rewards/margins": 5.09375,
+      "rewards/rejected": -14.25,
       "step": 1660
     },
     {
       "epoch": 1.7486910994764397,
-      "grad_norm": 3.4442192607623094,
+      "grad_norm": 2.531490447860375,
       "learning_rate": 2.214251060741528e-05,
-      "logits/chosen": -7.0,
-      "logits/rejected": -5.25,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -1960.0,
-      "loss": 0.1181,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.375,
-      "rewards/margins": 5.65625,
-      "rewards/rejected": -14.0,
+      "logits/chosen": -9.875,
+      "logits/rejected": -8.25,
+      "logps/chosen": -1304.0,
+      "logps/rejected": -1992.0,
+      "loss": 0.113,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -8.6875,
+      "rewards/margins": 5.625,
+      "rewards/rejected": -14.375,
       "step": 1670
     },
     {
       "epoch": 1.7591623036649215,
-      "grad_norm": 3.0489214321117983,
+      "grad_norm": 4.537536543034156,
       "learning_rate": 2.1840072811859302e-05,
-      "logits/chosen": -7.5625,
-      "logits/rejected": -6.0,
-      "logps/chosen": -1152.0,
-      "logps/rejected": -1696.0,
-      "loss": 0.1487,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.28125,
-      "rewards/margins": 4.75,
-      "rewards/rejected": -12.0,
+      "logits/chosen": -10.0,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -1792.0,
+      "loss": 0.1461,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -12.9375,
       "step": 1680
     },
     {
       "epoch": 1.7696335078534031,
-      "grad_norm": 1.4893787543913555,
+      "grad_norm": 1.856416784416,
       "learning_rate": 2.1538104267750287e-05,
-      "logits/chosen": -6.90625,
-      "logits/rejected": -5.71875,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -1536.0,
-      "loss": 0.1197,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -6.625,
-      "rewards/margins": 3.90625,
-      "rewards/rejected": -10.5625,
+      "logits/chosen": -10.6875,
+      "logits/rejected": -9.625,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1608.0,
+      "loss": 0.1157,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.15625,
+      "rewards/margins": 4.0625,
+      "rewards/rejected": -11.25,
       "step": 1690
     },
     {
       "epoch": 1.7801047120418847,
-      "grad_norm": 1.5551381990151674,
+      "grad_norm": 1.706387116915449,
       "learning_rate": 2.1236649817632952e-05,
-      "logits/chosen": -6.46875,
-      "logits/rejected": -5.53125,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -1592.0,
-      "loss": 0.0967,
-      "rewards/accuracies": 0.949999988079071,
+      "logits/chosen": -11.0625,
+      "logits/rejected": -9.75,
+      "logps/chosen": -1064.0,
+      "logps/rejected": -1584.0,
+      "loss": 0.1007,
+      "rewards/accuracies": 0.9312499761581421,
       "rewards/chosen": -6.96875,
-      "rewards/margins": 4.09375,
-      "rewards/rejected": -11.0625,
+      "rewards/margins": 4.03125,
+      "rewards/rejected": -11.0,
       "step": 1700
     },
     {
       "epoch": 1.7801047120418847,
-      "eval_logits/chosen": -3.640625,
-      "eval_logits/rejected": -1.6328125,
-      "eval_logps/chosen": -1320.0,
-      "eval_logps/rejected": -1400.0,
-      "eval_loss": 1.1723437309265137,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -10.0,
-      "eval_rewards/margins": 1.0859375,
-      "eval_rewards/rejected": -11.125,
-      "eval_runtime": 46.5631,
-      "eval_samples_per_second": 42.952,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -9.3125,
+      "eval_logits/rejected": -7.75,
+      "eval_logps/chosen": -1272.0,
+      "eval_logps/rejected": -1344.0,
+      "eval_loss": 1.1275156736373901,
+      "eval_rewards/accuracies": 0.583984375,
+      "eval_rewards/chosen": -9.5,
+      "eval_rewards/margins": 1.0625,
+      "eval_rewards/rejected": -10.5625,
+      "eval_runtime": 57.8872,
+      "eval_samples_per_second": 34.55,
+      "eval_steps_per_second": 0.553,
       "step": 1700
     },
     {
       "epoch": 1.7905759162303665,
-      "grad_norm": 1.547652500674168,
+      "grad_norm": 2.936619907557649,
       "learning_rate": 2.093575422770872e-05,
-      "logits/chosen": -4.6875,
-      "logits/rejected": -3.703125,
-      "logps/chosen": -1216.0,
+      "logits/chosen": -9.0,
+      "logits/rejected": -7.90625,
+      "logps/chosen": -1224.0,
       "logps/rejected": -1824.0,
-      "loss": 0.1009,
+      "loss": 0.1112,
       "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 5.125,
+      "rewards/chosen": -8.4375,
+      "rewards/margins": 5.03125,
       "rewards/rejected": -13.4375,
       "step": 1710
     },
     {
       "epoch": 1.8010471204188483,
-      "grad_norm": 3.4345059678794057,
+      "grad_norm": 3.0277077683524403,
       "learning_rate": 2.0635462181187827e-05,
-      "logits/chosen": -5.125,
-      "logits/rejected": -3.609375,
+      "logits/chosen": -8.5625,
+      "logits/rejected": -6.9375,
       "logps/chosen": -1400.0,
-      "logps/rejected": -2160.0,
-      "loss": 0.1417,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -9.8125,
-      "rewards/margins": 5.75,
+      "logps/rejected": -2144.0,
+      "loss": 0.1469,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -9.75,
+      "rewards/margins": 5.6875,
       "rewards/rejected": -15.5,
       "step": 1720
     },
     {
       "epoch": 1.8115183246073299,
-      "grad_norm": 3.2447122902916936,
+      "grad_norm": 1.0267853340157587,
       "learning_rate": 2.0335818271653917e-05,
-      "logits/chosen": -6.6875,
-      "logits/rejected": -5.0625,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -1800.0,
-      "loss": 0.1351,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 4.75,
-      "rewards/rejected": -13.0,
+      "logits/chosen": -10.3125,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -1704.0,
+      "loss": 0.126,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -7.625,
+      "rewards/margins": 4.34375,
+      "rewards/rejected": -11.9375,
       "step": 1730
     },
     {
       "epoch": 1.8219895287958114,
-      "grad_norm": 2.3132618273893777,
+      "grad_norm": 2.6992969882488778,
       "learning_rate": 2.0036866996441815e-05,
-      "logits/chosen": -8.5,
-      "logits/rejected": -6.25,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -1680.0,
-      "loss": 0.1272,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.34375,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -12.0,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1656.0,
+      "loss": 0.1248,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.0,
+      "rewards/margins": 4.75,
+      "rewards/rejected": -11.75,
       "step": 1740
     },
     {
       "epoch": 1.8324607329842932,
-      "grad_norm": 2.7870304641557104,
+      "grad_norm": 2.300269381680004,
       "learning_rate": 1.9738652750029677e-05,
-      "logits/chosen": -6.84375,
-      "logits/rejected": -5.09375,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -1776.0,
-      "loss": 0.1158,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.9375,
-      "rewards/margins": 4.65625,
-      "rewards/rejected": -12.625,
+      "logits/chosen": -9.75,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -1784.0,
+      "loss": 0.1181,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.90625,
+      "rewards/margins": 4.78125,
+      "rewards/rejected": -12.6875,
       "step": 1750
     },
     {
       "epoch": 1.8429319371727748,
-      "grad_norm": 2.3858729248154393,
+      "grad_norm": 3.0524414940745137,
       "learning_rate": 1.944121981744631e-05,
-      "logits/chosen": -7.5625,
-      "logits/rejected": -5.96875,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -1848.0,
-      "loss": 0.1255,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -12.9375,
+      "logits/chosen": -9.375,
+      "logits/rejected": -8.0,
+      "logps/chosen": -1352.0,
+      "logps/rejected": -1944.0,
+      "loss": 0.1196,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.9375,
+      "rewards/margins": 5.0,
+      "rewards/rejected": -13.9375,
       "step": 1760
     },
     {
       "epoch": 1.8534031413612566,
-      "grad_norm": 1.9956403937758487,
+      "grad_norm": 1.538285181072194,
       "learning_rate": 1.9144612367694887e-05,
-      "logits/chosen": -7.3125,
-      "logits/rejected": -5.40625,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -1864.0,
-      "loss": 0.1149,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 4.84375,
-      "rewards/rejected": -13.0,
+      "logits/chosen": -9.1875,
+      "logits/rejected": -7.34375,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -1928.0,
+      "loss": 0.1201,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -8.625,
+      "rewards/margins": 5.0625,
+      "rewards/rejected": -13.6875,
       "step": 1770
     },
     {
       "epoch": 1.8638743455497382,
-      "grad_norm": 1.8551612665866088,
+      "grad_norm": 2.915144006251418,
       "learning_rate": 1.8848874447193803e-05,
-      "logits/chosen": -5.75,
-      "logits/rejected": -4.34375,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.1109,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.375,
-      "rewards/margins": 4.96875,
-      "rewards/rejected": -13.3125,
+      "logits/chosen": -8.25,
+      "logits/rejected": -6.90625,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -1728.0,
+      "loss": 0.1263,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.125,
+      "rewards/margins": 4.6875,
+      "rewards/rejected": -12.8125,
       "step": 1780
     },
     {
       "epoch": 1.8743455497382198,
-      "grad_norm": 2.9298110556336168,
+      "grad_norm": 1.938433915812453,
       "learning_rate": 1.855404997323571e-05,
-      "logits/chosen": -6.4375,
-      "logits/rejected": -4.65625,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -1912.0,
-      "loss": 0.1357,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -13.875,
+      "logits/chosen": -9.4375,
+      "logits/rejected": -7.78125,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -1832.0,
+      "loss": 0.1278,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.125,
+      "rewards/margins": 4.90625,
+      "rewards/rejected": -13.0625,
       "step": 1790
     },
     {
       "epoch": 1.8848167539267016,
-      "grad_norm": 1.7055774841267966,
+      "grad_norm": 1.7873328966113902,
       "learning_rate": 1.82601827274658e-05,
-      "logits/chosen": -7.21875,
-      "logits/rejected": -5.28125,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1752.0,
-      "loss": 0.112,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 4.78125,
-      "rewards/rejected": -12.875,
+      "logits/chosen": -9.625,
+      "logits/rejected": -7.84375,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -1760.0,
+      "loss": 0.1186,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 4.6875,
+      "rewards/rejected": -12.9375,
       "step": 1800
     },
     {
       "epoch": 1.8848167539267016,
-      "eval_logits/chosen": -4.5,
-      "eval_logits/rejected": -2.5,
-      "eval_logps/chosen": -1296.0,
-      "eval_logps/rejected": -1360.0,
-      "eval_loss": 1.0807499885559082,
-      "eval_rewards/accuracies": 0.58984375,
-      "eval_rewards/chosen": -9.75,
-      "eval_rewards/margins": 0.98046875,
-      "eval_rewards/rejected": -10.75,
-      "eval_runtime": 46.5692,
-      "eval_samples_per_second": 42.947,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -7.4375,
+      "eval_logits/rejected": -5.71875,
+      "eval_logps/chosen": -1312.0,
+      "eval_logps/rejected": -1392.0,
+      "eval_loss": 1.1385468244552612,
+      "eval_rewards/accuracies": 0.5703125,
+      "eval_rewards/chosen": -9.9375,
+      "eval_rewards/margins": 1.0546875,
+      "eval_rewards/rejected": -11.0,
+      "eval_runtime": 57.8674,
+      "eval_samples_per_second": 34.562,
+      "eval_steps_per_second": 0.553,
       "step": 1800
     },
     {
       "epoch": 1.8952879581151834,
-      "grad_norm": 2.921748420079359,
+      "grad_norm": 2.6426468394629246,
