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": 1.5146546699768613,
+      "grad_norm": 1.5146951048181116,
       "learning_rate": 1.7421602787456448e-07,
       "logits/chosen": -11.625,
       "logits/rejected": -11.5625,
@@ -25,4750 +25,4750 @@
     },
     {
       "epoch": 0.010471204188481676,
-      "grad_norm": 1.3103312123640602,
+      "grad_norm": 1.3146134391384272,
       "learning_rate": 1.7421602787456445e-06,
       "logits/chosen": -11.5,
       "logits/rejected": -11.0625,
       "logps/chosen": -456.0,
       "logps/rejected": -452.0,
-      "loss": 0.6919,
-      "rewards/accuracies": 0.1458333283662796,
-      "rewards/chosen": -0.0002307891845703125,
-      "rewards/margins": 2.7060508728027344e-05,
-      "rewards/rejected": -0.0002574920654296875,
+      "loss": 0.6921,
+      "rewards/accuracies": 0.1388888955116272,
+      "rewards/chosen": 0.00104522705078125,
+      "rewards/margins": 8.20159912109375e-05,
+      "rewards/rejected": 0.0009613037109375,
       "step": 10
     },
     {
       "epoch": 0.020942408376963352,
-      "grad_norm": 1.3168976332024394,
+      "grad_norm": 1.31748305020593,
       "learning_rate": 3.484320557491289e-06,
       "logits/chosen": -11.1875,
       "logits/rejected": -11.0,
       "logps/chosen": -410.0,
       "logps/rejected": -400.0,
-      "loss": 0.6913,
-      "rewards/accuracies": 0.2874999940395355,
-      "rewards/chosen": -0.002685546875,
-      "rewards/margins": 0.003509521484375,
-      "rewards/rejected": -0.006195068359375,
+      "loss": 0.6914,
+      "rewards/accuracies": 0.30000001192092896,
+      "rewards/chosen": -0.0032501220703125,
+      "rewards/margins": 0.0023651123046875,
+      "rewards/rejected": -0.005615234375,
       "step": 20
     },
     {
       "epoch": 0.031413612565445025,
-      "grad_norm": 1.3671543225461076,
+      "grad_norm": 1.3660963804119755,
       "learning_rate": 5.226480836236934e-06,
       "logits/chosen": -11.625,
       "logits/rejected": -11.4375,
       "logps/chosen": -444.0,
-      "logps/rejected": -430.0,
-      "loss": 0.6876,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": -0.0245361328125,
-      "rewards/margins": 0.00909423828125,
-      "rewards/rejected": -0.03369140625,
+      "logps/rejected": -428.0,
+      "loss": 0.6874,
+      "rewards/accuracies": 0.39375001192092896,
+      "rewards/chosen": -0.0240478515625,
+      "rewards/margins": 0.00836181640625,
+      "rewards/rejected": -0.032470703125,
       "step": 30
     },
     {
       "epoch": 0.041884816753926704,
-      "grad_norm": 1.5871791398049164,
+      "grad_norm": 1.5844964719771166,
       "learning_rate": 6.968641114982578e-06,
-      "logits/chosen": -12.0625,
-      "logits/rejected": -11.75,
+      "logits/chosen": -12.0,
+      "logits/rejected": -11.6875,
       "logps/chosen": -446.0,
       "logps/rejected": -428.0,
-      "loss": 0.675,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": -0.08837890625,
-      "rewards/margins": 0.03955078125,
-      "rewards/rejected": -0.1279296875,
+      "loss": 0.6754,
+      "rewards/accuracies": 0.59375,
+      "rewards/chosen": -0.0888671875,
+      "rewards/margins": 0.037841796875,
+      "rewards/rejected": -0.126953125,
       "step": 40
     },
     {
       "epoch": 0.05235602094240838,
-      "grad_norm": 1.8702430641118655,
+      "grad_norm": 1.9114079133692858,
       "learning_rate": 8.710801393728225e-06,
       "logits/chosen": -12.25,
-      "logits/rejected": -11.875,
+      "logits/rejected": -11.8125,
       "logps/chosen": -460.0,
       "logps/rejected": -480.0,
-      "loss": 0.6556,
+      "loss": 0.6554,
       "rewards/accuracies": 0.6312500238418579,
-      "rewards/chosen": -0.2021484375,
-      "rewards/margins": 0.10302734375,
-      "rewards/rejected": -0.3046875,
+      "rewards/chosen": -0.197265625,
+      "rewards/margins": 0.10498046875,
+      "rewards/rejected": -0.30078125,
       "step": 50
     },
     {
       "epoch": 0.06282722513089005,
-      "grad_norm": 1.9069564602113085,
+      "grad_norm": 1.9327882365035682,
       "learning_rate": 1.0452961672473868e-05,
       "logits/chosen": -12.0,
       "logits/rejected": -11.8125,
       "logps/chosen": -404.0,
       "logps/rejected": -402.0,
-      "loss": 0.6357,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -0.46875,
-      "rewards/margins": 0.11328125,
+      "loss": 0.6352,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.46484375,
+      "rewards/margins": 0.115234375,
       "rewards/rejected": -0.58203125,
       "step": 60
     },
     {
       "epoch": 0.07329842931937172,
-      "grad_norm": 1.9385328367549584,
+      "grad_norm": 1.7699236839683066,
       "learning_rate": 1.2195121951219513e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.375,
-      "logps/chosen": -504.0,
-      "logps/rejected": -548.0,
-      "loss": 0.6102,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -0.58203125,
-      "rewards/margins": 0.267578125,
-      "rewards/rejected": -0.84765625,
+      "logits/chosen": -10.9375,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -506.0,
+      "logps/rejected": -552.0,
+      "loss": 0.6104,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.6015625,
+      "rewards/margins": 0.2734375,
+      "rewards/rejected": -0.875,
       "step": 70
     },
     {
       "epoch": 0.08376963350785341,
-      "grad_norm": 2.2856327662560347,
+      "grad_norm": 1.9937556983566087,
       "learning_rate": 1.3937282229965156e-05,
-      "logits/chosen": -11.375,
-      "logits/rejected": -10.9375,
-      "logps/chosen": -478.0,
-      "logps/rejected": -540.0,
-      "loss": 0.5984,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.71484375,
-      "rewards/margins": 0.333984375,
-      "rewards/rejected": -1.046875,
+      "logits/chosen": -10.9375,
+      "logits/rejected": -10.5,
+      "logps/chosen": -486.0,
+      "logps/rejected": -548.0,
+      "loss": 0.5978,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -0.796875,
+      "rewards/margins": 0.330078125,
+      "rewards/rejected": -1.125,
       "step": 80
     },
     {
       "epoch": 0.09424083769633508,
-      "grad_norm": 2.499797489171207,
+      "grad_norm": 2.385939809622327,
       "learning_rate": 1.56794425087108e-05,
-      "logits/chosen": -12.125,
-      "logits/rejected": -11.8125,
-      "logps/chosen": -482.0,
-      "logps/rejected": -506.0,
-      "loss": 0.6089,
-      "rewards/accuracies": 0.5562499761581421,
-      "rewards/chosen": -1.1015625,
-      "rewards/margins": 0.220703125,
-      "rewards/rejected": -1.3203125,
+      "logits/chosen": -10.5625,
+      "logits/rejected": -10.125,
+      "logps/chosen": -506.0,
+      "logps/rejected": -532.0,
+      "loss": 0.5996,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -1.328125,
+      "rewards/margins": 0.2578125,
+      "rewards/rejected": -1.59375,
       "step": 90
     },
     {
       "epoch": 0.10471204188481675,
-      "grad_norm": 2.0157046034481607,
+      "grad_norm": 2.2746479695155495,
       "learning_rate": 1.742160278745645e-05,
-      "logits/chosen": -12.75,
-      "logits/rejected": -12.375,
-      "logps/chosen": -628.0,
-      "logps/rejected": -676.0,
-      "loss": 0.5851,
-      "rewards/accuracies": 0.65625,
-      "rewards/chosen": -1.28125,
-      "rewards/margins": 0.455078125,
-      "rewards/rejected": -1.734375,
+      "logits/chosen": -10.8125,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -664.0,
+      "logps/rejected": -716.0,
+      "loss": 0.5786,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -1.640625,
+      "rewards/margins": 0.50390625,
+      "rewards/rejected": -2.140625,
       "step": 100
     },
     {
       "epoch": 0.10471204188481675,
-      "eval_logits/chosen": -12.0625,
-      "eval_logits/rejected": -11.75,
-      "eval_logps/chosen": -456.0,
-      "eval_logps/rejected": -450.0,
-      "eval_loss": 0.6820625066757202,
-      "eval_rewards/accuracies": 0.607421875,
-      "eval_rewards/chosen": -1.375,
-      "eval_rewards/margins": 0.2353515625,
-      "eval_rewards/rejected": -1.609375,
-      "eval_runtime": 48.7134,
-      "eval_samples_per_second": 41.056,
-      "eval_steps_per_second": 0.657,
+      "eval_logits/chosen": -9.75,
+      "eval_logits/rejected": -9.25,
+      "eval_logps/chosen": -500.0,
+      "eval_logps/rejected": -494.0,
+      "eval_loss": 0.6689062714576721,
+      "eval_rewards/accuracies": 0.609375,
+      "eval_rewards/chosen": -1.8203125,
+      "eval_rewards/margins": 0.2373046875,
+      "eval_rewards/rejected": -2.0625,
+      "eval_runtime": 48.3864,
+      "eval_samples_per_second": 41.334,
+      "eval_steps_per_second": 0.661,
       "step": 100
     },
     {
       "epoch": 0.11518324607329843,
-      "grad_norm": 1.8809166613965007,
+      "grad_norm": 2.1372183582369817,
       "learning_rate": 1.9163763066202093e-05,
-      "logits/chosen": -11.5,
-      "logits/rejected": -11.25,
-      "logps/chosen": -604.0,
-      "logps/rejected": -620.0,
-      "loss": 0.5825,
-      "rewards/accuracies": 0.65625,
-      "rewards/chosen": -1.6328125,
-      "rewards/margins": 0.349609375,
-      "rewards/rejected": -1.984375,
+      "logits/chosen": -10.5,
+      "logits/rejected": -10.125,
+      "logps/chosen": -616.0,
+      "logps/rejected": -636.0,
+      "loss": 0.5762,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -1.78125,
+      "rewards/margins": 0.3515625,
+      "rewards/rejected": -2.125,
       "step": 110
     },
     {
       "epoch": 0.1256544502617801,
-      "grad_norm": 2.397815402887626,
+      "grad_norm": 2.237693941993736,
       "learning_rate": 2.0905923344947736e-05,
-      "logits/chosen": -10.6875,
-      "logits/rejected": -10.25,
-      "logps/chosen": -576.0,
-      "logps/rejected": -628.0,
-      "loss": 0.5604,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": -1.875,
-      "rewards/margins": 0.4375,
-      "rewards/rejected": -2.3125,
+      "logits/chosen": -9.9375,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -620.0,
+      "logps/rejected": -676.0,
+      "loss": 0.5563,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -2.3125,
+      "rewards/margins": 0.470703125,
+      "rewards/rejected": -2.78125,
       "step": 120
     },
     {
       "epoch": 0.13612565445026178,
-      "grad_norm": 2.6998928722302153,
+      "grad_norm": 2.4884459680204736,
       "learning_rate": 2.264808362369338e-05,
-      "logits/chosen": -10.375,
-      "logits/rejected": -9.75,
-      "logps/chosen": -624.0,
-      "logps/rejected": -676.0,
-      "loss": 0.5531,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -1.859375,
-      "rewards/margins": 0.5,
-      "rewards/rejected": -2.359375,
+      "logits/chosen": -11.0625,
+      "logits/rejected": -10.375,
+      "logps/chosen": -584.0,
+      "logps/rejected": -628.0,
+      "loss": 0.5687,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": -1.4765625,
+      "rewards/margins": 0.421875,
+      "rewards/rejected": -1.8984375,
       "step": 130
     },
     {
       "epoch": 0.14659685863874344,
-      "grad_norm": 2.8681729887428618,
+      "grad_norm": 2.1374795132077806,
       "learning_rate": 2.4390243902439026e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -708.0,
-      "logps/rejected": -744.0,
-      "loss": 0.5476,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -2.359375,
-      "rewards/margins": 0.6328125,
-      "rewards/rejected": -2.984375,
+      "logits/chosen": -12.125,
+      "logits/rejected": -11.5,
+      "logps/chosen": -624.0,
+      "logps/rejected": -648.0,
+      "loss": 0.5633,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.515625,
+      "rewards/margins": 0.482421875,
+      "rewards/rejected": -2.0,
       "step": 140
     },
     {
       "epoch": 0.15706806282722513,
-      "grad_norm": 2.1948797770311304,
+      "grad_norm": 2.1931080488148313,
       "learning_rate": 2.6132404181184672e-05,
-      "logits/chosen": -11.375,
-      "logits/rejected": -10.75,
-      "logps/chosen": -600.0,
-      "logps/rejected": -668.0,
-      "loss": 0.5682,
-      "rewards/accuracies": 0.65625,
-      "rewards/chosen": -1.4921875,
-      "rewards/margins": 0.462890625,
-      "rewards/rejected": -1.953125,
+      "logits/chosen": -10.375,
+      "logits/rejected": -9.5,
+      "logps/chosen": -676.0,
+      "logps/rejected": -752.0,
+      "loss": 0.5506,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -2.21875,
+      "rewards/margins": 0.58984375,
+      "rewards/rejected": -2.8125,
       "step": 150
     },
     {
       "epoch": 0.16753926701570682,
-      "grad_norm": 2.494444171952847,
+      "grad_norm": 2.4556173388149687,
       "learning_rate": 2.7874564459930312e-05,
-      "logits/chosen": -8.875,
-      "logits/rejected": -7.84375,
+      "logits/chosen": -10.1875,
+      "logits/rejected": -9.0625,
       "logps/chosen": -700.0,
-      "logps/rejected": -784.0,
-      "loss": 0.5255,
-      "rewards/accuracies": 0.731249988079071,
+      "logps/rejected": -780.0,
+      "loss": 0.5239,
+      "rewards/accuracies": 0.7124999761581421,
       "rewards/chosen": -2.84375,
-      "rewards/margins": 0.76953125,
-      "rewards/rejected": -3.625,
+      "rewards/margins": 0.7421875,
+      "rewards/rejected": -3.59375,
       "step": 160
     },
     {
       "epoch": 0.17801047120418848,
-      "grad_norm": 1.859186034402492,
+      "grad_norm": 2.3699115190175664,
       "learning_rate": 2.9616724738675962e-05,
-      "logits/chosen": -8.1875,
-      "logits/rejected": -7.1875,
-      "logps/chosen": -808.0,
-      "logps/rejected": -900.0,
-      "loss": 0.5377,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 0.8046875,
-      "rewards/rejected": -4.3125,
+      "logits/chosen": -8.875,
+      "logits/rejected": -7.75,
+      "logps/chosen": -752.0,
+      "logps/rejected": -848.0,
+      "loss": 0.5376,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 0.8359375,
+      "rewards/rejected": -3.796875,
       "step": 170
     },
     {
       "epoch": 0.18848167539267016,
-      "grad_norm": 2.844887972165444,
+      "grad_norm": 2.0383070066302653,
       "learning_rate": 3.13588850174216e-05,
-      "logits/chosen": -8.3125,
-      "logits/rejected": -7.28125,
-      "logps/chosen": -692.0,
-      "logps/rejected": -748.0,
-      "loss": 0.5485,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -2.34375,
-      "rewards/margins": 0.65234375,
-      "rewards/rejected": -2.984375,
+      "logits/chosen": -9.0625,
+      "logits/rejected": -8.0,
+      "logps/chosen": -672.0,
+      "logps/rejected": -724.0,
+      "loss": 0.5456,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -2.140625,
+      "rewards/margins": 0.625,
+      "rewards/rejected": -2.765625,
       "step": 180
     },
     {
       "epoch": 0.19895287958115182,
-      "grad_norm": 3.3952197285586228,
+      "grad_norm": 2.4415281399319366,
       "learning_rate": 3.310104529616725e-05,
-      "logits/chosen": -8.75,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -824.0,
-      "logps/rejected": -908.0,
-      "loss": 0.5442,
-      "rewards/accuracies": 0.6312500238418579,
-      "rewards/chosen": -4.15625,
-      "rewards/margins": 0.63671875,
-      "rewards/rejected": -4.78125,
+      "logits/chosen": -7.0,
+      "logits/rejected": -6.0,
+      "logps/chosen": -672.0,
+      "logps/rejected": -752.0,
+      "loss": 0.5605,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -2.625,
+      "rewards/margins": 0.609375,
+      "rewards/rejected": -3.21875,
       "step": 190
     },
     {
       "epoch": 0.2094240837696335,
-      "grad_norm": 1.872136277367507,
+      "grad_norm": 1.8935163555003156,
       "learning_rate": 3.48432055749129e-05,
-      "logits/chosen": -8.5625,
-      "logits/rejected": -7.5,
-      "logps/chosen": -788.0,
-      "logps/rejected": -860.0,
-      "loss": 0.5386,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -3.171875,
-      "rewards/margins": 0.7421875,
-      "rewards/rejected": -3.921875,
+      "logits/chosen": -4.4375,
+      "logits/rejected": -3.0625,
+      "logps/chosen": -856.0,
+      "logps/rejected": -940.0,
+      "loss": 0.5359,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -3.859375,
+      "rewards/margins": 0.84765625,
+      "rewards/rejected": -4.71875,
       "step": 200
     },
     {
       "epoch": 0.2094240837696335,
-      "eval_logits/chosen": -6.53125,
-      "eval_logits/rejected": -5.65625,
-      "eval_logps/chosen": -636.0,
-      "eval_logps/rejected": -648.0,
-      "eval_loss": 0.6998124718666077,
-      "eval_rewards/accuracies": 0.578125,
-      "eval_rewards/chosen": -3.1875,
-      "eval_rewards/margins": 0.416015625,
-      "eval_rewards/rejected": -3.609375,
-      "eval_runtime": 46.8487,
-      "eval_samples_per_second": 42.691,
-      "eval_steps_per_second": 0.683,
+      "eval_logits/chosen": -2.890625,
+      "eval_logits/rejected": -1.78125,
+      "eval_logps/chosen": -652.0,
+      "eval_logps/rejected": -672.0,
+      "eval_loss": 0.736578106880188,
+      "eval_rewards/accuracies": 0.58984375,
+      "eval_rewards/chosen": -3.328125,
+      "eval_rewards/margins": 0.482421875,
+      "eval_rewards/rejected": -3.8125,
+      "eval_runtime": 46.6414,
+      "eval_samples_per_second": 42.88,
+      "eval_steps_per_second": 0.686,
       "step": 200
     },
     {
       "epoch": 0.2198952879581152,
-      "grad_norm": 2.335873088339001,
+      "grad_norm": 2.3667936980293987,
       "learning_rate": 3.6585365853658535e-05,
-      "logits/chosen": -7.375,
-      "logits/rejected": -6.34375,
-      "logps/chosen": -656.0,
-      "logps/rejected": -744.0,
-      "loss": 0.5282,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -2.71875,
-      "rewards/margins": 0.8125,
-      "rewards/rejected": -3.53125,
+      "logits/chosen": -5.34375,
+      "logits/rejected": -4.4375,
+      "logps/chosen": -664.0,
+      "logps/rejected": -748.0,
+      "loss": 0.5607,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -2.8125,
+      "rewards/margins": 0.734375,
+      "rewards/rejected": -3.546875,
       "step": 210
     },
     {
       "epoch": 0.23036649214659685,
-      "grad_norm": 2.186335333393021,
+      "grad_norm": 2.0646242132084125,
       "learning_rate": 3.8327526132404185e-05,
-      "logits/chosen": -8.375,
-      "logits/rejected": -7.15625,
-      "logps/chosen": -736.0,
-      "logps/rejected": -792.0,
-      "loss": 0.5189,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -2.984375,
-      "rewards/margins": 0.76953125,
-      "rewards/rejected": -3.75,
+      "logits/chosen": -7.96875,
+      "logits/rejected": -6.875,
+      "logps/chosen": -704.0,
+      "logps/rejected": -756.0,
+      "loss": 0.527,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -2.671875,
+      "rewards/margins": 0.734375,
+      "rewards/rejected": -3.40625,
       "step": 220
     },
     {
       "epoch": 0.24083769633507854,
-      "grad_norm": 1.9823213488525824,
+      "grad_norm": 2.5005731416323465,
       "learning_rate": 4.006968641114983e-05,
-      "logits/chosen": -7.59375,
-      "logits/rejected": -6.53125,
-      "logps/chosen": -676.0,
-      "logps/rejected": -760.0,
-      "loss": 0.516,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -2.796875,
-      "rewards/margins": 0.86328125,
-      "rewards/rejected": -3.65625,
+      "logits/chosen": -7.65625,
+      "logits/rejected": -6.6875,
+      "logps/chosen": -744.0,
+      "logps/rejected": -828.0,
+      "loss": 0.5118,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -3.46875,
+      "rewards/margins": 0.859375,
+      "rewards/rejected": -4.34375,
       "step": 230
     },
     {
       "epoch": 0.2513089005235602,
-      "grad_norm": 2.1858802805465247,
+      "grad_norm": 1.9776261092191019,
       "learning_rate": 4.181184668989547e-05,
-      "logits/chosen": -7.71875,
-      "logits/rejected": -6.84375,
-      "logps/chosen": -812.0,
-      "logps/rejected": -892.0,
-      "loss": 0.5124,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -4.21875,
-      "rewards/margins": 0.80859375,
-      "rewards/rejected": -5.03125,
+      "logits/chosen": -7.8125,
+      "logits/rejected": -6.875,
+      "logps/chosen": -772.0,
+      "logps/rejected": -848.0,
+      "loss": 0.5303,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -3.828125,
+      "rewards/margins": 0.75,
+      "rewards/rejected": -4.59375,
       "step": 240
     },
     {
       "epoch": 0.2617801047120419,
-      "grad_norm": 2.070066051247774,
+      "grad_norm": 2.0616220990657514,
       "learning_rate": 4.3554006968641115e-05,
-      "logits/chosen": -8.25,
-      "logits/rejected": -6.96875,
-      "logps/chosen": -900.0,
-      "logps/rejected": -1016.0,
-      "loss": 0.5121,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -4.28125,
-      "rewards/margins": 1.078125,
-      "rewards/rejected": -5.34375,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -7.59375,
+      "logps/chosen": -808.0,
+      "logps/rejected": -912.0,
+      "loss": 0.5141,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 1.0078125,
+      "rewards/rejected": -4.34375,
       "step": 250
     },
     {
       "epoch": 0.27225130890052357,
-      "grad_norm": 2.1265433747281617,
+      "grad_norm": 2.129464825834919,
       "learning_rate": 4.529616724738676e-05,
-      "logits/chosen": -7.125,
-      "logits/rejected": -5.5625,
-      "logps/chosen": -696.0,
-      "logps/rejected": -756.0,
-      "loss": 0.5203,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -3.0625,
-      "rewards/margins": 0.79296875,
-      "rewards/rejected": -3.859375,
+      "logits/chosen": -6.3125,
+      "logits/rejected": -4.9375,
+      "logps/chosen": -732.0,
+      "logps/rejected": -792.0,
+      "loss": 0.5192,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 0.80859375,
+      "rewards/rejected": -4.25,
       "step": 260
     },
     {
       "epoch": 0.28272251308900526,
-      "grad_norm": 1.9498355382790653,
+      "grad_norm": 2.3181331810357975,
       "learning_rate": 4.703832752613241e-05,
-      "logits/chosen": -9.25,
-      "logits/rejected": -8.25,
-      "logps/chosen": -780.0,
-      "logps/rejected": -844.0,
-      "loss": 0.5128,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -3.859375,
-      "rewards/margins": 0.68359375,
-      "rewards/rejected": -4.53125,
+      "logits/chosen": -6.125,
+      "logits/rejected": -4.875,
+      "logps/chosen": -832.0,
+      "logps/rejected": -892.0,
+      "loss": 0.5281,
+      "rewards/accuracies": 0.6187499761581421,
+      "rewards/chosen": -4.375,
+      "rewards/margins": 0.63671875,
+      "rewards/rejected": -5.03125,
       "step": 270
     },
     {
       "epoch": 0.2931937172774869,
-      "grad_norm": 2.3383305268219736,
+      "grad_norm": 2.1349422205950255,
       "learning_rate": 4.878048780487805e-05,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -12.75,
-      "logps/chosen": -688.0,
-      "logps/rejected": -728.0,
-      "loss": 0.5285,
+      "logits/chosen": -9.75,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -716.0,
+      "logps/rejected": -768.0,
+      "loss": 0.5355,
       "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 0.54296875,
-      "rewards/rejected": -3.6875,
+      "rewards/chosen": -3.421875,
+      "rewards/margins": 0.65625,
+      "rewards/rejected": -4.0625,
       "step": 280
     },
     {
       "epoch": 0.3036649214659686,
-      "grad_norm": 2.0436073696107275,
+      "grad_norm": 2.6845002357725756,
       "learning_rate": 4.999983293481418e-05,
-      "logits/chosen": -8.875,
-      "logits/rejected": -7.65625,
-      "logps/chosen": -768.0,
+      "logits/chosen": -8.0,
+      "logits/rejected": -6.5,
+      "logps/chosen": -764.0,
       "logps/rejected": -840.0,
-      "loss": 0.5213,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -3.71875,
-      "rewards/margins": 0.6484375,
+      "loss": 0.5184,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -3.65625,
+      "rewards/margins": 0.7109375,
       "rewards/rejected": -4.375,
       "step": 290
     },
     {
       "epoch": 0.31413612565445026,
-      "grad_norm": 2.092747096852258,
+      "grad_norm": 2.550168306465655,
       "learning_rate": 4.9996862949181595e-05,
-      "logits/chosen": -9.75,
-      "logits/rejected": -9.0,
-      "logps/chosen": -760.0,
-      "logps/rejected": -792.0,
-      "loss": 0.5183,
-      "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -3.390625,
-      "rewards/margins": 0.6484375,
-      "rewards/rejected": -4.03125,
+      "logits/chosen": -7.8125,
+      "logits/rejected": -6.625,
+      "logps/chosen": -788.0,
+      "logps/rejected": -828.0,
+      "loss": 0.5163,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -3.703125,
+      "rewards/margins": 0.7265625,
+      "rewards/rejected": -4.4375,
       "step": 300
     },
     {
       "epoch": 0.31413612565445026,
-      "eval_logits/chosen": -8.0625,
-      "eval_logits/rejected": -7.0625,
-      "eval_logps/chosen": -792.0,
-      "eval_logps/rejected": -820.0,
-      "eval_loss": 0.71875,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -4.75,
-      "eval_rewards/margins": 0.58203125,
-      "eval_rewards/rejected": -5.3125,
-      "eval_runtime": 46.8576,
-      "eval_samples_per_second": 42.683,
-      "eval_steps_per_second": 0.683,
+      "eval_logits/chosen": -6.75,
+      "eval_logits/rejected": -5.46875,
+      "eval_logps/chosen": -744.0,
+      "eval_logps/rejected": -776.0,
+      "eval_loss": 0.697390615940094,
+      "eval_rewards/accuracies": 0.642578125,
+      "eval_rewards/chosen": -4.25,
+      "eval_rewards/margins": 0.6015625,
+      "eval_rewards/rejected": -4.84375,
+      "eval_runtime": 46.6231,
+      "eval_samples_per_second": 42.897,
+      "eval_steps_per_second": 0.686,
       "step": 300
     },
     {
       "epoch": 0.32460732984293195,
-      "grad_norm": 2.0712771317275562,
+      "grad_norm": 1.9912878846017152,
       "learning_rate": 4.9990180911526896e-05,
-      "logits/chosen": -8.5625,
-      "logits/rejected": -7.9375,
-      "logps/chosen": -764.0,
-      "logps/rejected": -828.0,
-      "loss": 0.5359,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -3.875,
-      "rewards/margins": 0.68359375,
-      "rewards/rejected": -4.5625,
+      "logits/chosen": -6.9375,
+      "logits/rejected": -6.0625,
+      "logps/chosen": -772.0,
+      "logps/rejected": -832.0,
+      "loss": 0.5373,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -3.9375,
+      "rewards/margins": 0.65625,
+      "rewards/rejected": -4.59375,
       "step": 310
     },
     {
       "epoch": 0.33507853403141363,
-      "grad_norm": 2.0342507128222445,
+      "grad_norm": 2.0015280089987093,
       "learning_rate": 4.997978781413745e-05,
-      "logits/chosen": -9.5,
-      "logits/rejected": -8.375,
-      "logps/chosen": -776.0,
-      "logps/rejected": -816.0,
-      "loss": 0.5204,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -3.234375,
-      "rewards/margins": 0.84765625,
-      "rewards/rejected": -4.09375,
+      "logits/chosen": -7.34375,
+      "logits/rejected": -5.4375,
+      "logps/chosen": -852.0,
+      "logps/rejected": -900.0,
+      "loss": 0.5128,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -4.0,
+      "rewards/margins": 0.94140625,
+      "rewards/rejected": -4.9375,
       "step": 320
     },
     {
       "epoch": 0.34554973821989526,
-      "grad_norm": 2.366077984609981,
+      "grad_norm": 2.641599539175695,
       "learning_rate": 4.996568520039565e-05,
-      "logits/chosen": -6.71875,
-      "logits/rejected": -5.53125,
-      "logps/chosen": -828.0,
-      "logps/rejected": -916.0,
-      "loss": 0.5321,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -3.859375,
-      "rewards/margins": 0.875,
-      "rewards/rejected": -4.71875,
+      "logits/chosen": -7.34375,
+      "logits/rejected": -6.0,
+      "logps/chosen": -820.0,
+      "logps/rejected": -908.0,
+      "loss": 0.5367,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -3.78125,
+      "rewards/margins": 0.8515625,
+      "rewards/rejected": -4.625,
       "step": 330
     },
     {
       "epoch": 0.35602094240837695,
-      "grad_norm": 2.6042554244870626,
+      "grad_norm": 2.3123766801719583,
       "learning_rate": 4.994787516454973e-05,
-      "logits/chosen": -6.625,
-      "logits/rejected": -5.25,
-      "logps/chosen": -816.0,
-      "logps/rejected": -892.0,
-      "loss": 0.5205,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -3.953125,
-      "rewards/margins": 0.8828125,
-      "rewards/rejected": -4.84375,
+      "logits/chosen": -6.4375,
+      "logits/rejected": -5.03125,
+      "logps/chosen": -908.0,
+      "logps/rejected": -972.0,
+      "loss": 0.5168,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -4.84375,
+      "rewards/margins": 0.8046875,
+      "rewards/rejected": -5.65625,
       "step": 340
     },
     {
       "epoch": 0.36649214659685864,
-      "grad_norm": 2.003748685917312,
+      "grad_norm": 2.835068285274983,
       "learning_rate": 4.992636035140273e-05,
-      "logits/chosen": -9.125,
-      "logits/rejected": -8.3125,
-      "logps/chosen": -740.0,
-      "logps/rejected": -824.0,
-      "loss": 0.5368,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 0.6328125,
-      "rewards/rejected": -4.1875,
+      "logits/chosen": -8.6875,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -808.0,
+      "logps/rejected": -900.0,
+      "loss": 0.5217,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -4.21875,
+      "rewards/margins": 0.70703125,
+      "rewards/rejected": -4.90625,
       "step": 350
     },
     {
       "epoch": 0.3769633507853403,
-      "grad_norm": 2.0548903262238847,
+      "grad_norm": 2.080215993679113,
       "learning_rate": 4.99011439559198e-05,
-      "logits/chosen": -7.75,
-      "logits/rejected": -6.375,
-      "logps/chosen": -996.0,
-      "logps/rejected": -1096.0,
-      "loss": 0.5107,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -5.8125,
