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": 4.721636388677752,
+      "grad_norm": 4.721639122004471,
       "learning_rate": 1.7421602787456448e-07,
       "logits/chosen": -12.0,
       "logits/rejected": -9.25,
@@ -25,73 +25,73 @@
     },
     {
       "epoch": 0.010471204188481676,
-      "grad_norm": 4.7912124622583,
+      "grad_norm": 4.783730795492032,
       "learning_rate": 1.7421602787456445e-06,
       "logits/chosen": -11.0,
       "logits/rejected": -8.5,
       "logps/chosen": -404.0,
       "logps/rejected": -1784.0,
-      "loss": 0.6909,
-      "rewards/accuracies": 0.1666666716337204,
-      "rewards/chosen": -0.0003147125244140625,
-      "rewards/margins": 0.000774383544921875,
-      "rewards/rejected": -0.0010986328125,
+      "loss": 0.6905,
+      "rewards/accuracies": 0.2013888955116272,
+      "rewards/chosen": 9.5367431640625e-05,
+      "rewards/margins": -0.00075531005859375,
+      "rewards/rejected": 0.0008544921875,
       "step": 10
     },
     {
       "epoch": 0.020942408376963352,
-      "grad_norm": 4.257119370385409,
+      "grad_norm": 4.256886972315482,
       "learning_rate": 3.484320557491289e-06,
       "logits/chosen": -10.8125,
       "logits/rejected": -8.6875,
-      "logps/chosen": -388.0,
+      "logps/chosen": -386.0,
       "logps/rejected": -1632.0,
-      "loss": 0.6427,
-      "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": -8.20159912109375e-05,
-      "rewards/margins": 0.1123046875,
-      "rewards/rejected": -0.11279296875,
+      "loss": 0.6422,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -0.0002117156982421875,
+      "rewards/margins": 0.11376953125,
+      "rewards/rejected": -0.11376953125,
       "step": 20
     },
     {
       "epoch": 0.031413612565445025,
-      "grad_norm": 1.9875037352174125,
+      "grad_norm": 1.9879354437192425,
       "learning_rate": 5.226480836236934e-06,
       "logits/chosen": -11.5625,
       "logits/rejected": -9.5625,
       "logps/chosen": -416.0,
       "logps/rejected": -1576.0,
-      "loss": 0.3857,
+      "loss": 0.3854,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -0.0986328125,
-      "rewards/margins": 0.796875,
+      "rewards/chosen": -0.09912109375,
+      "rewards/margins": 0.80078125,
       "rewards/rejected": -0.8984375,
       "step": 30
     },
     {
       "epoch": 0.041884816753926704,
-      "grad_norm": 1.3579028290896005,
+      "grad_norm": 1.3556998476691806,
       "learning_rate": 6.968641114982578e-06,
       "logits/chosen": -12.25,
       "logits/rejected": -10.625,
       "logps/chosen": -480.0,
       "logps/rejected": -1904.0,
-      "loss": 0.1615,
+      "loss": 0.1612,
       "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -0.671875,
+      "rewards/chosen": -0.66796875,
       "rewards/margins": 3.859375,
       "rewards/rejected": -4.53125,
       "step": 40
     },
     {
       "epoch": 0.05235602094240838,
-      "grad_norm": 0.6278937338590627,
+      "grad_norm": 0.6312032220700643,
       "learning_rate": 8.710801393728225e-06,
       "logits/chosen": -14.75,
       "logits/rejected": -12.875,
       "logps/chosen": -528.0,
       "logps/rejected": -2544.0,
-      "loss": 0.0646,
+      "loss": 0.0647,
       "rewards/accuracies": 0.96875,
       "rewards/chosen": -1.21875,
       "rewards/margins": 8.75,
@@ -100,4675 +100,4675 @@
     },
     {
       "epoch": 0.06282722513089005,
-      "grad_norm": 0.7680884333794784,
+      "grad_norm": 0.7616948494279886,
       "learning_rate": 1.0452961672473868e-05,
       "logits/chosen": -15.5625,
       "logits/rejected": -13.8125,
       "logps/chosen": -436.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0811,
+      "logps/rejected": -2688.0,
+      "loss": 0.0808,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -1.0546875,
+      "rewards/chosen": -1.0625,
       "rewards/margins": 10.1875,
       "rewards/rejected": -11.25,
       "step": 60
     },
     {
       "epoch": 0.07329842931937172,
-      "grad_norm": 0.5065499432998848,
+      "grad_norm": 0.49976608571115705,
       "learning_rate": 1.2195121951219513e-05,
-      "logits/chosen": -15.25,
-      "logits/rejected": -14.0,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -13.9375,
       "logps/chosen": -556.0,
-      "logps/rejected": -2944.0,
-      "loss": 0.0397,
+      "logps/rejected": -2928.0,
+      "loss": 0.0402,
       "rewards/accuracies": 0.987500011920929,
       "rewards/chosen": -1.3828125,
-      "rewards/margins": 12.125,
-      "rewards/rejected": -13.5,
+      "rewards/margins": 12.0,
+      "rewards/rejected": -13.4375,
       "step": 70
     },
     {
       "epoch": 0.08376963350785341,
-      "grad_norm": 0.5989205693429206,
+      "grad_norm": 0.6215709805393562,
       "learning_rate": 1.3937282229965156e-05,
       "logits/chosen": -14.25,
       "logits/rejected": -13.3125,
-      "logps/chosen": -536.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.0434,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -1.4921875,
-      "rewards/margins": 11.625,
-      "rewards/rejected": -13.0625,
+      "logps/chosen": -532.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.0431,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.4765625,
+      "rewards/margins": 11.4375,
+      "rewards/rejected": -12.9375,
       "step": 80
     },
     {
       "epoch": 0.09424083769633508,
-      "grad_norm": 7.138491484450438,
+      "grad_norm": 4.289613250011983,
       "learning_rate": 1.56794425087108e-05,
       "logits/chosen": -13.125,
-      "logits/rejected": -12.3125,
-      "logps/chosen": -540.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0287,
+      "logits/rejected": -12.375,
+      "logps/chosen": -580.0,
+      "logps/rejected": -3488.0,
+      "loss": 0.0293,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.875,
-      "rewards/margins": 16.0,
-      "rewards/rejected": -17.875,
+      "rewards/chosen": -2.28125,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -19.875,
       "step": 90
     },
     {
       "epoch": 0.10471204188481675,
-      "grad_norm": 0.8254009160671442,
+      "grad_norm": 0.6991052418132925,
       "learning_rate": 1.742160278745645e-05,
-      "logits/chosen": -11.0625,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -640.0,
-      "logps/rejected": -3440.0,
-      "loss": 0.05,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -1.859375,
-      "rewards/margins": 17.125,
-      "rewards/rejected": -19.0,
+      "logits/chosen": -10.625,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -616.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0562,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.6015625,
+      "rewards/margins": 15.0,
+      "rewards/rejected": -16.625,
       "step": 100
     },
     {
       "epoch": 0.10471204188481675,
       "eval_logits/chosen": -9.5625,
       "eval_logits/rejected": -9.3125,
-      "eval_logps/chosen": -524.0,
-      "eval_logps/rejected": -528.0,
-      "eval_loss": 0.8594218492507935,
-      "eval_rewards/accuracies": 0.548828125,
-      "eval_rewards/chosen": -2.046875,
-      "eval_rewards/margins": 0.349609375,
-      "eval_rewards/rejected": -2.390625,
-      "eval_runtime": 48.3634,
-      "eval_samples_per_second": 41.354,
-      "eval_steps_per_second": 0.662,
+      "eval_logps/chosen": -444.0,
+      "eval_logps/rejected": -446.0,
+      "eval_loss": 0.6970937252044678,
+      "eval_rewards/accuracies": 0.576171875,
+      "eval_rewards/chosen": -1.2578125,
+      "eval_rewards/margins": 0.314453125,
+      "eval_rewards/rejected": -1.5703125,
+      "eval_runtime": 60.4666,
+      "eval_samples_per_second": 33.076,
+      "eval_steps_per_second": 0.529,
       "step": 100
     },
     {
       "epoch": 0.11518324607329843,
-      "grad_norm": 1.6208913047456113,
+      "grad_norm": 0.6921214701569128,
       "learning_rate": 1.9163763066202093e-05,
-      "logits/chosen": -11.0625,
-      "logits/rejected": -10.125,
-      "logps/chosen": -516.0,
-      "logps/rejected": -3408.0,
-      "loss": 0.0393,
+      "logits/chosen": -11.0,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -484.0,
+      "logps/rejected": -3392.0,
+      "loss": 0.0303,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -1.296875,
-      "rewards/margins": 15.1875,
-      "rewards/rejected": -16.5,
+      "rewards/chosen": -0.953125,
+      "rewards/margins": 15.375,