       "learning_rate": 1.7967316349380213e-05,
-      "logits/chosen": -6.4375,
-      "logits/rejected": -5.09375,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -1720.0,
-      "loss": 0.1233,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.0,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -12.375,
+      "logits/chosen": -9.0625,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -1760.0,
+      "loss": 0.1156,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.125,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -12.8125,
       "step": 1810
     },
     {
       "epoch": 1.905759162303665,
-      "grad_norm": 2.5937134944860896,
+      "grad_norm": 2.853432330383538,
       "learning_rate": 1.7675494329845516e-05,
-      "logits/chosen": -5.84375,
-      "logits/rejected": -4.0625,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.1158,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 4.625,
-      "rewards/rejected": -12.9375,
+      "logits/chosen": -7.9375,
+      "logits/rejected": -6.375,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -1816.0,
+      "loss": 0.1126,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.4375,
+      "rewards/margins": 4.875,
+      "rewards/rejected": -13.3125,
       "step": 1820
     },
     {
       "epoch": 1.9162303664921465,
-      "grad_norm": 1.7402476104330142,
+      "grad_norm": 1.5697712431401951,
       "learning_rate": 1.738476000464026e-05,
-      "logits/chosen": -5.4375,
-      "logits/rejected": -3.765625,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -1920.0,
-      "loss": 0.1079,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 5.1875,
-      "rewards/rejected": -13.875,
+      "logits/chosen": -7.09375,
+      "logits/rejected": -5.3125,
+      "logps/chosen": -1312.0,
+      "logps/rejected": -1976.0,
+      "loss": 0.1069,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -9.0625,
+      "rewards/margins": 5.375,
+      "rewards/rejected": -14.4375,
       "step": 1830
     },
     {
       "epoch": 1.9267015706806283,
-      "grad_norm": 1.5783685974843171,
+      "grad_norm": 1.3785117778060572,
       "learning_rate": 1.7095156548019646e-05,
-      "logits/chosen": -6.375,
-      "logits/rejected": -4.8125,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -1944.0,
-      "loss": 0.1135,
+      "logits/chosen": -7.625,
+      "logits/rejected": -5.875,
+      "logps/chosen": -1304.0,
+      "logps/rejected": -2024.0,
+      "loss": 0.1049,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 5.28125,
-      "rewards/rejected": -14.0,
+      "rewards/chosen": -9.1875,
+      "rewards/margins": 5.625,
+      "rewards/rejected": -14.8125,
       "step": 1840
     },
     {
       "epoch": 1.93717277486911,
-      "grad_norm": 1.9116972331714166,
+      "grad_norm": 2.197993418395338,
       "learning_rate": 1.6806726966304064e-05,
-      "logits/chosen": -7.3125,
-      "logits/rejected": -5.5,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -1840.0,
-      "loss": 0.1255,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -8.375,
-      "rewards/margins": 4.90625,
-      "rewards/rejected": -13.25,
+      "logits/chosen": -8.5,
+      "logits/rejected": -7.0,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -1920.0,
+      "loss": 0.132,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -9.0625,
+      "rewards/margins": 5.0,
+      "rewards/rejected": -14.0,
       "step": 1850
     },
     {
       "epoch": 1.9476439790575917,
-      "grad_norm": 4.529527873403766,
+      "grad_norm": 4.1431824026162,
       "learning_rate": 1.6519514091492625e-05,
-      "logits/chosen": -6.25,
-      "logits/rejected": -4.5625,
+      "logits/chosen": -8.3125,
+      "logits/rejected": -6.8125,
       "logps/chosen": -1232.0,
       "logps/rejected": -1800.0,
-      "loss": 0.1634,
-      "rewards/accuracies": 0.8999999761581421,
+      "loss": 0.1646,
+      "rewards/accuracies": 0.90625,
       "rewards/chosen": -8.625,
       "rewards/margins": 4.65625,
-      "rewards/rejected": -13.3125,
+      "rewards/rejected": -13.25,
       "step": 1860
     },
     {
       "epoch": 1.9581151832460733,
-      "grad_norm": 2.1749682781100876,
+      "grad_norm": 1.9546843932245153,
       "learning_rate": 1.6233560574902627e-05,
-      "logits/chosen": -5.59375,
-      "logits/rejected": -4.03125,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -1888.0,
-      "loss": 0.1068,
+      "logits/chosen": -8.3125,
+      "logits/rejected": -6.78125,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -1792.0,
+      "loss": 0.1178,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 5.125,
-      "rewards/rejected": -14.125,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 4.8125,
+      "rewards/rejected": -13.125,
       "step": 1870
     },
     {
       "epoch": 1.9685863874345548,
-      "grad_norm": 1.48735883497607,
+      "grad_norm": 1.9598201634837193,
       "learning_rate": 1.5948908880835753e-05,
-      "logits/chosen": -6.09375,
-      "logits/rejected": -4.03125,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -1888.0,
-      "loss": 0.1198,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 5.21875,
-      "rewards/rejected": -14.125,
+      "logits/chosen": -8.625,
+      "logits/rejected": -6.59375,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -1792.0,
+      "loss": 0.116,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 5.03125,
+      "rewards/rejected": -13.25,
       "step": 1880
     },
     {
       "epoch": 1.9790575916230366,
-      "grad_norm": 1.7245129771828225,
+      "grad_norm": 2.026733376867381,
       "learning_rate": 1.5665601280272124e-05,
-      "logits/chosen": -6.34375,
-      "logits/rejected": -4.375,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -1824.0,
-      "loss": 0.1273,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -9.1875,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -13.6875,
+      "logits/chosen": -7.59375,
+      "logits/rejected": -5.6875,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -1848.0,
+      "loss": 0.1319,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -9.375,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -13.875,
       "step": 1890
     },
     {
       "epoch": 1.9895287958115184,
-      "grad_norm": 2.747494052999277,
+      "grad_norm": 2.9552700374595062,
       "learning_rate": 1.538367984459298e-05,
-      "logits/chosen": -5.5,
-      "logits/rejected": -3.484375,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -1856.0,
-      "loss": 0.1158,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 5.0625,
-      "rewards/rejected": -13.875,
+      "logits/chosen": -6.625,
+      "logits/rejected": -4.4375,
+      "logps/chosen": -1328.0,
+      "logps/rejected": -1944.0,
+      "loss": 0.1098,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -9.4375,
+      "rewards/margins": 5.375,
+      "rewards/rejected": -14.8125,
       "step": 1900
     },
     {
       "epoch": 1.9895287958115184,
-      "eval_logits/chosen": -3.5625,
-      "eval_logits/rejected": -1.5390625,
-      "eval_logps/chosen": -1400.0,
-      "eval_logps/rejected": -1496.0,
-      "eval_loss": 1.146984338760376,
-      "eval_rewards/accuracies": 0.59375,
-      "eval_rewards/chosen": -10.875,
-      "eval_rewards/margins": 1.2109375,
-      "eval_rewards/rejected": -12.0625,
-      "eval_runtime": 46.5545,
-      "eval_samples_per_second": 42.96,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -4.6875,
+      "eval_logits/rejected": -2.703125,
+      "eval_logps/chosen": -1512.0,
+      "eval_logps/rejected": -1616.0,
+      "eval_loss": 1.2802969217300415,
+      "eval_rewards/accuracies": 0.587890625,
+      "eval_rewards/chosen": -11.9375,
+      "eval_rewards/margins": 1.3359375,
+      "eval_rewards/rejected": -13.25,
+      "eval_runtime": 57.9536,
+      "eval_samples_per_second": 34.51,
+      "eval_steps_per_second": 0.552,
       "step": 1900
     },
     {
       "epoch": 2.0,
-      "grad_norm": 1.3120366069737726,
+      "grad_norm": 1.451883298265358,
       "learning_rate": 1.5103186439333133e-05,
-      "logits/chosen": -5.1875,
-      "logits/rejected": -3.53125,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -1920.0,
-      "loss": 0.1021,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 5.3125,
-      "rewards/rejected": -14.0,
+      "logits/chosen": -6.625,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1296.0,
+      "logps/rejected": -2000.0,
+      "loss": 0.1127,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -9.125,
+      "rewards/margins": 5.625,
+      "rewards/rejected": -14.8125,
       "step": 1910
     },
     {
       "epoch": 2.0104712041884816,
-      "grad_norm": 0.29837410864758246,
+      "grad_norm": 0.7732912036946348,
       "learning_rate": 1.4824162717963829e-05,
-      "logits/chosen": -5.625,
-      "logits/rejected": -2.671875,
-      "logps/chosen": -1216.0,
-      "logps/rejected": -1928.0,
-      "loss": 0.0202,
+      "logits/chosen": -7.75,
+      "logits/rejected": -4.625,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -1920.0,
+      "loss": 0.0215,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 6.5,
-      "rewards/rejected": -14.75,
+      "rewards/chosen": -8.375,
+      "rewards/margins": 6.3125,
+      "rewards/rejected": -14.6875,
       "step": 1920
     },
     {
       "epoch": 2.020942408376963,
-      "grad_norm": 0.2507344747486213,
+      "grad_norm": 0.1751649382536878,
       "learning_rate": 1.4546650115707192e-05,
-      "logits/chosen": -5.21875,
-      "logits/rejected": -3.046875,
+      "logits/chosen": -6.96875,
+      "logits/rejected": -4.5,
       "logps/chosen": -1296.0,
-      "logps/rejected": -2080.0,
-      "loss": 0.018,
-      "rewards/accuracies": 0.9937499761581421,
+      "logps/rejected": -2096.0,
+      "loss": 0.0206,
+      "rewards/accuracies": 0.987500011920929,
       "rewards/chosen": -9.1875,
-      "rewards/margins": 6.96875,
-      "rewards/rejected": -16.125,
+      "rewards/margins": 7.1875,
+      "rewards/rejected": -16.375,
       "step": 1930
     },
     {
       "epoch": 2.031413612565445,
-      "grad_norm": 0.14184261301934967,
+      "grad_norm": 0.12815805373907574,
       "learning_rate": 1.4270689843383111e-05,
-      "logits/chosen": -4.15625,
-      "logits/rejected": -1.515625,
-      "logps/chosen": -1384.0,
-      "logps/rejected": -2208.0,
-      "loss": 0.0304,
-      "rewards/accuracies": 0.987500011920929,
+      "logits/chosen": -5.46875,
+      "logits/rejected": -2.625,
+      "logps/chosen": -1376.0,
+      "logps/rejected": -2240.0,
+      "loss": 0.0206,
+      "rewards/accuracies": 0.9937499761581421,
       "rewards/chosen": -9.75,
-      "rewards/margins": 7.3125,
-      "rewards/rejected": -17.0,
+      "rewards/margins": 7.53125,
+      "rewards/rejected": -17.25,
       "step": 1940
     },
     {
       "epoch": 2.0418848167539267,
-      "grad_norm": 0.7208264026884581,
-      "learning_rate": 1.3996322881289347e-05,
-      "logits/chosen": -4.9375,
-      "logits/rejected": -1.984375,
-      "logps/chosen": -1352.0,
-      "logps/rejected": -2256.0,
-      "loss": 0.0223,
+      "grad_norm": 2.0165174468191545,
+      "learning_rate": 1.3996322881289347e-05,
+      "logits/chosen": -5.90625,
+      "logits/rejected": -2.9375,
+      "logps/chosen": -1400.0,
+      "logps/rejected": -2288.0,
+      "loss": 0.0231,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.8125,
-      "rewards/margins": 7.84375,
-      "rewards/rejected": -17.625,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 7.78125,
+      "rewards/rejected": -18.125,
       "step": 1950
     },
     {
       "epoch": 2.0523560209424083,
-      "grad_norm": 0.2216880720179684,
+      "grad_norm": 0.07821153721819314,
       "learning_rate": 1.3723589973115963e-05,
-      "logits/chosen": -4.90625,
-      "logits/rejected": -2.015625,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -2176.0,
-      "loss": 0.02,
+      "logits/chosen": -5.90625,