-      "rewards/margins": 1.0390625,
-      "rewards/rejected": -6.875,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -6.84375,
+      "logps/chosen": -872.0,
+      "logps/rejected": -968.0,
+      "loss": 0.4957,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -4.5625,
+      "rewards/margins": 1.0,
+      "rewards/rejected": -5.5625,
       "step": 360
     },
     {
       "epoch": 0.387434554973822,
-      "grad_norm": 2.3446809862023557,
+      "grad_norm": 3.1739166263176273,
       "learning_rate": 4.987222972275369e-05,
-      "logits/chosen": -8.0625,
-      "logits/rejected": -6.46875,
-      "logps/chosen": -940.0,
-      "logps/rejected": -1064.0,
-      "loss": 0.515,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -5.21875,
-      "rewards/margins": 1.1171875,
-      "rewards/rejected": -6.3125,
+      "logits/chosen": -9.375,
+      "logits/rejected": -7.84375,
+      "logps/chosen": -844.0,
+      "logps/rejected": -968.0,
+      "loss": 0.5208,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -4.21875,
+      "rewards/margins": 1.1640625,
+      "rewards/rejected": -5.40625,
       "step": 370
     },
     {
       "epoch": 0.39790575916230364,
-      "grad_norm": 1.797674609218526,
+      "grad_norm": 1.9772179200085087,
       "learning_rate": 4.983962194568872e-05,
-      "logits/chosen": -7.46875,
-      "logits/rejected": -5.625,
-      "logps/chosen": -968.0,
-      "logps/rejected": -1088.0,
-      "loss": 0.4877,
+      "logits/chosen": -7.71875,
+      "logits/rejected": -5.65625,
+      "logps/chosen": -948.0,
+      "logps/rejected": -1072.0,
+      "loss": 0.4866,
       "rewards/accuracies": 0.75,
-      "rewards/chosen": -5.28125,
-      "rewards/margins": 1.15625,
-      "rewards/rejected": -6.4375,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 1.234375,
+      "rewards/rejected": -6.34375,
       "step": 380
     },
     {
       "epoch": 0.4083769633507853,
-      "grad_norm": 2.199429821772013,
+      "grad_norm": 2.4997397069627705,
       "learning_rate": 4.9803325467003084e-05,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -10.25,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -1208.0,
-      "loss": 0.5017,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -6.25,
-      "rewards/margins": 1.09375,
-      "rewards/rejected": -7.34375,
+      "logits/chosen": -10.375,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -948.0,
+      "logps/rejected": -1080.0,
+      "loss": 0.4988,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -5.0625,
+      "rewards/margins": 1.0234375,
+      "rewards/rejected": -6.09375,
       "step": 390
     },
     {
       "epoch": 0.418848167539267,
-      "grad_norm": 2.179880939171049,
+      "grad_norm": 1.604137359816435,
       "learning_rate": 4.9763345676749816e-05,
-      "logits/chosen": -12.1875,
-      "logits/rejected": -11.0625,
-      "logps/chosen": -1008.0,
-      "logps/rejected": -1080.0,
-      "loss": 0.4924,
+      "logits/chosen": -10.5,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -948.0,
+      "logps/rejected": -1024.0,
+      "loss": 0.5127,
       "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -5.96875,
-      "rewards/margins": 0.9765625,
-      "rewards/rejected": -6.9375,
+      "rewards/chosen": -5.375,
+      "rewards/margins": 1.0,
+      "rewards/rejected": -6.375,
       "step": 400
     },
     {
       "epoch": 0.418848167539267,
-      "eval_logits/chosen": -9.5,
-      "eval_logits/rejected": -7.59375,
-      "eval_logps/chosen": -952.0,
-      "eval_logps/rejected": -996.0,
-      "eval_loss": 0.8316875100135803,
-      "eval_rewards/accuracies": 0.591796875,
-      "eval_rewards/chosen": -6.34375,
-      "eval_rewards/margins": 0.72265625,
-      "eval_rewards/rejected": -7.0625,
-      "eval_runtime": 46.8264,
-      "eval_samples_per_second": 42.711,
-      "eval_steps_per_second": 0.683,
+      "eval_logits/chosen": -9.125,
+      "eval_logits/rejected": -7.9375,
+      "eval_logps/chosen": -856.0,
+      "eval_logps/rejected": -896.0,
+      "eval_loss": 0.7937187552452087,
+      "eval_rewards/accuracies": 0.6015625,
+      "eval_rewards/chosen": -5.375,
+      "eval_rewards/margins": 0.6796875,
+      "eval_rewards/rejected": -6.0625,
+      "eval_runtime": 46.6339,
+      "eval_samples_per_second": 42.887,
+      "eval_steps_per_second": 0.686,
       "step": 400
     },
     {
       "epoch": 0.4293193717277487,
-      "grad_norm": 2.3338751029426197,
+      "grad_norm": 2.335114869498187,
       "learning_rate": 4.9719688511956374e-05,
-      "logits/chosen": -9.375,
-      "logits/rejected": -7.75,
-      "logps/chosen": -952.0,
-      "logps/rejected": -1080.0,
-      "loss": 0.5123,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -5.03125,
-      "rewards/margins": 1.0546875,
-      "rewards/rejected": -6.0625,
+      "logits/chosen": -9.125,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -936.0,
+      "logps/rejected": -1064.0,
+      "loss": 0.53,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 1.046875,
+      "rewards/rejected": -5.90625,
       "step": 410
     },
     {
       "epoch": 0.4397905759162304,
-      "grad_norm": 2.0363983435539037,
+      "grad_norm": 2.187805687086,
       "learning_rate": 4.967236045574294e-05,
-      "logits/chosen": -6.59375,
-      "logits/rejected": -4.53125,
-      "logps/chosen": -936.0,
-      "logps/rejected": -1004.0,
-      "loss": 0.5119,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -5.125,
-      "rewards/margins": 0.90234375,
-      "rewards/rejected": -6.0,
-      "step": 420
-    },
-    {
-      "epoch": 0.450261780104712,
-      "grad_norm": 2.0918225529260317,
-      "learning_rate": 4.96213685363597e-05,
-      "logits/chosen": -8.9375,
-      "logits/rejected": -7.3125,
-      "logps/chosen": -824.0,
-      "logps/rejected": -924.0,
-      "loss": 0.5007,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -3.953125,
-      "rewards/margins": 0.984375,
-      "rewards/rejected": -4.9375,
+      "logits/chosen": -8.0625,
+      "logits/rejected": -6.0,
+      "logps/chosen": -844.0,
+      "logps/rejected": -924.0,
+      "loss": 0.4913,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -4.15625,
+      "rewards/margins": 1.03125,
+      "rewards/rejected": -5.21875,
+      "step": 420
+    },
+    {
+      "epoch": 0.450261780104712,
+      "grad_norm": 2.042533983434017,
+      "learning_rate": 4.96213685363597e-05,
+      "logits/chosen": -9.4375,
+      "logits/rejected": -7.21875,
+      "logps/chosen": -772.0,
+      "logps/rejected": -868.0,
+      "loss": 0.4962,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 0.95703125,
+      "rewards/rejected": -4.40625,
       "step": 430
     },
     {
       "epoch": 0.4607329842931937,
-      "grad_norm": 4.601984632508327,
+      "grad_norm": 2.47817739262179,
       "learning_rate": 4.956672032614315e-05,
-      "logits/chosen": -8.625,
-      "logits/rejected": -6.90625,
-      "logps/chosen": -880.0,
+      "logits/chosen": -8.0625,
+      "logits/rejected": -5.9375,
+      "logps/chosen": -876.0,
       "logps/rejected": -992.0,
-      "loss": 0.4944,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -4.46875,
-      "rewards/margins": 1.03125,
-      "rewards/rejected": -5.5,
+      "loss": 0.4747,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -4.40625,
+      "rewards/margins": 1.0703125,
+      "rewards/rejected": -5.46875,
       "step": 440
     },
     {
       "epoch": 0.4712041884816754,
-      "grad_norm": 2.028435305057175,
+      "grad_norm": 2.245158133298755,
       "learning_rate": 4.950842394039157e-05,
-      "logits/chosen": -8.625,
-      "logits/rejected": -7.3125,
+      "logits/chosen": -7.65625,
+      "logits/rejected": -6.0,
       "logps/chosen": -856.0,
-      "logps/rejected": -960.0,
-      "loss": 0.5064,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -4.53125,
-      "rewards/margins": 0.87109375,
-      "rewards/rejected": -5.40625,
+      "logps/rejected": -964.0,
+      "loss": 0.4914,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -4.5625,
+      "rewards/margins": 0.9375,
+      "rewards/rejected": -5.5,
       "step": 450
     },
     {
       "epoch": 0.4816753926701571,
-      "grad_norm": 1.61887330624493,
+      "grad_norm": 1.6742862305781647,
       "learning_rate": 4.944648803615994e-05,
-      "logits/chosen": -8.125,
-      "logits/rejected": -6.0,
-      "logps/chosen": -952.0,
-      "logps/rejected": -1080.0,
-      "loss": 0.4975,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -5.25,
-      "rewards/margins": 1.109375,
-      "rewards/rejected": -6.375,
+      "logits/chosen": -6.90625,
+      "logits/rejected": -4.34375,
+      "logps/chosen": -948.0,
+      "logps/rejected": -1096.0,
+      "loss": 0.4897,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -5.21875,
+      "rewards/margins": 1.3203125,
+      "rewards/rejected": -6.53125,
       "step": 460
     },
     {
       "epoch": 0.49214659685863876,
-      "grad_norm": 2.1744031798777024,
+      "grad_norm": 2.260093723856323,
       "learning_rate": 4.938092181097433e-05,
-      "logits/chosen": -9.5625,
-      "logits/rejected": -7.78125,
-      "logps/chosen": -836.0,
-      "logps/rejected": -908.0,
-      "loss": 0.4768,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -4.21875,
-      "rewards/margins": 1.0078125,
-      "rewards/rejected": -5.21875,
+      "logits/chosen": -9.4375,
+      "logits/rejected": -7.625,
+      "logps/chosen": -860.0,
+      "logps/rejected": -928.0,
+      "loss": 0.4787,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -4.46875,
+      "rewards/margins": 0.921875,
+      "rewards/rejected": -5.375,
       "step": 470
     },
     {
       "epoch": 0.5026178010471204,
-      "grad_norm": 3.3076901057594172,
+      "grad_norm": 3.263482530946283,
       "learning_rate": 4.931173500146607e-05,
-      "logits/chosen": -8.3125,
-      "logits/rejected": -6.6875,
-      "logps/chosen": -1024.0,
-      "logps/rejected": -1216.0,
-      "loss": 0.4901,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -6.28125,
-      "rewards/margins": 1.546875,
-      "rewards/rejected": -7.84375,
+      "logits/chosen": -5.78125,
+      "logits/rejected": -3.90625,
+      "logps/chosen": -940.0,
+      "logps/rejected": -1120.0,
+      "loss": 0.4785,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -5.46875,
+      "rewards/margins": 1.4453125,
+      "rewards/rejected": -6.9375,
       "step": 480
     },
     {
       "epoch": 0.5130890052356021,
-      "grad_norm": 2.0369537118023637,
+      "grad_norm": 2.0699989387956705,
       "learning_rate": 4.923893788192582e-05,
-      "logits/chosen": -11.6875,
-      "logits/rejected": -10.625,
-      "logps/chosen": -872.0,
-      "logps/rejected": -1056.0,
-      "loss": 0.4741,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -4.34375,
-      "rewards/margins": 1.3515625,
-      "rewards/rejected": -5.71875,
+      "logits/chosen": -5.5,
+      "logits/rejected": -3.34375,
+      "logps/chosen": -912.0,
+      "logps/rejected": -1096.0,
+      "loss": 0.4789,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -4.75,
+      "rewards/margins": 1.3671875,
+      "rewards/rejected": -6.09375,
       "step": 490
     },
     {
       "epoch": 0.5235602094240838,
-      "grad_norm": 2.057437462412614,
+      "grad_norm": 2.302061806227037,
       "learning_rate": 4.9162541262777906e-05,
-      "logits/chosen": -12.0,
-      "logits/rejected": -11.25,
-      "logps/chosen": -872.0,
-      "logps/rejected": -996.0,
-      "loss": 0.5057,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -4.3125,
-      "rewards/margins": 1.125,
-      "rewards/rejected": -5.46875,
+      "logits/chosen": -6.71875,
+      "logits/rejected": -5.0,
+      "logps/chosen": -892.0,
+      "logps/rejected": -1008.0,
+      "loss": 0.5047,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 1.0859375,
+      "rewards/rejected": -5.59375,
       "step": 500
     },
     {
       "epoch": 0.5235602094240838,
-      "eval_logits/chosen": -11.0,
-      "eval_logits/rejected": -9.75,
-      "eval_logps/chosen": -828.0,
-      "eval_logps/rejected": -872.0,
-      "eval_loss": 0.7777031064033508,
-      "eval_rewards/accuracies": 0.591796875,
-      "eval_rewards/chosen": -5.125,
-      "eval_rewards/margins": 0.70703125,
-      "eval_rewards/rejected": -5.8125,
-      "eval_runtime": 46.8187,
-      "eval_samples_per_second": 42.718,
-      "eval_steps_per_second": 0.683,
+      "eval_logits/chosen": -5.59375,
+      "eval_logits/rejected": -3.71875,
+      "eval_logps/chosen": -776.0,
+      "eval_logps/rejected": -812.0,
+      "eval_loss": 0.7908594012260437,
+      "eval_rewards/accuracies": 0.5703125,
+      "eval_rewards/chosen": -4.59375,
+      "eval_rewards/margins": 0.65234375,
+      "eval_rewards/rejected": -5.21875,
+      "eval_runtime": 46.6406,
+      "eval_samples_per_second": 42.881,
+      "eval_steps_per_second": 0.686,
       "step": 500
     },
     {
       "epoch": 0.5340314136125655,
-      "grad_norm": 1.984532947602431,
+      "grad_norm": 2.9007407978674142,
       "learning_rate": 4.90825564889749e-05,
-      "logits/chosen": -9.5625,
-      "logits/rejected": -8.25,
-      "logps/chosen": -964.0,
-      "logps/rejected": -1048.0,
-      "loss": 0.4685,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -5.125,
-      "rewards/margins": 1.078125,
-      "rewards/rejected": -6.1875,
+      "logits/chosen": -5.59375,
+      "logits/rejected": -3.859375,
+      "logps/chosen": -984.0,
+      "logps/rejected": -1056.0,
+      "loss": 0.4814,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -5.3125,
+      "rewards/margins": 0.984375,
+      "rewards/rejected": -6.28125,
       "step": 510
     },
     {
       "epoch": 0.5445026178010471,
-      "grad_norm": 2.064108317793805,
+      "grad_norm": 1.8672919815129103,
       "learning_rate": 4.899899543831288e-05,
-      "logits/chosen": -8.8125,
-      "logits/rejected": -7.4375,
-      "logps/chosen": -900.0,
-      "logps/rejected": -1008.0,
-      "loss": 0.4968,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -4.46875,
-      "rewards/margins": 1.171875,
-      "rewards/rejected": -5.65625,
+      "logits/chosen": -6.4375,
+      "logits/rejected": -4.625,
+      "logps/chosen": -936.0,
+      "logps/rejected": -1040.0,
+      "loss": 0.5257,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -4.8125,
+      "rewards/margins": 1.1015625,
+      "rewards/rejected": -5.9375,
       "step": 520
     },
     {
       "epoch": 0.5549738219895288,
-      "grad_norm": 2.4878903419415646,
+      "grad_norm": 2.0436049019093545,
       "learning_rate": 4.891187051966765e-05,
-      "logits/chosen": -11.125,
-      "logits/rejected": -10.4375,
-      "logps/chosen": -868.0,
-      "logps/rejected": -908.0,
-      "loss": 0.5013,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -4.25,
-      "rewards/margins": 0.91015625,
-      "rewards/rejected": -5.15625,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -4.84375,
+      "logps/chosen": -944.0,
+      "logps/rejected": -976.0,
+      "loss": 0.5174,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 0.8046875,
+      "rewards/rejected": -5.84375,
       "step": 530
     },
     {
       "epoch": 0.5654450261780105,
-      "grad_norm": 2.1489671994319215,
+      "grad_norm": 2.2588681558073596,
       "learning_rate": 4.882119467115191e-05,
-      "logits/chosen": -10.5625,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -964.0,
-      "logps/rejected": -1048.0,
-      "loss": 0.4953,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -5.4375,
-      "rewards/margins": 0.9375,
-      "rewards/rejected": -6.375,
+      "logits/chosen": -5.96875,
+      "logits/rejected": -4.1875,
+      "logps/chosen": -976.0,
+      "logps/rejected": -1064.0,
+      "loss": 0.5109,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -5.59375,
+      "rewards/margins": 0.9296875,
+      "rewards/rejected": -6.53125,
       "step": 540
     },
     {
       "epoch": 0.5759162303664922,
-      "grad_norm": 2.6333038384528833,
+      "grad_norm": 2.165127223564882,
       "learning_rate": 4.872698135819402e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -9.125,
-      "logps/chosen": -884.0,
-      "logps/rejected": -972.0,
-      "loss": 0.5323,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -4.65625,
-      "rewards/margins": 0.8828125,
-      "rewards/rejected": -5.53125,
+      "logits/chosen": -6.625,
+      "logits/rejected": -5.03125,
+      "logps/chosen": -944.0,
+      "logps/rejected": -1032.0,
+      "loss": 0.5198,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -5.28125,
+      "rewards/margins": 0.86328125,
+      "rewards/rejected": -6.125,
       "step": 550
     },
     {
       "epoch": 0.5863874345549738,
-      "grad_norm": 2.1137260079514832,
+      "grad_norm": 2.3207232690882194,
       "learning_rate": 4.862924457153831e-05,
-      "logits/chosen": -7.8125,
-      "logits/rejected": -6.1875,
-      "logps/chosen": -904.0,
-      "logps/rejected": -1048.0,
-      "loss": 0.4906,
-      "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -5.46875,
-      "rewards/margins": 1.0703125,
-      "rewards/rejected": -6.53125,
+      "logits/chosen": -4.59375,
+      "logits/rejected": -2.78125,
+      "logps/chosen": -996.0,
+      "logps/rejected": -1152.0,
+      "loss": 0.4851,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": -6.40625,
+      "rewards/margins": 1.1484375,
+      "rewards/rejected": -7.5625,
       "step": 560
     },
     {
       "epoch": 0.5968586387434555,
-      "grad_norm": 2.2688188952354267,
+      "grad_norm": 2.078433174332932,
       "learning_rate": 4.8527998825167484e-05,
-      "logits/chosen": -6.6875,
-      "logits/rejected": -5.34375,
-      "logps/chosen": -924.0,
-      "logps/rejected": -1048.0,
-      "loss": 0.4994,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -5.34375,
-      "rewards/margins": 1.25,
-      "rewards/rejected": -6.59375,
+      "logits/chosen": -6.59375,
+      "logits/rejected": -5.0625,
+      "logps/chosen": -936.0,
+      "logps/rejected": -1072.0,
+      "loss": 0.52,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -5.46875,
+      "rewards/margins": 1.3671875,
+      "rewards/rejected": -6.84375,
       "step": 570
     },
     {
       "epoch": 0.6073298429319371,
-      "grad_norm": 1.7710515581532253,
+      "grad_norm": 2.0563118442066273,
       "learning_rate": 4.842325915414728e-05,
-      "logits/chosen": -7.96875,
-      "logits/rejected": -6.5,
-      "logps/chosen": -900.0,
-      "logps/rejected": -992.0,
-      "loss": 0.4782,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -5.1875,
-      "rewards/margins": 1.0,
-      "rewards/rejected": -6.1875,
+      "logits/chosen": -7.0625,
+      "logits/rejected": -5.5,
+      "logps/chosen": -848.0,
+      "logps/rejected": -940.0,
+      "loss": 0.4775,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -4.65625,
+      "rewards/margins": 0.9765625,
+      "rewards/rejected": -5.65625,
       "step": 580
     },
     {
       "epoch": 0.6178010471204188,
-      "grad_norm": 2.4631746889663964,
+      "grad_norm": 2.4482509452202836,
       "learning_rate": 4.831504111239373e-05,
-      "logits/chosen": -9.5,
-      "logits/rejected": -8.0,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -1152.0,
-      "loss": 0.4935,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -6.0625,
-      "rewards/margins": 1.09375,
-      "rewards/rejected": -7.15625,
+      "logits/chosen": -7.0,
+      "logits/rejected": -4.96875,
+      "logps/chosen": -960.0,
+      "logps/rejected": -1056.0,
+      "loss": 0.4993,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -5.1875,
+      "rewards/margins": 1.0703125,
+      "rewards/rejected": -6.25,
       "step": 590
     },
     {
       "epoch": 0.6282722513089005,
-      "grad_norm": 2.4145697631480085,
+      "grad_norm": 2.4396274469831924,
       "learning_rate": 4.8203360770363425e-05,
-      "logits/chosen": -11.4375,
-      "logits/rejected": -10.0,
-      "logps/chosen": -876.0,
-      "logps/rejected": -1032.0,
-      "loss": 0.5085,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -4.46875,
-      "rewards/margins": 1.203125,
-      "rewards/rejected": -5.65625,
+      "logits/chosen": -8.125,
+      "logits/rejected": -5.78125,
+      "logps/chosen": -912.0,
+      "logps/rejected": -1088.0,
+      "loss": 0.5057,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -4.8125,
+      "rewards/margins": 1.390625,
+      "rewards/rejected": -6.1875,
       "step": 600
     },
     {
       "epoch": 0.6282722513089005,
-      "eval_logits/chosen": -10.25,
-      "eval_logits/rejected": -8.625,
-      "eval_logps/chosen": -848.0,
-      "eval_logps/rejected": -896.0,
-      "eval_loss": 0.7982968688011169,
+      "eval_logits/chosen": -6.40625,
+      "eval_logits/rejected": -4.5,
+      "eval_logps/chosen": -856.0,
+      "eval_logps/rejected": -904.0,
+      "eval_loss": 0.8287500143051147,
       "eval_rewards/accuracies": 0.591796875,
-      "eval_rewards/chosen": -5.28125,
-      "eval_rewards/margins": 0.7890625,
-      "eval_rewards/rejected": -6.09375,
-      "eval_runtime": 46.8448,
-      "eval_samples_per_second": 42.694,
-      "eval_steps_per_second": 0.683,
+      "eval_rewards/chosen": -5.375,
+      "eval_rewards/margins": 0.75390625,
+      "eval_rewards/rejected": -6.125,
+      "eval_runtime": 46.6385,
+      "eval_samples_per_second": 42.883,
+      "eval_steps_per_second": 0.686,
       "step": 600
     },
     {
       "epoch": 0.6387434554973822,
-      "grad_norm": 3.0304262349850895,
+      "grad_norm": 2.640576850504263,
       "learning_rate": 4.808823471266702e-05,
-      "logits/chosen": -8.8125,
-      "logits/rejected": -7.34375,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -1152.0,
-      "loss": 0.5075,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -6.4375,
+      "logits/chosen": -6.375,
+      "logits/rejected": -4.4375,
+      "logps/chosen": -960.0,
+      "logps/rejected": -1048.0,
+      "loss": 0.5007,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -5.375,
       "rewards/margins": 1.109375,
-      "rewards/rejected": -7.53125,
+      "rewards/rejected": -6.5,
       "step": 610
     },
     {
       "epoch": 0.6492146596858639,
-      "grad_norm": 2.0325140126241723,
+      "grad_norm": 1.936158779559134,
       "learning_rate": 4.796968003560639e-05,
-      "logits/chosen": -9.4375,
-      "logits/rejected": -7.90625,
+      "logits/chosen": -6.8125,
+      "logits/rejected": -4.84375,
       "logps/chosen": -928.0,
-      "logps/rejected": -1040.0,
-      "loss": 0.4988,
-      "rewards/accuracies": 0.768750011920929,
+      "logps/rejected": -1032.0,
+      "loss": 0.4956,
+      "rewards/accuracies": 0.7749999761581421,
       "rewards/chosen": -5.03125,
-      "rewards/margins": 1.1875,
-      "rewards/rejected": -6.21875,
+      "rewards/margins": 1.140625,
+      "rewards/rejected": -6.1875,
       "step": 620
     },
     {
       "epoch": 0.6596858638743456,
-      "grad_norm": 1.8505767173658088,
+      "grad_norm": 1.6282187758984752,
       "learning_rate": 4.7847714344635864e-05,
-      "logits/chosen": -7.59375,
-      "logits/rejected": -6.28125,
-      "logps/chosen": -908.0,
-      "logps/rejected": -1000.0,
-      "loss": 0.486,
+      "logits/chosen": -8.5625,
+      "logits/rejected": -7.125,
+      "logps/chosen": -816.0,
+      "logps/rejected": -916.0,
+      "loss": 0.4696,
       "rewards/accuracies": 0.75,
-      "rewards/chosen": -5.09375,
-      "rewards/margins": 0.94140625,
-      "rewards/rejected": -6.03125,
+      "rewards/chosen": -4.15625,
+      "rewards/margins": 1.0703125,
+      "rewards/rejected": -5.21875,
       "step": 630
     },
     {
       "epoch": 0.6701570680628273,
-      "grad_norm": 2.042867330097134,
+      "grad_norm": 2.2543844317208293,
       "learning_rate": 4.772235575174776e-05,
-      "logits/chosen": -7.8125,
-      "logits/rejected": -5.8125,
-      "logps/chosen": -964.0,
-      "logps/rejected": -1096.0,
-      "loss": 0.467,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -5.125,
-      "rewards/margins": 1.3046875,
-      "rewards/rejected": -6.4375,
+      "logits/chosen": -9.0625,
+      "logits/rejected": -6.75,
+      "logps/chosen": -936.0,
+      "logps/rejected": -1072.0,
+      "loss": 0.4684,
+      "rewards/accuracies": 0.793749988079071,
+      "rewards/chosen": -4.8125,
+      "rewards/margins": 1.390625,
+      "rewards/rejected": -6.21875,
       "step": 640
     },
     {
       "epoch": 0.680628272251309,
-      "grad_norm": 2.6835153152939384,
+      "grad_norm": 2.752482803490422,
       "learning_rate": 4.7593622872782754e-05,
-      "logits/chosen": -7.0,
-      "logits/rejected": -5.6875,
-      "logps/chosen": -1012.0,
-      "logps/rejected": -1088.0,
-      "loss": 0.4874,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -6.15625,
-      "rewards/margins": 1.0234375,
-      "rewards/rejected": -7.15625,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -7.21875,
+      "logps/chosen": -936.0,
+      "logps/rejected": -1040.0,
+      "loss": 0.4653,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -5.40625,
+      "rewards/margins": 1.28125,
+      "rewards/rejected": -6.6875,
       "step": 650
     },
     {
       "epoch": 0.6910994764397905,
-      "grad_norm": 1.8936031067381187,
+      "grad_norm": 1.8719202734305553,
       "learning_rate": 4.746153482466542e-05,
-      "logits/chosen": -8.125,
-      "logits/rejected": -6.71875,
-      "logps/chosen": -944.0,
-      "logps/rejected": -1040.0,
-      "loss": 0.4728,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -5.375,
-      "rewards/margins": 1.015625,
-      "rewards/rejected": -6.40625,
+      "logits/chosen": -7.78125,
+      "logits/rejected": -5.65625,
+      "logps/chosen": -1000.0,
+      "logps/rejected": -1128.0,
+      "loss": 0.4574,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -5.9375,
+      "rewards/margins": 1.328125,
+      "rewards/rejected": -7.25,
       "step": 660
     },
     {
       "epoch": 0.7015706806282722,
-      "grad_norm": 1.8549373178514739,
+      "grad_norm": 1.8131013462815513,
       "learning_rate": 4.732611122256537e-05,
-      "logits/chosen": -6.9375,
-      "logits/rejected": -5.3125,
-      "logps/chosen": -996.0,
-      "logps/rejected": -1104.0,
-      "loss": 0.438,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -5.375,
-      "rewards/margins": 1.21875,
-      "rewards/rejected": -6.59375,
+      "logits/chosen": -6.125,
+      "logits/rejected": -4.21875,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1152.0,
+      "loss": 0.4378,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -5.8125,
+      "rewards/margins": 1.2421875,
+      "rewards/rejected": -7.03125,
       "step": 670
     },
     {
       "epoch": 0.7120418848167539,
-      "grad_norm": 1.85147231213006,
+      "grad_norm": 1.9834672184693278,
       "learning_rate": 4.718737217698435e-05,
-      "logits/chosen": -8.125,
-      "logits/rejected": -6.125,
-      "logps/chosen": -824.0,
-      "logps/rejected": -916.0,
-      "loss": 0.5123,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -4.3125,
-      "rewards/margins": 1.234375,
-      "rewards/rejected": -5.5625,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -6.1875,
+      "logps/chosen": -808.0,
+      "logps/rejected": -896.0,
+      "loss": 0.5036,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -4.15625,
+      "rewards/margins": 1.203125,
+      "rewards/rejected": -5.375,
       "step": 680
     },
     {
       "epoch": 0.7225130890052356,
-      "grad_norm": 2.053229306430636,
+      "grad_norm": 2.0241055799554917,
       "learning_rate": 4.704533829076982e-05,
-      "logits/chosen": -6.375,
-      "logits/rejected": -4.84375,
-      "logps/chosen": -960.0,
-      "logps/rejected": -1120.0,
-      "loss": 0.463,
+      "logits/chosen": -5.5,
+      "logits/rejected": -3.46875,
+      "logps/chosen": -1008.0,
+      "logps/rejected": -1176.0,
+      "loss": 0.4577,
       "rewards/accuracies": 0.71875,
-      "rewards/chosen": -5.5,
-      "rewards/margins": 1.09375,
-      "rewards/rejected": -6.59375,
+      "rewards/chosen": -6.0,
+      "rewards/margins": 1.109375,
+      "rewards/rejected": -7.09375,
       "step": 690
     },
     {
       "epoch": 0.7329842931937173,
-      "grad_norm": 2.0316797892565015,
+      "grad_norm": 2.064745047450795,
       "learning_rate": 4.6900030656055455e-05,
-      "logits/chosen": -8.875,
-      "logits/rejected": -7.53125,
-      "logps/chosen": -840.0,
-      "logps/rejected": -928.0,
-      "loss": 0.4655,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -4.15625,
-      "rewards/margins": 1.1328125,
-      "rewards/rejected": -5.28125,
+      "logits/chosen": -5.4375,
+      "logits/rejected": -3.453125,
+      "logps/chosen": -996.0,
+      "logps/rejected": -1088.0,
+      "loss": 0.48,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -5.71875,
+      "rewards/margins": 1.125,
+      "rewards/rejected": -6.84375,
       "step": 700
     },
     {
       "epoch": 0.7329842931937173,
-      "eval_logits/chosen": -10.375,
-      "eval_logits/rejected": -8.75,
-      "eval_logps/chosen": -712.0,
-      "eval_logps/rejected": -768.0,
-      "eval_loss": 0.8071874976158142,
-      "eval_rewards/accuracies": 0.625,
-      "eval_rewards/chosen": -3.9375,
-      "eval_rewards/margins": 0.8515625,
-      "eval_rewards/rejected": -4.78125,
-      "eval_runtime": 46.807,
-      "eval_samples_per_second": 42.729,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": -5.65625,
+      "eval_logits/rejected": -3.84375,
+      "eval_logps/chosen": -872.0,
+      "eval_logps/rejected": -928.0,
+      "eval_loss": 0.798703134059906,
+      "eval_rewards/accuracies": 0.62890625,
+      "eval_rewards/chosen": -5.53125,