+      "rewards/rejected": -16.375,
       "step": 110
     },
     {
       "epoch": 0.1256544502617801,
-      "grad_norm": 0.5008886836049369,
+      "grad_norm": 0.5385006510657429,
       "learning_rate": 2.0905923344947736e-05,
-      "logits/chosen": -13.125,
-      "logits/rejected": -12.0625,
-      "logps/chosen": -446.0,
-      "logps/rejected": -2944.0,
-      "loss": 0.0251,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -0.96875,
-      "rewards/margins": 13.5,
-      "rewards/rejected": -14.4375,
+      "logits/chosen": -13.875,
+      "logits/rejected": -13.0,
+      "logps/chosen": -512.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0243,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -1.6328125,
+      "rewards/margins": 17.125,
+      "rewards/rejected": -18.75,
       "step": 120
     },
     {
       "epoch": 0.13612565445026178,
-      "grad_norm": 0.5663279938832929,
+      "grad_norm": 0.5460928985623315,
       "learning_rate": 2.264808362369338e-05,
-      "logits/chosen": -13.125,
-      "logits/rejected": -12.5,
-      "logps/chosen": -624.0,
-      "logps/rejected": -4096.0,
-      "loss": 0.0671,
+      "logits/chosen": -15.0625,
+      "logits/rejected": -14.0625,
+      "logps/chosen": -564.0,
+      "logps/rejected": -3520.0,
+      "loss": 0.0594,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.328125,
-      "rewards/margins": 23.375,
-      "rewards/rejected": -25.75,
+      "rewards/chosen": -1.7421875,
+      "rewards/margins": 18.125,
+      "rewards/rejected": -19.875,
       "step": 130
     },
     {
       "epoch": 0.14659685863874344,
-      "grad_norm": 0.2471992305103558,
+      "grad_norm": 0.2904576131081844,
       "learning_rate": 2.4390243902439026e-05,
-      "logits/chosen": -13.875,
-      "logits/rejected": -12.6875,
-      "logps/chosen": -564.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0521,
+      "logits/chosen": -16.75,
+      "logits/rejected": -15.375,
+      "logps/chosen": -600.0,
+      "logps/rejected": -3440.0,
+      "loss": 0.0568,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -1.296875,
-      "rewards/margins": 10.75,
-      "rewards/rejected": -12.0625,
+      "rewards/chosen": -1.640625,
+      "rewards/margins": 17.875,
+      "rewards/rejected": -19.625,
       "step": 140
     },
     {
       "epoch": 0.15706806282722513,
-      "grad_norm": 3.2454461879887218,
+      "grad_norm": 0.5242219439068009,
       "learning_rate": 2.6132404181184672e-05,
-      "logits/chosen": -13.8125,
-      "logits/rejected": -12.6875,
-      "logps/chosen": -556.0,
-      "logps/rejected": -3152.0,
-      "loss": 0.0286,
+      "logits/chosen": -16.0,
+      "logits/rejected": -14.25,
+      "logps/chosen": -684.0,
+      "logps/rejected": -4256.0,
+      "loss": 0.0379,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.453125,
-      "rewards/margins": 13.75,
-      "rewards/rejected": -15.25,
+      "rewards/chosen": -2.75,
+      "rewards/margins": 23.75,
+      "rewards/rejected": -26.5,
       "step": 150
     },
     {
       "epoch": 0.16753926701570682,
-      "grad_norm": 3.3144620620064904,
+      "grad_norm": 2.8762586405003283,
       "learning_rate": 2.7874564459930312e-05,
-      "logits/chosen": -13.9375,
-      "logits/rejected": -12.5,
-      "logps/chosen": -500.0,
-      "logps/rejected": -3056.0,
-      "loss": 0.0696,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.1796875,
-      "rewards/margins": 13.125,
-      "rewards/rejected": -14.3125,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -12.375,
+      "logps/chosen": -490.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.0873,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -1.0859375,
+      "rewards/margins": 10.3125,
+      "rewards/rejected": -11.375,
       "step": 160
     },
     {
       "epoch": 0.17801047120418848,
-      "grad_norm": 0.5783916742126871,
+      "grad_norm": 0.6828534029794374,
       "learning_rate": 2.9616724738675962e-05,
-      "logits/chosen": -13.0,
-      "logits/rejected": -12.3125,
-      "logps/chosen": -524.0,
-      "logps/rejected": -3104.0,
-      "loss": 0.0233,
+      "logits/chosen": -14.0625,
+      "logits/rejected": -13.125,
+      "logps/chosen": -540.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.0238,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -1.1328125,
-      "rewards/margins": 13.875,
-      "rewards/rejected": -15.0,
+      "rewards/chosen": -1.2734375,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -13.5,
       "step": 170
     },
     {
       "epoch": 0.18848167539267016,
-      "grad_norm": 1.4466407764510198,
+      "grad_norm": 0.9030686349341729,
       "learning_rate": 3.13588850174216e-05,
-      "logits/chosen": -12.5,
-      "logits/rejected": -11.875,
-      "logps/chosen": -540.0,
-      "logps/rejected": -3184.0,
-      "loss": 0.0538,
+      "logits/chosen": -14.375,
+      "logits/rejected": -13.625,
+      "logps/chosen": -628.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0571,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -1.2421875,
-      "rewards/margins": 14.25,
-      "rewards/rejected": -15.5,
+      "rewards/chosen": -2.09375,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -21.0,
       "step": 180
     },
     {
       "epoch": 0.19895287958115182,
-      "grad_norm": 0.5267904603473407,
+      "grad_norm": 0.4122349277479674,
       "learning_rate": 3.310104529616725e-05,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -11.9375,
-      "logps/chosen": -512.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.053,
+      "logits/chosen": -13.875,
+      "logits/rejected": -12.5,
+      "logps/chosen": -540.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.0447,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -1.3828125,
-      "rewards/margins": 13.8125,
-      "rewards/rejected": -15.1875,
+      "rewards/chosen": -1.671875,
+      "rewards/margins": 14.25,
+      "rewards/rejected": -15.9375,
       "step": 190
     },
     {
       "epoch": 0.2094240837696335,
-      "grad_norm": 0.7105167936150414,
+      "grad_norm": 0.5232184397543935,
       "learning_rate": 3.48432055749129e-05,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -12.25,
-      "logps/chosen": -648.0,
-      "logps/rejected": -3536.0,
-      "loss": 0.0695,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -12.8125,
+      "logps/chosen": -604.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.0394,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.1875,
-      "rewards/margins": 17.25,
-      "rewards/rejected": -19.5,
+      "rewards/chosen": -1.71875,
+      "rewards/margins": 11.9375,
+      "rewards/rejected": -13.625,
       "step": 200
     },
     {
       "epoch": 0.2094240837696335,
-      "eval_logits/chosen": -13.3125,
-      "eval_logits/rejected": -12.9375,
-      "eval_logps/chosen": -438.0,
-      "eval_logps/rejected": -428.0,
-      "eval_loss": 0.7368593811988831,
-      "eval_rewards/accuracies": 0.48828125,
-      "eval_rewards/chosen": -1.203125,
-      "eval_rewards/margins": 0.1953125,
-      "eval_rewards/rejected": -1.3984375,
-      "eval_runtime": 46.645,
-      "eval_samples_per_second": 42.877,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -12.75,
+      "eval_logits/rejected": -12.3125,
+      "eval_logps/chosen": -404.0,
+      "eval_logps/rejected": -390.0,
+      "eval_loss": 0.7478749752044678,
+      "eval_rewards/accuracies": 0.51953125,
+      "eval_rewards/chosen": -0.8515625,
+      "eval_rewards/margins": 0.1572265625,
+      "eval_rewards/rejected": -1.0078125,
+      "eval_runtime": 58.2524,
+      "eval_samples_per_second": 34.333,
+      "eval_steps_per_second": 0.549,
       "step": 200
     },
     {
       "epoch": 0.2198952879581152,
-      "grad_norm": 0.7665795490640238,
+      "grad_norm": 0.6713539915249732,
       "learning_rate": 3.6585365853658535e-05,
-      "logits/chosen": -13.0625,
-      "logits/rejected": -11.5,
-      "logps/chosen": -462.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.0294,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.0546875,
-      "rewards/margins": 10.6875,
-      "rewards/rejected": -11.75,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -11.3125,
+      "logps/chosen": -472.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.03,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -1.15625,