+      "logits/rejected": -2.875,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -2208.0,
+      "loss": 0.0173,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 7.78125,
-      "rewards/rejected": -17.125,
+      "rewards/chosen": -9.9375,
+      "rewards/margins": 7.59375,
+      "rewards/rejected": -17.5,
       "step": 1960
     },
     {
       "epoch": 2.06282722513089,
-      "grad_norm": 0.8990335577264101,
+      "grad_norm": 0.3947686892949864,
       "learning_rate": 1.3452531619894837e-05,
-      "logits/chosen": -4.125,
-      "logits/rejected": -1.8515625,
-      "logps/chosen": -1416.0,
-      "logps/rejected": -2368.0,
-      "loss": 0.0241,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.25,
-      "rewards/margins": 8.0,
-      "rewards/rejected": -18.25,
+      "logits/chosen": -5.53125,
+      "logits/rejected": -3.265625,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -2400.0,
+      "loss": 0.0218,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -18.625,
       "step": 1970
     },
     {
       "epoch": 2.073298429319372,
-      "grad_norm": 0.10990873267220386,
+      "grad_norm": 0.5280253525402364,
       "learning_rate": 1.3183188073985245e-05,
-      "logits/chosen": -4.28125,
-      "logits/rejected": -2.078125,
-      "logps/chosen": -1488.0,
-      "logps/rejected": -2464.0,
-      "loss": 0.0205,
+      "logits/chosen": -6.1875,
+      "logits/rejected": -3.15625,
+      "logps/chosen": -1512.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0192,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.9375,
-      "rewards/margins": 8.5,
-      "rewards/rejected": -19.5,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 8.9375,
+      "rewards/rejected": -20.125,
       "step": 1980
     },
     {
       "epoch": 2.0837696335078535,
-      "grad_norm": 0.23621469902619202,
+      "grad_norm": 0.3924118831485485,
       "learning_rate": 1.2915599333096351e-05,
-      "logits/chosen": -4.03125,
-      "logits/rejected": -2.375,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0152,
+      "logits/chosen": -5.6875,
+      "logits/rejected": -3.734375,
+      "logps/chosen": -1632.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.0138,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.5625,
-      "rewards/margins": 8.125,
-      "rewards/rejected": -19.75,
+      "rewards/chosen": -12.125,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -20.625,
       "step": 1990
     },
     {
       "epoch": 2.094240837696335,
-      "grad_norm": 0.18460184801846521,
+      "grad_norm": 0.3069112396911554,
       "learning_rate": 1.264980513434752e-05,
-      "logits/chosen": -3.9375,
-      "logits/rejected": -1.828125,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0172,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.375,
-      "rewards/margins": 8.5,
-      "rewards/rejected": -19.875,
+      "logits/chosen": -6.1875,
+      "logits/rejected": -3.59375,
+      "logps/chosen": -1608.0,
+      "logps/rejected": -2576.0,
+      "loss": 0.0179,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -11.6875,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -20.375,
       "step": 2000
     },
     {
       "epoch": 2.094240837696335,
-      "eval_logits/chosen": -1.1171875,
-      "eval_logits/rejected": 0.84375,
-      "eval_logps/chosen": -1736.0,
-      "eval_logps/rejected": -1864.0,
-      "eval_loss": 1.6191718578338623,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -14.1875,
-      "eval_rewards/margins": 1.5078125,
-      "eval_rewards/rejected": -15.6875,
-      "eval_runtime": 46.5554,
-      "eval_samples_per_second": 42.96,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -3.640625,
+      "eval_logits/rejected": -1.5078125,
+      "eval_logps/chosen": -1768.0,
+      "eval_logps/rejected": -1896.0,
+      "eval_loss": 1.7014062404632568,
+      "eval_rewards/accuracies": 0.58203125,
+      "eval_rewards/chosen": -14.5,
+      "eval_rewards/margins": 1.59375,
+      "eval_rewards/rejected": -16.0,
+      "eval_runtime": 57.9165,
+      "eval_samples_per_second": 34.532,
+      "eval_steps_per_second": 0.553,
       "step": 2000
     },
     {
       "epoch": 2.1047120418848166,
-      "grad_norm": 0.5058350758325395,
+      "grad_norm": 0.2459800194556874,
       "learning_rate": 1.2385844948367324e-05,
-      "logits/chosen": -3.71875,
-      "logits/rejected": -1.71875,
-      "logps/chosen": -1472.0,
-      "logps/rejected": -2320.0,
-      "loss": 0.0272,
+      "logits/chosen": -5.75,
+      "logits/rejected": -3.46875,
+      "logps/chosen": -1528.0,
+      "logps/rejected": -2368.0,
+      "loss": 0.0278,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.0,
-      "rewards/margins": 7.8125,
-      "rewards/rejected": -18.875,
+      "rewards/chosen": -11.5625,
+      "rewards/margins": 7.78125,
+      "rewards/rejected": -19.25,
       "step": 2010
     },
     {
       "epoch": 2.115183246073298,
-      "grad_norm": 0.17799734937667397,
+      "grad_norm": 0.15810265345486685,
       "learning_rate": 1.2123757973432114e-05,
-      "logits/chosen": -4.40625,
-      "logits/rejected": -2.21875,
-      "logps/chosen": -1448.0,
-      "logps/rejected": -2288.0,
-      "loss": 0.0168,
+      "logits/chosen": -6.59375,
+      "logits/rejected": -4.03125,
+      "logps/chosen": -1552.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.018,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -10.3125,
-      "rewards/margins": 7.84375,
-      "rewards/rejected": -18.125,
+      "rewards/chosen": -11.3125,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -19.625,
       "step": 2020
     },
     {
       "epoch": 2.1256544502617802,
-      "grad_norm": 0.1833178438854485,
+      "grad_norm": 0.20117084297273355,
       "learning_rate": 1.1863583129645106e-05,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -1.859375,
-      "logps/chosen": -1456.0,
-      "logps/rejected": -2384.0,
-      "loss": 0.0226,
+      "logits/chosen": -6.8125,
+      "logits/rejected": -4.15625,
+      "logps/chosen": -1480.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0237,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -10.625,
-      "rewards/margins": 8.125,
-      "rewards/rejected": -18.75,
+      "rewards/chosen": -10.875,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -19.625,
       "step": 2030
     },
     {
       "epoch": 2.136125654450262,
-      "grad_norm": 0.4497694542685086,
+      "grad_norm": 0.6551481489101396,
       "learning_rate": 1.1605359053156607e-05,
-      "logits/chosen": -4.5,
-      "logits/rejected": -2.015625,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2448.0,
-      "loss": 0.0152,
+      "logits/chosen": -6.8125,
+      "logits/rejected": -3.84375,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0184,
       "rewards/accuracies": 0.9937499761581421,
       "rewards/chosen": -10.75,
-      "rewards/margins": 8.375,
-      "rewards/rejected": -19.125,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -19.25,
       "step": 2040
     },
     {
       "epoch": 2.1465968586387434,
-      "grad_norm": 0.1976775875814716,
+      "grad_norm": 0.6107731998650321,
       "learning_rate": 1.1349124090426644e-05,
-      "logits/chosen": -4.3125,
-      "logits/rejected": -1.859375,
+      "logits/chosen": -6.625,
+      "logits/rejected": -3.828125,
       "logps/chosen": -1440.0,
-      "logps/rejected": -2384.0,
-      "loss": 0.0157,
+      "logps/rejected": -2352.0,
+      "loss": 0.0165,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.3125,
-      "rewards/margins": 8.6875,
-      "rewards/rejected": -19.0,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -18.625,
       "step": 2050
     },
     {
       "epoch": 2.157068062827225,
-      "grad_norm": 0.12996888241369076,
+      "grad_norm": 0.9846004109600391,
       "learning_rate": 1.1094916292530406e-05,
-      "logits/chosen": -4.40625,
-      "logits/rejected": -2.28125,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -2384.0,
-      "loss": 0.0192,
+      "logits/chosen": -6.5,
+      "logits/rejected": -4.09375,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -2352.0,
+      "loss": 0.0218,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.875,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -18.625,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -18.375,
       "step": 2060
     },
     {
       "epoch": 2.167539267015707,
-      "grad_norm": 0.3918329152692046,
+      "grad_norm": 0.36498227115395293,
       "learning_rate": 1.0842773409507623e-05,
-      "logits/chosen": -4.0625,
-      "logits/rejected": -1.8203125,
-      "logps/chosen": -1352.0,
-      "logps/rejected": -2288.0,
-      "loss": 0.0191,
+      "logits/chosen": -6.03125,
+      "logits/rejected": -3.390625,
+      "logps/chosen": -1408.0,
+      "logps/rejected": -2384.0,
+      "loss": 0.0203,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.8125,
-      "rewards/margins": 8.3125,
-      "rewards/rejected": -18.125,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -19.0,
       "step": 2070
     },
     {
       "epoch": 2.1780104712041886,
-      "grad_norm": 0.2312966320880158,
+      "grad_norm": 0.2554659312143972,
       "learning_rate": 1.0592732884756754e-05,
-      "logits/chosen": -4.53125,
-      "logits/rejected": -2.28125,
-      "logps/chosen": -1552.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0126,
+      "logits/chosen": -6.96875,
+      "logits/rejected": -4.28125,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0147,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.3125,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -20.5,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 9.4375,
+      "rewards/rejected": -20.625,
       "step": 2080
     },
     {
       "epoch": 2.18848167539267,
-      "grad_norm": 0.6614939144300226,
+      "grad_norm": 0.42233308208813325,
       "learning_rate": 1.0344831849474507e-05,
-      "logits/chosen": -3.53125,
-      "logits/rejected": -1.3671875,
-      "logps/chosen": -1616.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0182,
+      "logits/chosen": -6.0,
+      "logits/rejected": -3.203125,
+      "logps/chosen": -1560.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.0174,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -12.0625,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -21.0,
+      "rewards/chosen": -11.5,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -20.5,
       "step": 2090
     },
     {
       "epoch": 2.1989528795811517,
-      "grad_norm": 0.25438731727877395,
+      "grad_norm": 0.32308181366050115,
       "learning_rate": 1.0099107117141879e-05,
-      "logits/chosen": -3.546875,
-      "logits/rejected": -1.5546875,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2544.0,
-      "loss": 0.012,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.4375,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -20.5,
+      "logits/chosen": -6.65625,
+      "logits/rejected": -3.984375,
+      "logps/chosen": -1464.0,
+      "logps/rejected": -2496.0,
+      "loss": 0.0165,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -10.8125,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -20.0,
       "step": 2100
     },
     {
       "epoch": 2.1989528795811517,
-      "eval_logits/chosen": -1.46875,
-      "eval_logits/rejected": 0.51171875,
-      "eval_logps/chosen": -1792.0,
-      "eval_logps/rejected": -1928.0,
-      "eval_loss": 1.7069687843322754,
-      "eval_rewards/accuracies": 0.6015625,
-      "eval_rewards/chosen": -14.6875,
-      "eval_rewards/margins": 1.6953125,
-      "eval_rewards/rejected": -16.375,
-      "eval_runtime": 46.5608,
-      "eval_samples_per_second": 42.955,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -4.0625,
+      "eval_logits/rejected": -1.953125,
+      "eval_logps/chosen": -1760.0,
+      "eval_logps/rejected": -1904.0,
+      "eval_loss": 1.726171851158142,
+      "eval_rewards/accuracies": 0.580078125,
+      "eval_rewards/chosen": -14.4375,