+      "eval_rewards/margins": 0.86328125,
+      "eval_rewards/rejected": -6.40625,
+      "eval_runtime": 46.6227,
+      "eval_samples_per_second": 42.898,
+      "eval_steps_per_second": 0.686,
       "step": 700
     },
     {
       "epoch": 0.743455497382199,
-      "grad_norm": 2.323908308980522,
+      "grad_norm": 1.9544102677581432,
       "learning_rate": 4.6751470851128906e-05,
-      "logits/chosen": -10.0625,
-      "logits/rejected": -8.375,
-      "logps/chosen": -872.0,
-      "logps/rejected": -1016.0,
-      "loss": 0.4862,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -4.1875,
-      "rewards/margins": 1.4140625,
-      "rewards/rejected": -5.59375,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -6.65625,
+      "logps/chosen": -900.0,
+      "logps/rejected": -1048.0,
+      "loss": 0.4794,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -4.4375,
+      "rewards/margins": 1.4609375,
+      "rewards/rejected": -5.90625,
       "step": 710
     },
     {
       "epoch": 0.7539267015706806,
-      "grad_norm": 2.0687875752659255,
+      "grad_norm": 2.3114249017309683,
       "learning_rate": 4.6599680937227454e-05,
-      "logits/chosen": -9.9375,
-      "logits/rejected": -8.125,
-      "logps/chosen": -860.0,
-      "logps/rejected": -1040.0,
-      "loss": 0.4902,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -4.21875,
-      "rewards/margins": 1.5,
-      "rewards/rejected": -5.71875,
+      "logits/chosen": -7.75,
+      "logits/rejected": -5.53125,
+      "logps/chosen": -952.0,
+      "logps/rejected": -1152.0,
+      "loss": 0.4804,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -5.1875,
+      "rewards/margins": 1.640625,
+      "rewards/rejected": -6.8125,
       "step": 720
     },
     {
       "epoch": 0.7643979057591623,
-      "grad_norm": 2.3681745055258223,
+      "grad_norm": 2.248882221015109,
       "learning_rate": 4.644468345526183e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -8.3125,
-      "logps/chosen": -788.0,
-      "logps/rejected": -996.0,
-      "loss": 0.4808,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 1.5625,
-      "rewards/rejected": -5.0625,
+      "logits/chosen": -10.0625,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -836.0,
+      "logps/rejected": -1048.0,
+      "loss": 0.4861,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -3.984375,
+      "rewards/margins": 1.5703125,
+      "rewards/rejected": -5.5625,
       "step": 730
     },
     {
       "epoch": 0.774869109947644,
-      "grad_norm": 2.3219112324645725,
+      "grad_norm": 2.07148756182618,
       "learning_rate": 4.628650142246894e-05,
-      "logits/chosen": -5.625,
-      "logits/rejected": -3.46875,
-      "logps/chosen": -1000.0,
-      "logps/rejected": -1136.0,
-      "loss": 0.4488,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -6.03125,
-      "rewards/margins": 1.453125,
-      "rewards/rejected": -7.46875,
+      "logits/chosen": -5.28125,
+      "logits/rejected": -3.28125,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -1328.0,
+      "loss": 0.4742,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -8.125,
+      "rewards/margins": 1.3046875,
+      "rewards/rejected": -9.375,
       "step": 740
     },
     {
       "epoch": 0.7853403141361257,
-      "grad_norm": 2.109011925862191,
+      "grad_norm": 1.969118276119175,
       "learning_rate": 4.6125158328993765e-05,
-      "logits/chosen": -7.0625,
-      "logits/rejected": -5.53125,
-      "logps/chosen": -880.0,
-      "logps/rejected": -1032.0,
-      "loss": 0.4863,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -4.8125,
-      "rewards/margins": 1.375,
-      "rewards/rejected": -6.1875,
+      "logits/chosen": -8.375,
+      "logits/rejected": -6.8125,
+      "logps/chosen": -924.0,
+      "logps/rejected": -1056.0,
+      "loss": 0.4827,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -5.28125,
+      "rewards/margins": 1.1640625,
+      "rewards/rejected": -6.4375,
       "step": 750
     },
     {
       "epoch": 0.7958115183246073,
-      "grad_norm": 1.7111513133718592,
+      "grad_norm": 2.1529001856162697,
       "learning_rate": 4.5960678134401064e-05,
-      "logits/chosen": -6.875,
-      "logits/rejected": -4.3125,
-      "logps/chosen": -936.0,
-      "logps/rejected": -1096.0,
-      "loss": 0.4657,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -5.03125,
-      "rewards/margins": 1.5,
-      "rewards/rejected": -6.5,
+      "logits/chosen": -7.0,
+      "logits/rejected": -4.25,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -1256.0,
+      "loss": 0.4736,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -6.40625,
+      "rewards/margins": 1.765625,
+      "rewards/rejected": -8.1875,
       "step": 760
     },
     {
       "epoch": 0.806282722513089,
-      "grad_norm": 2.3113065774807495,
+      "grad_norm": 2.3761455467015504,
       "learning_rate": 4.579308526411734e-05,
-      "logits/chosen": -6.46875,
-      "logits/rejected": -4.75,
-      "logps/chosen": -896.0,
-      "logps/rejected": -1048.0,
-      "loss": 0.4674,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -4.78125,
-      "rewards/margins": 1.515625,
-      "rewards/rejected": -6.3125,
+      "logits/chosen": -5.59375,
+      "logits/rejected": -3.609375,
+      "logps/chosen": -1024.0,
+      "logps/rejected": -1152.0,
+      "loss": 0.471,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -6.0625,
+      "rewards/margins": 1.2421875,
+      "rewards/rejected": -7.3125,
       "step": 770
     },
     {
       "epoch": 0.8167539267015707,
-      "grad_norm": 2.5129511978349344,
+      "grad_norm": 2.2008730462901203,
       "learning_rate": 4.562240460580367e-05,
-      "logits/chosen": -7.15625,
-      "logits/rejected": -5.4375,
-      "logps/chosen": -944.0,
-      "logps/rejected": -1096.0,
-      "loss": 0.4768,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -5.28125,
-      "rewards/margins": 1.390625,
-      "rewards/rejected": -6.6875,
+      "logits/chosen": -8.375,
+      "logits/rejected": -6.53125,
+      "logps/chosen": -916.0,
+      "logps/rejected": -1064.0,
+      "loss": 0.4717,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -5.0,
+      "rewards/margins": 1.359375,
+      "rewards/rejected": -6.375,
       "step": 780
     },
     {
       "epoch": 0.8272251308900523,
-      "grad_norm": 1.8459629070783456,
+      "grad_norm": 2.052875771656185,
       "learning_rate": 4.544866150565985e-05,
-      "logits/chosen": -7.46875,
-      "logits/rejected": -5.90625,
-      "logps/chosen": -920.0,
-      "logps/rejected": -1032.0,
-      "loss": 0.4841,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -4.96875,
-      "rewards/margins": 1.09375,
-      "rewards/rejected": -6.0625,
+      "logits/chosen": -8.1875,
+      "logits/rejected": -6.625,
+      "logps/chosen": -996.0,
+      "logps/rejected": -1112.0,
+      "loss": 0.488,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -5.75,
+      "rewards/margins": 1.15625,
+      "rewards/rejected": -6.90625,
       "step": 790
     },
     {
       "epoch": 0.837696335078534,
-      "grad_norm": 1.8707582040737978,
+      "grad_norm": 1.9167385400996355,
       "learning_rate": 4.527188176466051e-05,
-      "logits/chosen": -7.65625,
-      "logits/rejected": -6.34375,
-      "logps/chosen": -920.0,
-      "logps/rejected": -1048.0,
-      "loss": 0.4638,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -4.71875,
-      "rewards/margins": 1.2421875,
-      "rewards/rejected": -5.9375,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -7.0,
+      "logps/chosen": -984.0,
+      "logps/rejected": -1112.0,
+      "loss": 0.4751,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -5.34375,
+      "rewards/margins": 1.3046875,
+      "rewards/rejected": -6.65625,
       "step": 800
     },
     {
       "epoch": 0.837696335078534,
-      "eval_logits/chosen": -3.96875,
-      "eval_logits/rejected": -2.546875,
-      "eval_logps/chosen": -1056.0,
-      "eval_logps/rejected": -1088.0,
-      "eval_loss": 0.8441874980926514,
-      "eval_rewards/accuracies": 0.578125,
-      "eval_rewards/chosen": -7.34375,
-      "eval_rewards/margins": 0.625,
-      "eval_rewards/rejected": -7.96875,
-      "eval_runtime": 46.8177,
-      "eval_samples_per_second": 42.719,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": -6.125,
+      "eval_logits/rejected": -4.3125,
+      "eval_logps/chosen": -1024.0,
+      "eval_logps/rejected": -1064.0,
+      "eval_loss": 0.8429531455039978,
+      "eval_rewards/accuracies": 0.55859375,
+      "eval_rewards/chosen": -7.0625,
+      "eval_rewards/margins": 0.70703125,
+      "eval_rewards/rejected": -7.78125,
+      "eval_runtime": 46.6149,
+      "eval_samples_per_second": 42.905,
+      "eval_steps_per_second": 0.686,
       "step": 800
     },
     {
       "epoch": 0.8481675392670157,
-      "grad_norm": 1.8167046893245546,
+      "grad_norm": 1.866566299494209,
       "learning_rate": 4.509209163472357e-05,
-      "logits/chosen": -7.28125,
-      "logits/rejected": -5.625,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -1160.0,
-      "loss": 0.461,
+      "logits/chosen": -9.375,
+      "logits/rejected": -7.625,
+      "logps/chosen": -1016.0,
+      "logps/rejected": -1152.0,
+      "loss": 0.4494,
       "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -6.0,
-      "rewards/margins": 1.1953125,
-      "rewards/rejected": -7.1875,
+      "rewards/chosen": -5.75,
+      "rewards/margins": 1.3984375,
+      "rewards/rejected": -7.15625,
       "step": 810
     },
     {
       "epoch": 0.8586387434554974,
-      "grad_norm": 1.609489211268817,
+      "grad_norm": 1.9244148132125336,
       "learning_rate": 4.4909317814811896e-05,
-      "logits/chosen": -8.8125,
-      "logits/rejected": -7.15625,
-      "logps/chosen": -944.0,
-      "logps/rejected": -1064.0,
-      "loss": 0.4629,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -5.15625,
-      "rewards/margins": 1.3125,
-      "rewards/rejected": -6.4375,
+      "logits/chosen": -10.0625,
+      "logits/rejected": -8.375,
+      "logps/chosen": -1000.0,
+      "logps/rejected": -1120.0,
+      "loss": 0.4699,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -5.71875,
+      "rewards/margins": 1.34375,
+      "rewards/rejected": -7.0625,
       "step": 820
     },
     {
       "epoch": 0.8691099476439791,
-      "grad_norm": 2.429530371269208,
+      "grad_norm": 2.11765377196965,
       "learning_rate": 4.472358744696845e-05,
-      "logits/chosen": -7.375,
-      "logits/rejected": -5.4375,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -1192.0,
-      "loss": 0.4669,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -6.25,
-      "rewards/margins": 1.4921875,
-      "rewards/rejected": -7.75,
+      "logits/chosen": -8.875,
+      "logits/rejected": -6.78125,
+      "logps/chosen": -1016.0,
+      "logps/rejected": -1152.0,
+      "loss": 0.4662,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -5.90625,
+      "rewards/margins": 1.4375,
+      "rewards/rejected": -7.375,
       "step": 830
     },
     {
       "epoch": 0.8795811518324608,
-      "grad_norm": 2.1567800284661414,
+      "grad_norm": 1.9748723869331066,
       "learning_rate": 4.4534928112285645e-05,
       "logits/chosen": -8.625,
-      "logits/rejected": -7.0625,
-      "logps/chosen": -980.0,
-      "logps/rejected": -1128.0,
-      "loss": 0.4789,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -5.5625,
-      "rewards/margins": 1.3671875,
-      "rewards/rejected": -6.9375,
+      "logits/rejected": -6.65625,
+      "logps/chosen": -1000.0,
+      "logps/rejected": -1152.0,
+      "loss": 0.4774,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -5.78125,
+      "rewards/margins": 1.3515625,
+      "rewards/rejected": -7.125,
       "step": 840
     },
     {
       "epoch": 0.8900523560209425,
-      "grad_norm": 1.9021537171078922,
+      "grad_norm": 2.0991841369745954,
       "learning_rate": 4.43433678268096e-05,
-      "logits/chosen": -10.125,
-      "logits/rejected": -8.75,
-      "logps/chosen": -912.0,
-      "logps/rejected": -1048.0,
-      "loss": 0.4599,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -4.9375,
-      "rewards/margins": 1.34375,
-      "rewards/rejected": -6.28125,
+      "logits/chosen": -10.25,
+      "logits/rejected": -8.625,
+      "logps/chosen": -936.0,
+      "logps/rejected": -1080.0,
+      "loss": 0.4433,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -5.15625,
+      "rewards/margins": 1.4375,
+      "rewards/rejected": -6.625,
       "step": 850
     },
     {
       "epoch": 0.900523560209424,
-      "grad_norm": 1.9592960880281511,
+      "grad_norm": 2.5067781793905484,
       "learning_rate": 4.414893503737968e-05,
-      "logits/chosen": -8.0625,
-      "logits/rejected": -6.1875,
-      "logps/chosen": -960.0,
-      "logps/rejected": -1112.0,
-      "loss": 0.4653,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -5.75,
-      "rewards/margins": 1.5703125,
-      "rewards/rejected": -7.34375,
+      "logits/chosen": -8.6875,
+      "logits/rejected": -6.53125,
+      "logps/chosen": -1020.0,
+      "logps/rejected": -1184.0,
+      "loss": 0.471,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -6.375,
+      "rewards/margins": 1.7265625,
+      "rewards/rejected": -8.0625,
       "step": 860
     },
     {
       "epoch": 0.9109947643979057,
-      "grad_norm": 1.9347482593304213,
+      "grad_norm": 1.9032588602342604,
       "learning_rate": 4.395165861740414e-05,
-      "logits/chosen": -9.1875,
-      "logits/rejected": -7.625,
-      "logps/chosen": -844.0,
-      "logps/rejected": -1000.0,
-      "loss": 0.4659,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -4.46875,
-      "rewards/margins": 1.46875,
-      "rewards/rejected": -5.9375,
+      "logits/chosen": -8.6875,
+      "logits/rejected": -6.9375,
+      "logps/chosen": -912.0,
+      "logps/rejected": -1072.0,
+      "loss": 0.4652,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -5.125,
+      "rewards/margins": 1.5,
+      "rewards/rejected": -6.625,
       "step": 870
     },
     {
       "epoch": 0.9214659685863874,
-      "grad_norm": 2.2325101896023685,
+      "grad_norm": 2.3070067235221257,
       "learning_rate": 4.375156786257241e-05,
-      "logits/chosen": -9.8125,
-      "logits/rejected": -8.375,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -1216.0,
-      "loss": 0.4768,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -5.75,
-      "rewards/margins": 1.6171875,
-      "rewards/rejected": -7.34375,
+      "logits/chosen": -9.1875,
+      "logits/rejected": -7.625,
+      "logps/chosen": -980.0,
+      "logps/rejected": -1144.0,
+      "loss": 0.4819,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -5.1875,
+      "rewards/margins": 1.4453125,
+      "rewards/rejected": -6.625,
       "step": 880
     },
     {
       "epoch": 0.9319371727748691,
-      "grad_norm": 2.1893634608015557,
+      "grad_norm": 2.281445982846512,
       "learning_rate": 4.354869248650466e-05,
-      "logits/chosen": -9.4375,
-      "logits/rejected": -8.125,
-      "logps/chosen": -1000.0,
-      "logps/rejected": -1144.0,
-      "loss": 0.4659,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -5.6875,
-      "rewards/margins": 1.3125,
-      "rewards/rejected": -7.0,
+      "logits/chosen": -8.0625,
+      "logits/rejected": -6.375,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -1272.0,
+      "loss": 0.4633,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -7.03125,
+      "rewards/margins": 1.265625,
+      "rewards/rejected": -8.3125,
       "step": 890
     },
     {
       "epoch": 0.9424083769633508,
-      "grad_norm": 2.748794788212845,
+      "grad_norm": 2.544870746118545,
       "learning_rate": 4.334306261633931e-05,
-      "logits/chosen": -8.1875,
-      "logits/rejected": -6.59375,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -1208.0,
-      "loss": 0.4265,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -5.84375,
-      "rewards/margins": 1.6953125,
-      "rewards/rejected": -7.53125,
+      "logits/chosen": -9.0,
+      "logits/rejected": -7.625,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -1264.0,
+      "loss": 0.4408,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -6.34375,
+      "rewards/margins": 1.7421875,
+      "rewards/rejected": -8.0625,
       "step": 900
     },
     {
       "epoch": 0.9424083769633508,
-      "eval_logits/chosen": -6.4375,
-      "eval_logits/rejected": -4.8125,
-      "eval_logps/chosen": -1120.0,
-      "eval_logps/rejected": -1184.0,
-      "eval_loss": 0.9620468616485596,
-      "eval_rewards/accuracies": 0.591796875,
-      "eval_rewards/chosen": -8.0,
+      "eval_logits/chosen": -8.1875,
+      "eval_logits/rejected": -6.34375,
+      "eval_logps/chosen": -1152.0,
+      "eval_logps/rejected": -1208.0,
+      "eval_loss": 0.897140622138977,
+      "eval_rewards/accuracies": 0.599609375,
+      "eval_rewards/chosen": -8.3125,
       "eval_rewards/margins": 0.90234375,
-      "eval_rewards/rejected": -8.9375,
-      "eval_runtime": 46.8058,
-      "eval_samples_per_second": 42.73,
-      "eval_steps_per_second": 0.684,
+      "eval_rewards/rejected": -9.1875,
+      "eval_runtime": 46.6384,
+      "eval_samples_per_second": 42.883,
+      "eval_steps_per_second": 0.686,
       "step": 900
     },
     {
       "epoch": 0.9528795811518325,
-      "grad_norm": 2.5576424411477148,
+      "grad_norm": 2.260602341318326,
       "learning_rate": 4.313470878825911e-05,
-      "logits/chosen": -7.3125,
-      "logits/rejected": -5.65625,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -1336.0,
-      "loss": 0.4734,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -7.8125,
-      "rewards/margins": 1.453125,
-      "rewards/rejected": -9.25,
+      "logits/chosen": -8.625,
+      "logits/rejected": -7.125,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -1296.0,
+      "loss": 0.4762,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -7.65625,
+      "rewards/margins": 1.2734375,
+      "rewards/rejected": -8.9375,
       "step": 910
     },
     {
       "epoch": 0.9633507853403142,
-      "grad_norm": 1.980336434781313,
+      "grad_norm": 2.161143254769399,
       "learning_rate": 4.292366194295652e-05,
-      "logits/chosen": -7.40625,
-      "logits/rejected": -5.71875,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1192.0,
-      "loss": 0.4573,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -6.65625,
-      "rewards/margins": 1.3671875,
-      "rewards/rejected": -8.0,
+      "logits/chosen": -9.0,
+      "logits/rejected": -7.46875,
+      "logps/chosen": -1016.0,
+      "logps/rejected": -1112.0,
+      "loss": 0.4709,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -5.84375,
+      "rewards/margins": 1.3984375,
+      "rewards/rejected": -7.25,
       "step": 920
     },
     {
       "epoch": 0.9738219895287958,
-      "grad_norm": 2.1970009626320803,
+      "grad_norm": 1.5835750970096396,
       "learning_rate": 4.2709953421038964e-05,
-      "logits/chosen": -6.65625,
-      "logits/rejected": -5.15625,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -1208.0,
-      "loss": 0.4714,
+      "logits/chosen": -7.5,
+      "logits/rejected": -5.6875,
+      "logps/chosen": -1024.0,
+      "logps/rejected": -1184.0,
+      "loss": 0.4707,
       "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -6.4375,
-      "rewards/margins": 1.3515625,
-      "rewards/rejected": -7.78125,
+      "rewards/chosen": -6.21875,
+      "rewards/margins": 1.328125,
+      "rewards/rejected": -7.5625,
       "step": 930
     },
     {
       "epoch": 0.9842931937172775,
-      "grad_norm": 2.037720592310999,
+      "grad_norm": 2.1528349869218286,
       "learning_rate": 4.249361495837478e-05,
-      "logits/chosen": -9.375,
-      "logits/rejected": -7.59375,
-      "logps/chosen": -816.0,
-      "logps/rejected": -912.0,
-      "loss": 0.4808,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -4.03125,
-      "rewards/margins": 1.2109375,
-      "rewards/rejected": -5.25,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -6.25,
+      "logps/chosen": -980.0,
+      "logps/rejected": -1088.0,
+      "loss": 0.4546,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -5.6875,
+      "rewards/margins": 1.3203125,
+      "rewards/rejected": -7.0,
       "step": 940
     },
     {
       "epoch": 0.9947643979057592,
-      "grad_norm": 2.5446821295358393,
+      "grad_norm": 2.4384117442928255,
       "learning_rate": 4.227467868138035e-05,
-      "logits/chosen": -5.25,
-      "logits/rejected": -3.71875,
-      "logps/chosen": -968.0,
-      "logps/rejected": -1104.0,
-      "loss": 0.462,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -5.40625,
-      "rewards/margins": 1.328125,
-      "rewards/rejected": -6.71875,
+      "logits/chosen": -5.53125,
+      "logits/rejected": -3.828125,
+      "logps/chosen": -1080.0,
+      "logps/rejected": -1232.0,
+      "loss": 0.4556,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -6.5625,
+      "rewards/margins": 1.453125,
+      "rewards/rejected": -8.0,
       "step": 950
     },
     {
       "epoch": 1.0052356020942408,
-      "grad_norm": 0.9414263330277836,
+      "grad_norm": 0.9748827169891753,
       "learning_rate": 4.205317710224938e-05,
-      "logits/chosen": -7.09375,
-      "logits/rejected": -5.40625,
-      "logps/chosen": -868.0,
-      "logps/rejected": -1112.0,
-      "loss": 0.3374,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -4.34375,
-      "rewards/margins": 2.171875,
-      "rewards/rejected": -6.53125,
+      "logits/chosen": -7.71875,
+      "logits/rejected": -5.90625,
+      "logps/chosen": -948.0,
+      "logps/rejected": -1184.0,
+      "loss": 0.3439,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": -5.1875,
+      "rewards/margins": 2.125,
+      "rewards/rejected": -7.3125,
       "step": 960
     },
     {
       "epoch": 1.0157068062827226,
-      "grad_norm": 2.378288905189939,
+      "grad_norm": 1.2101418275414364,
       "learning_rate": 4.1829143114124733e-05,
-      "logits/chosen": -7.3125,
-      "logits/rejected": -4.78125,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -1432.0,
-      "loss": 0.1963,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -6.34375,
-      "rewards/margins": 3.734375,
-      "rewards/rejected": -10.0625,
+      "logits/chosen": -7.6875,
+      "logits/rejected": -4.53125,
+      "logps/chosen": -1080.0,
+      "logps/rejected": -1440.0,
+      "loss": 0.1724,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -6.5,
+      "rewards/margins": 3.703125,
+      "rewards/rejected": -10.1875,
       "step": 970
     },
     {
       "epoch": 1.0261780104712042,
-      "grad_norm": 1.18128860592773,
+      "grad_norm": 1.1806220535156653,
       "learning_rate": 4.160260998621387e-05,
-      "logits/chosen": -8.3125,
-      "logits/rejected": -5.21875,
-      "logps/chosen": -1120.0,
-      "logps/rejected": -1544.0,
-      "loss": 0.181,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -6.4375,
-      "rewards/margins": 4.5625,
-      "rewards/rejected": -11.0,
+      "logits/chosen": -8.375,
+      "logits/rejected": -4.78125,
+      "logps/chosen": -1296.0,
+      "logps/rejected": -1736.0,
+      "loss": 0.1643,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -12.9375,
       "step": 980
     },
     {
       "epoch": 1.036649214659686,
-      "grad_norm": 1.357575691400835,
+      "grad_norm": 1.3506350144839012,
       "learning_rate": 4.137361135884824e-05,
-      "logits/chosen": -7.1875,
-      "logits/rejected": -4.875,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -1424.0,
-      "loss": 0.1675,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -6.09375,
-      "rewards/margins": 3.8125,
-      "rewards/rejected": -9.875,
+      "logits/chosen": -7.15625,
+      "logits/rejected": -4.1875,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -1704.0,
+      "loss": 0.1759,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 4.34375,
+      "rewards/rejected": -12.6875,
       "step": 990
     },
     {
       "epoch": 1.0471204188481675,
-      "grad_norm": 1.729346907182029,
+      "grad_norm": 1.6674718407067428,
       "learning_rate": 4.114218123848777e-05,
-      "logits/chosen": -5.65625,
-      "logits/rejected": -2.53125,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -1520.0,
-      "loss": 0.1656,
+      "logits/chosen": -7.46875,
+      "logits/rejected": -3.578125,
+      "logps/chosen": -984.0,
+      "logps/rejected": -1376.0,
+      "loss": 0.1609,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -6.65625,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -11.0625,
+      "rewards/chosen": -5.78125,
+      "rewards/margins": 3.859375,
+      "rewards/rejected": -9.625,
       "step": 1000
     },
     {
       "epoch": 1.0471204188481675,
-      "eval_logits/chosen": -3.34375,
-      "eval_logits/rejected": -1.5234375,
-      "eval_logps/chosen": -1160.0,
-      "eval_logps/rejected": -1248.0,
-      "eval_loss": 0.9980000257492065,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -8.4375,
-      "eval_rewards/margins": 1.1953125,
-      "eval_rewards/rejected": -9.625,
-      "eval_runtime": 46.8106,
-      "eval_samples_per_second": 42.725,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": -3.765625,
+      "eval_logits/rejected": -1.7734375,
+      "eval_logps/chosen": -1136.0,
+      "eval_logps/rejected": -1208.0,
+      "eval_loss": 0.9796249866485596,
+      "eval_rewards/accuracies": 0.599609375,
+      "eval_rewards/chosen": -8.1875,
+      "eval_rewards/margins": 1.015625,
+      "eval_rewards/rejected": -9.1875,
+      "eval_runtime": 46.6368,
+      "eval_samples_per_second": 42.885,
+      "eval_steps_per_second": 0.686,
       "step": 1000
     },
     {
       "epoch": 1.057591623036649,
-      "grad_norm": 2.1436923948987543,
+      "grad_norm": 1.8667489331972587,
       "learning_rate": 4.0908353992670824e-05,
-      "logits/chosen": -5.71875,
-      "logits/rejected": -3.171875,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -1536.0,
-      "loss": 0.1687,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.0,
-      "rewards/margins": 4.25,
-      "rewards/rejected": -11.25,
+      "logits/chosen": -6.5625,
+      "logits/rejected": -3.234375,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -1496.0,
+      "loss": 0.1774,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -6.5625,
+      "rewards/margins": 4.21875,
+      "rewards/rejected": -10.8125,
       "step": 1010
     },
     {
       "epoch": 1.068062827225131,
-      "grad_norm": 1.976972317816638,
+      "grad_norm": 2.5365811583073854,
       "learning_rate": 4.067216434491059e-05,
-      "logits/chosen": -4.65625,
-      "logits/rejected": -1.796875,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -1680.0,
-      "loss": 0.1525,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -12.625,
+      "logits/chosen": -7.28125,
+      "logits/rejected": -3.71875,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1600.0,
+      "loss": 0.1451,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 4.5625,
+      "rewards/rejected": -11.8125,
       "step": 1020
     },
     {
       "epoch": 1.0785340314136125,
-      "grad_norm": 1.2782877692773384,
+      "grad_norm": 1.3972188688907148,
       "learning_rate": 4.043364736953866e-05,
-      "logits/chosen": -5.6875,
-      "logits/rejected": -2.453125,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -1584.0,
-      "loss": 0.1546,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.96875,
-      "rewards/margins": 4.15625,
-      "rewards/rejected": -11.125,
+      "logits/chosen": -6.5,
+      "logits/rejected": -2.96875,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -1640.0,
+      "loss": 0.1507,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -7.375,
+      "rewards/margins": 4.34375,
+      "rewards/rejected": -11.6875,
       "step": 1030
     },
     {
       "epoch": 1.0890052356020943,
-      "grad_norm": 2.5777401889905978,
+      "grad_norm": 2.278875673898671,
       "learning_rate": 4.019283848649643e-05,
-      "logits/chosen": -4.9375,
-      "logits/rejected": -2.203125,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -1832.0,
-      "loss": 0.1841,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 4.25,
-      "rewards/rejected": -13.875,
+      "logits/chosen": -7.0,
+      "logits/rejected": -3.203125,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -1712.0,
+      "loss": 0.1824,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 4.34375,
+      "rewards/rejected": -12.6875,
       "step": 1040
     },
     {
       "epoch": 1.0994764397905759,
-      "grad_norm": 1.0856741621771475,
+      "grad_norm": 2.3045804904765346,
       "learning_rate": 3.9949773456075215e-05,
-      "logits/chosen": -6.03125,
-      "logits/rejected": -4.15625,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -1472.0,
-      "loss": 0.1433,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -6.8125,
-      "rewards/margins": 3.984375,
-      "rewards/rejected": -10.8125,
+      "logits/chosen": -6.625,
+      "logits/rejected": -4.1875,
+      "logps/chosen": -1000.0,
+      "logps/rejected": -1392.0,
+      "loss": 0.1514,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -6.09375,
+      "rewards/margins": 3.90625,
+      "rewards/rejected": -10.0,
       "step": 1050
     },
     {
       "epoch": 1.1099476439790577,
-      "grad_norm": 1.5573378033953031,
+      "grad_norm": 4.222589649284655,
       "learning_rate": 3.970448837360585e-05,
-      "logits/chosen": -7.0625,
-      "logits/rejected": -4.71875,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -1744.0,
-      "loss": 0.1392,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -12.6875,
+      "logits/chosen": -7.65625,
+      "logits/rejected": -4.59375,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -1704.0,