+      "rewards/margins": 10.5,
+      "rewards/rejected": -11.625,
       "step": 210
     },
     {
       "epoch": 0.23036649214659685,
-      "grad_norm": 3.2472849133898176,
+      "grad_norm": 2.11344232375256,
       "learning_rate": 3.8327526132404185e-05,
-      "logits/chosen": -13.6875,
-      "logits/rejected": -11.6875,
-      "logps/chosen": -536.0,
-      "logps/rejected": -2800.0,
-      "loss": 0.0573,
+      "logits/chosen": -14.625,
+      "logits/rejected": -12.375,
+      "logps/chosen": -556.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0527,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -1.1796875,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -13.5,
+      "rewards/chosen": -1.390625,
+      "rewards/margins": 10.875,
+      "rewards/rejected": -12.25,
       "step": 220
     },
     {
       "epoch": 0.24083769633507854,
-      "grad_norm": 0.4829358332641341,
+      "grad_norm": 0.2994869398036509,
       "learning_rate": 4.006968641114983e-05,
-      "logits/chosen": -13.25,
-      "logits/rejected": -11.5,
-      "logps/chosen": -544.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.0251,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -1.7578125,
-      "rewards/margins": 15.4375,
-      "rewards/rejected": -17.25,
+      "logits/chosen": -15.625,
+      "logits/rejected": -12.9375,
+      "logps/chosen": -568.0,
+      "logps/rejected": -2976.0,
+      "loss": 0.0286,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -1.9921875,
+      "rewards/margins": 12.75,
+      "rewards/rejected": -14.75,
       "step": 230
     },
     {
       "epoch": 0.2513089005235602,
-      "grad_norm": 0.8419965359719804,
+      "grad_norm": 0.698709730544559,
       "learning_rate": 4.181184668989547e-05,
-      "logits/chosen": -12.875,
-      "logits/rejected": -11.5,
-      "logps/chosen": -498.0,
-      "logps/rejected": -3440.0,
-      "loss": 0.0875,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -1.359375,
-      "rewards/margins": 18.25,
-      "rewards/rejected": -19.625,
+      "logits/chosen": -14.125,
+      "logits/rejected": -11.625,
+      "logps/chosen": -612.0,
+      "logps/rejected": -3360.0,
+      "loss": 0.0436,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -2.5,
+      "rewards/margins": 16.25,
+      "rewards/rejected": -18.75,
       "step": 240
     },
     {
       "epoch": 0.2617801047120419,
-      "grad_norm": 0.3895064561891479,
+      "grad_norm": 0.8020044433623945,
       "learning_rate": 4.3554006968641115e-05,
-      "logits/chosen": -13.0625,
-      "logits/rejected": -11.9375,
-      "logps/chosen": -584.0,
-      "logps/rejected": -3488.0,
-      "loss": 0.0361,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -628.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.026,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -1.609375,
-      "rewards/margins": 18.75,
-      "rewards/rejected": -20.375,
+      "rewards/chosen": -2.078125,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -18.375,
       "step": 250
     },
     {
       "epoch": 0.27225130890052357,
-      "grad_norm": 0.49754806212910385,
+      "grad_norm": 0.4893002600991188,
       "learning_rate": 4.529616724738676e-05,
-      "logits/chosen": -11.9375,
-      "logits/rejected": -11.0,
-      "logps/chosen": -560.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.0299,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -1.8984375,
-      "rewards/margins": 11.625,
-      "rewards/rejected": -13.5,
+      "logits/chosen": -10.6875,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -544.0,
+      "logps/rejected": -2992.0,
+      "loss": 0.026,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -1.7578125,
+      "rewards/margins": 13.4375,
+      "rewards/rejected": -15.1875,
       "step": 260
     },
     {
       "epoch": 0.28272251308900526,
-      "grad_norm": 0.37978493222236703,
+      "grad_norm": 0.5797178090531139,
       "learning_rate": 4.703832752613241e-05,
-      "logits/chosen": -12.8125,
-      "logits/rejected": -11.8125,
-      "logps/chosen": -640.0,
-      "logps/rejected": -3456.0,
-      "loss": 0.0878,
+      "logits/chosen": -11.8125,
+      "logits/rejected": -10.0,
+      "logps/chosen": -506.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.0509,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.765625,
-      "rewards/margins": 16.75,
-      "rewards/rejected": -19.625,
+      "rewards/chosen": -1.4296875,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -14.0625,
       "step": 270
     },
     {
       "epoch": 0.2931937172774869,
-      "grad_norm": 0.3621231821290884,
+      "grad_norm": 0.4707786527508873,
       "learning_rate": 4.878048780487805e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.375,
-      "logps/chosen": -556.0,
-      "logps/rejected": -2928.0,
-      "loss": 0.0401,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.109375,
-      "rewards/margins": 12.4375,
-      "rewards/rejected": -14.5,
+      "logits/chosen": -13.875,
+      "logits/rejected": -11.5625,
+      "logps/chosen": -520.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.0359,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -1.7265625,
+      "rewards/margins": 11.0625,
+      "rewards/rejected": -12.75,
       "step": 280
     },
     {
       "epoch": 0.3036649214659686,
-      "grad_norm": 0.4111282833499949,
+      "grad_norm": 0.1671536030644796,
       "learning_rate": 4.999983293481418e-05,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -10.5,
-      "logps/chosen": -720.0,
-      "logps/rejected": -3648.0,
-      "loss": 0.0194,
+      "logits/chosen": -14.25,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -632.0,
+      "logps/rejected": -3104.0,
+      "loss": 0.0196,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 18.625,
-      "rewards/rejected": -22.125,
+      "rewards/chosen": -2.5625,
+      "rewards/margins": 14.125,
+      "rewards/rejected": -16.625,
       "step": 290
     },
     {
       "epoch": 0.31413612565445026,
-      "grad_norm": 0.48546810256802125,
+      "grad_norm": 1.2072006061135843,
       "learning_rate": 4.9996862949181595e-05,
-      "logits/chosen": -10.5625,
-      "logits/rejected": -9.875,
-      "logps/chosen": -724.0,
-      "logps/rejected": -3568.0,
-      "loss": 0.0402,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 17.75,
-      "rewards/rejected": -21.125,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -12.25,
+      "logps/chosen": -592.0,
+      "logps/rejected": -3104.0,
+      "loss": 0.0487,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -2.046875,
+      "rewards/margins": 14.375,
+      "rewards/rejected": -16.375,
       "step": 300
     },
     {
       "epoch": 0.31413612565445026,
-      "eval_logits/chosen": -9.75,
-      "eval_logits/rejected": -9.125,
-      "eval_logps/chosen": -776.0,
-      "eval_logps/rejected": -792.0,
-      "eval_loss": 1.4284374713897705,
-      "eval_rewards/accuracies": 0.51953125,
-      "eval_rewards/chosen": -4.59375,
-      "eval_rewards/margins": 0.4296875,
-      "eval_rewards/rejected": -5.0,
-      "eval_runtime": 46.6493,
-      "eval_samples_per_second": 42.873,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -13.6875,
+      "eval_logits/rejected": -13.4375,
+      "eval_logps/chosen": -516.0,
+      "eval_logps/rejected": -520.0,
+      "eval_loss": 0.9194687604904175,
+      "eval_rewards/accuracies": 0.517578125,
+      "eval_rewards/chosen": -1.9921875,
+      "eval_rewards/margins": 0.3203125,
+      "eval_rewards/rejected": -2.3125,
+      "eval_runtime": 58.2421,
+      "eval_samples_per_second": 34.339,
+      "eval_steps_per_second": 0.549,
       "step": 300
     },
     {
       "epoch": 0.32460732984293195,
-      "grad_norm": 0.3578998742883041,
+      "grad_norm": 0.25631935493523433,
       "learning_rate": 4.9990180911526896e-05,
-      "logits/chosen": -9.875,
-      "logits/rejected": -8.5,
-      "logps/chosen": -700.0,
-      "logps/rejected": -3744.0,
-      "loss": 0.029,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -10.125,
+      "logps/chosen": -512.0,
+      "logps/rejected": -3072.0,
+      "loss": 0.02,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.46875,
-      "rewards/margins": 16.75,
-      "rewards/rejected": -20.25,
+      "rewards/chosen": -1.59375,