+      "eval_rewards/margins": 1.6796875,
+      "eval_rewards/rejected": -16.125,
+      "eval_runtime": 57.9361,
+      "eval_samples_per_second": 34.521,
+      "eval_steps_per_second": 0.552,
       "step": 2100
     },
     {
       "epoch": 2.2094240837696333,
-      "grad_norm": 0.37647610997440734,
+      "grad_norm": 0.3197947633834769,
       "learning_rate": 9.855595178057334e-06,
-      "logits/chosen": -4.25,
-      "logits/rejected": -2.265625,
-      "logps/chosen": -1544.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0167,
+      "logits/chosen": -6.5625,
+      "logits/rejected": -4.09375,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0179,
       "rewards/accuracies": 0.981249988079071,
       "rewards/chosen": -11.1875,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -20.125,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -20.25,
       "step": 2110
     },
     {
       "epoch": 2.2198952879581153,
-      "grad_norm": 0.2606530948971035,
+      "grad_norm": 0.6866344841080878,
       "learning_rate": 9.614332193917949e-06,
-      "logits/chosen": -4.625,
-      "logits/rejected": -1.8359375,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0202,
+      "logits/chosen": -6.34375,
+      "logits/rejected": -3.265625,
+      "logps/chosen": -1664.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0199,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -20.5,
+      "rewards/chosen": -12.3125,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -21.5,
       "step": 2120
     },
     {
       "epoch": 2.230366492146597,
-      "grad_norm": 0.37568539050562694,
+      "grad_norm": 0.4758065971615745,
       "learning_rate": 9.375353992449385e-06,
-      "logits/chosen": -4.6875,
-      "logits/rejected": -2.796875,
-      "logps/chosen": -1552.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0131,
+      "logits/chosen": -6.84375,
+      "logits/rejected": -4.375,
+      "logps/chosen": -1528.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0112,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.5625,
-      "rewards/margins": 8.875,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 9.125,
       "rewards/rejected": -20.375,
       "step": 2130
     },
     {
       "epoch": 2.2408376963350785,
-      "grad_norm": 0.1678625123954523,
+      "grad_norm": 0.22115870446522254,
       "learning_rate": 9.138696062085442e-06,
-      "logits/chosen": -4.53125,
-      "logits/rejected": -2.140625,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0168,
+      "logits/chosen": -6.625,
+      "logits/rejected": -3.765625,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0155,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -20.0,
+      "rewards/chosen": -10.9375,
+      "rewards/margins": 8.75,
+      "rewards/rejected": -19.625,
       "step": 2140
     },
     {
       "epoch": 2.25130890052356,
-      "grad_norm": 0.35581490046465525,
+      "grad_norm": 0.3460349861948145,
       "learning_rate": 8.904393546698007e-06,
-      "logits/chosen": -4.5625,
-      "logits/rejected": -2.46875,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.012,
+      "logits/chosen": -6.0625,
+      "logits/rejected": -3.71875,
+      "logps/chosen": -1560.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0132,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 9.625,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -11.5,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -21.0,
       "step": 2150
     },
     {
       "epoch": 2.261780104712042,
-      "grad_norm": 0.9239441901116889,
+      "grad_norm": 0.5306187508744994,
       "learning_rate": 8.672481240378142e-06,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -1.125,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0219,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.5625,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -20.375,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -2.59375,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.0284,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -11.3125,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -19.875,
       "step": 2160
     },
     {
       "epoch": 2.2722513089005236,
-      "grad_norm": 0.753587173568858,
+      "grad_norm": 0.34178650041415626,
       "learning_rate": 8.44299358226919e-06,
-      "logits/chosen": -3.703125,
-      "logits/rejected": -0.94140625,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.021,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -12.0625,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -21.0,
+      "logits/chosen": -5.59375,
+      "logits/rejected": -2.078125,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0213,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -11.5625,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -20.625,
       "step": 2170
     },
     {
       "epoch": 2.282722513089005,
-      "grad_norm": 0.22400444817590756,
+      "grad_norm": 0.2523239648176281,
       "learning_rate": 8.215964651452457e-06,
-      "logits/chosen": -3.875,
-      "logits/rejected": -1.375,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0158,
+      "logits/chosen": -5.625,
+      "logits/rejected": -2.484375,
+      "logps/chosen": -1568.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0143,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -11.1875,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -20.375,
+      "rewards/chosen": -11.6875,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -21.0,
       "step": 2180
     },
     {
       "epoch": 2.2931937172774868,
-      "grad_norm": 0.15059768658422779,
+      "grad_norm": 0.25765959795691407,
       "learning_rate": 7.991428161886502e-06,
-      "logits/chosen": -4.09375,
-      "logits/rejected": -1.890625,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2512.0,
-      "loss": 0.0146,
+      "logits/chosen": -5.84375,
+      "logits/rejected": -3.078125,
+      "logps/chosen": -1568.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0131,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.9375,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -19.875,
+      "rewards/chosen": -11.375,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -20.75,
       "step": 2190
     },
     {
       "epoch": 2.303664921465969,
-      "grad_norm": 0.25924991959215704,
+      "grad_norm": 0.41599379520889174,
       "learning_rate": 7.76941745740061e-06,
-      "logits/chosen": -4.5,
-      "logits/rejected": -2.375,
-      "logps/chosen": -1560.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0145,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -3.546875,
+      "logps/chosen": -1608.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0158,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -11.3125,
+      "rewards/chosen": -11.8125,
       "rewards/margins": 8.875,
-      "rewards/rejected": -20.25,
+      "rewards/rejected": -20.75,
       "step": 2200
     },
     {
       "epoch": 2.303664921465969,
-      "eval_logits/chosen": -1.2890625,
-      "eval_logits/rejected": 0.6875,
-      "eval_logps/chosen": -1728.0,
-      "eval_logps/rejected": -1856.0,
-      "eval_loss": 1.6656562089920044,
-      "eval_rewards/accuracies": 0.595703125,
-      "eval_rewards/chosen": -14.0625,
-      "eval_rewards/margins": 1.5546875,
-      "eval_rewards/rejected": -15.625,
-      "eval_runtime": 46.5684,
-      "eval_samples_per_second": 42.948,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -3.359375,
+      "eval_logits/rejected": -1.234375,
+      "eval_logps/chosen": -1744.0,
+      "eval_logps/rejected": -1872.0,
+      "eval_loss": 1.7524374723434448,
+      "eval_rewards/accuracies": 0.576171875,
+      "eval_rewards/chosen": -14.25,
+      "eval_rewards/margins": 1.5703125,
+      "eval_rewards/rejected": -15.8125,
+      "eval_runtime": 57.9357,
+      "eval_samples_per_second": 34.521,
+      "eval_steps_per_second": 0.552,
       "step": 2200
     },
     {
       "epoch": 2.3141361256544504,
-      "grad_norm": 0.08370834341178045,
+      "grad_norm": 0.3153588187679405,
       "learning_rate": 7.549965506743123e-06,
-      "logits/chosen": -3.78125,
-      "logits/rejected": -1.8046875,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2384.0,
-      "loss": 0.023,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.1875,
-      "rewards/margins": 8.25,
-      "rewards/rejected": -19.375,
+      "logits/chosen": -6.125,
+      "logits/rejected": -3.46875,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -2400.0,
+      "loss": 0.0257,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -11.125,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -19.625,
       "step": 2210
     },
     {
       "epoch": 2.324607329842932,
-      "grad_norm": 0.09462548060973694,
+      "grad_norm": 0.21587810252220344,
       "learning_rate": 7.333104898685631e-06,
-      "logits/chosen": -4.09375,
-      "logits/rejected": -1.9296875,
-      "logps/chosen": -1512.0,
-      "logps/rejected": -2544.0,
-      "loss": 0.0238,
+      "logits/chosen": -6.40625,
+      "logits/rejected": -3.734375,
+      "logps/chosen": -1464.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0209,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 9.0625,
-      "rewards/rejected": -20.125,
+      "rewards/chosen": -10.625,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -19.375,
       "step": 2220
     },
     {
       "epoch": 2.3350785340314135,
-      "grad_norm": 0.3196668367874345,
+      "grad_norm": 0.23997722848341535,
       "learning_rate": 7.118867837183518e-06,
-      "logits/chosen": -4.625,
-      "logits/rejected": -2.078125,
-      "logps/chosen": -1608.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0136,
+      "logits/chosen": -7.03125,
+      "logits/rejected": -3.9375,
+      "logps/chosen": -1504.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0148,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.3125,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -19.75,
       "step": 2230
     },
     {
       "epoch": 2.345549738219895,
-      "grad_norm": 0.15629179933871978,
+      "grad_norm": 0.21092212413197486,
       "learning_rate": 6.9072861365936045e-06,
-      "logits/chosen": -4.84375,
-      "logits/rejected": -2.734375,
-      "logps/chosen": -1648.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0131,
+      "logits/chosen": -7.15625,
+      "logits/rejected": -4.40625,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.0166,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.8125,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -10.75,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -19.375,
       "step": 2240
     },
     {
       "epoch": 2.356020942408377,
-      "grad_norm": 0.42196270416296183,
+      "grad_norm": 0.3540464715144953,
       "learning_rate": 6.6983912169497005e-06,
-      "logits/chosen": -4.375,
-      "logits/rejected": -1.8984375,
-      "logps/chosen": -1560.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.02,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.1875,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -20.125,
+      "logits/chosen": -6.0625,
+      "logits/rejected": -3.296875,
+      "logps/chosen": -1504.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.015,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.6875,
+      "rewards/margins": 8.375,
+      "rewards/rejected": -19.125,
       "step": 2250
     },
     {
       "epoch": 2.3664921465968587,
-      "grad_norm": 0.12157178108969681,
+      "grad_norm": 0.2621236756500411,
       "learning_rate": 6.49221409929677e-06,
-      "logits/chosen": -3.71875,
-      "logits/rejected": -1.2109375,
-      "logps/chosen": -1488.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0116,
+      "logits/chosen": -5.59375,
+      "logits/rejected": -2.75,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.0112,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.0625,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -20.125,