+      "loss": 0.1404,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -7.90625,
+      "rewards/margins": 4.34375,
+      "rewards/rejected": -12.25,
       "step": 1060
     },
     {
       "epoch": 1.1204188481675392,
-      "grad_norm": 2.3041891051124956,
+      "grad_norm": 2.1593949026258166,
       "learning_rate": 3.945701966409846e-05,
-      "logits/chosen": -6.40625,
-      "logits/rejected": -4.03125,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -1904.0,
-      "loss": 0.1715,
+      "logits/chosen": -6.34375,
+      "logits/rejected": -3.046875,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -1864.0,
+      "loss": 0.1796,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 4.9375,
-      "rewards/rejected": -14.5,
+      "rewards/chosen": -9.25,
+      "rewards/margins": 4.90625,
+      "rewards/rejected": -14.125,
       "step": 1070
     },
     {
       "epoch": 1.130890052356021,
-      "grad_norm": 1.8314854946983867,
+      "grad_norm": 1.287975844826753,
       "learning_rate": 3.9207404076833374e-05,
-      "logits/chosen": -7.625,
-      "logits/rejected": -5.5,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -1632.0,
-      "loss": 0.1613,
+      "logits/chosen": -7.6875,
+      "logits/rejected": -5.25,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -1584.0,
+      "loss": 0.1489,
       "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.8125,
-      "rewards/margins": 4.125,
-      "rewards/rejected": -11.9375,
+      "rewards/chosen": -7.40625,
+      "rewards/margins": 4.03125,
+      "rewards/rejected": -11.4375,
       "step": 1080
     },
     {
       "epoch": 1.1413612565445026,
-      "grad_norm": 1.684240142160982,
+      "grad_norm": 1.1478443493194328,
       "learning_rate": 3.895567867990379e-05,
-      "logits/chosen": -7.15625,
-      "logits/rejected": -4.5625,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -1800.0,
-      "loss": 0.1448,
+      "logits/chosen": -6.875,
+      "logits/rejected": -3.953125,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -1784.0,
+      "loss": 0.1408,
       "rewards/accuracies": 0.9375,
-      "rewards/chosen": -8.8125,
+      "rewards/chosen": -8.625,
       "rewards/margins": 4.3125,
-      "rewards/rejected": -13.125,
+      "rewards/rejected": -12.9375,
       "step": 1090
     },
     {
       "epoch": 1.1518324607329844,
-      "grad_norm": 1.2786082174077242,
+      "grad_norm": 1.6786808826424053,
       "learning_rate": 3.870188085471114e-05,
-      "logits/chosen": -7.5625,
-      "logits/rejected": -5.125,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -1600.0,
-      "loss": 0.1481,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.59375,
-      "rewards/margins": 4.125,
-      "rewards/rejected": -11.75,
+      "logits/chosen": -6.25,
+      "logits/rejected": -3.421875,
+      "logps/chosen": -1312.0,
+      "logps/rejected": -1744.0,
+      "loss": 0.1551,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -8.8125,
+      "rewards/margins": 4.375,
+      "rewards/rejected": -13.1875,
       "step": 1100
     },
     {
       "epoch": 1.1518324607329844,
-      "eval_logits/chosen": -6.09375,
-      "eval_logits/rejected": -4.28125,
-      "eval_logps/chosen": -1280.0,
-      "eval_logps/rejected": -1368.0,
-      "eval_loss": 1.0423359870910645,
-      "eval_rewards/accuracies": 0.591796875,
-      "eval_rewards/chosen": -9.625,
-      "eval_rewards/margins": 1.1640625,
-      "eval_rewards/rejected": -10.8125,
-      "eval_runtime": 46.8214,
-      "eval_samples_per_second": 42.716,
-      "eval_steps_per_second": 0.683,
+      "eval_logits/chosen": -2.03125,
+      "eval_logits/rejected": -0.26171875,
+      "eval_logps/chosen": -1704.0,
+      "eval_logps/rejected": -1792.0,
+      "eval_loss": 1.2333672046661377,
+      "eval_rewards/accuracies": 0.59375,
+      "eval_rewards/chosen": -13.8125,
+      "eval_rewards/margins": 1.2421875,
+      "eval_rewards/rejected": -15.0625,
+      "eval_runtime": 46.6255,
+      "eval_samples_per_second": 42.895,
+      "eval_steps_per_second": 0.686,
       "step": 1100
     },
     {
       "epoch": 1.162303664921466,
-      "grad_norm": 2.9795825520268795,
+      "grad_norm": 3.0513992031733497,
       "learning_rate": 3.844604829041395e-05,
-      "logits/chosen": -7.6875,
-      "logits/rejected": -5.03125,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -1680.0,
-      "loss": 0.1804,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -7.9375,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -12.4375,
+      "logits/chosen": -4.875,
+      "logits/rejected": -1.9609375,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -1976.0,
+      "loss": 0.169,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 5.03125,
+      "rewards/rejected": -15.3125,
       "step": 1110
     },
     {
       "epoch": 1.1727748691099475,
-      "grad_norm": 1.9045242597462804,
+      "grad_norm": 2.4356387016062038,
       "learning_rate": 3.818821897833095e-05,
-      "logits/chosen": -7.0,
-      "logits/rejected": -4.96875,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -1672.0,
-      "loss": 0.1512,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 3.9375,
-      "rewards/rejected": -12.4375,
+      "logits/chosen": -5.90625,
+      "logits/rejected": -3.21875,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -1688.0,
+      "loss": 0.1551,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 4.21875,
+      "rewards/rejected": -12.5625,
       "step": 1120
     },
     {
       "epoch": 1.1832460732984293,
-      "grad_norm": 3.6971662075228244,
+      "grad_norm": 3.510978560271283,
       "learning_rate": 3.792843120629935e-05,
-      "logits/chosen": -7.875,
-      "logits/rejected": -5.65625,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -1728.0,
-      "loss": 0.1582,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.375,
-      "rewards/margins": 4.65625,
-      "rewards/rejected": -13.0,
+      "logits/chosen": -5.09375,
+      "logits/rejected": -2.5,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -1816.0,
+      "loss": 0.1815,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -9.4375,
+      "rewards/margins": 4.46875,
+      "rewards/rejected": -13.875,
       "step": 1130
     },
     {
       "epoch": 1.193717277486911,
-      "grad_norm": 2.9243706518547095,
+      "grad_norm": 2.4597822586801623,
       "learning_rate": 3.766672355298907e-05,
-      "logits/chosen": -8.75,
-      "logits/rejected": -6.0625,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1664.0,
-      "loss": 0.1857,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.0,
-      "rewards/margins": 4.53125,
-      "rewards/rejected": -12.5,
+      "logits/chosen": -6.8125,
+      "logits/rejected": -4.15625,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -1504.0,
+      "loss": 0.177,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -7.09375,
+      "rewards/margins": 3.84375,
+      "rewards/rejected": -10.9375,
       "step": 1140
     },
     {
       "epoch": 1.2041884816753927,
-      "grad_norm": 0.8842993926639197,
+      "grad_norm": 1.2283444944199513,
       "learning_rate": 3.740313488217373e-05,
-      "logits/chosen": -6.9375,
-      "logits/rejected": -4.4375,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -1752.0,
-      "loss": 0.1577,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -8.75,
-      "rewards/margins": 4.53125,
-      "rewards/rejected": -13.25,
+      "logits/chosen": -7.8125,
+      "logits/rejected": -5.0,
+      "logps/chosen": -984.0,
+      "logps/rejected": -1424.0,
+      "loss": 0.1684,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -5.9375,
+      "rewards/margins": 4.0625,
+      "rewards/rejected": -10.0,
       "step": 1150
     },
     {
       "epoch": 1.2146596858638743,
-      "grad_norm": 3.3942778253534795,
+      "grad_norm": 2.513482786361478,
       "learning_rate": 3.7137704336959464e-05,
-      "logits/chosen": -5.09375,
-      "logits/rejected": -2.90625,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -1720.0,
-      "loss": 0.1562,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 4.1875,
-      "rewards/rejected": -13.125,
+      "logits/chosen": -4.96875,
+      "logits/rejected": -2.5,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -1584.0,
+      "loss": 0.1587,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -7.96875,
+      "rewards/margins": 3.796875,
+      "rewards/rejected": -11.75,
       "step": 1160
     },
     {
       "epoch": 1.225130890052356,
-      "grad_norm": 2.882375390249507,
+      "grad_norm": 2.389371344591705,
       "learning_rate": 3.687047133397201e-05,
-      "logits/chosen": -6.96875,
-      "logits/rejected": -4.375,
-      "logps/chosen": -1216.0,
-      "logps/rejected": -1616.0,
-      "loss": 0.1662,
+      "logits/chosen": -6.125,
+      "logits/rejected": -3.09375,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1560.0,
+      "loss": 0.1574,
       "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 4.125,
-      "rewards/rejected": -12.25,
+      "rewards/chosen": -7.5,
+      "rewards/margins": 4.1875,
+      "rewards/rejected": -11.6875,
       "step": 1170
     },
     {
       "epoch": 1.2356020942408377,
-      "grad_norm": 1.8036707306247453,
+      "grad_norm": 1.5623653069027057,
       "learning_rate": 3.6601475557503415e-05,
-      "logits/chosen": -6.0625,
-      "logits/rejected": -3.578125,
-      "logps/chosen": -1336.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.1693,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -9.125,
-      "rewards/margins": 4.21875,
-      "rewards/rejected": -13.3125,
+      "logits/chosen": -5.9375,
+      "logits/rejected": -2.859375,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -1584.0,
+      "loss": 0.1733,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -7.40625,
+      "rewards/margins": 4.1875,
+      "rewards/rejected": -11.625,
       "step": 1180
     },
     {
       "epoch": 1.2460732984293195,
-      "grad_norm": 1.6944393520102101,
+      "grad_norm": 1.7825194236860602,
       "learning_rate": 3.633075695361882e-05,
-      "logits/chosen": -6.34375,
-      "logits/rejected": -3.890625,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -1632.0,
-      "loss": 0.1607,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 4.25,
-      "rewards/rejected": -12.375,
+      "logits/chosen": -6.03125,
+      "logits/rejected": -3.046875,
+      "logps/chosen": -1064.0,
+      "logps/rejected": -1488.0,
+      "loss": 0.1695,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -6.6875,
+      "rewards/margins": 4.1875,
+      "rewards/rejected": -10.875,
       "step": 1190
     },
     {
       "epoch": 1.256544502617801,
-      "grad_norm": 1.5505814489660843,
+      "grad_norm": 2.4188875583998124,
       "learning_rate": 3.6058355724224477e-05,
-      "logits/chosen": -7.1875,
-      "logits/rejected": -4.90625,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -1688.0,
-      "loss": 0.1547,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -8.4375,
+      "logits/chosen": -6.46875,
+      "logits/rejected": -3.6875,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -1592.0,
+      "loss": 0.1584,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -7.4375,
       "rewards/margins": 4.28125,
-      "rewards/rejected": -12.6875,
+      "rewards/rejected": -11.6875,
       "step": 1200
     },
     {
       "epoch": 1.256544502617801,
-      "eval_logits/chosen": -4.625,
-      "eval_logits/rejected": -3.171875,
-      "eval_logps/chosen": -1480.0,
-      "eval_logps/rejected": -1560.0,
-      "eval_loss": 1.09389066696167,
-      "eval_rewards/accuracies": 0.595703125,
-      "eval_rewards/chosen": -11.625,
-      "eval_rewards/margins": 1.0859375,
-      "eval_rewards/rejected": -12.6875,
-      "eval_runtime": 46.7961,
-      "eval_samples_per_second": 42.739,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": -3.984375,
+      "eval_logits/rejected": -2.1875,
+      "eval_logps/chosen": -1360.0,
+      "eval_logps/rejected": -1440.0,
+      "eval_loss": 1.064203143119812,
+      "eval_rewards/accuracies": 0.591796875,
+      "eval_rewards/chosen": -10.375,
+      "eval_rewards/margins": 1.1640625,
+      "eval_rewards/rejected": -11.5625,
+      "eval_runtime": 46.6215,
+      "eval_samples_per_second": 42.899,
+      "eval_steps_per_second": 0.686,
       "step": 1200
     },
     {
       "epoch": 1.2670157068062826,
-      "grad_norm": 3.0693158748211804,
+      "grad_norm": 3.391591679680584,
       "learning_rate": 3.5784312321097744e-05,
-      "logits/chosen": -7.1875,
-      "logits/rejected": -5.03125,
-      "logps/chosen": -1328.0,
-      "logps/rejected": -1744.0,
-      "loss": 0.1474,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 4.25,
-      "rewards/rejected": -13.125,
+      "logits/chosen": -7.03125,
+      "logits/rejected": -4.3125,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -1696.0,
+      "loss": 0.1463,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.0625,
+      "rewards/margins": 4.59375,
+      "rewards/rejected": -12.6875,
       "step": 1210
     },
     {
       "epoch": 1.2774869109947644,
-      "grad_norm": 3.497546271262803,
+      "grad_norm": 2.9789324967308324,
       "learning_rate": 3.550866743987994e-05,
-      "logits/chosen": -7.78125,
-      "logits/rejected": -5.28125,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -1584.0,
-      "loss": 0.1583,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -7.5,
-      "rewards/margins": 4.0,
-      "rewards/rejected": -11.5,
+      "logits/chosen": -7.34375,
+      "logits/rejected": -4.53125,
+      "logps/chosen": -1224.0,
+      "logps/rejected": -1648.0,
+      "loss": 0.1702,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.65625,
+      "rewards/margins": 4.5,
+      "rewards/rejected": -12.125,
       "step": 1220
     },
     {
       "epoch": 1.2879581151832462,
-      "grad_norm": 1.572498337125938,
+      "grad_norm": 2.345263458285364,
       "learning_rate": 3.5231462014033024e-05,
-      "logits/chosen": -6.5625,
-      "logits/rejected": -4.28125,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -1640.0,
-      "loss": 0.1322,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 4.15625,
-      "rewards/rejected": -12.1875,
+      "logits/chosen": -7.46875,
+      "logits/rejected": -4.8125,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1480.0,
+      "loss": 0.1261,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -6.34375,
+      "rewards/margins": 4.25,
+      "rewards/rejected": -10.625,
       "step": 1230
     },
     {
       "epoch": 1.2984293193717278,
-      "grad_norm": 2.9226828422842175,
+      "grad_norm": 2.4569951385805284,
       "learning_rate": 3.495273720876095e-05,
-      "logits/chosen": -4.84375,
-      "logits/rejected": -3.015625,
-      "logps/chosen": -1336.0,
-      "logps/rejected": -1776.0,
-      "loss": 0.1429,
+      "logits/chosen": -4.75,
+      "logits/rejected": -2.671875,
+      "logps/chosen": -1264.0,
+      "logps/rejected": -1680.0,
+      "loss": 0.1589,
       "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 4.34375,
-      "rewards/rejected": -13.375,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 4.125,
+      "rewards/rejected": -12.4375,
       "step": 1240
     },
     {
       "epoch": 1.3089005235602094,
-      "grad_norm": 2.042289043023032,
+      "grad_norm": 2.1463092406286473,
       "learning_rate": 3.4672534414896616e-05,
-      "logits/chosen": -6.4375,
-      "logits/rejected": -3.6875,
-      "logps/chosen": -1352.0,
-      "logps/rejected": -1768.0,
-      "loss": 0.1393,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 4.4375,
-      "rewards/rejected": -13.375,
+      "logits/chosen": -7.0,
+      "logits/rejected": -3.890625,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -1648.0,
+      "loss": 0.148,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -7.875,
+      "rewards/margins": 4.3125,
+      "rewards/rejected": -12.1875,
       "step": 1250
     },
     {
       "epoch": 1.3193717277486912,
-      "grad_norm": 1.542039668656754,
+      "grad_norm": 1.5712653637719576,
       "learning_rate": 3.439089524275526e-05,
-      "logits/chosen": -4.5625,
-      "logits/rejected": -2.203125,
-      "logps/chosen": -1328.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.1443,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 4.4375,
-      "rewards/rejected": -13.5,
+      "logits/chosen": -5.3125,
+      "logits/rejected": -2.828125,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -1648.0,
+      "loss": 0.1541,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -7.9375,
+      "rewards/margins": 4.1875,
+      "rewards/rejected": -12.125,
       "step": 1260
     },
     {
       "epoch": 1.3298429319371727,
-      "grad_norm": 3.616863006487765,
+      "grad_norm": 1.991384724745942,
       "learning_rate": 3.410786151595533e-05,
-      "logits/chosen": -4.0625,
-      "logits/rejected": -1.8828125,
-      "logps/chosen": -1384.0,
-      "logps/rejected": -1920.0,
-      "loss": 0.1239,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -9.9375,
-      "rewards/margins": 5.1875,
-      "rewards/rejected": -15.125,
+      "logits/chosen": -4.34375,
+      "logits/rejected": -1.6328125,
+      "logps/chosen": -1320.0,
+      "logps/rejected": -1816.0,
+      "loss": 0.1289,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -9.25,
+      "rewards/margins": 4.875,
+      "rewards/rejected": -14.125,
       "step": 1270
     },
     {
       "epoch": 1.3403141361256545,
-      "grad_norm": 4.830752508061788,
+      "grad_norm": 5.409574503294168,
       "learning_rate": 3.382347526520765e-05,
-      "logits/chosen": -5.46875,
-      "logits/rejected": -2.890625,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -1824.0,
-      "loss": 0.1415,
+      "logits/chosen": -4.21875,
+      "logits/rejected": -1.15625,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -1992.0,
+      "loss": 0.1387,
       "rewards/accuracies": 0.90625,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 5.0,
-      "rewards/rejected": -14.0,
+      "rewards/chosen": -10.5,
+      "rewards/margins": 5.21875,
+      "rewards/rejected": -15.75,
       "step": 1280
     },
     {
       "epoch": 1.350785340314136,
-      "grad_norm": 3.2228882753649297,
+      "grad_norm": 3.5857255278730014,
       "learning_rate": 3.353777872207381e-05,
-      "logits/chosen": -4.0,
-      "logits/rejected": -1.3984375,
-      "logps/chosen": -1536.0,
-      "logps/rejected": -2064.0,
-      "loss": 0.1407,
+      "logits/chosen": -3.984375,
+      "logits/rejected": -0.72265625,
+      "logps/chosen": -1512.0,
+      "logps/rejected": -2016.0,
+      "loss": 0.1659,
       "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -11.0625,
-      "rewards/margins": 5.46875,
-      "rewards/rejected": -16.5,
+      "rewards/chosen": -10.8125,
+      "rewards/margins": 5.1875,
+      "rewards/rejected": -16.0,
       "step": 1290
     },
     {
       "epoch": 1.3612565445026177,
-      "grad_norm": 2.2242283994311163,
+      "grad_norm": 1.9959244910488476,
       "learning_rate": 3.325081431269475e-05,
-      "logits/chosen": -3.40625,
-      "logits/rejected": -1.0390625,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -1728.0,
-      "loss": 0.1577,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 4.65625,
-      "rewards/rejected": -13.75,
+      "logits/chosen": -3.984375,
+      "logits/rejected": -0.9453125,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -1512.0,
+      "loss": 0.1618,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -7.53125,
+      "rewards/margins": 4.0625,
+      "rewards/rejected": -11.5625,
       "step": 1300
     },
     {
       "epoch": 1.3612565445026177,
-      "eval_logits/chosen": -2.328125,
-      "eval_logits/rejected": -0.75,
-      "eval_logps/chosen": -1400.0,
-      "eval_logps/rejected": -1488.0,
-      "eval_loss": 1.0584686994552612,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -10.8125,
-      "eval_rewards/margins": 1.2265625,
-      "eval_rewards/rejected": -12.0,
-      "eval_runtime": 46.8081,
-      "eval_samples_per_second": 42.728,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": -3.078125,
+      "eval_logits/rejected": -1.25,
+      "eval_logps/chosen": -1240.0,
+      "eval_logps/rejected": -1320.0,
+      "eval_loss": 0.9749531149864197,
+      "eval_rewards/accuracies": 0.62109375,
+      "eval_rewards/chosen": -9.1875,
+      "eval_rewards/margins": 1.1484375,
+      "eval_rewards/rejected": -10.3125,
+      "eval_runtime": 46.6407,
+      "eval_samples_per_second": 42.881,
+      "eval_steps_per_second": 0.686,
       "step": 1300
     },
     {
       "epoch": 1.3717277486910995,
-      "grad_norm": 2.7062531346086205,
+      "grad_norm": 1.4874181135744604,
       "learning_rate": 3.296262465149046e-05,
-      "logits/chosen": -5.8125,
-      "logits/rejected": -2.796875,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -1752.0,
-      "loss": 0.129,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -12.875,
+      "logits/chosen": -6.8125,
+      "logits/rejected": -3.46875,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -1584.0,
+      "loss": 0.131,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.78125,
+      "rewards/margins": 4.3125,
+      "rewards/rejected": -11.125,
       "step": 1310
     },
     {
       "epoch": 1.3821989528795813,
-      "grad_norm": 1.4074181417128708,
+      "grad_norm": 2.1975648179713723,
       "learning_rate": 3.2673252534831687e-05,
       "logits/chosen": -5.34375,
-      "logits/rejected": -2.8125,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -1704.0,
-      "loss": 0.1434,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 4.75,
-      "rewards/rejected": -12.9375,
+      "logits/rejected": -2.578125,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -1696.0,
+      "loss": 0.1563,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 4.5625,
+      "rewards/rejected": -12.8125,
       "step": 1320
     },
     {
       "epoch": 1.3926701570680629,
-      "grad_norm": 2.112444793359292,
+      "grad_norm": 2.6476881391743383,
       "learning_rate": 3.2382740934684696e-05,
-      "logits/chosen": -4.375,
-      "logits/rejected": -1.8125,
-      "logps/chosen": -1352.0,
-      "logps/rejected": -1944.0,
-      "loss": 0.1577,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 5.21875,
-      "rewards/rejected": -14.625,
+      "logits/chosen": -5.09375,
+      "logits/rejected": -2.046875,
+      "logps/chosen": -1312.0,
+      "logps/rejected": -1872.0,
+      "loss": 0.1694,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -8.9375,
+      "rewards/margins": 5.03125,
+      "rewards/rejected": -13.9375,
       "step": 1330
     },
     {
       "epoch": 1.4031413612565444,
-      "grad_norm": 2.182136302958407,
+      "grad_norm": 2.254726033959155,
       "learning_rate": 3.209113299222982e-05,
-      "logits/chosen": -4.625,
-      "logits/rejected": -2.53125,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -1568.0,
-      "loss": 0.1452,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.53125,
-      "rewards/margins": 4.3125,
-      "rewards/rejected": -11.8125,
+      "logits/chosen": -4.9375,
+      "logits/rejected": -2.609375,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -1648.0,
+      "loss": 0.1422,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.25,
+      "rewards/margins": 4.4375,
+      "rewards/rejected": -12.6875,
       "step": 1340
     },
     {
       "epoch": 1.4136125654450262,
-      "grad_norm": 3.1420580028507037,
+      "grad_norm": 3.3422295861830493,
       "learning_rate": 3.179847201145505e-05,
-      "logits/chosen": -4.40625,
-      "logits/rejected": -2.1875,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -1568.0,
-      "loss": 0.1602,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 3.859375,
-      "rewards/rejected": -11.4375,
+      "logits/chosen": -4.25,
+      "logits/rejected": -1.46875,
+      "logps/chosen": -1296.0,
+      "logps/rejected": -1696.0,
+      "loss": 0.1652,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -8.5625,
+      "rewards/margins": 4.21875,
+      "rewards/rejected": -12.75,
       "step": 1350
     },
     {
       "epoch": 1.4240837696335078,
-      "grad_norm": 2.1989133284238287,
+      "grad_norm": 2.574727135764431,
       "learning_rate": 3.150480145272528e-05,
-      "logits/chosen": -4.59375,
-      "logits/rejected": -2.234375,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -1632.0,
-      "loss": 0.1666,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 3.875,
-      "rewards/rejected": -12.125,
+      "logits/chosen": -5.28125,
+      "logits/rejected": -2.59375,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -1592.0,
+      "loss": 0.1647,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -7.96875,
+      "rewards/margins": 3.796875,
+      "rewards/rejected": -11.75,
       "step": 1360
     },
     {
       "epoch": 1.4345549738219896,
-      "grad_norm": 1.851908132670553,
+      "grad_norm": 1.3621427947882565,
       "learning_rate": 3.121016492632848e-05,
-      "logits/chosen": -5.46875,
-      "logits/rejected": -2.953125,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -1624.0,
-      "loss": 0.1515,
+      "logits/chosen": -6.0625,
+      "logits/rejected": -3.21875,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -1536.0,
+      "loss": 0.1523,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 4.15625,
-      "rewards/rejected": -12.1875,
+      "rewards/chosen": -7.40625,
+      "rewards/margins": 3.9375,
+      "rewards/rejected": -11.3125,
       "step": 1370
     },
     {
       "epoch": 1.4450261780104712,
-      "grad_norm": 1.4539073836793057,
+      "grad_norm": 1.6003453036418098,
       "learning_rate": 3.091460618599951e-05,
-      "logits/chosen": -3.640625,
-      "logits/rejected": -0.9375,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -1920.0,
-      "loss": 0.1506,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -9.8125,
-      "rewards/margins": 4.84375,
-      "rewards/rejected": -14.625,
+      "logits/chosen": -4.28125,
+      "logits/rejected": -1.0625,
+      "logps/chosen": -1328.0,
+      "logps/rejected": -1832.0,
+      "loss": 0.1813,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -9.0,
+      "rewards/margins": 4.8125,
+      "rewards/rejected": -13.8125,
       "step": 1380
     },
     {
       "epoch": 1.4554973821989527,
-      "grad_norm": 2.0547677772931476,
+      "grad_norm": 1.7189208339778697,
       "learning_rate": 3.061816912242265e-05,
-      "logits/chosen": -6.03125,
-      "logits/rejected": -3.375,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1600.0,
-      "loss": 0.1324,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.59375,
-      "rewards/margins": 4.34375,
-      "rewards/rejected": -11.9375,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -2.328125,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -1608.0,
+      "loss": 0.1393,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -8.0,
+      "rewards/margins": 4.0,
+      "rewards/rejected": -12.0,
       "step": 1390
     },
     {
       "epoch": 1.4659685863874345,
-      "grad_norm": 2.81998032503535,
+      "grad_norm": 3.1544567700193253,
       "learning_rate": 3.0320897756713785e-05,
-      "logits/chosen": -6.0,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -1752.0,
-      "loss": 0.1773,
+      "logits/chosen": -5.625,
+      "logits/rejected": -2.4375,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -1696.0,
+      "loss": 0.1667,
       "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 4.53125,
-      "rewards/rejected": -12.8125,
+      "rewards/chosen": -7.6875,
+      "rewards/margins": 4.46875,
+      "rewards/rejected": -12.1875,
       "step": 1400
     },
     {
       "epoch": 1.4659685863874345,
-      "eval_logits/chosen": -2.734375,
-      "eval_logits/rejected": -1.1328125,
-      "eval_logps/chosen": -1432.0,
-      "eval_logps/rejected": -1512.0,
-      "eval_loss": 1.0706093311309814,
-      "eval_rewards/accuracies": 0.59375,
-      "eval_rewards/chosen": -11.125,
-      "eval_rewards/margins": 1.140625,
-      "eval_rewards/rejected": -12.25,
-      "eval_runtime": 46.7949,
-      "eval_samples_per_second": 42.74,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": -3.1875,
+      "eval_logits/rejected": -1.109375,
+      "eval_logps/chosen": -1296.0,
+      "eval_logps/rejected": -1400.0,
+      "eval_loss": 1.0400781631469727,
+      "eval_rewards/accuracies": 0.619140625,
+      "eval_rewards/chosen": -9.75,
+      "eval_rewards/margins": 1.3125,
+      "eval_rewards/rejected": -11.125,
+      "eval_runtime": 46.6552,
+      "eval_samples_per_second": 42.868,
+      "eval_steps_per_second": 0.686,
       "step": 1400
     },
     {
       "epoch": 1.4764397905759163,
-      "grad_norm": 1.6542715982950615,
+      "grad_norm": 1.9049073862561965,
       "learning_rate": 3.002283623388332e-05,
-      "logits/chosen": -5.1875,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -1616.0,
-      "loss": 0.1402,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -8.375,
-      "rewards/margins": 3.859375,
-      "rewards/rejected": -12.25,
+      "logits/chosen": -6.03125,
+      "logits/rejected": -3.578125,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -1480.0,
+      "loss": 0.1554,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -6.875,
+      "rewards/margins": 3.984375,
+      "rewards/rejected": -10.875,
       "step": 1410
     },
     {
       "epoch": 1.486910994764398,
-      "grad_norm": 4.719307296141394,
+      "grad_norm": 3.3864975503793335,
       "learning_rate": 2.9724028816280507e-05,
-      "logits/chosen": -5.53125,
-      "logits/rejected": -3.28125,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -1664.0,