+      "rewards/margins": 11.875,
+      "rewards/rejected": -13.5,
       "step": 310
     },
     {
       "epoch": 0.33507853403141363,
-      "grad_norm": 0.4278547531900684,
+      "grad_norm": 0.4566448070635291,
       "learning_rate": 4.997978781413745e-05,
-      "logits/chosen": -8.9375,
-      "logits/rejected": -7.71875,
-      "logps/chosen": -752.0,
-      "logps/rejected": -3184.0,
-      "loss": 0.0597,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -3.234375,
-      "rewards/margins": 14.0625,
-      "rewards/rejected": -17.25,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3472.0,
+      "loss": 0.075,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.765625,
+      "rewards/margins": 17.375,
+      "rewards/rejected": -20.25,
       "step": 320
     },
     {
       "epoch": 0.34554973821989526,
-      "grad_norm": 2.198346399823717,
+      "grad_norm": 8.541477273721926,
       "learning_rate": 4.996568520039565e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -8.375,
-      "logps/chosen": -652.0,
-      "logps/rejected": -2912.0,
-      "loss": 0.0745,
+      "logits/chosen": -11.875,
+      "logits/rejected": -10.625,
+      "logps/chosen": -644.0,
+      "logps/rejected": -2992.0,
+      "loss": 0.063,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.375,
-      "rewards/margins": 12.5,
-      "rewards/rejected": -14.875,
+      "rewards/chosen": -2.3125,
+      "rewards/margins": 13.5,
+      "rewards/rejected": -15.8125,
       "step": 330
     },
     {
       "epoch": 0.35602094240837695,
-      "grad_norm": 0.25614963523411827,
+      "grad_norm": 0.1634236198155656,
       "learning_rate": 4.994787516454973e-05,
-      "logits/chosen": -11.125,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -676.0,
-      "logps/rejected": -3280.0,
-      "loss": 0.023,
+      "logits/chosen": -13.125,
+      "logits/rejected": -10.375,
+      "logps/chosen": -572.0,
+      "logps/rejected": -2976.0,
+      "loss": 0.0172,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.828125,
-      "rewards/margins": 13.5625,
-      "rewards/rejected": -16.375,
+      "rewards/chosen": -1.8203125,
+      "rewards/margins": 11.4375,
+      "rewards/rejected": -13.25,
       "step": 340
     },
     {
       "epoch": 0.36649214659685864,
-      "grad_norm": 0.7492165789300889,
+      "grad_norm": 1.872265339611973,
       "learning_rate": 4.992636035140273e-05,
-      "logits/chosen": -11.375,
-      "logits/rejected": -8.125,
-      "logps/chosen": -556.0,
-      "logps/rejected": -2816.0,
-      "loss": 0.0357,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -2.0625,
-      "rewards/margins": 10.1875,
-      "rewards/rejected": -12.25,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -11.0,
+      "logps/chosen": -560.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0382,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -2.125,
+      "rewards/margins": 15.0625,
+      "rewards/rejected": -17.125,
       "step": 350
     },
     {
       "epoch": 0.3769633507853403,
-      "grad_norm": 0.9137368941008791,
+      "grad_norm": 1.3397949238764175,
       "learning_rate": 4.99011439559198e-05,
-      "logits/chosen": -12.125,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -628.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.5,
+      "logps/chosen": -608.0,
       "logps/rejected": -3248.0,
-      "loss": 0.0342,
+      "loss": 0.0435,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.40625,
-      "rewards/margins": 14.75,
+      "rewards/chosen": -2.21875,
+      "rewards/margins": 14.9375,
       "rewards/rejected": -17.125,
       "step": 360
     },
     {
       "epoch": 0.387434554973822,
-      "grad_norm": 0.34423819856048526,
+      "grad_norm": 0.3560828237569494,
       "learning_rate": 4.987222972275369e-05,
-      "logits/chosen": -13.0,
-      "logits/rejected": -9.6875,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -9.5625,
       "logps/chosen": -636.0,
-      "logps/rejected": -3424.0,
-      "loss": 0.0281,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.46875,
-      "rewards/margins": 16.875,
-      "rewards/rejected": -19.375,
+      "logps/rejected": -3328.0,
+      "loss": 0.0248,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -2.484375,
+      "rewards/margins": 15.8125,
+      "rewards/rejected": -18.25,
       "step": 370
     },
     {
       "epoch": 0.39790575916230364,
-      "grad_norm": 0.12361471414333342,
+      "grad_norm": 0.08880095633365019,
       "learning_rate": 4.983962194568872e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -10.125,
-      "logps/chosen": -672.0,
-      "logps/rejected": -3632.0,
-      "loss": 0.0234,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.71875,
-      "rewards/margins": 18.75,
-      "rewards/rejected": -21.375,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -10.0,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0247,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 19.0,
+      "rewards/rejected": -22.125,
       "step": 380
     },
     {
       "epoch": 0.4083769633507853,
-      "grad_norm": 1.1359733512059913,
+      "grad_norm": 0.3501787780604538,
       "learning_rate": 4.9803325467003084e-05,
-      "logits/chosen": -12.9375,
-      "logits/rejected": -9.625,
-      "logps/chosen": -696.0,
-      "logps/rejected": -3760.0,
-      "loss": 0.0262,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.984375,
-      "rewards/margins": 17.875,
-      "rewards/rejected": -20.75,
+      "logits/chosen": -13.875,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -720.0,
+      "logps/rejected": -3936.0,
+      "loss": 0.0434,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.5,
       "step": 390
     },
     {
       "epoch": 0.418848167539267,
-      "grad_norm": 0.8273270619407733,
+      "grad_norm": 2.4077134956891815,
       "learning_rate": 4.9763345676749816e-05,
-      "logits/chosen": -12.1875,
-      "logits/rejected": -10.0,
-      "logps/chosen": -592.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0373,
+      "logits/chosen": -14.875,
+      "logits/rejected": -10.75,
+      "logps/chosen": -532.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.0454,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.171875,
-      "rewards/margins": 9.9375,
-      "rewards/rejected": -12.125,
+      "rewards/chosen": -1.578125,
+      "rewards/margins": 11.0625,
+      "rewards/rejected": -12.625,
       "step": 400
     },
     {
       "epoch": 0.418848167539267,
-      "eval_logits/chosen": -13.5625,
-      "eval_logits/rejected": -13.375,
-      "eval_logps/chosen": -556.0,
-      "eval_logps/rejected": -540.0,
-      "eval_loss": 0.8731562495231628,
-      "eval_rewards/accuracies": 0.490234375,
-      "eval_rewards/chosen": -2.390625,
-      "eval_rewards/margins": 0.1279296875,
-      "eval_rewards/rejected": -2.515625,
-      "eval_runtime": 46.6392,
-      "eval_samples_per_second": 42.882,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -15.75,
+      "eval_logits/rejected": -15.625,
+      "eval_logps/chosen": -462.0,
+      "eval_logps/rejected": -448.0,
+      "eval_loss": 0.8308749794960022,
+      "eval_rewards/accuracies": 0.49609375,
+      "eval_rewards/chosen": -1.4453125,
+      "eval_rewards/margins": 0.154296875,
+      "eval_rewards/rejected": -1.6015625,
+      "eval_runtime": 58.2749,
+      "eval_samples_per_second": 34.32,
+      "eval_steps_per_second": 0.549,
       "step": 400
     },
     {
       "epoch": 0.4293193717277487,
-      "grad_norm": 0.8442250459098813,
+      "grad_norm": 0.3406945596885454,
       "learning_rate": 4.9719688511956374e-05,
-      "logits/chosen": -14.0,
-      "logits/rejected": -10.75,
-      "logps/chosen": -648.0,
-      "logps/rejected": -3184.0,
-      "loss": 0.0435,
+      "logits/chosen": -15.25,
+      "logits/rejected": -10.625,
+      "logps/chosen": -568.0,
+      "logps/rejected": -2768.0,
+      "loss": 0.0557,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.4375,
-      "rewards/margins": 12.5625,
-      "rewards/rejected": -15.0,
+      "rewards/chosen": -1.6328125,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -10.8125,
       "step": 410
     },
     {
       "epoch": 0.4397905759162304,
-      "grad_norm": 0.46132531204367994,
+      "grad_norm": 1.1720431755211176,