+      "rewards/chosen": -10.5,
+      "rewards/margins": 8.9375,
+      "rewards/rejected": -19.375,
       "step": 2260
     },
     {
       "epoch": 2.3769633507853403,
-      "grad_norm": 0.36179508809937333,
+      "grad_norm": 0.29664420130713154,
       "learning_rate": 6.2887854010842075e-06,
-      "logits/chosen": -3.609375,
-      "logits/rejected": -1.9453125,
-      "logps/chosen": -1552.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0161,
+      "logits/chosen": -5.71875,
+      "logits/rejected": -3.421875,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0142,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 9.5,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -20.0,
       "step": 2270
     },
     {
       "epoch": 2.387434554973822,
-      "grad_norm": 0.08044049955945219,
+      "grad_norm": 0.4991480587718821,
       "learning_rate": 6.0881353316191385e-06,
-      "logits/chosen": -4.09375,
-      "logits/rejected": -2.234375,
-      "logps/chosen": -1584.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0115,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -20.5,
+      "logits/chosen": -5.34375,
+      "logits/rejected": -3.296875,
+      "logps/chosen": -1552.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.017,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -20.375,
       "step": 2280
     },
     {
       "epoch": 2.3979057591623034,
-      "grad_norm": 0.230858695402202,
+      "grad_norm": 0.5326786045280899,
       "learning_rate": 5.890293687580381e-06,
-      "logits/chosen": -3.375,
-      "logits/rejected": -1.0859375,
-      "logps/chosen": -1608.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0144,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.875,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -21.0,
+      "logits/chosen": -5.15625,
+      "logits/rejected": -2.375,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.0168,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 9.4375,
+      "rewards/rejected": -20.625,
       "step": 2290
     },
     {
       "epoch": 2.4083769633507854,
-      "grad_norm": 0.1766856084233999,
+      "grad_norm": 0.1908744652930913,
       "learning_rate": 5.695289848593532e-06,
-      "logits/chosen": -3.796875,
-      "logits/rejected": -2.390625,
-      "logps/chosen": -1632.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0161,
+      "logits/chosen": -5.59375,
+      "logits/rejected": -3.53125,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2496.0,
+      "loss": 0.0199,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -12.375,
-      "rewards/margins": 8.6875,
-      "rewards/rejected": -21.0,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -20.0,
       "step": 2300
     },
     {
       "epoch": 2.4083769633507854,
-      "eval_logits/chosen": -0.9140625,
-      "eval_logits/rejected": 1.0,
-      "eval_logps/chosen": -1872.0,
-      "eval_logps/rejected": -2016.0,
-      "eval_loss": 1.8216874599456787,
-      "eval_rewards/accuracies": 0.603515625,
-      "eval_rewards/chosen": -15.5625,
-      "eval_rewards/margins": 1.734375,
-      "eval_rewards/rejected": -17.25,
-      "eval_runtime": 46.559,
-      "eval_samples_per_second": 42.956,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -2.6875,
+      "eval_logits/rejected": -0.62109375,
+      "eval_logps/chosen": -1760.0,
+      "eval_logps/rejected": -1888.0,
+      "eval_loss": 1.7305155992507935,
+      "eval_rewards/accuracies": 0.583984375,
+      "eval_rewards/chosen": -14.4375,
+      "eval_rewards/margins": 1.5390625,
+      "eval_rewards/rejected": -15.9375,
+      "eval_runtime": 57.9482,
+      "eval_samples_per_second": 34.514,
+      "eval_steps_per_second": 0.552,
       "step": 2300
     },
     {
       "epoch": 2.418848167539267,
-      "grad_norm": 0.0904673525173586,
+      "grad_norm": 0.30952435326563377,
       "learning_rate": 5.50315277286812e-06,
-      "logits/chosen": -4.25,
-      "logits/rejected": -2.078125,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0158,
+      "logits/chosen": -5.6875,
+      "logits/rejected": -3.140625,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0168,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.125,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -11.3125,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -20.125,
       "step": 2310
     },
     {
       "epoch": 2.4293193717277486,
-      "grad_norm": 0.4816670275642321,
+      "grad_norm": 1.4335202796110207,
       "learning_rate": 5.313910992897281e-06,
-      "logits/chosen": -3.390625,
-      "logits/rejected": -1.125,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0116,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 9.4375,
-      "rewards/rejected": -20.5,
+      "logits/chosen": -5.0625,
+      "logits/rejected": -2.484375,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0123,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 8.9375,
+      "rewards/rejected": -19.375,
       "step": 2320
     },
     {
       "epoch": 2.4397905759162306,
-      "grad_norm": 1.3831830674406869,
+      "grad_norm": 0.21426981202790238,
       "learning_rate": 5.127592611220658e-06,
-      "logits/chosen": -4.15625,
-      "logits/rejected": -1.703125,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0274,
+      "logits/chosen": -5.53125,
+      "logits/rejected": -2.796875,
+      "logps/chosen": -1552.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.023,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -11.6875,
-      "rewards/margins": 10.0,
-      "rewards/rejected": -21.75,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 9.3125,
+      "rewards/rejected": -20.625,
       "step": 2330
     },
     {
       "epoch": 2.450261780104712,
-      "grad_norm": 0.1997773428003881,
+      "grad_norm": 0.250774626118941,
       "learning_rate": 4.944225296251159e-06,
-      "logits/chosen": -4.03125,
-      "logits/rejected": -1.90625,
-      "logps/chosen": -1432.0,
-      "logps/rejected": -2464.0,
-      "loss": 0.0171,
+      "logits/chosen": -5.5625,
+      "logits/rejected": -3.25,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -2400.0,
+      "loss": 0.0138,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.75,
-      "rewards/margins": 9.375,
-      "rewards/rejected": -20.125,
+      "rewards/chosen": -10.625,
+      "rewards/margins": 8.875,
+      "rewards/rejected": -19.5,
       "step": 2340
     },
     {
       "epoch": 2.4607329842931938,
-      "grad_norm": 0.8334352967218054,
+      "grad_norm": 0.5344280884049774,
       "learning_rate": 4.763836278166172e-06,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -0.97265625,
-      "logps/chosen": -1496.0,
-      "logps/rejected": -2432.0,
-      "loss": 0.0145,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -10.875,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -19.75,
+      "logits/chosen": -5.8125,
+      "logits/rejected": -2.84375,
+      "logps/chosen": -1480.0,
+      "logps/rejected": -2400.0,
+      "loss": 0.0116,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.75,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -19.375,
       "step": 2350
     },
     {
       "epoch": 2.4712041884816753,
-      "grad_norm": 0.14740320592874556,
+      "grad_norm": 0.13987698833697737,
       "learning_rate": 4.586452344863876e-06,
-      "logits/chosen": -3.5625,
-      "logits/rejected": -1.625,
-      "logps/chosen": -1488.0,
+      "logits/chosen": -5.84375,
+      "logits/rejected": -3.171875,
+      "logps/chosen": -1520.0,
       "logps/rejected": -2496.0,
-      "loss": 0.0164,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -10.8125,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -20.0,
+      "loss": 0.0181,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -19.875,
       "step": 2360
     },
     {
       "epoch": 2.481675392670157,
-      "grad_norm": 0.20786847833989067,
+      "grad_norm": 0.18830002344534685,
       "learning_rate": 4.412099837985192e-06,
-      "logits/chosen": -4.0,
-      "logits/rejected": -1.3203125,
-      "logps/chosen": -1512.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0128,
+      "logits/chosen": -5.5,
+      "logits/rejected": -2.4375,
+      "logps/chosen": -1552.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.0143,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.875,
-      "rewards/margins": 9.875,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -11.3125,
+      "rewards/margins": 10.125,
+      "rewards/rejected": -21.375,
       "step": 2370
     },
     {
       "epoch": 2.492146596858639,
-      "grad_norm": 0.35346365866241725,
+      "grad_norm": 0.2780070982338975,
       "learning_rate": 4.240804649002089e-06,
-      "logits/chosen": -4.0625,
-      "logits/rejected": -1.796875,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -2416.0,
-      "loss": 0.0196,
+      "logits/chosen": -5.65625,
+      "logits/rejected": -3.09375,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.023,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -10.375,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -19.375,
+      "rewards/chosen": -10.75,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -20.0,
       "step": 2380
     },
     {
       "epoch": 2.5026178010471205,
-      "grad_norm": 1.0131439276730239,
+      "grad_norm": 0.7724186453967887,
       "learning_rate": 4.072592215372598e-06,
-      "logits/chosen": -3.90625,
-      "logits/rejected": -1.84375,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.016,
+      "logits/chosen": -6.03125,
+      "logits/rejected": -3.421875,
+      "logps/chosen": -1608.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.02,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 9.3125,
+      "rewards/chosen": -11.6875,
+      "rewards/margins": 9.125,
       "rewards/rejected": -20.875,
       "step": 2390
     },
     {
       "epoch": 2.513089005235602,
-      "grad_norm": 0.5288387915834758,
+      "grad_norm": 0.12627377668940776,
       "learning_rate": 3.907487516763389e-06,
-      "logits/chosen": -3.21875,
-      "logits/rejected": -1.234375,
-      "logps/chosen": -1536.0,
-      "logps/rejected": -2464.0,
-      "loss": 0.0161,
+      "logits/chosen": -5.21875,
+      "logits/rejected": -2.953125,
+      "logps/chosen": -1512.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0172,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.375,
-      "rewards/margins": 8.5625,
-      "rewards/rejected": -20.0,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -19.75,
       "step": 2400
     },
     {
       "epoch": 2.513089005235602,
-      "eval_logits/chosen": -0.2470703125,
-      "eval_logits/rejected": 1.6328125,
-      "eval_logps/chosen": -1824.0,
-      "eval_logps/rejected": -1960.0,
-      "eval_loss": 1.78515625,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -15.0,
-      "eval_rewards/margins": 1.6640625,
-      "eval_rewards/rejected": -16.625,
-      "eval_runtime": 46.5719,
-      "eval_samples_per_second": 42.944,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -2.390625,
+      "eval_logits/rejected": -0.31640625,
+      "eval_logps/chosen": -1776.0,
+      "eval_logps/rejected": -1904.0,
+      "eval_loss": 1.739078164100647,
+      "eval_rewards/accuracies": 0.58203125,
+      "eval_rewards/chosen": -14.5625,
+      "eval_rewards/margins": 1.6015625,
+      "eval_rewards/rejected": -16.125,
+      "eval_runtime": 57.9858,
+      "eval_samples_per_second": 34.491,
+      "eval_steps_per_second": 0.552,
       "step": 2400
     },
     {
       "epoch": 2.5235602094240837,
-      "grad_norm": 0.19563847298675266,
+      "grad_norm": 0.18043395628045303,
       "learning_rate": 3.745515071340286e-06,
-      "logits/chosen": -3.15625,
-      "logits/rejected": -0.609375,
-      "logps/chosen": -1584.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0137,
+      "logits/chosen": -5.21875,
+      "logits/rejected": -2.15625,