-      "loss": 0.1459,
+      "logits/chosen": -5.90625,
+      "logits/rejected": -3.125,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -1600.0,
+      "loss": 0.136,
       "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 4.125,
-      "rewards/rejected": -12.625,
+      "rewards/chosen": -7.625,
+      "rewards/margins": 4.3125,
+      "rewards/rejected": -11.9375,
       "step": 1420
     },
     {
       "epoch": 1.4973821989528795,
-      "grad_norm": 2.6809702386732814,
+      "grad_norm": 1.8682813458473093,
       "learning_rate": 2.9424519877020524e-05,
-      "logits/chosen": -6.4375,
-      "logits/rejected": -4.125,
-      "logps/chosen": -1448.0,
-      "logps/rejected": -1936.0,
-      "loss": 0.153,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -9.8125,
-      "rewards/margins": 4.75,
-      "rewards/rejected": -14.5625,
+      "logits/chosen": -6.40625,
+      "logits/rejected": -3.59375,
+      "logps/chosen": -1312.0,
+      "logps/rejected": -1792.0,
+      "loss": 0.1522,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -8.4375,
+      "rewards/margins": 4.6875,
+      "rewards/rejected": -13.1875,
       "step": 1430
     },
     {
       "epoch": 1.5078534031413613,
-      "grad_norm": 2.429438908518276,
+      "grad_norm": 2.033291759819221,
       "learning_rate": 2.912435389339504e-05,
-      "logits/chosen": -6.3125,
-      "logits/rejected": -3.6875,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -1744.0,
-      "loss": 0.1215,
+      "logits/chosen": -5.71875,
+      "logits/rejected": -2.59375,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -1600.0,
+      "loss": 0.1151,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -9.25,
-      "rewards/margins": 4.375,
-      "rewards/rejected": -13.625,
+      "rewards/chosen": -7.78125,
+      "rewards/margins": 4.40625,
+      "rewards/rejected": -12.1875,
       "step": 1440
     },
     {
       "epoch": 1.518324607329843,
-      "grad_norm": 4.241122860130129,
+      "grad_norm": 3.1186557570388747,
       "learning_rate": 2.8823575440267226e-05,
-      "logits/chosen": -5.4375,
-      "logits/rejected": -2.953125,
-      "logps/chosen": -1536.0,
-      "logps/rejected": -2000.0,
-      "loss": 0.1648,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -11.1875,
-      "rewards/margins": 4.59375,
-      "rewards/rejected": -15.8125,
+      "logits/chosen": -4.96875,
+      "logits/rejected": -2.140625,
+      "logps/chosen": -1368.0,
+      "logps/rejected": -1824.0,
+      "loss": 0.1732,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -9.5625,
+      "rewards/margins": 4.46875,
+      "rewards/rejected": -14.0,
       "step": 1450
     },
     {
       "epoch": 1.5287958115183247,
-      "grad_norm": 1.7926270337651842,
+      "grad_norm": 2.3154355538311813,
       "learning_rate": 2.8522229183452465e-05,
-      "logits/chosen": -5.75,
-      "logits/rejected": -3.078125,
-      "logps/chosen": -1384.0,
-      "logps/rejected": -1856.0,
-      "loss": 0.1746,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -14.25,
+      "logits/chosen": -5.46875,
+      "logits/rejected": -2.421875,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -1824.0,
+      "loss": 0.1881,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -9.125,
+      "rewards/margins": 4.875,
+      "rewards/rejected": -14.0,
       "step": 1460
     },
     {
       "epoch": 1.5392670157068062,
-      "grad_norm": 1.8020308858430636,
+      "grad_norm": 2.1943723944707285,
       "learning_rate": 2.8220359873085373e-05,
-      "logits/chosen": -5.625,
-      "logits/rejected": -3.09375,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -1656.0,
-      "loss": 0.1403,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.0,
-      "rewards/margins": 4.375,
-      "rewards/rejected": -12.375,
+      "logits/chosen": -5.0,
+      "logits/rejected": -2.34375,
+      "logps/chosen": -1288.0,
+      "logps/rejected": -1720.0,
+      "loss": 0.1472,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -8.625,
+      "rewards/margins": 4.3125,
+      "rewards/rejected": -13.0,
       "step": 1470
     },
     {
       "epoch": 1.5497382198952878,
-      "grad_norm": 2.6121815673596687,
+      "grad_norm": 1.8457557037157377,
       "learning_rate": 2.7918012336974388e-05,
-      "logits/chosen": -4.6875,
-      "logits/rejected": -2.25,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -1680.0,
-      "loss": 0.1369,
+      "logits/chosen": -4.9375,
+      "logits/rejected": -2.265625,
+      "logps/chosen": -1328.0,
+      "logps/rejected": -1712.0,
+      "loss": 0.1365,
       "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 4.3125,
-      "rewards/rejected": -13.0,
+      "rewards/chosen": -9.125,
+      "rewards/margins": 4.21875,
+      "rewards/rejected": -13.3125,
       "step": 1480
     },
     {
       "epoch": 1.5602094240837696,
-      "grad_norm": 2.079992177774595,
+      "grad_norm": 1.5255346704917392,
       "learning_rate": 2.7615231473944815e-05,
-      "logits/chosen": -3.609375,
-      "logits/rejected": -1.0078125,
-      "logps/chosen": -1368.0,
-      "logps/rejected": -1800.0,
-      "loss": 0.1468,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -14.125,
+      "logits/chosen": -4.71875,
+      "logits/rejected": -1.75,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -1656.0,
+      "loss": 0.1493,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -8.625,
+      "rewards/margins": 4.15625,
+      "rewards/rejected": -12.75,
       "step": 1490
     },
     {
       "epoch": 1.5706806282722514,
-      "grad_norm": 2.759103381952096,
+      "grad_norm": 3.3102397848829463,
       "learning_rate": 2.7312062247171327e-05,
-      "logits/chosen": -5.5,
-      "logits/rejected": -3.0,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -1720.0,
-      "loss": 0.1675,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -8.6875,
+      "logits/chosen": -5.75,
+      "logits/rejected": -2.890625,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -1664.0,
+      "loss": 0.1714,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -8.1875,
       "rewards/margins": 4.3125,
-      "rewards/rejected": -13.0,
+      "rewards/rejected": -12.5,
       "step": 1500
     },
     {
       "epoch": 1.5706806282722514,
-      "eval_logits/chosen": -2.375,
-      "eval_logits/rejected": -0.73828125,
-      "eval_logps/chosen": -1464.0,
-      "eval_logps/rejected": -1560.0,
-      "eval_loss": 1.0755624771118164,
-      "eval_rewards/accuracies": 0.61328125,
-      "eval_rewards/chosen": -11.4375,
-      "eval_rewards/margins": 1.3125,
-      "eval_rewards/rejected": -12.75,
-      "eval_runtime": 46.8082,
-      "eval_samples_per_second": 42.728,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": -2.171875,
+      "eval_logits/rejected": -0.2578125,
+      "eval_logps/chosen": -1392.0,
+      "eval_logps/rejected": -1496.0,
+      "eval_loss": 1.0380312204360962,
+      "eval_rewards/accuracies": 0.623046875,
+      "eval_rewards/chosen": -10.6875,
+      "eval_rewards/margins": 1.34375,
+      "eval_rewards/rejected": -12.0625,
+      "eval_runtime": 46.6409,
+      "eval_samples_per_second": 42.881,
+      "eval_steps_per_second": 0.686,
       "step": 1500
     },
     {
       "epoch": 1.581151832460733,
-      "grad_norm": 3.604332986695593,
+      "grad_norm": 3.249043344401916,
       "learning_rate": 2.7008549677500876e-05,
-      "logits/chosen": -5.375,
-      "logits/rejected": -2.796875,
-      "logps/chosen": -1448.0,
-      "logps/rejected": -1968.0,
-      "loss": 0.1421,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -9.9375,
-      "rewards/margins": 4.9375,
-      "rewards/rejected": -14.875,
+      "logits/chosen": -5.21875,
+      "logits/rejected": -2.25,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -1848.0,
+      "loss": 0.1386,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -8.9375,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -13.625,
       "step": 1510
     },
     {
       "epoch": 1.5916230366492146,
-      "grad_norm": 2.165110468879555,
+      "grad_norm": 1.2209038858939858,
       "learning_rate": 2.670473883676709e-05,
-      "logits/chosen": -4.0,
-      "logits/rejected": -1.8984375,
-      "logps/chosen": -1384.0,
-      "logps/rejected": -1872.0,
-      "loss": 0.1243,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 4.78125,
-      "rewards/rejected": -14.375,
+      "logits/chosen": -4.1875,
+      "logits/rejected": -1.6015625,
+      "logps/chosen": -1264.0,
+      "logps/rejected": -1728.0,
+      "loss": 0.1334,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -8.375,
+      "rewards/margins": 4.59375,
+      "rewards/rejected": -13.0,
       "step": 1520
     },
     {
       "epoch": 1.6020942408376964,
-      "grad_norm": 3.4827422443230063,
+      "grad_norm": 3.3795368252885187,
       "learning_rate": 2.6400674841097074e-05,
-      "logits/chosen": -5.75,
-      "logits/rejected": -3.1875,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -1856.0,
-      "loss": 0.1473,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -14.1875,
+      "logits/chosen": -5.6875,
+      "logits/rejected": -2.8125,
+      "logps/chosen": -1280.0,
+      "logps/rejected": -1696.0,
+      "loss": 0.1397,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -8.4375,
+      "rewards/margins": 4.21875,
+      "rewards/rejected": -12.625,
       "step": 1530
     },
     {
       "epoch": 1.6125654450261782,
-      "grad_norm": 4.146372832508028,
+      "grad_norm": 3.438374741347147,
       "learning_rate": 2.609640284421158e-05,
-      "logits/chosen": -6.15625,
-      "logits/rejected": -3.03125,
-      "logps/chosen": -1368.0,
-      "logps/rejected": -1824.0,
-      "loss": 0.156,
+      "logits/chosen": -5.25,
+      "logits/rejected": -1.5,
+      "logps/chosen": -1400.0,
+      "logps/rejected": -1888.0,
+      "loss": 0.1635,
       "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -9.4375,
-      "rewards/margins": 4.59375,
-      "rewards/rejected": -14.0625,
+      "rewards/chosen": -9.75,
+      "rewards/margins": 4.90625,
+      "rewards/rejected": -14.625,
       "step": 1540
     },
     {
       "epoch": 1.6230366492146597,
-      "grad_norm": 2.217195969337078,
+      "grad_norm": 2.469835833336815,
       "learning_rate": 2.5791968030719726e-05,
-      "logits/chosen": -6.15625,
-      "logits/rejected": -3.546875,
-      "logps/chosen": -1328.0,
-      "logps/rejected": -1864.0,
-      "loss": 0.1455,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 5.15625,
-      "rewards/rejected": -13.875,
+      "logits/chosen": -5.21875,
+      "logits/rejected": -2.234375,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -1880.0,
+      "loss": 0.1466,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -8.9375,
+      "rewards/margins": 5.125,
+      "rewards/rejected": -14.0625,
       "step": 1550
     },
     {
       "epoch": 1.6335078534031413,
-      "grad_norm": 1.7618050879630325,
+      "grad_norm": 1.68551638404653,
       "learning_rate": 2.548741560940902e-05,
-      "logits/chosen": -5.28125,
-      "logits/rejected": -2.921875,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -1744.0,
-      "loss": 0.1544,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 4.3125,
-      "rewards/rejected": -13.3125,
+      "logits/chosen": -3.90625,
+      "logits/rejected": -1.2421875,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -1816.0,
+      "loss": 0.1404,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -9.5625,
+      "rewards/margins": 4.40625,
+      "rewards/rejected": -13.9375,
       "step": 1560
     },
     {
       "epoch": 1.6439790575916229,
-      "grad_norm": 3.5351899289784123,
+      "grad_norm": 2.649643781408795,
       "learning_rate": 2.5182790806531787e-05,
-      "logits/chosen": -4.875,
-      "logits/rejected": -2.25,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -1736.0,
-      "loss": 0.1654,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 4.15625,
-      "rewards/rejected": -13.25,
+      "logits/chosen": -3.84375,
+      "logits/rejected": -0.8515625,
+      "logps/chosen": -1352.0,
+      "logps/rejected": -1792.0,
+      "loss": 0.1733,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -9.1875,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -13.75,
       "step": 1570
     },
     {
       "epoch": 1.6544502617801047,
-      "grad_norm": 3.6271779731834384,
+      "grad_norm": 2.9151261226755647,
       "learning_rate": 2.4878138859089074e-05,
-      "logits/chosen": -6.0,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1664.0,
-      "loss": 0.1231,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 4.625,
-      "rewards/rejected": -12.25,
+      "logits/chosen": -4.4375,
+      "logits/rejected": -1.4765625,
+      "logps/chosen": -1224.0,
+      "logps/rejected": -1720.0,
+      "loss": 0.1208,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -7.96875,
+      "rewards/margins": 4.78125,
+      "rewards/rejected": -12.75,
       "step": 1580
     },
     {
       "epoch": 1.6649214659685865,
-      "grad_norm": 1.3241592049252036,
+      "grad_norm": 1.3953626342342078,
       "learning_rate": 2.4573505008112924e-05,
-      "logits/chosen": -5.5,
-      "logits/rejected": -2.703125,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -1848.0,
-      "loss": 0.1395,
+      "logits/chosen": -4.28125,
+      "logits/rejected": -1.359375,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -1888.0,
+      "loss": 0.134,
       "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -9.5625,
-      "rewards/margins": 4.65625,
-      "rewards/rejected": -14.1875,
+      "rewards/chosen": -9.75,
+      "rewards/margins": 4.9375,
+      "rewards/rejected": -14.625,
       "step": 1590
     },
     {
       "epoch": 1.675392670157068,
-      "grad_norm": 1.204792115339624,
+      "grad_norm": 1.7312201580744693,
       "learning_rate": 2.4268934491948032e-05,
-      "logits/chosen": -5.6875,
-      "logits/rejected": -2.8125,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -1696.0,
-      "loss": 0.1329,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -8.375,
-      "rewards/margins": 4.34375,
-      "rewards/rejected": -12.6875,
+      "logits/chosen": -4.75,
+      "logits/rejected": -1.71875,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -1752.0,
+      "loss": 0.1406,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -9.0,
+      "rewards/margins": 4.3125,
+      "rewards/rejected": -13.25,
       "step": 1600
     },
     {
       "epoch": 1.675392670157068,
-      "eval_logits/chosen": -3.796875,
-      "eval_logits/rejected": -1.796875,
-      "eval_logps/chosen": -1304.0,
-      "eval_logps/rejected": -1424.0,
-      "eval_loss": 1.039609432220459,
-      "eval_rewards/accuracies": 0.63671875,
-      "eval_rewards/chosen": -9.875,
-      "eval_rewards/margins": 1.453125,
-      "eval_rewards/rejected": -11.3125,
-      "eval_runtime": 46.8141,
-      "eval_samples_per_second": 42.722,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": -2.046875,
+      "eval_logits/rejected": -0.08740234375,
+      "eval_logps/chosen": -1440.0,
+      "eval_logps/rejected": -1552.0,
+      "eval_loss": 1.0427031517028809,
+      "eval_rewards/accuracies": 0.62109375,
+      "eval_rewards/chosen": -11.25,
+      "eval_rewards/margins": 1.375,
+      "eval_rewards/rejected": -12.625,
+      "eval_runtime": 46.6309,
+      "eval_samples_per_second": 42.89,
+      "eval_steps_per_second": 0.686,
       "step": 1600
     },
     {
       "epoch": 1.6858638743455496,
-      "grad_norm": 1.5579122096112112,
+      "grad_norm": 2.3044317264244887,
       "learning_rate": 2.396447253953385e-05,
-      "logits/chosen": -7.21875,
-      "logits/rejected": -4.375,
-      "logps/chosen": -1216.0,
-      "logps/rejected": -1688.0,
-      "loss": 0.1463,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.625,
+      "logits/chosen": -6.4375,
+      "logits/rejected": -3.609375,
+      "logps/chosen": -1264.0,
+      "logps/rejected": -1736.0,
+      "loss": 0.1494,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.125,
       "rewards/margins": 4.5625,
-      "rewards/rejected": -12.1875,
+      "rewards/rejected": -12.6875,
       "step": 1610
     },
     {
       "epoch": 1.6963350785340314,
-      "grad_norm": 1.8217060286418512,
+      "grad_norm": 1.4671041015006738,
       "learning_rate": 2.3660164363687998e-05,
-      "logits/chosen": -5.96875,
-      "logits/rejected": -3.390625,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -1752.0,
-      "loss": 0.1526,
+      "logits/chosen": -6.46875,
+      "logits/rejected": -3.96875,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -1672.0,
+      "loss": 0.1522,
       "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 4.5625,
-      "rewards/rejected": -13.0625,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -12.25,
       "step": 1620
     },
     {
       "epoch": 1.7068062827225132,
-      "grad_norm": 2.3575311152847744,
+      "grad_norm": 3.114952122051232,
       "learning_rate": 2.33560551543922e-05,
-      "logits/chosen": -6.5625,
-      "logits/rejected": -3.546875,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.1449,
+      "logits/chosen": -7.1875,
+      "logits/rejected": -4.09375,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -1648.0,
+      "loss": 0.1492,
       "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.5625,
-      "rewards/margins": 4.21875,
-      "rewards/rejected": -12.8125,
+      "rewards/chosen": -7.90625,
+      "rewards/margins": 4.1875,
+      "rewards/rejected": -12.125,
       "step": 1630
     },
     {
       "epoch": 1.7172774869109948,
-      "grad_norm": 3.3229953904575664,
+      "grad_norm": 2.7599339393418942,
       "learning_rate": 2.3052190072081493e-05,
-      "logits/chosen": -7.15625,
-      "logits/rejected": -3.984375,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -1752.0,
-      "loss": 0.1321,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 4.75,
-      "rewards/rejected": -12.8125,
+      "logits/chosen": -6.6875,
+      "logits/rejected": -3.625,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -1776.0,
+      "loss": 0.1414,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 4.84375,
+      "rewards/rejected": -13.125,
       "step": 1640
     },
     {
       "epoch": 1.7277486910994764,
-      "grad_norm": 3.1935648487407766,
+      "grad_norm": 2.4093424178053984,
       "learning_rate": 2.2748614240937864e-05,
-      "logits/chosen": -6.1875,
-      "logits/rejected": -3.890625,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -1656.0,
-      "loss": 0.143,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -8.5625,
-      "rewards/margins": 4.09375,
-      "rewards/rejected": -12.6875,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -3.59375,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -1608.0,
+      "loss": 0.1362,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 4.0,
+      "rewards/rejected": -12.25,
       "step": 1650
     },
     {
       "epoch": 1.738219895287958,
-      "grad_norm": 2.305956076392004,
+      "grad_norm": 1.3945011847925282,
       "learning_rate": 2.2445372742189334e-05,
-      "logits/chosen": -6.28125,
-      "logits/rejected": -4.0,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -1728.0,
-      "loss": 0.1479,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 4.5625,
-      "rewards/rejected": -12.875,
+      "logits/chosen": -5.0625,
+      "logits/rejected": -2.671875,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -1824.0,
+      "loss": 0.1323,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -9.0,
+      "rewards/margins": 4.8125,
+      "rewards/rejected": -13.8125,
       "step": 1660
     },
     {
       "epoch": 1.7486910994764397,
-      "grad_norm": 1.6851086548136442,
+      "grad_norm": 2.9256677107569873,
       "learning_rate": 2.214251060741528e-05,
-      "logits/chosen": -5.40625,
-      "logits/rejected": -2.953125,
-      "logps/chosen": -1352.0,
-      "logps/rejected": -1848.0,
-      "loss": 0.1057,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 4.84375,
-      "rewards/rejected": -13.9375,
+      "logits/chosen": -3.96875,
+      "logits/rejected": -1.3984375,
+      "logps/chosen": -1552.0,
+      "logps/rejected": -2064.0,
+      "loss": 0.1135,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 5.0625,
+      "rewards/rejected": -16.125,
       "step": 1670
     },
     {
       "epoch": 1.7591623036649215,
-      "grad_norm": 2.483965808700773,
+      "grad_norm": 2.908077954075383,
       "learning_rate": 2.1840072811859302e-05,
-      "logits/chosen": -5.21875,
-      "logits/rejected": -2.21875,
-      "logps/chosen": -1336.0,
-      "logps/rejected": -1840.0,
-      "loss": 0.1345,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 4.96875,
-      "rewards/rejected": -13.875,
+      "logits/chosen": -4.8125,
+      "logits/rejected": -1.6015625,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -1936.0,
+      "loss": 0.1472,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 4.8125,
+      "rewards/rejected": -14.8125,
       "step": 1680
     },
     {
       "epoch": 1.7696335078534031,
-      "grad_norm": 1.1267047333368767,
+      "grad_norm": 1.2142501833909511,
       "learning_rate": 2.1538104267750287e-05,
-      "logits/chosen": -4.625,
-      "logits/rejected": -2.046875,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -1768.0,
-      "loss": 0.1273,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -13.375,
+      "logits/chosen": -4.4375,
+      "logits/rejected": -1.5390625,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -1824.0,
+      "loss": 0.1195,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -9.4375,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -13.9375,
       "step": 1690
     },
     {
       "epoch": 1.7801047120418847,
-      "grad_norm": 1.538161640883356,
+      "grad_norm": 2.0254080715916616,
       "learning_rate": 2.1236649817632952e-05,
-      "logits/chosen": -4.8125,
-      "logits/rejected": -2.59375,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -1736.0,
-      "loss": 0.1055,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -13.3125,
+      "logits/chosen": -4.15625,
+      "logits/rejected": -1.7890625,
+      "logps/chosen": -1320.0,
+      "logps/rejected": -1840.0,
+      "loss": 0.1195,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -9.4375,
+      "rewards/margins": 4.90625,
+      "rewards/rejected": -14.3125,
       "step": 1700
     },
     {
       "epoch": 1.7801047120418847,
-      "eval_logits/chosen": -2.265625,
-      "eval_logits/rejected": -0.57421875,
-      "eval_logps/chosen": -1472.0,
-      "eval_logps/rejected": -1584.0,
-      "eval_loss": 1.108328104019165,
-      "eval_rewards/accuracies": 0.611328125,
-      "eval_rewards/chosen": -11.5,
-      "eval_rewards/margins": 1.4375,
-      "eval_rewards/rejected": -12.9375,
-      "eval_runtime": 46.8054,
-      "eval_samples_per_second": 42.73,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": -2.1875,
+      "eval_logits/rejected": -0.431640625,
+      "eval_logps/chosen": -1544.0,
+      "eval_logps/rejected": -1648.0,
+      "eval_loss": 1.137374997138977,
+      "eval_rewards/accuracies": 0.61328125,
+      "eval_rewards/chosen": -12.25,
+      "eval_rewards/margins": 1.390625,
+      "eval_rewards/rejected": -13.625,
+      "eval_runtime": 46.6236,
+      "eval_samples_per_second": 42.897,
+      "eval_steps_per_second": 0.686,
       "step": 1700
     },
     {
       "epoch": 1.7905759162303665,
-      "grad_norm": 1.8485766974813609,
+      "grad_norm": 2.0706788707724355,
       "learning_rate": 2.093575422770872e-05,
-      "logits/chosen": -4.21875,
-      "logits/rejected": -1.78125,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -1728.0,
-      "loss": 0.1254,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -8.75,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -13.375,
+      "logits/chosen": -4.03125,
+      "logits/rejected": -1.6328125,
+      "logps/chosen": -1320.0,
+      "logps/rejected": -1776.0,
+      "loss": 0.1342,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -9.375,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -13.9375,
       "step": 1710
     },
     {
       "epoch": 1.8010471204188483,
-      "grad_norm": 2.309921958800081,
+      "grad_norm": 2.781000122507841,
       "learning_rate": 2.0635462181187827e-05,
-      "logits/chosen": -5.21875,
-      "logits/rejected": -2.453125,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -1928.0,
-      "loss": 0.1144,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -9.8125,
-      "rewards/margins": 5.25,
-      "rewards/rejected": -15.0625,
+      "logits/chosen": -5.28125,
+      "logits/rejected": -2.421875,
+      "logps/chosen": -1456.0,
+      "logps/rejected": -1944.0,
+      "loss": 0.1236,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 4.96875,
+      "rewards/rejected": -15.25,
       "step": 1720
     },
     {
       "epoch": 1.8115183246073299,
-      "grad_norm": 2.9543744707766204,
+      "grad_norm": 1.9181349512843588,
       "learning_rate": 2.0335818271653917e-05,
-      "logits/chosen": -5.1875,
-      "logits/rejected": -2.234375,
-      "logps/chosen": -1336.0,
-      "logps/rejected": -1904.0,
-      "loss": 0.1558,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 5.4375,
-      "rewards/rejected": -14.5,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -2.515625,
+      "logps/chosen": -1320.0,
+      "logps/rejected": -1848.0,
+      "loss": 0.1372,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -8.875,
+      "rewards/margins": 5.09375,
+      "rewards/rejected": -14.0,
       "step": 1730
     },
     {
       "epoch": 1.8219895287958114,
-      "grad_norm": 4.298930561988656,
+      "grad_norm": 3.7275535913955262,
       "learning_rate": 2.0036866996441815e-05,
-      "logits/chosen": -4.9375,
-      "logits/rejected": -1.609375,
-      "logps/chosen": -1328.0,
-      "logps/rejected": -1912.0,
-      "loss": 0.1468,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -9.125,
-      "rewards/margins": 5.65625,
-      "rewards/rejected": -14.75,
+      "logits/chosen": -5.46875,
+      "logits/rejected": -1.984375,
+      "logps/chosen": -1352.0,
+      "logps/rejected": -1928.0,
+      "loss": 0.1421,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -9.375,
+      "rewards/margins": 5.53125,
+      "rewards/rejected": -14.9375,
       "step": 1740
     },
     {
       "epoch": 1.8324607329842932,
-      "grad_norm": 3.3330150896418074,
+      "grad_norm": 2.25436276227202,
       "learning_rate": 1.9738652750029677e-05,
-      "logits/chosen": -3.84375,
-      "logits/rejected": -1.484375,
-      "logps/chosen": -1328.0,
-      "logps/rejected": -1856.0,
-      "loss": 0.1313,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 4.84375,
-      "rewards/rejected": -13.8125,
+      "logits/chosen": -3.75,
+      "logits/rejected": -1.3046875,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -1952.0,
+      "loss": 0.1412,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -9.875,
+      "rewards/margins": 4.90625,
+      "rewards/rejected": -14.8125,
       "step": 1750
     },
     {
       "epoch": 1.8429319371727748,
-      "grad_norm": 3.269744327289175,
+      "grad_norm": 2.473446434384911,
       "learning_rate": 1.944121981744631e-05,
-      "logits/chosen": -4.90625,
-      "logits/rejected": -2.03125,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -1848.0,
-      "loss": 0.131,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 5.15625,
-      "rewards/rejected": -13.5,
+      "logits/chosen": -5.09375,
+      "logits/rejected": -2.234375,
+      "logps/chosen": -1368.0,
+      "logps/rejected": -1880.0,
+      "loss": 0.1306,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -8.9375,
+      "rewards/margins": 4.875,
+      "rewards/rejected": -13.8125,
       "step": 1760
     },
     {
       "epoch": 1.8534031413612566,
-      "grad_norm": 1.4384045373801313,
+      "grad_norm": 1.5930976659112075,
       "learning_rate": 1.9144612367694887e-05,
-      "logits/chosen": -4.75,
-      "logits/rejected": -2.046875,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -1728.0,
-      "loss": 0.1215,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 4.75,
-      "rewards/rejected": -12.875,
+      "logits/chosen": -5.78125,
+      "logits/rejected": -3.1875,
+      "logps/chosen": -1224.0,
+      "logps/rejected": -1664.0,
+      "loss": 0.1278,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -7.96875,
+      "rewards/margins": 4.375,
+      "rewards/rejected": -12.3125,
       "step": 1770
     },
     {
       "epoch": 1.8638743455497382,
-      "grad_norm": 2.1160082201871377,
+      "grad_norm": 1.4843070114957237,
       "learning_rate": 1.8848874447193803e-05,
-      "logits/chosen": -4.15625,
-      "logits/rejected": -1.875,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.1287,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -8.5625,
-      "rewards/margins": 4.90625,
-      "rewards/rejected": -13.5,
+      "logits/chosen": -5.1875,
+      "logits/rejected": -2.84375,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -1696.0,
+      "loss": 0.1321,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -8.4375,
+      "rewards/margins": 4.40625,
+      "rewards/rejected": -12.875,
       "step": 1780
     },
     {
       "epoch": 1.8743455497382198,
-      "grad_norm": 2.0963468182332257,
+      "grad_norm": 2.5695490970538666,
       "learning_rate": 1.855404997323571e-05,
-      "logits/chosen": -4.40625,
-      "logits/rejected": -1.6875,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -1856.0,
-      "loss": 0.1422,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 4.96875,
-      "rewards/rejected": -14.3125,