       "learning_rate": 4.967236045574294e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -656.0,
-      "logps/rejected": -3392.0,
-      "loss": 0.0788,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.546875,
-      "rewards/margins": 16.375,
-      "rewards/rejected": -18.875,
+      "logits/chosen": -13.875,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -596.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.0276,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -1.9375,
+      "rewards/margins": 9.625,
+      "rewards/rejected": -11.5625,
       "step": 420
     },
     {
       "epoch": 0.450261780104712,
-      "grad_norm": 0.11309333711192368,
+      "grad_norm": 2.760614816757791,
       "learning_rate": 4.96213685363597e-05,
-      "logits/chosen": -14.5,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -688.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0714,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -3.109375,
-      "rewards/margins": 24.125,
-      "rewards/rejected": -27.25,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -10.5,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0672,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 14.625,
+      "rewards/rejected": -17.625,
       "step": 430
     },
     {
       "epoch": 0.4607329842931937,
-      "grad_norm": 4.2649476255015735,
+      "grad_norm": 1.2650121419821303,
       "learning_rate": 4.956672032614315e-05,
-      "logits/chosen": -14.1875,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -700.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.085,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 19.125,
-      "rewards/rejected": -22.125,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.25,
+      "logps/chosen": -616.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.0396,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.046875,
+      "rewards/margins": 13.125,
+      "rewards/rejected": -15.1875,
       "step": 440
     },
     {
       "epoch": 0.4712041884816754,
-      "grad_norm": 1.6047074467868236,
+      "grad_norm": 0.8530532879086257,
       "learning_rate": 4.950842394039157e-05,
-      "logits/chosen": -14.6875,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -632.0,
-      "logps/rejected": -3488.0,
-      "loss": 0.0282,
+      "logits/chosen": -15.5625,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -612.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0233,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.625,
-      "rewards/margins": 15.75,
-      "rewards/rejected": -18.375,
+      "rewards/chosen": -2.4375,
+      "rewards/margins": 15.625,
+      "rewards/rejected": -18.0,
       "step": 450
     },
     {
       "epoch": 0.4816753926701571,
-      "grad_norm": 0.37580755764222806,
+      "grad_norm": 0.24566228991385503,
       "learning_rate": 4.944648803615994e-05,
-      "logits/chosen": -14.0,
-      "logits/rejected": -9.125,
-      "logps/chosen": -724.0,
-      "logps/rejected": -3520.0,
-      "loss": 0.036,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -3.25,
-      "rewards/margins": 16.375,
-      "rewards/rejected": -19.625,
+      "logits/chosen": -15.75,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -648.0,
+      "logps/rejected": -3072.0,
+      "loss": 0.0347,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -2.484375,
+      "rewards/margins": 12.6875,
+      "rewards/rejected": -15.125,
       "step": 460
     },
     {
       "epoch": 0.49214659685863876,
-      "grad_norm": 0.12418907524389593,
+      "grad_norm": 0.13237800791684495,
       "learning_rate": 4.938092181097433e-05,
-      "logits/chosen": -14.6875,
-      "logits/rejected": -10.5,
-      "logps/chosen": -672.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0412,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.765625,
-      "rewards/margins": 14.75,
-      "rewards/rejected": -17.5,
+      "logits/chosen": -14.375,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -640.0,
+      "logps/rejected": -3088.0,
+      "loss": 0.1102,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.4375,
+      "rewards/margins": 14.0,
+      "rewards/rejected": -16.375,
       "step": 470
     },
     {
       "epoch": 0.5026178010471204,
-      "grad_norm": 2.339293514334623,
+      "grad_norm": 1.317983178223874,
       "learning_rate": 4.931173500146607e-05,
-      "logits/chosen": -15.875,
-      "logits/rejected": -11.25,
-      "logps/chosen": -632.0,
-      "logps/rejected": -3616.0,
-      "loss": 0.0382,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -644.0,
+      "logps/rejected": -3072.0,
+      "loss": 0.0418,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.640625,
-      "rewards/margins": 17.125,
-      "rewards/rejected": -19.75,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -14.25,
       "step": 480
     },
     {
       "epoch": 0.5130890052356021,
-      "grad_norm": 0.7658652485594658,
+      "grad_norm": 0.6286965245916285,
       "learning_rate": 4.923893788192582e-05,
-      "logits/chosen": -15.0625,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -696.0,
-      "logps/rejected": -3552.0,
-      "loss": 0.0639,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.046875,
-      "rewards/margins": 17.25,
-      "rewards/rejected": -20.25,
+      "logits/chosen": -14.4375,
+      "logits/rejected": -10.25,
+      "logps/chosen": -644.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.0266,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.515625,
+      "rewards/margins": 11.0625,
+      "rewards/rejected": -13.5625,
       "step": 490
     },
     {
       "epoch": 0.5235602094240838,
-      "grad_norm": 0.4646506214373773,
+      "grad_norm": 0.5000012569443792,
       "learning_rate": 4.9162541262777906e-05,
-      "logits/chosen": -15.5,
-      "logits/rejected": -10.125,
-      "logps/chosen": -704.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0289,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.046875,
-      "rewards/margins": 14.3125,
-      "rewards/rejected": -17.375,
+      "logits/chosen": -15.625,
+      "logits/rejected": -11.6875,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3136.0,
+      "loss": 0.0297,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -2.828125,
+      "rewards/margins": 13.0,
+      "rewards/rejected": -15.8125,
       "step": 500
     },
     {
       "epoch": 0.5235602094240838,
-      "eval_logits/chosen": -15.25,
-      "eval_logits/rejected": -15.0,
-      "eval_logps/chosen": -680.0,
-      "eval_logps/rejected": -676.0,
-      "eval_loss": 0.9760624766349792,
-      "eval_rewards/accuracies": 0.490234375,
-      "eval_rewards/chosen": -3.59375,
-      "eval_rewards/margins": 0.287109375,
-      "eval_rewards/rejected": -3.890625,
-      "eval_runtime": 46.6371,
-      "eval_samples_per_second": 42.884,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -15.6875,
+      "eval_logits/rejected": -15.5,
+      "eval_logps/chosen": -628.0,
+      "eval_logps/rejected": -628.0,
+      "eval_loss": 0.8325625061988831,
+      "eval_rewards/accuracies": 0.50390625,
+      "eval_rewards/chosen": -3.109375,
+      "eval_rewards/margins": 0.2734375,
+      "eval_rewards/rejected": -3.375,
+      "eval_runtime": 58.3297,
+      "eval_samples_per_second": 34.288,
+      "eval_steps_per_second": 0.549,
       "step": 500
     },
     {
       "epoch": 0.5340314136125655,
-      "grad_norm": 1.658962222219197,
+      "grad_norm": 0.9475821620338638,
       "learning_rate": 4.90825564889749e-05,
-      "logits/chosen": -15.125,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -816.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0245,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.9375,
+      "logits/chosen": -15.8125,
+      "logits/rejected": -12.25,
+      "logps/chosen": -760.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0194,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.34375,
       "rewards/margins": 17.125,
-      "rewards/rejected": -21.125,
+      "rewards/rejected": -20.375,
       "step": 510
     },
     {
       "epoch": 0.5445026178010471,
-      "grad_norm": 1.9016287614380438,
+      "grad_norm": 2.3871753445572863,
       "learning_rate": 4.899899543831288e-05,
-      "logits/chosen": -15.3125,
-      "logits/rejected": -8.625,
-      "logps/chosen": -856.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.031,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -4.3125,