+      "logps/chosen": -1560.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0156,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.625,
+      "rewards/chosen": -11.375,
       "rewards/margins": 9.4375,
-      "rewards/rejected": -21.125,
+      "rewards/rejected": -20.75,
       "step": 2410
     },
     {
       "epoch": 2.5340314136125652,
-      "grad_norm": 0.12243899586004475,
+      "grad_norm": 0.1419659676529431,
       "learning_rate": 3.586698932127236e-06,
-      "logits/chosen": -3.75,
-      "logits/rejected": -1.3125,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0102,
+      "logits/chosen": -6.0,
+      "logits/rejected": -2.9375,
+      "logps/chosen": -1480.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.0083,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.0,
-      "rewards/margins": 9.0625,
-      "rewards/rejected": -20.0,
+      "rewards/chosen": -10.75,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -19.5,
       "step": 2420
     },
     {
       "epoch": 2.5445026178010473,
-      "grad_norm": 0.5367341081272166,
+      "grad_norm": 1.1670975345173005,
       "learning_rate": 3.4310626834344744e-06,
-      "logits/chosen": -3.578125,
-      "logits/rejected": -1.875,
-      "logps/chosen": -1608.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0125,
+      "logits/chosen": -5.46875,
+      "logits/rejected": -3.296875,
+      "logps/chosen": -1576.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0147,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.9375,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -21.0,
+      "rewards/chosen": -11.5625,
+      "rewards/margins": 9.3125,
+      "rewards/rejected": -20.875,
       "step": 2430
     },
     {
       "epoch": 2.554973821989529,
-      "grad_norm": 0.24505059064259707,
+      "grad_norm": 0.06181980800362782,
       "learning_rate": 3.2786294373562343e-06,
-      "logits/chosen": -3.125,
-      "logits/rejected": -0.83203125,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0079,
+      "logits/chosen": -4.96875,
+      "logits/rejected": -2.25,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0066,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -20.875,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -20.375,
       "step": 2440
     },
     {
       "epoch": 2.5654450261780104,
-      "grad_norm": 0.9727496148171099,
+      "grad_norm": 0.3568883503963858,
       "learning_rate": 3.1294218303385526e-06,
-      "logits/chosen": -3.09375,
-      "logits/rejected": -0.75,
-      "logps/chosen": -1648.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0137,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.0625,
-      "rewards/margins": 9.4375,
-      "rewards/rejected": -21.5,
+      "logits/chosen": -5.0,
+      "logits/rejected": -2.1875,
+      "logps/chosen": -1616.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.016,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -11.75,
+      "rewards/margins": 9.3125,
+      "rewards/rejected": -21.125,
       "step": 2450
     },
     {
       "epoch": 2.5759162303664924,
-      "grad_norm": 0.20002759818224256,
+      "grad_norm": 0.6141722648321841,
       "learning_rate": 2.9834620198177483e-06,
-      "logits/chosen": -2.96875,
-      "logits/rejected": -1.359375,
-      "logps/chosen": -1624.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0199,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.25,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -21.5,
+      "logits/chosen": -4.59375,
+      "logits/rejected": -2.671875,
+      "logps/chosen": -1608.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0224,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -12.0625,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -21.125,
       "step": 2460
     },
     {
       "epoch": 2.5863874345549736,
-      "grad_norm": 0.35416210030895906,
+      "grad_norm": 0.44219411244755574,
       "learning_rate": 2.840771680930068e-06,
-      "logits/chosen": -2.953125,
-      "logits/rejected": -1.375,
-      "logps/chosen": -1704.0,
-      "logps/rejected": -2880.0,
-      "loss": 0.0112,
+      "logits/chosen": -4.9375,
+      "logits/rejected": -2.71875,
+      "logps/chosen": -1648.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0113,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.75,
-      "rewards/margins": 10.0625,
-      "rewards/rejected": -22.875,
+      "rewards/chosen": -12.1875,
+      "rewards/margins": 9.9375,
+      "rewards/rejected": -22.125,
       "step": 2470
     },
     {
       "epoch": 2.5968586387434556,
-      "grad_norm": 0.3056516930631035,
+      "grad_norm": 0.2374185193713288,
       "learning_rate": 2.701372003292868e-06,
-      "logits/chosen": -2.4375,
-      "logits/rejected": -0.72265625,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0149,
+      "logits/chosen": -4.65625,
+      "logits/rejected": -2.4375,
+      "logps/chosen": -1528.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.0141,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.875,
-      "rewards/margins": 9.625,
-      "rewards/rejected": -21.5,
+      "rewards/chosen": -11.3125,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -20.5,
       "step": 2480
     },
     {
       "epoch": 2.607329842931937,
-      "grad_norm": 0.3449391905763446,
+      "grad_norm": 0.902013066758525,
       "learning_rate": 2.56528368785795e-06,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -0.64453125,
-      "logps/chosen": -1592.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0124,
+      "logits/chosen": -5.375,
+      "logits/rejected": -2.0,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.0132,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.75,
-      "rewards/margins": 9.5,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -20.375,
       "step": 2490
     },
     {
       "epoch": 2.6178010471204187,
-      "grad_norm": 0.40689389569572226,
+      "grad_norm": 0.10769579699362837,
       "learning_rate": 2.432526943837471e-06,
-      "logits/chosen": -2.9375,
-      "logits/rejected": -0.2294921875,
-      "logps/chosen": -1552.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.0182,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.75,
-      "rewards/margins": 10.5,
-      "rewards/rejected": -22.25,
+      "logits/chosen": -4.8125,
+      "logits/rejected": -1.5703125,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0162,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -11.375,
+      "rewards/margins": 9.4375,
+      "rewards/rejected": -20.875,
       "step": 2500
     },
     {
       "epoch": 2.6178010471204187,
-      "eval_logits/chosen": -0.10888671875,
-      "eval_logits/rejected": 1.7578125,
-      "eval_logps/chosen": -1952.0,
-      "eval_logps/rejected": -2096.0,
-      "eval_loss": 1.9599530696868896,
-      "eval_rewards/accuracies": 0.595703125,
-      "eval_rewards/chosen": -16.25,
-      "eval_rewards/margins": 1.8125,
-      "eval_rewards/rejected": -18.125,
-      "eval_runtime": 46.5669,
-      "eval_samples_per_second": 42.949,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -1.921875,
+      "eval_logits/rejected": 0.126953125,
+      "eval_logps/chosen": -1872.0,
+      "eval_logps/rejected": -2008.0,
+      "eval_loss": 1.84556245803833,
+      "eval_rewards/accuracies": 0.58984375,
+      "eval_rewards/chosen": -15.5,
+      "eval_rewards/margins": 1.703125,
+      "eval_rewards/rejected": -17.25,
+      "eval_runtime": 57.933,
+      "eval_samples_per_second": 34.523,
+      "eval_steps_per_second": 0.552,
       "step": 2500
     },
     {
       "epoch": 2.6282722513089007,
-      "grad_norm": 0.1186257541219406,
+      "grad_norm": 0.10206750330442171,
       "learning_rate": 2.3031214857028446e-06,
-      "logits/chosen": -2.625,
-      "logits/rejected": -0.5,
-      "logps/chosen": -1664.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.0122,
+      "logits/chosen": -4.78125,
+      "logits/rejected": -2.109375,
+      "logps/chosen": -1592.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0139,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.375,
-      "rewards/margins": 10.0625,
-      "rewards/rejected": -22.375,
+      "rewards/chosen": -11.625,
+      "rewards/margins": 9.5625,
+      "rewards/rejected": -21.125,
       "step": 2510
     },
     {
       "epoch": 2.6387434554973823,
-      "grad_norm": 0.12010593888186323,
+      "grad_norm": 0.3443173238213999,
       "learning_rate": 2.177086530257139e-06,
-      "logits/chosen": -3.09375,
-      "logits/rejected": -0.65234375,
-      "logps/chosen": -1592.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0165,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.0625,
-      "rewards/margins": 9.6875,
-      "rewards/rejected": -21.75,
+      "logits/chosen": -5.125,
+      "logits/rejected": -2.1875,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0206,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -11.5625,
+      "rewards/margins": 9.3125,
+      "rewards/rejected": -20.875,
       "step": 2520
     },
     {
       "epoch": 2.649214659685864,
-      "grad_norm": 0.26201962820398345,
+      "grad_norm": 0.23702883208543327,
       "learning_rate": 2.054440793781354e-06,
-      "logits/chosen": -3.578125,
-      "logits/rejected": -1.0,
-      "logps/chosen": -1560.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0211,
+      "logits/chosen": -5.5,
+      "logits/rejected": -2.515625,
+      "logps/chosen": -1528.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0194,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.5625,
-      "rewards/margins": 10.0625,
-      "rewards/rejected": -21.625,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 9.75,
+      "rewards/rejected": -21.0,
       "step": 2530
     },
     {
       "epoch": 2.6596858638743455,
-      "grad_norm": 0.15358240942057458,
+      "grad_norm": 0.1406573155365173,
       "learning_rate": 1.9352024892550587e-06,
-      "logits/chosen": -3.171875,
-      "logits/rejected": -1.140625,
-      "logps/chosen": -1584.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0136,
+      "logits/chosen": -4.8125,
+      "logits/rejected": -2.40625,
+      "logps/chosen": -1568.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0139,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.75,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -21.625,
+      "rewards/chosen": -11.5625,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -21.0,
       "step": 2540
     },
     {
       "epoch": 2.670157068062827,
-      "grad_norm": 0.3219251642745328,
+      "grad_norm": 0.35741526447364347,
       "learning_rate": 1.8193893236517379e-06,
-      "logits/chosen": -3.40625,
-      "logits/rejected": -1.15625,
-      "logps/chosen": -1632.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0174,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.125,
-      "rewards/margins": 9.6875,
-      "rewards/rejected": -21.75,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -2.34375,
+      "logps/chosen": -1600.0,
+      "logps/rejected": -2576.0,
+      "loss": 0.0121,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -11.75,
+      "rewards/margins": 9.4375,
+      "rewards/rejected": -21.25,
       "step": 2550
     },
     {
       "epoch": 2.680628272251309,
-      "grad_norm": 0.29245893784851185,
+      "grad_norm": 0.5271373631573718,
       "learning_rate": 1.7070184953092956e-06,
-      "logits/chosen": -3.578125,
-      "logits/rejected": -1.4453125,
+      "logits/chosen": -5.21875,
+      "logits/rejected": -2.78125,
       "logps/chosen": -1528.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0121,
-      "rewards/accuracies": 0.9937499761581421,
+      "logps/rejected": -2544.0,
+      "loss": 0.0127,
+      "rewards/accuracies": 1.0,
       "rewards/chosen": -11.25,
-      "rewards/margins": 9.5625,
-      "rewards/rejected": -20.875,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -20.375,
       "step": 2560
     },
     {
       "epoch": 2.6910994764397906,
-      "grad_norm": 1.29600446209877,