+      "logits/chosen": -5.25,
+      "logits/rejected": -2.40625,
+      "logps/chosen": -1328.0,
+      "logps/rejected": -1792.0,
+      "loss": 0.133,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -9.125,
+      "rewards/margins": 4.59375,
+      "rewards/rejected": -13.6875,
       "step": 1790
     },
     {
       "epoch": 1.8848167539267016,
-      "grad_norm": 3.0332665135833583,
+      "grad_norm": 2.4698832399999504,
       "learning_rate": 1.82601827274658e-05,
-      "logits/chosen": -3.921875,
-      "logits/rejected": -1.359375,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.1226,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -13.875,
+      "logits/chosen": -4.46875,
+      "logits/rejected": -1.5,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -1824.0,
+      "loss": 0.1291,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -9.5625,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -14.25,
       "step": 1800
     },
     {
       "epoch": 1.8848167539267016,
-      "eval_logits/chosen": -1.515625,
-      "eval_logits/rejected": 0.0625,
-      "eval_logps/chosen": -1408.0,
-      "eval_logps/rejected": -1520.0,
-      "eval_loss": 1.0952812433242798,
-      "eval_rewards/accuracies": 0.609375,
-      "eval_rewards/chosen": -10.9375,
-      "eval_rewards/margins": 1.3671875,
-      "eval_rewards/rejected": -12.3125,
-      "eval_runtime": 46.8204,
-      "eval_samples_per_second": 42.716,
-      "eval_steps_per_second": 0.683,
+      "eval_logits/chosen": -1.734375,
+      "eval_logits/rejected": 0.030517578125,
+      "eval_logps/chosen": -1488.0,
+      "eval_logps/rejected": -1592.0,
+      "eval_loss": 1.0742031335830688,
+      "eval_rewards/accuracies": 0.59375,
+      "eval_rewards/chosen": -11.6875,
+      "eval_rewards/margins": 1.34375,
+      "eval_rewards/rejected": -13.0625,
+      "eval_runtime": 46.6394,
+      "eval_samples_per_second": 42.882,
+      "eval_steps_per_second": 0.686,
       "step": 1800
     },
     {
       "epoch": 1.8952879581151834,
-      "grad_norm": 3.0116690073400845,
+      "grad_norm": 3.7269652465740744,
       "learning_rate": 1.7967316349380213e-05,
-      "logits/chosen": -4.59375,
-      "logits/rejected": -1.9296875,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -1664.0,
+      "logits/chosen": -4.53125,
+      "logits/rejected": -1.7421875,
+      "logps/chosen": -1280.0,
+      "logps/rejected": -1768.0,
       "loss": 0.1313,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.9375,
-      "rewards/margins": 4.84375,
-      "rewards/rejected": -12.75,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -8.9375,
+      "rewards/margins": 4.9375,
+      "rewards/rejected": -13.875,
       "step": 1810
     },
     {
       "epoch": 1.905759162303665,
-      "grad_norm": 2.213869463364088,
+      "grad_norm": 1.5545291973730322,
       "learning_rate": 1.7675494329845516e-05,
-      "logits/chosen": -4.1875,
-      "logits/rejected": -1.4296875,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1680.0,
-      "loss": 0.1187,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -7.875,
-      "rewards/margins": 4.78125,
-      "rewards/rejected": -12.6875,
+      "logits/chosen": -4.34375,
+      "logits/rejected": -1.5,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -1712.0,
+      "loss": 0.1175,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 4.5,
+      "rewards/rejected": -13.0,
       "step": 1820
     },
     {
       "epoch": 1.9162303664921465,
-      "grad_norm": 1.9798498679520493,
+      "grad_norm": 2.1446005790842912,
       "learning_rate": 1.738476000464026e-05,
-      "logits/chosen": -3.0625,
-      "logits/rejected": -0.6640625,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.1536,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 5.03125,
-      "rewards/rejected": -13.3125,
+      "logits/chosen": -3.453125,
+      "logits/rejected": -0.9453125,
+      "logps/chosen": -1320.0,
+      "logps/rejected": -1832.0,
+      "loss": 0.1446,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -9.125,
+      "rewards/margins": 4.96875,
+      "rewards/rejected": -14.0625,
       "step": 1830
     },
     {
       "epoch": 1.9267015706806283,
-      "grad_norm": 1.8057750936680306,
+      "grad_norm": 1.751151763299605,
       "learning_rate": 1.7095156548019646e-05,
-      "logits/chosen": -4.0625,
-      "logits/rejected": -1.6953125,
+      "logits/chosen": -5.15625,
+      "logits/rejected": -2.625,
       "logps/chosen": -1216.0,
-      "logps/rejected": -1720.0,
-      "loss": 0.1117,
-      "rewards/accuracies": 0.9437500238418579,
+      "logps/rejected": -1704.0,
+      "loss": 0.1099,
+      "rewards/accuracies": 0.918749988079071,
       "rewards/chosen": -8.1875,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -13.0,
+      "rewards/margins": 4.6875,
+      "rewards/rejected": -12.875,
       "step": 1840
     },
     {
       "epoch": 1.93717277486911,
-      "grad_norm": 2.4484918260650588,
+      "grad_norm": 2.1670770309595158,
       "learning_rate": 1.6806726966304064e-05,
-      "logits/chosen": -4.125,
-      "logits/rejected": -1.4609375,
-      "logps/chosen": -1400.0,
+      "logits/chosen": -4.875,
+      "logits/rejected": -2.0,
+      "logps/chosen": -1416.0,
       "logps/rejected": -1912.0,
-      "loss": 0.1561,
+      "loss": 0.1571,
       "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 5.0625,
+      "rewards/chosen": -9.625,
+      "rewards/margins": 4.9375,
       "rewards/rejected": -14.5625,
       "step": 1850
     },
     {
       "epoch": 1.9476439790575917,
-      "grad_norm": 1.4733067776561461,
+      "grad_norm": 1.0872004050858182,
       "learning_rate": 1.6519514091492625e-05,
-      "logits/chosen": -4.53125,
-      "logits/rejected": -1.8984375,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -1688.0,
-      "loss": 0.1283,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -8.625,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -13.125,
+      "logits/chosen": -4.5,
+      "logits/rejected": -1.8359375,
+      "logps/chosen": -1304.0,
+      "logps/rejected": -1752.0,
+      "loss": 0.1255,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -9.1875,
+      "rewards/margins": 4.5625,
+      "rewards/rejected": -13.75,
       "step": 1860
     },
     {
       "epoch": 1.9581151832460733,
-      "grad_norm": 2.1564465953023904,
+      "grad_norm": 1.6030246134507422,
       "learning_rate": 1.6233560574902627e-05,
-      "logits/chosen": -4.625,
-      "logits/rejected": -2.234375,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -1664.0,
-      "loss": 0.1337,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.65625,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -12.5,
+      "logits/chosen": -4.34375,
+      "logits/rejected": -2.0,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -1744.0,
+      "loss": 0.1533,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -8.5625,
+      "rewards/margins": 4.75,
+      "rewards/rejected": -13.3125,
       "step": 1870
     },
     {
       "epoch": 1.9685863874345548,
-      "grad_norm": 1.7491550190549408,
-      "learning_rate": 1.5948908880835753e-05,
-      "logits/chosen": -3.609375,
-      "logits/rejected": -0.84375,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -1776.0,
-      "loss": 0.1356,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 4.9375,
-      "rewards/rejected": -13.875,
+      "grad_norm": 2.5820815597909115,
+      "learning_rate": 1.5948908880835753e-05,
+      "logits/chosen": -4.15625,
+      "logits/rejected": -1.5546875,
+      "logps/chosen": -1312.0,
+      "logps/rejected": -1768.0,
+      "loss": 0.1399,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -9.0625,
+      "rewards/margins": 4.75,
+      "rewards/rejected": -13.75,
       "step": 1880
     },
     {
       "epoch": 1.9790575916230366,
-      "grad_norm": 2.79035843498615,
+      "grad_norm": 1.7259447152708345,
       "learning_rate": 1.5665601280272124e-05,
-      "logits/chosen": -4.03125,
-      "logits/rejected": -1.265625,
-      "logps/chosen": -1336.0,
-      "logps/rejected": -1792.0,
-      "loss": 0.1456,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -9.3125,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -14.0,
+      "logits/chosen": -4.875,
+      "logits/rejected": -2.21875,
+      "logps/chosen": -1304.0,
+      "logps/rejected": -1744.0,
+      "loss": 0.1389,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -9.0,
+      "rewards/margins": 4.5625,
+      "rewards/rejected": -13.5625,
       "step": 1890
     },
     {
       "epoch": 1.9895287958115184,
-      "grad_norm": 1.397705745303225,
+      "grad_norm": 2.1763448966537386,
       "learning_rate": 1.538367984459298e-05,
-      "logits/chosen": -3.8125,
-      "logits/rejected": -1.0,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -1728.0,
-      "loss": 0.1211,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -8.375,
-      "rewards/margins": 4.875,
-      "rewards/rejected": -13.25,
+      "logits/chosen": -3.78125,
+      "logits/rejected": -0.79296875,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -1840.0,
+      "loss": 0.1236,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -9.4375,
+      "rewards/margins": 4.9375,
+      "rewards/rejected": -14.375,
       "step": 1900
     },
     {
       "epoch": 1.9895287958115184,
-      "eval_logits/chosen": -0.96484375,
-      "eval_logits/rejected": 0.67578125,
-      "eval_logps/chosen": -1456.0,
-      "eval_logps/rejected": -1568.0,
-      "eval_loss": 1.0709062814712524,
-      "eval_rewards/accuracies": 0.61328125,
-      "eval_rewards/chosen": -11.375,
-      "eval_rewards/margins": 1.421875,
-      "eval_rewards/rejected": -12.75,
-      "eval_runtime": 46.8333,
-      "eval_samples_per_second": 42.705,
-      "eval_steps_per_second": 0.683,
+      "eval_logits/chosen": -0.97265625,
+      "eval_logits/rejected": 0.73828125,
+      "eval_logps/chosen": -1616.0,
+      "eval_logps/rejected": -1728.0,
+      "eval_loss": 1.153906226158142,
+      "eval_rewards/accuracies": 0.583984375,
+      "eval_rewards/chosen": -13.0,
+      "eval_rewards/margins": 1.3984375,
+      "eval_rewards/rejected": -14.375,
+      "eval_runtime": 46.6316,
+      "eval_samples_per_second": 42.889,
+      "eval_steps_per_second": 0.686,
       "step": 1900
     },
     {
       "epoch": 2.0,
-      "grad_norm": 2.977695694298433,
+      "grad_norm": 2.1345604487099634,
       "learning_rate": 1.5103186439333133e-05,
-      "logits/chosen": -3.640625,
-      "logits/rejected": -0.78125,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.1375,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 4.84375,
-      "rewards/rejected": -13.3125,
+      "logits/chosen": -4.03125,
+      "logits/rejected": -1.0390625,
+      "logps/chosen": -1304.0,
+      "logps/rejected": -1808.0,
+      "loss": 0.1131,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -9.125,
+      "rewards/margins": 5.15625,
+      "rewards/rejected": -14.3125,
       "step": 1910
     },
     {
       "epoch": 2.0104712041884816,
-      "grad_norm": 0.22429573423936316,
+      "grad_norm": 0.6297402051542761,
       "learning_rate": 1.4824162717963829e-05,
-      "logits/chosen": -4.15625,
-      "logits/rejected": -0.291015625,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -1984.0,
-      "loss": 0.0304,
+      "logits/chosen": -5.625,
+      "logits/rejected": -1.1875,
+      "logps/chosen": -1264.0,
+      "logps/rejected": -1976.0,
+      "loss": 0.0299,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.4375,
-      "rewards/margins": 7.1875,
-      "rewards/rejected": -15.625,
+      "rewards/chosen": -8.5625,
+      "rewards/margins": 7.0,
+      "rewards/rejected": -15.5625,
       "step": 1920
     },
     {
       "epoch": 2.020942408376963,
-      "grad_norm": 0.31787124977716114,
+      "grad_norm": 0.2477707121871446,
       "learning_rate": 1.4546650115707192e-05,
-      "logits/chosen": -4.0,
-      "logits/rejected": -0.84375,
-      "logps/chosen": -1328.0,
-      "logps/rejected": -2064.0,
-      "loss": 0.0347,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -9.4375,
-      "rewards/margins": 7.25,
-      "rewards/rejected": -16.625,
+      "logits/chosen": -4.40625,
+      "logits/rejected": -1.046875,
+      "logps/chosen": -1408.0,
+      "logps/rejected": -2144.0,
+      "loss": 0.0387,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 7.28125,
+      "rewards/rejected": -17.5,
       "step": 1930
     },
     {
       "epoch": 2.031413612565445,
-      "grad_norm": 0.2526825155410066,
+      "grad_norm": 0.4017483385038123,
       "learning_rate": 1.4270689843383111e-05,
-      "logits/chosen": -2.84375,
-      "logits/rejected": 0.130859375,
-      "logps/chosen": -1496.0,
-      "logps/rejected": -2224.0,
-      "loss": 0.0332,
+      "logits/chosen": -3.21875,
+      "logits/rejected": 0.10546875,
+      "logps/chosen": -1624.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.0333,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -10.75,
-      "rewards/margins": 7.46875,
-      "rewards/rejected": -18.25,
+      "rewards/chosen": -12.0,
+      "rewards/margins": 8.0625,
+      "rewards/rejected": -20.125,
       "step": 1940
     },
     {
       "epoch": 2.0418848167539267,
-      "grad_norm": 0.30384723123700136,
+      "grad_norm": 0.25424741448721505,
       "learning_rate": 1.3996322881289347e-05,
-      "logits/chosen": -3.25,
-      "logits/rejected": -0.017333984375,
-      "logps/chosen": -1528.0,
-      "logps/rejected": -2336.0,
-      "loss": 0.0306,
+      "logits/chosen": -3.34375,
+      "logits/rejected": 0.212890625,
+      "logps/chosen": -1680.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0333,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -11.4375,
-      "rewards/margins": 7.75,
-      "rewards/rejected": -19.125,
+      "rewards/chosen": -12.875,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -21.25,
       "step": 1950
     },
     {
       "epoch": 2.0523560209424083,
-      "grad_norm": 0.32148883765329156,
+      "grad_norm": 1.4519731776449065,
       "learning_rate": 1.3723589973115963e-05,
-      "logits/chosen": -2.546875,
-      "logits/rejected": 0.8671875,
-      "logps/chosen": -1552.0,
-      "logps/rejected": -2432.0,
-      "loss": 0.042,
+      "logits/chosen": -3.125,
+      "logits/rejected": 0.73828125,
+      "logps/chosen": -1648.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0466,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 8.5,
-      "rewards/rejected": -20.125,
+      "rewards/chosen": -12.5625,
+      "rewards/margins": 9.25,
+      "rewards/rejected": -21.75,
       "step": 1960
     },
     {
       "epoch": 2.06282722513089,
-      "grad_norm": 0.40648777654978546,
+      "grad_norm": 0.2835475178639754,
       "learning_rate": 1.3452531619894837e-05,
-      "logits/chosen": -3.03125,
-      "logits/rejected": 0.56640625,
-      "logps/chosen": -1624.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0275,
+      "logits/chosen": -4.125,
+      "logits/rejected": 0.01251220703125,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0286,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -12.0,
-      "rewards/margins": 8.5,
-      "rewards/rejected": -20.5,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -20.125,
       "step": 1970
     },
     {
       "epoch": 2.073298429319372,
-      "grad_norm": 0.1570121050738589,
+      "grad_norm": 0.2288758910760686,
       "learning_rate": 1.3183188073985245e-05,
-      "logits/chosen": -3.953125,
-      "logits/rejected": 0.224609375,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2432.0,
-      "loss": 0.037,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -11.8125,
-      "rewards/margins": 8.375,
-      "rewards/rejected": -20.125,
+      "logits/chosen": -5.28125,
+      "logits/rejected": -0.9453125,
+      "logps/chosen": -1456.0,
+      "logps/rejected": -2304.0,
+      "loss": 0.0372,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -18.875,
       "step": 1980
     },
     {
       "epoch": 2.0837696335078535,
-      "grad_norm": 0.0878090386039078,
+      "grad_norm": 0.09249295129833393,
       "learning_rate": 1.2915599333096351e-05,
-      "logits/chosen": -3.828125,
-      "logits/rejected": -0.703125,
-      "logps/chosen": -1624.0,
-      "logps/rejected": -2384.0,
-      "loss": 0.0283,
+      "logits/chosen": -5.09375,
+      "logits/rejected": -1.6640625,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -2272.0,
+      "loss": 0.0255,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -12.0,
-      "rewards/margins": 7.625,
-      "rewards/rejected": -19.625,
+      "rewards/chosen": -10.6875,
+      "rewards/margins": 7.75,
+      "rewards/rejected": -18.5,
       "step": 1990
     },
     {
       "epoch": 2.094240837696335,
-      "grad_norm": 0.39648521204815446,
+      "grad_norm": 0.21134274298594663,
       "learning_rate": 1.264980513434752e-05,
-      "logits/chosen": -3.609375,
-      "logits/rejected": -0.1005859375,
-      "logps/chosen": -1616.0,
-      "logps/rejected": -2416.0,
-      "loss": 0.0277,
+      "logits/chosen": -4.625,
+      "logits/rejected": -0.8203125,
+      "logps/chosen": -1584.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.0264,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 8.0625,
-      "rewards/rejected": -19.625,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 8.4375,
+      "rewards/rejected": -19.75,
       "step": 2000
     },
     {
       "epoch": 2.094240837696335,
-      "eval_logits/chosen": 0.431640625,
-      "eval_logits/rejected": 2.09375,
-      "eval_logps/chosen": -1912.0,
-      "eval_logps/rejected": -2064.0,
-      "eval_loss": 1.4782344102859497,
-      "eval_rewards/accuracies": 0.6015625,
-      "eval_rewards/chosen": -15.9375,
-      "eval_rewards/margins": 1.7890625,
-      "eval_rewards/rejected": -17.75,
-      "eval_runtime": 46.8017,
-      "eval_samples_per_second": 42.733,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": -0.625,
+      "eval_logits/rejected": 1.15625,
+      "eval_logps/chosen": -1968.0,
+      "eval_logps/rejected": -2112.0,
+      "eval_loss": 1.5532655715942383,
+      "eval_rewards/accuracies": 0.583984375,
+      "eval_rewards/chosen": -16.5,
+      "eval_rewards/margins": 1.75,
+      "eval_rewards/rejected": -18.25,
+      "eval_runtime": 46.6508,
+      "eval_samples_per_second": 42.872,
+      "eval_steps_per_second": 0.686,
       "step": 2000
     },
     {
       "epoch": 2.1047120418848166,
-      "grad_norm": 0.30760323238772863,
+      "grad_norm": 0.38557099137534334,
       "learning_rate": 1.2385844948367324e-05,
-      "logits/chosen": -2.921875,
-      "logits/rejected": 0.32421875,
-      "logps/chosen": -1616.0,
-      "logps/rejected": -2416.0,
-      "loss": 0.0438,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -12.25,
-      "rewards/margins": 8.0,
-      "rewards/rejected": -20.25,
+      "logits/chosen": -3.703125,
+      "logits/rejected": -0.27734375,
+      "logps/chosen": -1688.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0398,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -12.9375,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -22.125,
       "step": 2010
     },
     {
       "epoch": 2.115183246073298,
-      "grad_norm": 0.23107736865431497,
+      "grad_norm": 0.3158617636120164,
       "learning_rate": 1.2123757973432114e-05,
-      "logits/chosen": -3.625,
-      "logits/rejected": -0.416015625,
-      "logps/chosen": -1704.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0259,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -12.6875,
+      "logits/chosen": -3.859375,
+      "logits/rejected": -0.6875,
+      "logps/chosen": -1816.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0451,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -13.75,
       "rewards/margins": 8.25,
-      "rewards/rejected": -21.0,
+      "rewards/rejected": -22.0,
       "step": 2020
     },
     {
       "epoch": 2.1256544502617802,
-      "grad_norm": 0.28427857137569634,
+      "grad_norm": 0.4686009463545712,
       "learning_rate": 1.1863583129645106e-05,
-      "logits/chosen": -2.984375,
-      "logits/rejected": 0.306640625,
-      "logps/chosen": -1648.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0347,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -12.375,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -21.625,
+      "logits/chosen": -5.0,
+      "logits/rejected": -1.2109375,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2336.0,
+      "loss": 0.0375,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -10.625,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -19.25,
       "step": 2030
     },
     {
       "epoch": 2.136125654450262,
-      "grad_norm": 0.4069304249040383,
+      "grad_norm": 0.6130902236522178,
       "learning_rate": 1.1605359053156607e-05,
-      "logits/chosen": -3.40625,
-      "logits/rejected": 0.3671875,
-      "logps/chosen": -1768.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0222,
+      "logits/chosen": -4.5625,
+      "logits/rejected": -0.52734375,
+      "logps/chosen": -1648.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.026,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -13.125,
-      "rewards/margins": 9.75,
-      "rewards/rejected": -22.875,
+      "rewards/chosen": -11.9375,
+      "rewards/margins": 8.9375,
+      "rewards/rejected": -20.875,
       "step": 2040
     },
     {
       "epoch": 2.1465968586387434,
-      "grad_norm": 1.2506944041557424,
+      "grad_norm": 0.30379072449398525,
       "learning_rate": 1.1349124090426644e-05,
-      "logits/chosen": -2.828125,
-      "logits/rejected": 0.36328125,
-      "logps/chosen": -1728.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0332,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -13.0,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -22.125,
+      "logits/chosen": -3.328125,
+      "logits/rejected": -0.16015625,
+      "logps/chosen": -1696.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0295,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -12.6875,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -21.75,
       "step": 2050
     },
     {
       "epoch": 2.157068062827225,
-      "grad_norm": 0.3674160896721416,
+      "grad_norm": 0.5456155880199106,
       "learning_rate": 1.1094916292530406e-05,
-      "logits/chosen": -2.953125,
-      "logits/rejected": 0.353515625,
-      "logps/chosen": -1568.0,
-      "logps/rejected": -2432.0,
-      "loss": 0.0272,
+      "logits/chosen": -3.546875,
+      "logits/rejected": 0.0322265625,
+      "logps/chosen": -1664.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0278,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -11.3125,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -20.25,
+      "rewards/chosen": -12.25,
+      "rewards/margins": 8.875,
+      "rewards/rejected": -21.125,
       "step": 2060
     },
     {
       "epoch": 2.167539267015707,
-      "grad_norm": 0.13265182006138093,
+      "grad_norm": 1.1182631661420612,
       "learning_rate": 1.0842773409507623e-05,
-      "logits/chosen": -3.09375,
-      "logits/rejected": 0.47265625,
-      "logps/chosen": -1440.0,
-      "logps/rejected": -2352.0,
-      "loss": 0.0321,
+      "logits/chosen": -3.921875,
+      "logits/rejected": 0.049072265625,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0338,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -10.5625,
-      "rewards/margins": 8.625,
-      "rewards/rejected": -19.25,
+      "rewards/chosen": -11.3125,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -20.375,
       "step": 2070
     },
     {
       "epoch": 2.1780104712041886,
-      "grad_norm": 0.9997096468554217,
+      "grad_norm": 0.7433052599448169,
       "learning_rate": 1.0592732884756754e-05,
-      "logits/chosen": -3.421875,
-      "logits/rejected": 0.2177734375,
-      "logps/chosen": -1704.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0243,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -12.625,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -21.625,
+      "logits/chosen": -4.59375,
+      "logits/rejected": -0.40625,
+      "logps/chosen": -1688.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0245,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -12.5,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -21.875,
       "step": 2080
     },
     {
       "epoch": 2.18848167539267,
-      "grad_norm": 0.30717249258992646,
+      "grad_norm": 0.34915585936617366,
       "learning_rate": 1.0344831849474507e-05,
-      "logits/chosen": -2.4375,
-      "logits/rejected": 0.78125,
-      "logps/chosen": -1672.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0294,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -12.5,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -21.25,
+      "logits/chosen": -3.515625,
+      "logits/rejected": 0.1533203125,
+      "logps/chosen": -1640.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.0271,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -12.1875,
+      "rewards/margins": 8.5625,
+      "rewards/rejected": -20.75,
       "step": 2090
     },
     {
       "epoch": 2.1989528795811517,
-      "grad_norm": 0.7727584262619247,
+      "grad_norm": 0.0797052605189764,
       "learning_rate": 1.0099107117141879e-05,
-      "logits/chosen": -1.4140625,
-      "logits/rejected": 1.9765625,
-      "logps/chosen": -1736.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0199,
+      "logits/chosen": -3.0625,
+      "logits/rejected": 1.0390625,
+      "logps/chosen": -1576.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0222,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -13.375,
-      "rewards/margins": 10.25,
-      "rewards/rejected": -23.625,
+      "rewards/chosen": -11.8125,
+      "rewards/margins": 10.0,
+      "rewards/rejected": -21.875,
       "step": 2100
     },
     {
       "epoch": 2.1989528795811517,
-      "eval_logits/chosen": 1.46875,
-      "eval_logits/rejected": 3.03125,
-      "eval_logps/chosen": -2192.0,
-      "eval_logps/rejected": -2368.0,
-      "eval_loss": 1.7629687786102295,
-      "eval_rewards/accuracies": 0.59765625,
-      "eval_rewards/chosen": -18.625,
-      "eval_rewards/margins": 2.109375,
-      "eval_rewards/rejected": -20.75,
-      "eval_runtime": 46.8157,
-      "eval_samples_per_second": 42.721,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": 0.310546875,
+      "eval_logits/rejected": 2.078125,
+      "eval_logps/chosen": -2064.0,
+      "eval_logps/rejected": -2224.0,
+      "eval_loss": 1.60532808303833,
+      "eval_rewards/accuracies": 0.595703125,
+      "eval_rewards/chosen": -17.375,
+      "eval_rewards/margins": 1.890625,
+      "eval_rewards/rejected": -19.25,
+      "eval_runtime": 46.6545,
+      "eval_samples_per_second": 42.868,
+      "eval_steps_per_second": 0.686,
       "step": 2100
     },
     {
       "epoch": 2.2094240837696333,
-      "grad_norm": 0.3601747376315074,
+      "grad_norm": 0.25151140789695786,
       "learning_rate": 9.855595178057334e-06,
-      "logits/chosen": -2.140625,
-      "logits/rejected": 1.0390625,
-      "logps/chosen": -1880.0,
-      "logps/rejected": -2976.0,
-      "loss": 0.0208,
+      "logits/chosen": -3.265625,
+      "logits/rejected": 0.490234375,
+      "logps/chosen": -1760.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.021,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -14.5,
-      "rewards/margins": 10.8125,
-      "rewards/rejected": -25.375,
+      "rewards/chosen": -13.3125,
+      "rewards/margins": 10.5,
+      "rewards/rejected": -23.75,
       "step": 2110
     },
     {
       "epoch": 2.2198952879581153,
-      "grad_norm": 0.24745131024845177,
+      "grad_norm": 0.5084634897780025,
       "learning_rate": 9.614332193917949e-06,
-      "logits/chosen": -2.15625,
-      "logits/rejected": 1.3984375,
-      "logps/chosen": -1888.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.044,
+      "logits/chosen": -3.34375,
+      "logits/rejected": 0.73828125,
+      "logps/chosen": -1776.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.0339,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -14.5,
-      "rewards/margins": 10.125,
-      "rewards/rejected": -24.625,
+      "rewards/chosen": -13.3125,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -23.125,
       "step": 2120
     },
     {
       "epoch": 2.230366492146597,
-      "grad_norm": 0.36788293427088864,
+      "grad_norm": 0.1323803050610093,
       "learning_rate": 9.375353992449385e-06,
-      "logits/chosen": -2.984375,
-      "logits/rejected": 0.4453125,
-      "logps/chosen": -1672.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0242,
+      "logits/chosen": -4.15625,
+      "logits/rejected": -0.357421875,
+      "logps/chosen": -1648.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0209,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.625,
-      "rewards/margins": 9.625,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -12.3125,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -21.875,
       "step": 2130
     },
     {
       "epoch": 2.2408376963350785,
-      "grad_norm": 2.1436243689208685,
+      "grad_norm": 0.5124130565842094,
       "learning_rate": 9.138696062085442e-06,
-      "logits/chosen": -3.203125,
-      "logits/rejected": 0.70703125,
-      "logps/chosen": -1624.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0242,
+      "logits/chosen": -4.75,
+      "logits/rejected": -0.2412109375,
+      "logps/chosen": -1608.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0243,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -12.125,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -21.375,
+      "rewards/chosen": -11.9375,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -21.75,
       "step": 2140
     },
     {