-      "rewards/margins": 18.375,
-      "rewards/rejected": -22.625,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -848.0,
+      "logps/rejected": -4016.0,
+      "loss": 0.0392,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -4.25,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -24.25,
       "step": 520
     },
     {
       "epoch": 0.5549738219895288,
-      "grad_norm": 1.5977212119003472,
+      "grad_norm": 1.5661147813033802,
       "learning_rate": 4.891187051966765e-05,
-      "logits/chosen": -15.875,
-      "logits/rejected": -10.375,
-      "logps/chosen": -816.0,
-      "logps/rejected": -3456.0,
-      "loss": 0.064,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -4.0625,
-      "rewards/margins": 15.6875,
-      "rewards/rejected": -19.75,
+      "logits/chosen": -14.25,
+      "logits/rejected": -9.0,
+      "logps/chosen": -788.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.051,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -3.765625,
+      "rewards/margins": 14.375,
+      "rewards/rejected": -18.125,
       "step": 530
     },
     {
       "epoch": 0.5654450261780105,
-      "grad_norm": 0.3639124672439281,
+      "grad_norm": 1.527702549119752,
       "learning_rate": 4.882119467115191e-05,
-      "logits/chosen": -15.9375,
-      "logits/rejected": -12.0625,
-      "logps/chosen": -588.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0271,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -1.9765625,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -10.875,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -9.375,
+      "logps/chosen": -592.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0272,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.015625,
+      "rewards/margins": 9.3125,
+      "rewards/rejected": -11.3125,
       "step": 540
     },
     {
       "epoch": 0.5759162303664922,
-      "grad_norm": 0.5200538251722615,
+      "grad_norm": 0.4147669721512114,
       "learning_rate": 4.872698135819402e-05,
-      "logits/chosen": -15.625,
-      "logits/rejected": -11.75,
-      "logps/chosen": -608.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.0293,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -2.28125,
-      "rewards/margins": 10.875,
-      "rewards/rejected": -13.1875,
-      "step": 550
-    },
-    {
+      "logits/chosen": -14.5,
+      "logits/rejected": -9.25,
+      "logps/chosen": -616.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0447,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.359375,
+      "rewards/margins": 10.375,
+      "rewards/rejected": -12.75,
+      "step": 550
+    },
+    {
       "epoch": 0.5863874345549738,
-      "grad_norm": 1.5488427572690442,
+      "grad_norm": 1.3754866372603125,
       "learning_rate": 4.862924457153831e-05,
-      "logits/chosen": -14.75,
-      "logits/rejected": -11.375,
+      "logits/chosen": -14.875,
+      "logits/rejected": -9.25,
       "logps/chosen": -604.0,
-      "logps/rejected": -3440.0,
-      "loss": 0.026,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.703125,
-      "rewards/margins": 15.25,
-      "rewards/rejected": -17.875,
+      "logps/rejected": -3552.0,
+      "loss": 0.0298,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 16.25,
+      "rewards/rejected": -19.0,
       "step": 560
     },
     {
       "epoch": 0.5968586387434555,
-      "grad_norm": 0.3394840961569088,
+      "grad_norm": 0.3693918071126619,
       "learning_rate": 4.8527998825167484e-05,
       "logits/chosen": -15.0,
-      "logits/rejected": -10.75,
-      "logps/chosen": -636.0,
-      "logps/rejected": -3456.0,
-      "loss": 0.0427,
+      "logits/rejected": -9.5,
+      "logps/chosen": -604.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.0374,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.671875,
-      "rewards/margins": 15.625,
-      "rewards/rejected": -18.25,
+      "rewards/chosen": -2.359375,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -20.375,
       "step": 570
     },
     {
       "epoch": 0.6073298429319371,
-      "grad_norm": 0.26440354078514694,
+      "grad_norm": 0.24806611469256337,
       "learning_rate": 4.842325915414728e-05,
-      "logits/chosen": -15.6875,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -616.0,
-      "logps/rejected": -3072.0,
-      "loss": 0.0353,
+      "logits/chosen": -14.8125,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -560.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0299,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.515625,
-      "rewards/margins": 13.75,
-      "rewards/rejected": -16.25,
+      "rewards/chosen": -1.9765625,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -18.5,
       "step": 580
     },
     {
       "epoch": 0.6178010471204188,
-      "grad_norm": 0.29217481796559885,
+      "grad_norm": 0.18855608462679527,
       "learning_rate": 4.831504111239373e-05,
-      "logits/chosen": -15.625,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -680.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0419,
+      "logits/chosen": -14.875,
+      "logits/rejected": -9.875,
+      "logps/chosen": -636.0,
+      "logps/rejected": -3536.0,
+      "loss": 0.0317,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.609375,
-      "rewards/margins": 14.75,
-      "rewards/rejected": -17.375,
+      "rewards/chosen": -2.1875,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -19.25,
       "step": 590
     },
     {
       "epoch": 0.6282722513089005,
-      "grad_norm": 3.259899312616823,
+      "grad_norm": 1.012379912025639,
       "learning_rate": 4.8203360770363425e-05,
-      "logits/chosen": -15.875,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -644.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.0549,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.40625,
-      "rewards/margins": 15.5625,
-      "rewards/rejected": -18.0,
+      "logits/chosen": -15.4375,
+      "logits/rejected": -9.25,
+      "logps/chosen": -616.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0434,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.140625,
+      "rewards/margins": 17.375,
+      "rewards/rejected": -19.5,
       "step": 600
     },
     {
       "epoch": 0.6282722513089005,
-      "eval_logits/chosen": -15.1875,
-      "eval_logits/rejected": -15.0625,
-      "eval_logps/chosen": -556.0,
-      "eval_logps/rejected": -548.0,
-      "eval_loss": 0.9004374742507935,
-      "eval_rewards/accuracies": 0.48046875,
-      "eval_rewards/chosen": -2.359375,
-      "eval_rewards/margins": 0.25390625,
-      "eval_rewards/rejected": -2.609375,
-      "eval_runtime": 46.6319,
-      "eval_samples_per_second": 42.889,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -15.25,
+      "eval_logits/rejected": -15.0,
+      "eval_logps/chosen": -488.0,
+      "eval_logps/rejected": -476.0,
+      "eval_loss": 0.8373281359672546,
+      "eval_rewards/accuracies": 0.494140625,
+      "eval_rewards/chosen": -1.6953125,
+      "eval_rewards/margins": 0.1826171875,
+      "eval_rewards/rejected": -1.875,
+      "eval_runtime": 58.2528,
+      "eval_samples_per_second": 34.333,
+      "eval_steps_per_second": 0.549,
       "step": 600
     },
     {
       "epoch": 0.6387434554973822,
-      "grad_norm": 0.3544827873176982,
+      "grad_norm": 0.17199746318590983,
       "learning_rate": 4.808823471266702e-05,
       "logits/chosen": -15.25,
       "logits/rejected": -9.5625,
-      "logps/chosen": -580.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.0297,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -1.953125,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -13.375,
+      "logps/chosen": -564.0,
+      "logps/rejected": -2768.0,
+      "loss": 0.0295,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -1.8125,
+      "rewards/margins": 11.0625,
+      "rewards/rejected": -12.875,
       "step": 610
     },
     {
       "epoch": 0.6492146596858639,
-      "grad_norm": 1.1808540060984913,
+      "grad_norm": 0.9790838433638168,
       "learning_rate": 4.796968003560639e-05,
-      "logits/chosen": -14.375,
-      "logits/rejected": -9.125,
+      "logits/chosen": -14.875,
+      "logits/rejected": -9.4375,
       "logps/chosen": -624.0,
-      "logps/rejected": -3264.0,
-      "loss": 0.0476,