+      "grad_norm": 0.8244537426727586,
       "learning_rate": 1.5981066913760918e-06,
-      "logits/chosen": -2.703125,
-      "logits/rejected": -0.7109375,
-      "logps/chosen": -1456.0,
-      "logps/rejected": -2544.0,
-      "loss": 0.0144,
+      "logits/chosen": -4.6875,
+      "logits/rejected": -1.9296875,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0161,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.875,
-      "rewards/margins": 9.375,
-      "rewards/rejected": -20.25,
+      "rewards/chosen": -10.75,
+      "rewards/margins": 9.3125,
+      "rewards/rejected": -20.0,
       "step": 2570
     },
     {
       "epoch": 2.701570680628272,
-      "grad_norm": 0.27955864510273215,
+      "grad_norm": 0.24408596038043956,
       "learning_rate": 1.492670085332884e-06,
-      "logits/chosen": -3.546875,
-      "logits/rejected": -1.375,
+      "logits/chosen": -5.28125,
+      "logits/rejected": -2.734375,
       "logps/chosen": -1584.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0179,
+      "logps/rejected": -2656.0,
+      "loss": 0.0167,
       "rewards/accuracies": 0.9937499761581421,
       "rewards/chosen": -11.75,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -21.625,
+      "rewards/margins": 9.4375,
+      "rewards/rejected": -21.25,
       "step": 2580
     },
     {
       "epoch": 2.712041884816754,
-      "grad_norm": 0.34893198860474606,
+      "grad_norm": 0.3892302724710085,
       "learning_rate": 1.3907243345910787e-06,
-      "logits/chosen": -3.40625,
-      "logits/rejected": -0.59375,
-      "logps/chosen": -1592.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0166,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -2.09375,
+      "logps/chosen": -1584.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0157,
       "rewards/accuracies": 1.0,
       "rewards/chosen": -11.4375,
-      "rewards/margins": 9.75,
-      "rewards/rejected": -21.25,
+      "rewards/margins": 9.3125,
+      "rewards/rejected": -20.75,
       "step": 2590
     },
     {
       "epoch": 2.7225130890052354,
-      "grad_norm": 0.3863183421211475,
+      "grad_norm": 0.18997197324364945,
       "learning_rate": 1.292284578167538e-06,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -1.625,
-      "logps/chosen": -1512.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0121,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.25,
-      "rewards/margins": 9.6875,
-      "rewards/rejected": -20.875,
+      "logits/chosen": -5.21875,
+      "logits/rejected": -2.75,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0128,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -11.0,
+      "rewards/margins": 9.5625,
+      "rewards/rejected": -20.625,
       "step": 2600
     },
     {
       "epoch": 2.7225130890052354,
-      "eval_logits/chosen": -0.423828125,
-      "eval_logits/rejected": 1.4921875,
-      "eval_logps/chosen": -1832.0,
-      "eval_logps/rejected": -1976.0,
-      "eval_loss": 1.8075780868530273,
-      "eval_rewards/accuracies": 0.611328125,
-      "eval_rewards/chosen": -15.125,
-      "eval_rewards/margins": 1.765625,
-      "eval_rewards/rejected": -16.875,
-      "eval_runtime": 46.5631,
-      "eval_samples_per_second": 42.952,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -2.203125,
+      "eval_logits/rejected": -0.12890625,
+      "eval_logps/chosen": -1824.0,
+      "eval_logps/rejected": -1960.0,
+      "eval_loss": 1.7974061965942383,
+      "eval_rewards/accuracies": 0.587890625,
+      "eval_rewards/chosen": -15.0625,
+      "eval_rewards/margins": 1.6796875,
+      "eval_rewards/rejected": -16.75,
+      "eval_runtime": 57.9021,
+      "eval_samples_per_second": 34.541,
+      "eval_steps_per_second": 0.553,
       "step": 2600
     },
     {
       "epoch": 2.7329842931937174,
-      "grad_norm": 0.5300519915329284,
+      "grad_norm": 0.24677045779539844,
       "learning_rate": 1.1973654344364925e-06,
-      "logits/chosen": -3.375,
-      "logits/rejected": -1.8203125,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0134,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -3.078125,
+      "logps/chosen": -1560.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0127,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.4375,
-      "rewards/margins": 9.375,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 9.4375,
       "rewards/rejected": -20.75,
       "step": 2610
     },
     {
       "epoch": 2.743455497382199,
-      "grad_norm": 0.13051651571039083,
+      "grad_norm": 0.21961455405126123,
       "learning_rate": 1.1059809989586583e-06,
-      "logits/chosen": -2.765625,
-      "logits/rejected": -1.0859375,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.015,
+      "logits/chosen": -4.625,
+      "logits/rejected": -2.578125,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2576.0,
+      "loss": 0.0134,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 9.6875,
-      "rewards/rejected": -20.875,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -20.625,
       "step": 2620
     },
     {
       "epoch": 2.7539267015706805,
-      "grad_norm": 0.42206777705554904,
+      "grad_norm": 0.7908248256092003,
       "learning_rate": 1.018144842388044e-06,
-      "logits/chosen": -2.921875,
-      "logits/rejected": -0.162109375,
-      "logps/chosen": -1568.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0137,
+      "logits/chosen": -4.6875,
+      "logits/rejected": -1.4375,
+      "logps/chosen": -1576.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.015,
       "rewards/accuracies": 0.987500011920929,
       "rewards/chosen": -11.3125,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -21.125,
+      "rewards/margins": 9.5625,
+      "rewards/rejected": -20.875,
       "step": 2630
     },
     {
       "epoch": 2.7643979057591626,
-      "grad_norm": 0.3958547909120679,
+      "grad_norm": 0.3139866693898622,
       "learning_rate": 9.338700084567109e-07,
-      "logits/chosen": -3.125,
-      "logits/rejected": -1.1875,
-      "logps/chosen": -1560.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0129,
+      "logits/chosen": -4.875,
+      "logits/rejected": -2.4375,
+      "logps/chosen": -1568.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0127,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.5,
-      "rewards/margins": 10.0,
-      "rewards/rejected": -21.5,
+      "rewards/chosen": -11.625,
+      "rewards/margins": 9.5625,
+      "rewards/rejected": -21.25,
       "step": 2640
     },
     {
       "epoch": 2.774869109947644,
-      "grad_norm": 1.0354100865961462,
+      "grad_norm": 0.5946615239289051,
       "learning_rate": 8.531690120377606e-07,
-      "logits/chosen": -3.203125,
-      "logits/rejected": -1.1484375,
-      "logps/chosen": -1728.0,
-      "logps/rejected": -2880.0,
-      "loss": 0.0188,
+      "logits/chosen": -4.84375,
+      "logits/rejected": -2.421875,
+      "logps/chosen": -1712.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.0149,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.75,
-      "rewards/margins": 10.1875,
-      "rewards/rejected": -22.875,
+      "rewards/chosen": -12.5625,
+      "rewards/margins": 10.0625,
+      "rewards/rejected": -22.625,
       "step": 2650
     },
     {
       "epoch": 2.7853403141361257,
-      "grad_norm": 0.6002724871336585,
+      "grad_norm": 0.3156636469510393,
       "learning_rate": 7.760538372868636e-07,
-      "logits/chosen": -3.25,
-      "logits/rejected": -1.0859375,
-      "logps/chosen": -1568.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0147,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 9.375,
-      "rewards/rejected": -21.0,
+      "logits/chosen": -5.375,
+      "logits/rejected": -2.671875,
+      "logps/chosen": -1552.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0162,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -11.4375,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -20.625,
       "step": 2660
     },
     {
       "epoch": 2.7958115183246073,
-      "grad_norm": 0.4813715307780967,
+      "grad_norm": 0.2264318718229844,
       "learning_rate": 7.025359358626166e-07,
-      "logits/chosen": -3.234375,
-      "logits/rejected": -0.482421875,
-      "logps/chosen": -1672.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0122,
+      "logits/chosen": -5.15625,
+      "logits/rejected": -1.9375,
+      "logps/chosen": -1648.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.0115,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.3125,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -22.125,
+      "rewards/chosen": -12.0625,
+      "rewards/margins": 9.875,
+      "rewards/rejected": -22.0,
       "step": 2670
     },
     {
       "epoch": 2.806282722513089,
-      "grad_norm": 0.6395583213821416,
+      "grad_norm": 0.5948872814005832,
       "learning_rate": 6.326262252259463e-07,
-      "logits/chosen": -2.828125,
-      "logits/rejected": -0.1357421875,
-      "logps/chosen": -1568.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0169,
+      "logits/chosen": -4.9375,
+      "logits/rejected": -1.5546875,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0158,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.4375,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 9.625,
+      "rewards/rejected": -20.75,
       "step": 2680
     },
     {
       "epoch": 2.816753926701571,
-      "grad_norm": 0.09057359130887888,
+      "grad_norm": 0.23918784681001795,
       "learning_rate": 5.66335087018871e-07,
-      "logits/chosen": -3.53125,
-      "logits/rejected": -0.953125,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -2.09375,
       "logps/chosen": -1584.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0075,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.5625,
-      "rewards/margins": 9.375,
-      "rewards/rejected": -21.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0098,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -11.5,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -20.25,
       "step": 2690
     },
     {
       "epoch": 2.8272251308900525,
-      "grad_norm": 0.1284187754148978,
+      "grad_norm": 0.24883934658919152,
       "learning_rate": 5.036723655228226e-07,
-      "logits/chosen": -3.203125,
-      "logits/rejected": -1.171875,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.016,
+      "logits/chosen": -5.09375,
+      "logits/rejected": -2.390625,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0168,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.4375,
-      "rewards/margins": 9.625,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -20.25,
       "step": 2700
     },
     {
       "epoch": 2.8272251308900525,
-      "eval_logits/chosen": -0.27734375,
-      "eval_logits/rejected": 1.6015625,
-      "eval_logps/chosen": -1872.0,
-      "eval_logps/rejected": -2016.0,
-      "eval_loss": 1.8344218730926514,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -15.5,
-      "eval_rewards/margins": 1.7890625,
-      "eval_rewards/rejected": -17.25,
-      "eval_runtime": 46.5807,
-      "eval_samples_per_second": 42.936,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -2.015625,
+      "eval_logits/rejected": 0.0458984375,
+      "eval_logps/chosen": -1840.0,
+      "eval_logps/rejected": -1976.0,
+      "eval_loss": 1.8012343645095825,
+      "eval_rewards/accuracies": 0.587890625,
+      "eval_rewards/chosen": -15.1875,
+      "eval_rewards/margins": 1.671875,
+      "eval_rewards/rejected": -16.875,
+      "eval_runtime": 57.913,
+      "eval_samples_per_second": 34.535,
+      "eval_steps_per_second": 0.553,
       "step": 2700
     },
     {
       "epoch": 2.837696335078534,
-      "grad_norm": 0.2855056620223909,
+      "grad_norm": 0.5011385654184746,
       "learning_rate": 4.446473661967432e-07,
-      "logits/chosen": -3.609375,
-      "logits/rejected": -1.4140625,
-      "logps/chosen": -1552.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0133,
+      "logits/chosen": -5.46875,