       "epoch": 2.25130890052356,
-      "grad_norm": 1.3302029470220504,
+      "grad_norm": 4.1560061547793365,
       "learning_rate": 8.904393546698007e-06,
-      "logits/chosen": -2.625,
-      "logits/rejected": 0.5703125,
-      "logps/chosen": -1656.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0254,
+      "logits/chosen": -3.203125,
+      "logits/rejected": 0.287109375,
+      "logps/chosen": -1736.0,
+      "logps/rejected": -2576.0,
+      "loss": 0.028,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.375,
-      "rewards/margins": 8.375,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -13.25,
+      "rewards/margins": 8.4375,
+      "rewards/rejected": -21.75,
       "step": 2150
     },
     {
       "epoch": 2.261780104712042,
-      "grad_norm": 0.9189421943677044,
+      "grad_norm": 0.3389015104765602,
       "learning_rate": 8.672481240378142e-06,
-      "logits/chosen": -2.359375,
-      "logits/rejected": 1.296875,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2368.0,
-      "loss": 0.0405,
+      "logits/chosen": -3.3125,
+      "logits/rejected": 0.734375,
+      "logps/chosen": -1592.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0411,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -11.3125,
-      "rewards/margins": 8.5625,
-      "rewards/rejected": -19.875,
+      "rewards/chosen": -12.1875,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -20.75,
       "step": 2160
     },
     {
       "epoch": 2.2722513089005236,
-      "grad_norm": 0.11235342210814132,
+      "grad_norm": 0.36514291342332295,
       "learning_rate": 8.44299358226919e-06,
-      "logits/chosen": -2.53125,
-      "logits/rejected": 1.0078125,
-      "logps/chosen": -1552.0,
-      "logps/rejected": -2384.0,
-      "loss": 0.032,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -11.4375,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -19.625,
+      "logits/chosen": -3.765625,
+      "logits/rejected": 0.1171875,
+      "logps/chosen": -1640.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.0315,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -12.3125,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -20.625,
       "step": 2170
     },
     {
       "epoch": 2.282722513089005,
-      "grad_norm": 0.597154587083781,
+      "grad_norm": 0.2405542472432919,
       "learning_rate": 8.215964651452457e-06,
-      "logits/chosen": -2.046875,
-      "logits/rejected": 1.4296875,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2512.0,
-      "loss": 0.0247,
+      "logits/chosen": -3.09375,
+      "logits/rejected": 0.80859375,
+      "logps/chosen": -1680.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.021,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -11.9375,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -12.75,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -21.75,
       "step": 2180
     },
     {
       "epoch": 2.2931937172774868,
-      "grad_norm": 0.0881368744954682,
+      "grad_norm": 0.45259155879111546,
       "learning_rate": 7.991428161886502e-06,
-      "logits/chosen": -2.59375,
-      "logits/rejected": 1.2109375,
-      "logps/chosen": -1616.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0254,
+      "logits/chosen": -4.0,
+      "logits/rejected": 0.30078125,
+      "logps/chosen": -1672.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.024,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.75,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -20.625,
+      "rewards/chosen": -12.3125,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -21.875,
       "step": 2190
     },
     {
       "epoch": 2.303664921465969,
-      "grad_norm": 0.9231213742476826,
+      "grad_norm": 0.5975393416602715,
       "learning_rate": 7.76941745740061e-06,
-      "logits/chosen": -2.484375,
-      "logits/rejected": 0.8828125,
-      "logps/chosen": -1712.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0298,
+      "logits/chosen": -3.421875,
+      "logits/rejected": 0.51171875,
+      "logps/chosen": -1792.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.0266,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -12.625,
-      "rewards/margins": 8.0625,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -13.5,
+      "rewards/margins": 9.3125,
+      "rewards/rejected": -22.75,
       "step": 2200
     },
     {
       "epoch": 2.303664921465969,
-      "eval_logits/chosen": 1.03125,
-      "eval_logits/rejected": 2.640625,
-      "eval_logps/chosen": -1920.0,
-      "eval_logps/rejected": -2080.0,
-      "eval_loss": 1.5056406259536743,
+      "eval_logits/chosen": 0.091796875,
+      "eval_logits/rejected": 1.9296875,
+      "eval_logps/chosen": -2032.0,
+      "eval_logps/rejected": -2192.0,
+      "eval_loss": 1.5843125581741333,
       "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -16.0,
-      "eval_rewards/margins": 1.8203125,
-      "eval_rewards/rejected": -17.875,
-      "eval_runtime": 46.8149,
-      "eval_samples_per_second": 42.721,
-      "eval_steps_per_second": 0.684,
+      "eval_rewards/chosen": -17.125,
+      "eval_rewards/margins": 1.8671875,
+      "eval_rewards/rejected": -19.0,
+      "eval_runtime": 46.6526,
+      "eval_samples_per_second": 42.87,
+      "eval_steps_per_second": 0.686,
       "step": 2200
     },
     {
       "epoch": 2.3141361256544504,
-      "grad_norm": 0.17487680838726483,
+      "grad_norm": 0.0950610613837811,
       "learning_rate": 7.549965506743123e-06,
-      "logits/chosen": -2.5625,
-      "logits/rejected": 0.84375,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0229,
+      "logits/chosen": -3.671875,
+      "logits/rejected": 0.29296875,
+      "logps/chosen": -1672.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0303,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.0625,
-      "rewards/margins": 8.6875,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -12.75,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -22.25,
       "step": 2210
     },
     {
       "epoch": 2.324607329842932,
-      "grad_norm": 0.6858996052111531,
+      "grad_norm": 0.29994880255980566,
       "learning_rate": 7.333104898685631e-06,
-      "logits/chosen": -2.765625,
-      "logits/rejected": 0.59765625,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0377,
+      "logits/chosen": -4.03125,
+      "logits/rejected": -0.05322265625,
+      "logps/chosen": -1616.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0353,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.8125,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -11.9375,
+      "rewards/margins": 9.3125,
+      "rewards/rejected": -21.25,
       "step": 2220
     },
     {
       "epoch": 2.3350785340314135,
-      "grad_norm": 0.2774198817019759,
+      "grad_norm": 0.28746319295299394,
       "learning_rate": 7.118867837183518e-06,
-      "logits/chosen": -2.9375,
-      "logits/rejected": 0.6328125,
-      "logps/chosen": -1736.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0231,
+      "logits/chosen": -3.6875,
+      "logits/rejected": 0.3984375,
+      "logps/chosen": -1832.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.0247,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -12.5625,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -22.375,
+      "rewards/chosen": -13.5625,
+      "rewards/margins": 10.3125,
+      "rewards/rejected": -23.875,
       "step": 2230
     },
     {
       "epoch": 2.345549738219895,
-      "grad_norm": 0.13346925938875664,
+      "grad_norm": 0.19181106284235644,
       "learning_rate": 6.9072861365936045e-06,
-      "logits/chosen": -3.109375,
-      "logits/rejected": 0.474609375,
-      "logps/chosen": -1920.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.0218,
+      "logits/chosen": -3.75,
+      "logits/rejected": 0.328125,
+      "logps/chosen": -2032.0,
+      "logps/rejected": -3072.0,
+      "loss": 0.0238,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -14.3125,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -23.5,
+      "rewards/chosen": -15.5,
+      "rewards/margins": 10.1875,
+      "rewards/rejected": -25.625,
       "step": 2240
     },
     {
       "epoch": 2.356020942408377,
-      "grad_norm": 0.221839104586537,
+      "grad_norm": 0.18249853484463424,
       "learning_rate": 6.6983912169497005e-06,
-      "logits/chosen": -2.484375,
-      "logits/rejected": 0.65234375,
-      "logps/chosen": -1712.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0251,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.4375,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -21.625,
+      "logits/chosen": -3.40625,
+      "logits/rejected": 0.333984375,
+      "logps/chosen": -1832.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.0282,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -13.6875,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -23.0,
       "step": 2250
     },
     {
       "epoch": 2.3664921465968587,
-      "grad_norm": 0.19698076400584638,
+      "grad_norm": 0.2760240564395289,
       "learning_rate": 6.49221409929677e-06,
-      "logits/chosen": -1.703125,
-      "logits/rejected": 1.390625,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0206,
+      "logits/chosen": -2.4375,
+      "logits/rejected": 1.2109375,
+      "logps/chosen": -1736.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.0225,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.875,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -20.875,
+      "rewards/chosen": -13.4375,
+      "rewards/margins": 9.625,
+      "rewards/rejected": -23.125,
       "step": 2260
     },
     {
       "epoch": 2.3769633507853403,
-      "grad_norm": 0.5629550693549261,
+      "grad_norm": 0.7505920652284434,
       "learning_rate": 6.2887854010842075e-06,
-      "logits/chosen": -1.7890625,
-      "logits/rejected": 1.2265625,
-      "logps/chosen": -1680.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0277,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -12.875,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -22.25,
+      "logits/chosen": -2.453125,
+      "logits/rejected": 0.76953125,
+      "logps/chosen": -1832.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.0296,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -14.375,
+      "rewards/margins": 9.25,
+      "rewards/rejected": -23.625,
       "step": 2270
     },
     {
       "epoch": 2.387434554973822,
-      "grad_norm": 0.830461475465937,
+      "grad_norm": 0.3118603023270945,
       "learning_rate": 6.0881353316191385e-06,
-      "logits/chosen": -2.21875,
-      "logits/rejected": 0.36328125,
-      "logps/chosen": -1640.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0224,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -12.0625,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -21.375,
+      "logits/chosen": -2.734375,
+      "logits/rejected": 0.061767578125,
+      "logps/chosen": -1744.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0231,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -13.125,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -22.5,
       "step": 2280
     },
     {
       "epoch": 2.3979057591623034,
-      "grad_norm": 0.5808046074627202,
+      "grad_norm": 0.3846925299917,
       "learning_rate": 5.890293687580381e-06,
-      "logits/chosen": -1.7890625,
-      "logits/rejected": 1.375,
-      "logps/chosen": -1728.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0317,
+      "logits/chosen": -2.796875,
+      "logits/rejected": 0.76171875,
+      "logps/chosen": -1760.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0338,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -13.0,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -22.125,
+      "rewards/chosen": -13.3125,
+      "rewards/margins": 8.9375,
+      "rewards/rejected": -22.25,
       "step": 2290
     },
     {
       "epoch": 2.4083769633507854,
-      "grad_norm": 0.34161772300288784,
+      "grad_norm": 0.7361746610607077,
       "learning_rate": 5.695289848593532e-06,
-      "logits/chosen": -2.453125,
-      "logits/rejected": 0.007354736328125,
-      "logps/chosen": -1680.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0278,
+      "logits/chosen": -3.78125,
+      "logits/rejected": -0.8828125,
+      "logps/chosen": -1672.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.0247,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -12.75,
-      "rewards/margins": 9.5,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -12.6875,
+      "rewards/margins": 9.875,
+      "rewards/rejected": -22.5,
       "step": 2300
     },
     {
       "epoch": 2.4083769633507854,
-      "eval_logits/chosen": 1.8125,
-      "eval_logits/rejected": 3.375,
-      "eval_logps/chosen": -2080.0,
-      "eval_logps/rejected": -2256.0,
-      "eval_loss": 1.6822500228881836,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -17.625,
-      "eval_rewards/margins": 1.9453125,
-      "eval_rewards/rejected": -19.625,
-      "eval_runtime": 46.815,
-      "eval_samples_per_second": 42.721,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": 0.365234375,
+      "eval_logits/rejected": 2.171875,
+      "eval_logps/chosen": -2112.0,
+      "eval_logps/rejected": -2288.0,
+      "eval_loss": 1.6308749914169312,
+      "eval_rewards/accuracies": 0.609375,
+      "eval_rewards/chosen": -17.875,
+      "eval_rewards/margins": 2.0,
+      "eval_rewards/rejected": -19.875,
+      "eval_runtime": 46.6423,
+      "eval_samples_per_second": 42.88,
+      "eval_steps_per_second": 0.686,
       "step": 2300
     },
     {
       "epoch": 2.418848167539267,
-      "grad_norm": 0.5094115006262872,
+      "grad_norm": 0.35499238637135305,
       "learning_rate": 5.50315277286812e-06,
-      "logits/chosen": -2.3125,
-      "logits/rejected": 0.8828125,
-      "logps/chosen": -1704.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0378,
+      "logits/chosen": -3.984375,
+      "logits/rejected": -0.244140625,
+      "logps/chosen": -1688.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0339,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -13.1875,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -22.0,
+      "rewards/chosen": -13.0,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -22.75,
       "step": 2310
     },
     {
       "epoch": 2.4293193717277486,
-      "grad_norm": 0.39380116759005684,
+      "grad_norm": 0.8134324029264798,
       "learning_rate": 5.313910992897281e-06,
-      "logits/chosen": -2.4375,
-      "logits/rejected": 0.92578125,
-      "logps/chosen": -1616.0,
-      "logps/rejected": -2432.0,
-      "loss": 0.0369,
+      "logits/chosen": -3.40625,
+      "logits/rejected": 0.380859375,
+      "logps/chosen": -1704.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0319,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -11.9375,
-      "rewards/margins": 8.4375,
-      "rewards/rejected": -20.375,
+      "rewards/chosen": -12.75,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -21.875,
       "step": 2320
     },
     {
       "epoch": 2.4397905759162306,
-      "grad_norm": 0.44961516539140206,
+      "grad_norm": 0.39129690589014837,
       "learning_rate": 5.127592611220658e-06,
-      "logits/chosen": -2.875,
-      "logits/rejected": 0.65625,
-      "logps/chosen": -1672.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0345,
+      "logits/chosen": -3.796875,
+      "logits/rejected": 0.05517578125,
+      "logps/chosen": -1744.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.0357,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -12.25,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -21.5,
+      "rewards/chosen": -13.0625,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -22.5,
       "step": 2330
     },
     {
       "epoch": 2.450261780104712,
-      "grad_norm": 0.5222419196563728,
+      "grad_norm": 0.3863005815246044,
       "learning_rate": 4.944225296251159e-06,
-      "logits/chosen": -2.0,
-      "logits/rejected": 1.265625,
-      "logps/chosen": -1632.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0292,
+      "logits/chosen": -3.625,
+      "logits/rejected": 0.431640625,
+      "logps/chosen": -1680.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0269,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -12.5625,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -21.75,
+      "rewards/chosen": -13.0625,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -23.0,
       "step": 2340
     },
     {
       "epoch": 2.4607329842931938,
-      "grad_norm": 0.23420407224804035,
+      "grad_norm": 0.3167141549340763,
       "learning_rate": 4.763836278166172e-06,
-      "logits/chosen": -2.375,
-      "logits/rejected": 1.2109375,
-      "logps/chosen": -1640.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0208,
+      "logits/chosen": -3.703125,
+      "logits/rejected": 0.5,
+      "logps/chosen": -1728.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0246,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -12.1875,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -21.0,
+      "rewards/chosen": -13.0625,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -22.125,
       "step": 2350
     },
     {
       "epoch": 2.4712041884816753,
-      "grad_norm": 0.2122553187884289,
+      "grad_norm": 0.6923763880128672,
       "learning_rate": 4.586452344863876e-06,
-      "logits/chosen": -2.828125,
-      "logits/rejected": 0.68359375,
-      "logps/chosen": -1616.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0234,
+      "logits/chosen": -3.8125,
+      "logits/rejected": 0.1982421875,
+      "logps/chosen": -1720.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.0236,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -11.9375,
-      "rewards/margins": 9.375,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -12.9375,
+      "rewards/margins": 9.875,
+      "rewards/rejected": -22.75,
       "step": 2360
     },
     {
       "epoch": 2.481675392670157,
-      "grad_norm": 0.3700830106322245,
+      "grad_norm": 2.1571331694748954,
       "learning_rate": 4.412099837985192e-06,
-      "logits/chosen": -2.75,
-      "logits/rejected": 1.0,
-      "logps/chosen": -1656.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0288,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -12.125,
+      "logits/chosen": -3.75,
+      "logits/rejected": 0.341796875,
+      "logps/chosen": -1720.0,
+      "logps/rejected": -2768.0,
+      "loss": 0.0277,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -12.8125,
       "rewards/margins": 9.9375,
-      "rewards/rejected": -22.0,
+      "rewards/rejected": -22.75,
       "step": 2370
     },
     {
       "epoch": 2.492146596858639,
-      "grad_norm": 0.09117590419878288,
+      "grad_norm": 1.863400889237136,
       "learning_rate": 4.240804649002089e-06,
-      "logits/chosen": -2.3125,
-      "logits/rejected": 0.9453125,
+      "logits/chosen": -3.828125,
+      "logits/rejected": -0.07568359375,
       "logps/chosen": -1576.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0319,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -12.0,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -20.875,
+      "logps/rejected": -2528.0,
+      "loss": 0.0317,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -11.9375,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -21.125,
       "step": 2380
     },
     {
       "epoch": 2.5026178010471205,
-      "grad_norm": 0.2071623065144461,
+      "grad_norm": 0.29341283259082807,
       "learning_rate": 4.072592215372598e-06,
-      "logits/chosen": -2.90625,
-      "logits/rejected": 0.296875,
-      "logps/chosen": -1744.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.021,
+      "logits/chosen": -4.1875,
+      "logits/rejected": -0.50390625,
+      "logps/chosen": -1784.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.0229,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -12.875,
-      "rewards/margins": 9.375,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -13.3125,
+      "rewards/margins": 9.5625,
+      "rewards/rejected": -22.875,
       "step": 2390
     },
     {
       "epoch": 2.513089005235602,
-      "grad_norm": 0.24368742644646574,
+      "grad_norm": 0.0876913105487749,
       "learning_rate": 3.907487516763389e-06,
-      "logits/chosen": -2.28125,
-      "logits/rejected": 0.875,
+      "logits/chosen": -3.703125,
+      "logits/rejected": 0.1259765625,
       "logps/chosen": -1664.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0401,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -12.5625,
-      "rewards/margins": 9.4375,
-      "rewards/rejected": -22.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.0381,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -12.625,
+      "rewards/margins": 9.5625,
+      "rewards/rejected": -22.25,
       "step": 2400
     },
     {
       "epoch": 2.513089005235602,
-      "eval_logits/chosen": 1.546875,
-      "eval_logits/rejected": 3.125,
-      "eval_logps/chosen": -2048.0,
-      "eval_logps/rejected": -2224.0,
-      "eval_loss": 1.6474374532699585,
+      "eval_logits/chosen": 0.2353515625,
+      "eval_logits/rejected": 2.0,
+      "eval_logps/chosen": -2096.0,
+      "eval_logps/rejected": -2256.0,
+      "eval_loss": 1.6237187385559082,
       "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -17.375,
-      "eval_rewards/margins": 2.078125,
-      "eval_rewards/rejected": -19.375,
-      "eval_runtime": 46.8112,
-      "eval_samples_per_second": 42.725,
-      "eval_steps_per_second": 0.684,
+      "eval_rewards/chosen": -17.75,
+      "eval_rewards/margins": 1.921875,
+      "eval_rewards/rejected": -19.625,
+      "eval_runtime": 46.6307,
+      "eval_samples_per_second": 42.89,
+      "eval_steps_per_second": 0.686,
       "step": 2400
     },
     {
       "epoch": 2.5235602094240837,
-      "grad_norm": 2.7452061423719685,
+      "grad_norm": 0.28244466669423757,
       "learning_rate": 3.745515071340286e-06,
-      "logits/chosen": -1.8984375,
-      "logits/rejected": 1.5546875,
-      "logps/chosen": -1768.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0327,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -13.375,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -22.75,
+      "logits/chosen": -3.078125,
+      "logits/rejected": 0.7578125,
+      "logps/chosen": -1808.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.0304,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -13.8125,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -23.125,
       "step": 2410
     },
     {
       "epoch": 2.5340314136125652,
-      "grad_norm": 0.48717812479280703,
+      "grad_norm": 0.19025743262987949,
       "learning_rate": 3.586698932127236e-06,
-      "logits/chosen": -2.765625,
-      "logits/rejected": 0.64453125,
-      "logps/chosen": -1728.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0248,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -13.0,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -22.25,
+      "logits/chosen": -4.15625,
+      "logits/rejected": -0.11572265625,
+      "logps/chosen": -1736.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0234,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -13.0625,
+      "rewards/margins": 9.6875,
+      "rewards/rejected": -22.75,
       "step": 2420
     },
     {
       "epoch": 2.5445026178010473,
-      "grad_norm": 0.2111359208717434,
+      "grad_norm": 1.144510545924364,
       "learning_rate": 3.4310626834344744e-06,
-      "logits/chosen": -2.921875,
-      "logits/rejected": 0.51171875,
-      "logps/chosen": -1800.0,
+      "logits/chosen": -4.34375,
+      "logits/rejected": -0.34375,
+      "logps/chosen": -1784.0,
       "logps/rejected": -2832.0,
-      "loss": 0.029,
+      "loss": 0.0285,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -13.5,
-      "rewards/margins": 10.1875,
+      "rewards/chosen": -13.3125,
+      "rewards/margins": 10.3125,
       "rewards/rejected": -23.625,
       "step": 2430
     },
     {
       "epoch": 2.554973821989529,
-      "grad_norm": 0.4222202079820167,
+      "grad_norm": 0.26600991618367276,
       "learning_rate": 3.2786294373562343e-06,
-      "logits/chosen": -2.359375,
-      "logits/rejected": 1.421875,
-      "logps/chosen": -1712.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0238,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -12.8125,
-      "rewards/margins": 9.6875,
-      "rewards/rejected": -22.5,
+      "logits/chosen": -3.734375,
+      "logits/rejected": 0.453125,
+      "logps/chosen": -1728.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0276,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -13.0,
+      "rewards/margins": 9.25,
+      "rewards/rejected": -22.25,
       "step": 2440
     },
     {
       "epoch": 2.5654450261780104,
-      "grad_norm": 0.4218087543558086,
+      "grad_norm": 0.4033871703021631,
       "learning_rate": 3.1294218303385526e-06,
-      "logits/chosen": -3.109375,
-      "logits/rejected": 0.6875,
+      "logits/chosen": -4.1875,
+      "logits/rejected": -0.2265625,
       "logps/chosen": -1744.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0241,
+      "logps/rejected": -2624.0,
+      "loss": 0.0232,
       "rewards/accuracies": 0.9750000238418579,
       "rewards/chosen": -12.8125,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -22.125,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -21.75,
       "step": 2450
     },
     {
       "epoch": 2.5759162303664924,
-      "grad_norm": 0.2511257145758431,
+      "grad_norm": 0.12978736083644618,
       "learning_rate": 2.9834620198177483e-06,
-      "logits/chosen": -2.375,
-      "logits/rejected": 0.61328125,
-      "logps/chosen": -1768.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0324,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -13.4375,
-      "rewards/margins": 9.4375,
-      "rewards/rejected": -22.875,
+      "logits/chosen": -3.578125,
+      "logits/rejected": -0.1923828125,
+      "logps/chosen": -1728.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0362,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -13.0625,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -22.25,
       "step": 2460
     },
     {
       "epoch": 2.5863874345549736,
-      "grad_norm": 0.7642494399971571,
+      "grad_norm": 0.26345895269249103,
       "learning_rate": 2.840771680930068e-06,
-      "logits/chosen": -2.53125,
-      "logits/rejected": 1.2578125,
-      "logps/chosen": -1864.0,
-      "logps/rejected": -2992.0,
-      "loss": 0.0262,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -14.0625,
-      "rewards/margins": 11.0625,
-      "rewards/rejected": -25.125,
+      "logits/chosen": -3.609375,
+      "logits/rejected": 0.64453125,
+      "logps/chosen": -1832.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.0184,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -13.8125,
+      "rewards/margins": 10.9375,
+      "rewards/rejected": -24.75,
       "step": 2470
     },
     {
       "epoch": 2.5968586387434556,
-      "grad_norm": 0.1120933507630596,
+      "grad_norm": 0.14338392227286292,
       "learning_rate": 2.701372003292868e-06,
-      "logits/chosen": -1.9453125,
-      "logits/rejected": 1.5703125,
+      "logits/chosen": -3.171875,
+      "logits/rejected": 0.82421875,
       "logps/chosen": -1664.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0309,
+      "logps/rejected": -2592.0,
+      "loss": 0.0291,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -12.625,
-      "rewards/margins": 9.625,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -12.6875,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -22.0,
       "step": 2480
     },
     {
       "epoch": 2.607329842931937,
-      "grad_norm": 0.31553739531202757,
+      "grad_norm": 0.15553631647257285,
       "learning_rate": 2.56528368785795e-06,
-      "logits/chosen": -2.703125,
-      "logits/rejected": 1.234375,
-      "logps/chosen": -1712.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0279,
+      "logits/chosen": -3.890625,
+      "logits/rejected": 0.53125,
+      "logps/chosen": -1696.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.0246,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -12.8125,
-      "rewards/margins": 9.875,
-      "rewards/rejected": -22.75,
+      "rewards/chosen": -12.625,
+      "rewards/margins": 9.75,
+      "rewards/rejected": -22.375,
       "step": 2490
     },
     {
       "epoch": 2.6178010471204187,
-      "grad_norm": 0.09561709666444017,
+      "grad_norm": 0.6226393701839031,
       "learning_rate": 2.432526943837471e-06,
-      "logits/chosen": -2.25,
-      "logits/rejected": 1.484375,
-      "logps/chosen": -1584.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.025,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.875,
+      "logits/chosen": -3.171875,
+      "logits/rejected": 0.984375,
+      "logps/chosen": -1648.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.0307,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -12.5625,
       "rewards/margins": 10.25,
-      "rewards/rejected": -22.125,
+      "rewards/rejected": -22.875,
       "step": 2500
     },
     {
       "epoch": 2.6178010471204187,
-      "eval_logits/chosen": 1.3125,
-      "eval_logits/rejected": 2.921875,
-      "eval_logps/chosen": -2048.0,
-      "eval_logps/rejected": -2240.0,
-      "eval_loss": 1.6600781679153442,
+      "eval_logits/chosen": 0.10693359375,
+      "eval_logits/rejected": 1.9140625,
+      "eval_logps/chosen": -2064.0,
+      "eval_logps/rejected": -2224.0,
+      "eval_loss": 1.6102187633514404,
       "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -17.25,
-      "eval_rewards/margins": 2.171875,
-      "eval_rewards/rejected": -19.5,
-      "eval_runtime": 46.8245,
-      "eval_samples_per_second": 42.713,
-      "eval_steps_per_second": 0.683,
+      "eval_rewards/chosen": -17.375,
+      "eval_rewards/margins": 2.015625,