-      "rewards/accuracies": 0.9750000238418579,
+      "logps/rejected": -3280.0,
+      "loss": 0.0573,
+      "rewards/accuracies": 0.9624999761581421,
       "rewards/chosen": -2.4375,
-      "rewards/margins": 13.4375,
-      "rewards/rejected": -15.875,
+      "rewards/margins": 13.5625,
+      "rewards/rejected": -16.0,
       "step": 620
     },
     {
       "epoch": 0.6596858638743456,
-      "grad_norm": 0.03589007701308354,
+      "grad_norm": 0.03826508900886177,
       "learning_rate": 4.7847714344635864e-05,
-      "logits/chosen": -12.8125,
-      "logits/rejected": -7.71875,
-      "logps/chosen": -680.0,
-      "logps/rejected": -3584.0,
-      "loss": 0.0266,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.09375,
-      "rewards/margins": 16.625,
-      "rewards/rejected": -19.75,
+      "logits/chosen": -14.5,
+      "logits/rejected": -8.75,
+      "logps/chosen": -668.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0508,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -22.0,
       "step": 630
     },
     {
       "epoch": 0.6701570680628273,
-      "grad_norm": 1.0794958469361786,
+      "grad_norm": 0.4703306335864405,
       "learning_rate": 4.772235575174776e-05,
-      "logits/chosen": -13.0,
-      "logits/rejected": -6.5625,
-      "logps/chosen": -800.0,
-      "logps/rejected": -4000.0,
-      "loss": 0.0598,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -3.796875,
-      "rewards/margins": 20.75,
-      "rewards/rejected": -24.5,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -8.0,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.0278,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.75,
+      "rewards/margins": 14.8125,
+      "rewards/rejected": -17.625,
       "step": 640
     },
     {
       "epoch": 0.680628272251309,
-      "grad_norm": 0.6307699713381634,
+      "grad_norm": 0.48548502123061704,
       "learning_rate": 4.7593622872782754e-05,
-      "logits/chosen": -14.0,
-      "logits/rejected": -7.625,
-      "logps/chosen": -676.0,
-      "logps/rejected": -3744.0,
-      "loss": 0.031,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -8.0,
+      "logps/chosen": -620.0,
+      "logps/rejected": -3120.0,
+      "loss": 0.0314,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.109375,
-      "rewards/margins": 18.125,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -2.5625,
+      "rewards/margins": 12.375,
+      "rewards/rejected": -14.9375,
       "step": 650
     },
     {
       "epoch": 0.6910994764397905,
-      "grad_norm": 0.19076274638651378,
+      "grad_norm": 0.21397405398592934,
       "learning_rate": 4.746153482466542e-05,
-      "logits/chosen": -15.9375,
-      "logits/rejected": -8.875,
-      "logps/chosen": -620.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0416,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.484375,
-      "rewards/margins": 16.0,
-      "rewards/rejected": -18.5,
+      "logits/chosen": -13.875,
+      "logits/rejected": -7.59375,
+      "logps/chosen": -696.0,
+      "logps/rejected": -3360.0,
+      "loss": 0.0331,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 15.9375,
+      "rewards/rejected": -19.125,
       "step": 660
     },
     {
       "epoch": 0.7015706806282722,
-      "grad_norm": 0.9650710991194229,
+      "grad_norm": 0.5277319568985792,
       "learning_rate": 4.732611122256537e-05,
-      "logits/chosen": -15.875,
-      "logits/rejected": -8.125,
-      "logps/chosen": -676.0,
-      "logps/rejected": -3456.0,
-      "loss": 0.0504,
+      "logits/chosen": -13.875,
+      "logits/rejected": -7.4375,
+      "logps/chosen": -776.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.034,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -2.578125,
-      "rewards/margins": 15.9375,
-      "rewards/rejected": -18.5,
+      "rewards/chosen": -3.578125,
+      "rewards/margins": 16.875,
+      "rewards/rejected": -20.5,
       "step": 670
     },
     {
       "epoch": 0.7120418848167539,
-      "grad_norm": 1.9064976963823153,
+      "grad_norm": 2.3313798887629744,
       "learning_rate": 4.718737217698435e-05,
-      "logits/chosen": -14.9375,
-      "logits/rejected": -8.9375,
-      "logps/chosen": -668.0,
-      "logps/rejected": -3072.0,
-      "loss": 0.127,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -2.921875,
-      "rewards/margins": 11.5,
-      "rewards/rejected": -14.375,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -7.3125,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3392.0,
+      "loss": 0.0476,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 14.25,
+      "rewards/rejected": -17.5,
       "step": 680
     },
     {
       "epoch": 0.7225130890052356,
-      "grad_norm": 0.47402113546008434,
+      "grad_norm": 0.3235411097828364,
       "learning_rate": 4.704533829076982e-05,
-      "logits/chosen": -14.9375,
-      "logits/rejected": -9.5,
-      "logps/chosen": -632.0,
+      "logits/chosen": -11.4375,
+      "logits/rejected": -6.375,
+      "logps/chosen": -700.0,
       "logps/rejected": -2832.0,
-      "loss": 0.0646,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.390625,
-      "rewards/margins": 10.3125,
-      "rewards/rejected": -12.6875,
+      "loss": 0.0533,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.078125,
+      "rewards/margins": 9.6875,
+      "rewards/rejected": -12.75,
       "step": 690
     },
     {
       "epoch": 0.7329842931937173,
-      "grad_norm": 0.11224459937713936,
+      "grad_norm": 0.12999533214563885,
       "learning_rate": 4.6900030656055455e-05,
-      "logits/chosen": -13.75,
-      "logits/rejected": -7.5625,
-      "logps/chosen": -672.0,
-      "logps/rejected": -2976.0,
-      "loss": 0.0385,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.65625,
-      "rewards/margins": 12.5,
-      "rewards/rejected": -15.125,
+      "logits/chosen": -9.625,
+      "logits/rejected": -6.90625,
+      "logps/chosen": -768.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0496,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.640625,
+      "rewards/margins": 9.75,
+      "rewards/rejected": -13.375,
       "step": 700
     },
     {
       "epoch": 0.7329842931937173,
-      "eval_logits/chosen": -11.875,
-      "eval_logits/rejected": -11.5,
-      "eval_logps/chosen": -632.0,
-      "eval_logps/rejected": -612.0,
-      "eval_loss": 0.9996874928474426,
-      "eval_rewards/accuracies": 0.474609375,
-      "eval_rewards/chosen": -3.125,
-      "eval_rewards/margins": 0.1298828125,
-      "eval_rewards/rejected": -3.25,
-      "eval_runtime": 46.613,
-      "eval_samples_per_second": 42.906,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -10.3125,
+      "eval_logits/rejected": -9.5625,
+      "eval_logps/chosen": -692.0,
+      "eval_logps/rejected": -684.0,
+      "eval_loss": 0.9407031536102295,
+      "eval_rewards/accuracies": 0.533203125,
+      "eval_rewards/chosen": -3.734375,
+      "eval_rewards/margins": 0.2236328125,
+      "eval_rewards/rejected": -3.96875,
+      "eval_runtime": 58.3078,
+      "eval_samples_per_second": 34.301,
+      "eval_steps_per_second": 0.549,
       "step": 700
     },
     {
       "epoch": 0.743455497382199,
-      "grad_norm": 0.5247872396658949,
+      "grad_norm": 0.7306916484252496,
       "learning_rate": 4.6751470851128906e-05,
-      "logits/chosen": -12.125,
-      "logits/rejected": -6.96875,
-      "logps/chosen": -696.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.0545,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.765625,
-      "rewards/margins": 11.25,
-      "rewards/rejected": -14.0,
+      "logits/chosen": -12.0,
+      "logits/rejected": -6.65625,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0451,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.328125,
+      "rewards/margins": 14.8125,
+      "rewards/rejected": -18.125,
       "step": 710
     },
     {
       "epoch": 0.7539267015706806,
-      "grad_norm": 0.3600493979910419,
+      "grad_norm": 0.2934151359487545,
       "learning_rate": 4.6599680937227454e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -6.75,
-      "logps/chosen": -660.0,
-      "logps/rejected": -3072.0,
-      "loss": 0.0182,
+      "logits/chosen": -12.875,
+      "logits/rejected": -7.125,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.0347,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.625,