+      "logits/rejected": -2.71875,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0123,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.5,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -20.5,
+      "rewards/chosen": -11.375,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -20.625,
       "step": 2710
     },
     {
       "epoch": 2.8481675392670156,
-      "grad_norm": 0.5377348487244686,
+      "grad_norm": 0.2805017291584205,
       "learning_rate": 3.892688542952411e-07,
-      "logits/chosen": -2.671875,
-      "logits/rejected": -0.50390625,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2512.0,
-      "loss": 0.0174,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -20.0,
+      "logits/chosen": -4.6875,
+      "logits/rejected": -2.0,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0171,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -10.8125,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -19.5,
       "step": 2720
     },
     {
       "epoch": 2.858638743455497,
-      "grad_norm": 0.42584412060885607,
+      "grad_norm": 1.8639522286247052,
       "learning_rate": 3.3754505356694e-07,
-      "logits/chosen": -3.796875,
-      "logits/rejected": -2.21875,
-      "logps/chosen": -1608.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0155,
+      "logits/chosen": -5.25,
+      "logits/rejected": -3.296875,
+      "logps/chosen": -1624.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0162,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.9375,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -12.125,
+      "rewards/margins": 8.875,
+      "rewards/rejected": -21.0,
       "step": 2730
     },
     {
       "epoch": 2.869109947643979,
-      "grad_norm": 0.20885810094615484,
+      "grad_norm": 0.20975765455233325,
       "learning_rate": 2.8948364503322276e-07,
-      "logits/chosen": -3.09375,
-      "logits/rejected": -1.203125,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0177,
+      "logits/chosen": -4.96875,
+      "logits/rejected": -2.484375,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0179,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.1875,
-      "rewards/margins": 9.5625,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -20.375,
       "step": 2740
     },
     {
       "epoch": 2.8795811518324608,
-      "grad_norm": 0.36471525053270304,
+      "grad_norm": 0.2167402110054763,
       "learning_rate": 2.4509176584762706e-07,
-      "logits/chosen": -3.3125,
-      "logits/rejected": -0.7734375,
-      "logps/chosen": -1536.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0123,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.3125,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -20.25,
+      "logits/chosen": -5.15625,
+      "logits/rejected": -2.125,
+      "logps/chosen": -1504.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.013,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -19.875,
       "step": 2750
     },
     {
       "epoch": 2.8900523560209423,
-      "grad_norm": 0.36096788508082134,
+      "grad_norm": 0.3409419957494618,
       "learning_rate": 2.0437600823595694e-07,
-      "logits/chosen": -3.625,
-      "logits/rejected": -1.3515625,
-      "logps/chosen": -1616.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.012,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.6875,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -21.0,
+      "logits/chosen": -5.25,
+      "logits/rejected": -2.53125,
+      "logps/chosen": -1600.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.021,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -11.5,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -20.5,
       "step": 2760
     },
     {
       "epoch": 2.900523560209424,
-      "grad_norm": 0.26491731503762794,
+      "grad_norm": 0.4930270526261948,
       "learning_rate": 1.6734241851733e-07,
-      "logits/chosen": -3.109375,
-      "logits/rejected": -0.90625,
-      "logps/chosen": -1696.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.0129,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.375,
-      "rewards/margins": 9.875,
-      "rewards/rejected": -22.25,
+      "logits/chosen": -5.0625,
+      "logits/rejected": -2.109375,
+      "logps/chosen": -1664.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0142,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -12.0625,
+      "rewards/margins": 9.625,
+      "rewards/rejected": -21.625,
       "step": 2770
     },
     {
       "epoch": 2.9109947643979055,
-      "grad_norm": 0.2810854233311721,
+      "grad_norm": 0.4501869241169962,
       "learning_rate": 1.339964962062984e-07,
-      "logits/chosen": -3.375,
-      "logits/rejected": -1.734375,
-      "logps/chosen": -1552.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0087,
+      "logits/chosen": -4.9375,
+      "logits/rejected": -2.75,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.0109,
       "rewards/accuracies": 0.9937499761581421,
       "rewards/chosen": -11.75,
-      "rewards/margins": 10.0,
-      "rewards/rejected": -21.75,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -21.5,
       "step": 2780
     },
     {
       "epoch": 2.9214659685863875,
-      "grad_norm": 0.17188412636063793,
+      "grad_norm": 0.17782552597098292,
       "learning_rate": 1.0434319319617136e-07,
-      "logits/chosen": -2.828125,
-      "logits/rejected": -0.63671875,
-      "logps/chosen": -1528.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0165,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -11.4375,
-      "rewards/margins": 9.5,
-      "rewards/rejected": -21.0,
+      "logits/chosen": -4.71875,
+      "logits/rejected": -1.828125,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0185,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -11.3125,
+      "rewards/margins": 9.25,
+      "rewards/rejected": -20.5,
       "step": 2790
     },
     {
       "epoch": 2.931937172774869,
-      "grad_norm": 0.1008720441863218,
+      "grad_norm": 0.9247880921767014,
       "learning_rate": 7.838691302365363e-08,
-      "logits/chosen": -3.25,
-      "logits/rejected": -1.0625,
-      "logps/chosen": -1528.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0144,
+      "logits/chosen": -4.8125,
+      "logits/rejected": -2.234375,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0171,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.25,
-      "rewards/margins": 9.875,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -20.5,
       "step": 2800
     },
     {
       "epoch": 2.931937172774869,
-      "eval_logits/chosen": -0.2451171875,
-      "eval_logits/rejected": 1.625,
-      "eval_logps/chosen": -1872.0,
-      "eval_logps/rejected": -2024.0,
-      "eval_loss": 1.8411874771118164,
-      "eval_rewards/accuracies": 0.603515625,
-      "eval_rewards/chosen": -15.5625,
-      "eval_rewards/margins": 1.7890625,
-      "eval_rewards/rejected": -17.375,
-      "eval_runtime": 46.5759,
-      "eval_samples_per_second": 42.941,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -1.96875,
+      "eval_logits/rejected": 0.08154296875,
+      "eval_logps/chosen": -1840.0,
+      "eval_logps/rejected": -1984.0,
+      "eval_loss": 1.809000015258789,
+      "eval_rewards/accuracies": 0.5859375,
+      "eval_rewards/chosen": -15.25,
+      "eval_rewards/margins": 1.671875,
+      "eval_rewards/rejected": -16.875,
+      "eval_runtime": 57.8949,
+      "eval_samples_per_second": 34.545,
+      "eval_steps_per_second": 0.553,
       "step": 2800
     },
     {
       "epoch": 2.9424083769633507,
-      "grad_norm": 0.20847873691535207,
+      "grad_norm": 0.4441844078136973,
       "learning_rate": 5.613151021490459e-08,
-      "logits/chosen": -2.34375,
-      "logits/rejected": -0.98828125,
-      "logps/chosen": -1480.0,
-      "logps/rejected": -2512.0,
-      "loss": 0.014,
+      "logits/chosen": -4.21875,
+      "logits/rejected": -2.4375,
+      "logps/chosen": -1464.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0117,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.375,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -20.5,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -19.875,
       "step": 2810
     },
     {
       "epoch": 2.9528795811518327,
-      "grad_norm": 1.270603721877663,
+      "grad_norm": 0.4173712761798942,
       "learning_rate": 3.7580289713159965e-08,
-      "logits/chosen": -3.078125,
-      "logits/rejected": -0.96484375,
-      "logps/chosen": -1632.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.016,
+      "logits/chosen": -4.84375,
+      "logits/rejected": -2.359375,
+      "logps/chosen": -1600.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0164,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.125,
+      "rewards/chosen": -11.8125,
       "rewards/margins": 9.6875,
-      "rewards/rejected": -21.875,
+      "rewards/rejected": -21.5,
       "step": 2820
     },
     {
       "epoch": 2.9633507853403143,
-      "grad_norm": 0.7366434851365266,
+      "grad_norm": 0.24913905453407612,
       "learning_rate": 2.2736006387927277e-08,
-      "logits/chosen": -3.25,
-      "logits/rejected": -1.2734375,
-      "logps/chosen": -1616.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0146,
+      "logits/chosen": -5.28125,
+      "logits/rejected": -2.46875,
+      "logps/chosen": -1568.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0135,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.9375,
+      "rewards/chosen": -11.4375,
       "rewards/margins": 9.375,
-      "rewards/rejected": -21.25,
+      "rewards/rejected": -20.875,
       "step": 2830
     },
     {
       "epoch": 2.973821989528796,
-      "grad_norm": 0.6464590934273693,
+      "grad_norm": 0.5579622297604556,
       "learning_rate": 1.1600864625893604e-08,
-      "logits/chosen": -3.09375,
-      "logits/rejected": -1.109375,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0128,
+      "logits/chosen": -5.03125,
+      "logits/rejected": -2.1875,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0139,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.8125,
-      "rewards/margins": 9.375,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -11.125,
+      "rewards/margins": 9.875,
+      "rewards/rejected": -21.0,
       "step": 2840
     },
     {
       "epoch": 2.9842931937172774,
-      "grad_norm": 0.3063168501519172,
+      "grad_norm": 0.40668501818874236,
       "learning_rate": 4.176518003573549e-09,
-      "logits/chosen": -3.21875,
-      "logits/rejected": -0.97265625,
+      "logits/chosen": -5.09375,
+      "logits/rejected": -2.453125,
       "logps/chosen": -1576.0,
       "logps/rejected": -2656.0,
-      "loss": 0.0212,
+      "loss": 0.0248,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.75,
-      "rewards/margins": 9.625,
-      "rewards/rejected": -21.375,
+      "rewards/chosen": -11.8125,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -21.25,
       "step": 2850
     },
     {
       "epoch": 2.994764397905759,
-      "grad_norm": 0.9739798622405211,
+      "grad_norm": 0.13149410839023332,
       "learning_rate": 4.640690417528903e-10,
-      "logits/chosen": -3.6875,
-      "logits/rejected": -1.515625,
-      "logps/chosen": -1544.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0189,
+      "logits/chosen": -5.3125,
+      "logits/rejected": -2.734375,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2496.0,
+      "loss": 0.017,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -20.875,
+      "rewards/chosen": -11.5625,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -20.5,
       "step": 2860
     },
     {
       "epoch": 3.0,
       "step": 2865,
       "total_flos": 0.0,
-      "train_loss": 0.22363413207193944,
-      "train_runtime": 12304.9,
-      "train_samples_per_second": 14.901,
-      "train_steps_per_second": 0.233
+      "train_loss": 0.2247055232888949,
+      "train_runtime": 12462.0729,
+      "train_samples_per_second": 14.713,
+      "train_steps_per_second": 0.23
     }
   ],
   "logging_steps": 10,