+      "eval_rewards/rejected": -19.375,
+      "eval_runtime": 46.6358,
+      "eval_samples_per_second": 42.886,
+      "eval_steps_per_second": 0.686,
       "step": 2500
     },
     {
       "epoch": 2.6282722513089007,
-      "grad_norm": 0.8970183997819912,
+      "grad_norm": 0.45618465921697904,
       "learning_rate": 2.3031214857028446e-06,
-      "logits/chosen": -1.9375,
-      "logits/rejected": 1.0390625,
-      "logps/chosen": -1672.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0218,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.375,
-      "rewards/margins": 9.4375,
-      "rewards/rejected": -21.875,
+      "logits/chosen": -3.265625,
+      "logits/rejected": 0.3359375,
+      "logps/chosen": -1680.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0229,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -12.4375,
+      "rewards/margins": 9.625,
+      "rewards/rejected": -22.0,
       "step": 2510
     },
     {
       "epoch": 2.6387434554973823,
-      "grad_norm": 2.0660270811286385,
+      "grad_norm": 0.6853609526157255,
       "learning_rate": 2.177086530257139e-06,
-      "logits/chosen": -2.375,
-      "logits/rejected": 1.0625,
-      "logps/chosen": -1680.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0366,
+      "logits/chosen": -3.625,
+      "logits/rejected": 0.431640625,
+      "logps/chosen": -1664.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.033,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -12.6875,
+      "rewards/chosen": -12.5625,
       "rewards/margins": 9.9375,
-      "rewards/rejected": -22.625,
+      "rewards/rejected": -22.5,
       "step": 2520
     },
     {
       "epoch": 2.649214659685864,
-      "grad_norm": 0.21016434853410138,
+      "grad_norm": 0.11240244342419557,
       "learning_rate": 2.054440793781354e-06,
-      "logits/chosen": -2.390625,
-      "logits/rejected": 1.015625,
-      "logps/chosen": -1720.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0257,
+      "logits/chosen": -3.578125,
+      "logits/rejected": 0.478515625,
+      "logps/chosen": -1728.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.0288,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -13.0,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -22.0,
+      "rewards/chosen": -13.0625,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -22.5,
       "step": 2530
     },
     {
       "epoch": 2.6596858638743455,
-      "grad_norm": 0.21720986949010146,
+      "grad_norm": 0.14381000921080447,
       "learning_rate": 1.9352024892550587e-06,
-      "logits/chosen": -2.25,
-      "logits/rejected": 1.3359375,
-      "logps/chosen": -1728.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0203,
+      "logits/chosen": -3.375,
+      "logits/rejected": 0.828125,
+      "logps/chosen": -1720.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0197,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -13.125,
-      "rewards/margins": 10.75,
-      "rewards/rejected": -23.875,
+      "rewards/chosen": -13.0625,
+      "rewards/margins": 10.9375,
+      "rewards/rejected": -24.0,
       "step": 2540
     },
     {
       "epoch": 2.670157068062827,
-      "grad_norm": 0.1783798429407388,
+      "grad_norm": 0.11795057403262292,
       "learning_rate": 1.8193893236517379e-06,
-      "logits/chosen": -2.546875,
-      "logits/rejected": 0.75,
-      "logps/chosen": -1736.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.0247,
+      "logits/chosen": -3.5625,
+      "logits/rejected": 0.15625,
+      "logps/chosen": -1744.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0232,
       "rewards/accuracies": 0.987500011920929,
       "rewards/chosen": -13.125,
-      "rewards/margins": 10.125,
-      "rewards/rejected": -23.25,
+      "rewards/margins": 10.4375,
+      "rewards/rejected": -23.5,
       "step": 2550
     },
     {
       "epoch": 2.680628272251309,
-      "grad_norm": 0.1012205603456733,
+      "grad_norm": 0.29103351604793165,
       "learning_rate": 1.7070184953092956e-06,
-      "logits/chosen": -2.3125,
-      "logits/rejected": 0.85546875,
-      "logps/chosen": -1664.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0316,
+      "logits/chosen": -3.46875,
+      "logits/rejected": 0.2275390625,
+      "logps/chosen": -1680.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0289,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -12.3125,
-      "rewards/margins": 9.5,
-      "rewards/rejected": -21.875,
+      "rewards/chosen": -12.4375,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -22.25,
       "step": 2560
     },
     {
       "epoch": 2.6910994764397906,
-      "grad_norm": 0.7097461794240564,
+      "grad_norm": 0.5703671496292979,
       "learning_rate": 1.5981066913760918e-06,
-      "logits/chosen": -1.609375,
-      "logits/rejected": 1.78125,
-      "logps/chosen": -1552.0,
+      "logits/chosen": -2.78125,
+      "logits/rejected": 1.0390625,
+      "logps/chosen": -1592.0,
       "logps/rejected": -2496.0,
-      "loss": 0.0235,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 9.375,
+      "loss": 0.0231,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -12.0625,
+      "rewards/margins": 8.875,
       "rewards/rejected": -21.0,
       "step": 2570
     },
     {
       "epoch": 2.701570680628272,
-      "grad_norm": 0.9132465657900399,
+      "grad_norm": 0.5976008660533613,
       "learning_rate": 1.492670085332884e-06,
-      "logits/chosen": -3.015625,
-      "logits/rejected": 0.625,
-      "logps/chosen": -1680.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0306,
+      "logits/chosen": -3.796875,
+      "logits/rejected": 0.318359375,
+      "logps/chosen": -1744.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.03,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -12.625,
-      "rewards/margins": 9.75,
-      "rewards/rejected": -22.375,
+      "rewards/chosen": -13.25,
+      "rewards/margins": 10.5625,
+      "rewards/rejected": -23.75,
       "step": 2580
     },
     {
       "epoch": 2.712041884816754,
-      "grad_norm": 0.2985849062866447,
+      "grad_norm": 0.4445613795286907,
       "learning_rate": 1.3907243345910787e-06,
-      "logits/chosen": -2.5,
-      "logits/rejected": 1.1484375,
-      "logps/chosen": -1696.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0246,
+      "logits/chosen": -3.640625,
+      "logits/rejected": 0.578125,
+      "logps/chosen": -1712.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0221,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -12.5,
-      "rewards/margins": 9.6875,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -12.625,
+      "rewards/margins": 9.9375,
+      "rewards/rejected": -22.5,
       "step": 2590
     },
     {
       "epoch": 2.7225130890052354,
-      "grad_norm": 0.22115985628486312,
+      "grad_norm": 1.3411216086802848,
       "learning_rate": 1.292284578167538e-06,
-      "logits/chosen": -2.40625,
-      "logits/rejected": 0.734375,
-      "logps/chosen": -1624.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0251,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -12.25,
-      "rewards/margins": 9.75,
-      "rewards/rejected": -22.0,
+      "logits/chosen": -3.28125,
+      "logits/rejected": 0.287109375,
+      "logps/chosen": -1688.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.0259,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -12.875,
+      "rewards/margins": 10.125,
+      "rewards/rejected": -23.0,
       "step": 2600
     },
     {
       "epoch": 2.7225130890052354,
-      "eval_logits/chosen": 1.3203125,
-      "eval_logits/rejected": 2.921875,
-      "eval_logps/chosen": -2032.0,
-      "eval_logps/rejected": -2224.0,
-      "eval_loss": 1.6497812271118164,
+      "eval_logits/chosen": 0.27734375,
+      "eval_logits/rejected": 2.046875,
+      "eval_logps/chosen": -2096.0,
+      "eval_logps/rejected": -2272.0,
+      "eval_loss": 1.6399062871932983,
       "eval_rewards/accuracies": 0.603515625,
-      "eval_rewards/chosen": -17.125,
-      "eval_rewards/margins": 2.125,
-      "eval_rewards/rejected": -19.25,
-      "eval_runtime": 46.8184,
-      "eval_samples_per_second": 42.718,
-      "eval_steps_per_second": 0.683,
+      "eval_rewards/chosen": -17.75,
+      "eval_rewards/margins": 2.046875,
+      "eval_rewards/rejected": -19.75,
+      "eval_runtime": 46.6455,
+      "eval_samples_per_second": 42.877,
+      "eval_steps_per_second": 0.686,
       "step": 2600
     },
     {
       "epoch": 2.7329842931937174,
-      "grad_norm": 1.9377379495194558,
+      "grad_norm": 0.31279988539220377,
       "learning_rate": 1.1973654344364925e-06,
-      "logits/chosen": -2.78125,
-      "logits/rejected": 0.5390625,
-      "logps/chosen": -1680.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0226,
+      "logits/chosen": -3.640625,
+      "logits/rejected": -0.076171875,
+      "logps/chosen": -1712.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0197,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -12.375,
-      "rewards/margins": 9.75,
-      "rewards/rejected": -22.125,
+      "rewards/chosen": -12.6875,
+      "rewards/margins": 9.625,
+      "rewards/rejected": -22.25,
       "step": 2610
     },
     {
       "epoch": 2.743455497382199,
-      "grad_norm": 0.282235404376457,
+      "grad_norm": 0.09969985938569956,
       "learning_rate": 1.1059809989586583e-06,
-      "logits/chosen": -2.359375,
-      "logits/rejected": 1.171875,
-      "logps/chosen": -1648.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0276,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -12.1875,
+      "logits/chosen": -3.5625,
+      "logits/rejected": 0.412109375,
+      "logps/chosen": -1680.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.0268,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -12.5,
       "rewards/margins": 10.125,
-      "rewards/rejected": -22.25,
+      "rewards/rejected": -22.625,
       "step": 2620
     },
     {
       "epoch": 2.7539267015706805,
-      "grad_norm": 0.4269936717603736,
+      "grad_norm": 0.41007592379375984,
       "learning_rate": 1.018144842388044e-06,
-      "logits/chosen": -1.953125,
-      "logits/rejected": 1.9609375,
-      "logps/chosen": -1680.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0245,
+      "logits/chosen": -3.15625,
+      "logits/rejected": 1.3828125,
+      "logps/chosen": -1728.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.0253,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -12.3125,
-      "rewards/margins": 10.0,
-      "rewards/rejected": -22.375,
+      "rewards/chosen": -12.75,
+      "rewards/margins": 10.3125,
+      "rewards/rejected": -23.0,
       "step": 2630
     },
     {
       "epoch": 2.7643979057591626,
-      "grad_norm": 0.20650252182549744,
+      "grad_norm": 0.1678536450707497,
       "learning_rate": 9.338700084567109e-07,
-      "logits/chosen": -2.25,
-      "logits/rejected": 0.73828125,
-      "logps/chosen": -1640.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.02,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.25,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -22.0,
+      "logits/chosen": -3.5625,
+      "logits/rejected": -0.01458740234375,
+      "logps/chosen": -1704.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0211,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -12.875,
+      "rewards/margins": 9.6875,
+      "rewards/rejected": -22.625,
       "step": 2640
     },
     {
       "epoch": 2.774869109947644,
-      "grad_norm": 0.7202991201979606,
+      "grad_norm": 0.6500866658902382,
       "learning_rate": 8.531690120377606e-07,
-      "logits/chosen": -2.5625,
-      "logits/rejected": 0.69140625,
+      "logits/chosen": -3.671875,
+      "logits/rejected": 0.0302734375,
       "logps/chosen": -1888.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.025,
+      "logps/rejected": -2784.0,
+      "loss": 0.0237,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -14.125,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -23.125,
+      "rewards/chosen": -14.0625,
+      "rewards/margins": 9.25,
+      "rewards/rejected": -23.375,
       "step": 2650
     },
     {
       "epoch": 2.7853403141361257,
-      "grad_norm": 0.5028047658223058,
+      "grad_norm": 0.478333809693891,
       "learning_rate": 7.760538372868636e-07,
-      "logits/chosen": -2.671875,
-      "logits/rejected": 0.97265625,
-      "logps/chosen": -1688.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0312,
+      "logits/chosen": -3.640625,
+      "logits/rejected": 0.55859375,
+      "logps/chosen": -1752.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.0299,
       "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -12.75,
-      "rewards/margins": 9.375,
-      "rewards/rejected": -22.125,
+      "rewards/chosen": -13.375,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -22.875,
       "step": 2660
     },
     {
       "epoch": 2.7958115183246073,
-      "grad_norm": 0.5962686525740096,
+      "grad_norm": 0.2898715307169193,
       "learning_rate": 7.025359358626166e-07,
-      "logits/chosen": -2.578125,
-      "logits/rejected": 1.1171875,
-      "logps/chosen": -1768.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.0224,
+      "logits/chosen": -3.78125,
+      "logits/rejected": 0.416015625,
+      "logps/chosen": -1776.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.023,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -13.1875,
-      "rewards/margins": 9.875,
-      "rewards/rejected": -23.0,
+      "rewards/chosen": -13.25,
+      "rewards/margins": 10.125,
+      "rewards/rejected": -23.375,
       "step": 2670
     },
     {
       "epoch": 2.806282722513089,
-      "grad_norm": 0.7183064592910384,
+      "grad_norm": 0.4940118516704974,
       "learning_rate": 6.326262252259463e-07,
-      "logits/chosen": -1.921875,
-      "logits/rejected": 1.2109375,
-      "logps/chosen": -1712.0,
+      "logits/chosen": -2.78125,
+      "logits/rejected": 0.6875,
+      "logps/chosen": -1736.0,
       "logps/rejected": -2624.0,
-      "loss": 0.0292,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -12.75,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -22.0,
+      "loss": 0.0272,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -13.0,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -22.125,
       "step": 2680
     },
     {
       "epoch": 2.816753926701571,
-      "grad_norm": 0.18857714938049122,
+      "grad_norm": 0.26721281618386505,
       "learning_rate": 5.66335087018871e-07,
-      "logits/chosen": -2.8125,
-      "logits/rejected": 0.5859375,
-      "logps/chosen": -1696.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0169,
+      "logits/chosen": -3.625,
+      "logits/rejected": 0.2001953125,
+      "logps/chosen": -1736.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.016,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.5625,
-      "rewards/margins": 9.9375,
-      "rewards/rejected": -22.5,
+      "rewards/chosen": -12.9375,
+      "rewards/margins": 9.6875,
+      "rewards/rejected": -22.625,
       "step": 2690
     },
     {
       "epoch": 2.8272251308900525,
-      "grad_norm": 0.6784887946565302,
+      "grad_norm": 0.24648438704540113,
       "learning_rate": 5.036723655228226e-07,
-      "logits/chosen": -2.171875,
-      "logits/rejected": 1.109375,
-      "logps/chosen": -1672.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0249,
+      "logits/chosen": -3.109375,
+      "logits/rejected": 0.60546875,
+      "logps/chosen": -1720.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.0279,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -12.8125,
-      "rewards/margins": 9.0625,
-      "rewards/rejected": -21.875,
+      "rewards/chosen": -13.25,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -22.75,
       "step": 2700
     },
     {
       "epoch": 2.8272251308900525,
-      "eval_logits/chosen": 1.3515625,
-      "eval_logits/rejected": 2.953125,
-      "eval_logps/chosen": -2040.0,
-      "eval_logps/rejected": -2224.0,
-      "eval_loss": 1.6540937423706055,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -17.25,
-      "eval_rewards/margins": 2.078125,
-      "eval_rewards/rejected": -19.25,
-      "eval_runtime": 46.8022,
-      "eval_samples_per_second": 42.733,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": 0.1533203125,
+      "eval_logits/rejected": 1.9609375,
+      "eval_logps/chosen": -2064.0,
+      "eval_logps/rejected": -2240.0,
+      "eval_loss": 1.6252031326293945,
+      "eval_rewards/accuracies": 0.607421875,
+      "eval_rewards/chosen": -17.5,
+      "eval_rewards/margins": 2.03125,
+      "eval_rewards/rejected": -19.5,
+      "eval_runtime": 46.645,
+      "eval_samples_per_second": 42.877,
+      "eval_steps_per_second": 0.686,
       "step": 2700
     },
     {
       "epoch": 2.837696335078534,
-      "grad_norm": 0.22002248974961663,
+      "grad_norm": 1.070044563904558,
       "learning_rate": 4.446473661967432e-07,
-      "logits/chosen": -2.46875,
-      "logits/rejected": 0.8984375,
-      "logps/chosen": -1696.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.023,
+      "logits/chosen": -4.0,
+      "logits/rejected": 0.1259765625,
+      "logps/chosen": -1736.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0241,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -12.75,
-      "rewards/margins": 9.0625,
-      "rewards/rejected": -21.875,
+      "rewards/chosen": -13.125,
+      "rewards/margins": 9.4375,
+      "rewards/rejected": -22.5,
       "step": 2710
     },
     {
       "epoch": 2.8481675392670156,
-      "grad_norm": 0.4133536359685962,
+      "grad_norm": 0.5571615107307087,
       "learning_rate": 3.892688542952411e-07,
-      "logits/chosen": -1.9140625,
-      "logits/rejected": 1.578125,
-      "logps/chosen": -1656.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0262,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -12.5,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -21.5,
+      "logits/chosen": -3.3125,
+      "logits/rejected": 0.84765625,
+      "logps/chosen": -1672.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.0293,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -12.625,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -21.75,
       "step": 2720
     },
     {
       "epoch": 2.858638743455497,
-      "grad_norm": 3.338381741744112,
+      "grad_norm": 0.14640094889226737,
       "learning_rate": 3.3754505356694e-07,
-      "logits/chosen": -3.078125,
-      "logits/rejected": 0.71484375,
-      "logps/chosen": -1760.0,
+      "logits/chosen": -4.375,
+      "logits/rejected": 0.035888671875,
+      "logps/chosen": -1776.0,
       "logps/rejected": -2688.0,
-      "loss": 0.0301,
+      "loss": 0.0238,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -13.25,
-      "rewards/margins": 9.5,
-      "rewards/rejected": -22.75,
+      "rewards/chosen": -13.375,
+      "rewards/margins": 9.25,
+      "rewards/rejected": -22.625,
       "step": 2730
     },
     {
       "epoch": 2.869109947643979,
-      "grad_norm": 0.13786835359334412,
+      "grad_norm": 0.12564531002245383,
       "learning_rate": 2.8948364503322276e-07,
-      "logits/chosen": -2.296875,
-      "logits/rejected": 0.859375,
-      "logps/chosen": -1624.0,
+      "logits/chosen": -3.578125,
+      "logits/rejected": 0.06298828125,
+      "logps/chosen": -1648.0,
       "logps/rejected": -2608.0,
-      "loss": 0.0248,
+      "loss": 0.0255,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.25,
-      "rewards/margins": 9.75,
-      "rewards/rejected": -22.0,
+      "rewards/chosen": -12.5625,
+      "rewards/margins": 9.5625,
+      "rewards/rejected": -22.125,
       "step": 2740
     },
     {
       "epoch": 2.8795811518324608,
-      "grad_norm": 1.0297915167140144,
+      "grad_norm": 0.7011203533821238,
       "learning_rate": 2.4509176584762706e-07,
-      "logits/chosen": -2.296875,
-      "logits/rejected": 1.234375,
-      "logps/chosen": -1608.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0237,
+      "logits/chosen": -3.578125,
+      "logits/rejected": 0.74609375,
+      "logps/chosen": -1656.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0233,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.9375,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -21.75,
+      "rewards/chosen": -12.4375,
+      "rewards/margins": 9.9375,
+      "rewards/rejected": -22.375,
       "step": 2750
     },
     {
       "epoch": 2.8900523560209423,
-      "grad_norm": 0.8359479493170136,
+      "grad_norm": 1.2042012720928126,
       "learning_rate": 2.0437600823595694e-07,
-      "logits/chosen": -3.03125,
-      "logits/rejected": 0.82421875,
-      "logps/chosen": -1720.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0323,
+      "logits/chosen": -4.0,
+      "logits/rejected": 0.263671875,
+      "logps/chosen": -1752.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.0325,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -12.6875,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -21.625,
+      "rewards/chosen": -13.0,
+      "rewards/margins": 9.5625,
+      "rewards/rejected": -22.625,
       "step": 2760
     },
     {
       "epoch": 2.900523560209424,
-      "grad_norm": 0.3111472814223367,
+      "grad_norm": 0.4096637893047813,
       "learning_rate": 1.6734241851733e-07,
-      "logits/chosen": -2.640625,
-      "logits/rejected": 1.1328125,
-      "logps/chosen": -1760.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0253,
+      "logits/chosen": -3.234375,
+      "logits/rejected": 0.66015625,
+      "logps/chosen": -1840.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0231,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -13.0,
+      "rewards/chosen": -13.875,
       "rewards/margins": 9.75,
-      "rewards/rejected": -22.75,
+      "rewards/rejected": -23.625,
       "step": 2770
     },
     {
       "epoch": 2.9109947643979055,
-      "grad_norm": 0.46000348007025493,
+      "grad_norm": 0.4031818005707523,
       "learning_rate": 1.339964962062984e-07,
-      "logits/chosen": -2.703125,
-      "logits/rejected": 0.51171875,
-      "logps/chosen": -1592.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.023,
+      "logits/chosen": -3.5625,
+      "logits/rejected": 0.171875,
+      "logps/chosen": -1640.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0221,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -12.0625,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -21.875,
+      "rewards/chosen": -12.5,
+      "rewards/margins": 9.875,
+      "rewards/rejected": -22.375,
       "step": 2780
     },
     {
       "epoch": 2.9214659685863875,
-      "grad_norm": 0.5161493819236583,
+      "grad_norm": 0.300703403173769,
       "learning_rate": 1.0434319319617136e-07,
-      "logits/chosen": -2.0625,
-      "logits/rejected": 1.25,
-      "logps/chosen": -1640.0,
-      "logps/rejected": -2544.0,
-      "loss": 0.0294,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -12.3125,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -21.5,
+      "logits/chosen": -3.328125,
+      "logits/rejected": 0.59375,
+      "logps/chosen": -1664.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0289,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -12.5,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -22.375,
       "step": 2790
     },
     {
       "epoch": 2.931937172774869,
-      "grad_norm": 0.34315545583861384,
+      "grad_norm": 0.21235456999843158,
       "learning_rate": 7.838691302365363e-08,
-      "logits/chosen": -2.875,
-      "logits/rejected": 0.7734375,
-      "logps/chosen": -1640.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0222,
+      "logits/chosen": -3.6875,
+      "logits/rejected": 0.55078125,
+      "logps/chosen": -1696.0,
+      "logps/rejected": -2768.0,
+      "loss": 0.0219,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -12.0625,
-      "rewards/margins": 10.0625,
-      "rewards/rejected": -22.125,
+      "rewards/chosen": -12.6875,
+      "rewards/margins": 10.375,
+      "rewards/rejected": -23.125,
       "step": 2800
     },
     {
       "epoch": 2.931937172774869,
-      "eval_logits/chosen": 1.3515625,
-      "eval_logits/rejected": 2.953125,
-      "eval_logps/chosen": -2048.0,
-      "eval_logps/rejected": -2224.0,
-      "eval_loss": 1.6638906002044678,
-      "eval_rewards/accuracies": 0.603515625,
-      "eval_rewards/chosen": -17.25,
-      "eval_rewards/margins": 2.125,
-      "eval_rewards/rejected": -19.375,
-      "eval_runtime": 46.8184,
-      "eval_samples_per_second": 42.718,
-      "eval_steps_per_second": 0.683,
+      "eval_logits/chosen": 0.2392578125,
+      "eval_logits/rejected": 2.03125,
+      "eval_logps/chosen": -2096.0,
+      "eval_logps/rejected": -2272.0,
+      "eval_loss": 1.646484375,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -17.75,
+      "eval_rewards/margins": 2.046875,
+      "eval_rewards/rejected": -19.75,
+      "eval_runtime": 46.6472,
+      "eval_samples_per_second": 42.875,
+      "eval_steps_per_second": 0.686,
       "step": 2800
     },
     {
       "epoch": 2.9424083769633507,
-      "grad_norm": 0.10915416871252168,
+      "grad_norm": 0.09069786021856656,
       "learning_rate": 5.613151021490459e-08,
-      "logits/chosen": -1.1484375,
-      "logits/rejected": 1.3046875,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2512.0,
-      "loss": 0.0268,
+      "logits/chosen": -2.53125,
+      "logits/rejected": 0.75,
+      "logps/chosen": -1616.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0255,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -12.5,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -21.375,
+      "rewards/chosen": -12.6875,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -21.75,
       "step": 2810
     },
     {
       "epoch": 2.9528795811518327,
-      "grad_norm": 0.18168106690669275,
+      "grad_norm": 1.5446549315860092,
       "learning_rate": 3.7580289713159965e-08,
-      "logits/chosen": -2.4375,
-      "logits/rejected": 1.1796875,
-      "logps/chosen": -1728.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.0282,
+      "logits/chosen": -3.734375,
+      "logits/rejected": 0.412109375,
+      "logps/chosen": -1744.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0256,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -12.8125,
-      "rewards/margins": 10.125,
-      "rewards/rejected": -22.875,
+      "rewards/chosen": -13.0,
+      "rewards/margins": 10.25,
+      "rewards/rejected": -23.25,
       "step": 2820
     },
     {
       "epoch": 2.9633507853403143,
-      "grad_norm": 0.5266962947543569,
+      "grad_norm": 0.28584523508727483,
       "learning_rate": 2.2736006387927277e-08,
-      "logits/chosen": -2.5,
-      "logits/rejected": 1.0859375,
-      "logps/chosen": -1696.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0219,
+      "logits/chosen": -3.234375,
+      "logits/rejected": 0.796875,
+      "logps/chosen": -1736.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0227,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -12.625,
-      "rewards/margins": 9.9375,
-      "rewards/rejected": -22.625,
+      "rewards/chosen": -13.0625,
+      "rewards/margins": 10.375,
+      "rewards/rejected": -23.375,
       "step": 2830
     },
     {
       "epoch": 2.973821989528796,
-      "grad_norm": 0.2588338578800026,
+      "grad_norm": 0.23716549429582878,
       "learning_rate": 1.1600864625893604e-08,
-      "logits/chosen": -2.515625,
-      "logits/rejected": 0.92578125,
-      "logps/chosen": -1736.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0242,
+      "logits/chosen": -3.59375,
+      "logits/rejected": 0.46875,
+      "logps/chosen": -1760.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0267,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -13.0625,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -22.875,
+      "rewards/chosen": -13.3125,
+      "rewards/margins": 10.3125,
+      "rewards/rejected": -23.625,
       "step": 2840
     },
     {
       "epoch": 2.9842931937172774,
-      "grad_norm": 1.1340319138875847,
+      "grad_norm": 0.33128225605516326,
       "learning_rate": 4.176518003573549e-09,
-      "logits/chosen": -2.875,
-      "logits/rejected": 1.078125,
-      "logps/chosen": -1720.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0281,
+      "logits/chosen": -3.9375,
+      "logits/rejected": 0.38671875,
+      "logps/chosen": -1752.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.0333,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -13.0625,
-      "rewards/margins": 9.75,
-      "rewards/rejected": -22.75,
+      "rewards/chosen": -13.4375,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -23.25,
       "step": 2850
     },
     {
       "epoch": 2.994764397905759,
-      "grad_norm": 0.18477884046605084,
+      "grad_norm": 0.31518849691570666,
       "learning_rate": 4.640690417528903e-10,
-      "logits/chosen": -2.453125,
-      "logits/rejected": 0.9921875,
-      "logps/chosen": -1680.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0268,
+      "logits/chosen": -3.875,
+      "logits/rejected": 0.36328125,
+      "logps/chosen": -1696.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.0278,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -12.875,
-      "rewards/margins": 9.75,
-      "rewards/rejected": -22.625,
+      "rewards/chosen": -13.0625,
+      "rewards/margins": 10.1875,
+      "rewards/rejected": -23.25,
       "step": 2860
     },
     {
       "epoch": 3.0,
       "step": 2865,
       "total_flos": 0.0,
-      "train_loss": 0.2292164348338911,
-      "train_runtime": 12142.2594,
-      "train_samples_per_second": 15.101,
-      "train_steps_per_second": 0.236
+      "train_loss": 0.22992089002752386,
+      "train_runtime": 12100.6201,
+      "train_samples_per_second": 15.153,
+      "train_steps_per_second": 0.237
     }
   ],
   "logging_steps": 10,