-      "rewards/margins": 12.5625,
-      "rewards/rejected": -15.1875,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 14.75,
+      "rewards/rejected": -17.75,
       "step": 720
     },
     {
       "epoch": 0.7643979057591623,
-      "grad_norm": 0.8635049635799732,
+      "grad_norm": 0.42725377485636507,
       "learning_rate": 4.644468345526183e-05,
-      "logits/chosen": -10.5625,
-      "logits/rejected": -5.5,
-      "logps/chosen": -748.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.0302,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.609375,
-      "rewards/margins": 19.5,
-      "rewards/rejected": -23.125,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -7.03125,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0277,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 13.6875,
+      "rewards/rejected": -16.75,
       "step": 730
     },
     {
       "epoch": 0.774869109947644,
-      "grad_norm": 0.21938844467123172,
+      "grad_norm": 0.8583644931223695,
       "learning_rate": 4.628650142246894e-05,
-      "logits/chosen": -10.5625,
-      "logits/rejected": -5.75,
-      "logps/chosen": -704.0,
-      "logps/rejected": -3440.0,
-      "loss": 0.0664,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 15.375,
-      "rewards/rejected": -18.75,
+      "logits/chosen": -12.3125,
+      "logits/rejected": -6.3125,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0458,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.359375,
+      "rewards/margins": 13.4375,
+      "rewards/rejected": -16.75,
       "step": 740
     },
     {
       "epoch": 0.7853403141361257,
-      "grad_norm": 0.7810167131254897,
+      "grad_norm": 1.0104169865573032,
       "learning_rate": 4.6125158328993765e-05,
-      "logits/chosen": -11.8125,
-      "logits/rejected": -6.03125,
-      "logps/chosen": -648.0,
-      "logps/rejected": -2912.0,
-      "loss": 0.0321,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.71875,
-      "rewards/margins": 10.75,
-      "rewards/rejected": -13.5,
+      "logits/chosen": -12.0,
+      "logits/rejected": -5.34375,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3264.0,
+      "loss": 0.0294,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -3.5625,
+      "rewards/margins": 13.3125,
+      "rewards/rejected": -16.875,
       "step": 750
     },
     {
       "epoch": 0.7958115183246073,
-      "grad_norm": 0.06721505269479468,
+      "grad_norm": 1.1518427256662063,
       "learning_rate": 4.5960678134401064e-05,
       "logits/chosen": -12.9375,
-      "logits/rejected": -5.71875,
-      "logps/chosen": -692.0,
-      "logps/rejected": -3024.0,
-      "loss": 0.0372,
+      "logits/rejected": -6.25,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3168.0,
+      "loss": 0.027,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.9375,
-      "rewards/margins": 13.0625,
-      "rewards/rejected": -16.0,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 14.1875,
+      "rewards/rejected": -17.5,
       "step": 760
     },
     {
       "epoch": 0.806282722513089,
-      "grad_norm": 0.01310499060692319,
+      "grad_norm": 0.02666845282273652,
       "learning_rate": 4.579308526411734e-05,
-      "logits/chosen": -11.9375,
-      "logits/rejected": -5.78125,
-      "logps/chosen": -728.0,
-      "logps/rejected": -3376.0,
-      "loss": 0.0244,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 14.875,
-      "rewards/rejected": -18.25,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -6.8125,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3264.0,
+      "loss": 0.018,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 14.25,
+      "rewards/rejected": -17.125,
       "step": 770
     },
     {
       "epoch": 0.8167539267015707,
-      "grad_norm": 0.29189299059789137,
+      "grad_norm": 0.37372728528493826,
       "learning_rate": 4.562240460580367e-05,
-      "logits/chosen": -12.75,
-      "logits/rejected": -6.78125,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -6.9375,
       "logps/chosen": -680.0,
-      "logps/rejected": -3472.0,
-      "loss": 0.027,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.84375,
-      "rewards/margins": 17.25,
-      "rewards/rejected": -20.125,
+      "logps/rejected": -3408.0,
+      "loss": 0.0345,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.828125,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -19.375,
       "step": 780
     },
     {
       "epoch": 0.8272251308900523,
-      "grad_norm": 0.06363200858633618,
+      "grad_norm": 0.08223850096766275,
       "learning_rate": 4.544866150565985e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -7.9375,
-      "logps/chosen": -620.0,
-      "logps/rejected": -3360.0,
-      "loss": 0.0314,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.34375,
-      "rewards/margins": 15.5,
-      "rewards/rejected": -17.875,
+      "logits/chosen": -13.625,
+      "logits/rejected": -6.8125,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3600.0,
+      "loss": 0.0644,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.0625,
+      "rewards/margins": 17.25,
+      "rewards/rejected": -20.375,
       "step": 790
     },
     {
       "epoch": 0.837696335078534,
-      "grad_norm": 0.5021321015036181,
+      "grad_norm": 0.4864416919113278,
       "learning_rate": 4.527188176466051e-05,
-      "logits/chosen": -13.8125,
-      "logits/rejected": -8.375,
-      "logps/chosen": -672.0,
+      "logits/chosen": -12.9375,
+      "logits/rejected": -7.34375,
+      "logps/chosen": -732.0,
       "logps/rejected": -3008.0,
-      "loss": 0.0303,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -2.59375,
-      "rewards/margins": 12.8125,
-      "rewards/rejected": -15.4375,
+      "loss": 0.0289,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -15.5,
       "step": 800
     },
     {
       "epoch": 0.837696335078534,
-      "eval_logits/chosen": -14.0625,
-      "eval_logits/rejected": -13.75,
-      "eval_logps/chosen": -596.0,
-      "eval_logps/rejected": -584.0,
-      "eval_loss": 1.0037187337875366,
-      "eval_rewards/accuracies": 0.478515625,
-      "eval_rewards/chosen": -2.765625,
-      "eval_rewards/margins": 0.1748046875,
-      "eval_rewards/rejected": -2.9375,
-      "eval_runtime": 46.747,
-      "eval_samples_per_second": 42.783,
-      "eval_steps_per_second": 0.685,
+      "eval_logits/chosen": -13.3125,
+      "eval_logits/rejected": -13.0625,
+      "eval_logps/chosen": -632.0,
+      "eval_logps/rejected": -612.0,
+      "eval_loss": 1.010812520980835,
+      "eval_rewards/accuracies": 0.470703125,
+      "eval_rewards/chosen": -3.140625,
+      "eval_rewards/margins": 0.09912109375,
+      "eval_rewards/rejected": -3.25,
+      "eval_runtime": 58.2962,
+      "eval_samples_per_second": 34.308,
+      "eval_steps_per_second": 0.549,
       "step": 800
     },
     {
       "epoch": 0.8481675392670157,
-      "grad_norm": 0.2036534321998557,
+      "grad_norm": 0.9738056878115542,
       "learning_rate": 4.509209163472357e-05,
-      "logits/chosen": -14.25,
-      "logits/rejected": -8.125,
-      "logps/chosen": -692.0,
-      "logps/rejected": -3120.0,
-      "loss": 0.0215,
+      "logits/chosen": -13.25,
+      "logits/rejected": -7.3125,
+      "logps/chosen": -688.0,
+      "logps/rejected": -2768.0,
+      "loss": 0.0207,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.875,
-      "rewards/margins": 13.0,
-      "rewards/rejected": -15.875,
+      "rewards/chosen": -2.828125,
+      "rewards/margins": 9.6875,
+      "rewards/rejected": -12.5,
       "step": 810
     },
     {
       "epoch": 0.8586387434554974,
-      "grad_norm": 1.8049084624492868,
+      "grad_norm": 2.1843742962997665,
       "learning_rate": 4.4909317814811896e-05,
-      "logits/chosen": -15.0625,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3248.0,
-      "loss": 0.0355,
+      "logits/chosen": -14.25,
+      "logits/rejected": -7.59375,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3424.0,
+      "loss": 0.0318,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -2.796875,
-      "rewards/margins": 13.5625,
-      "rewards/rejected": -16.375,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 15.0,
+      "rewards/rejected": -18.125,
       "step": 820
     },
     {
       "epoch": 0.8691099476439791,
-      "grad_norm": 0.814631487758852,
+      "grad_norm": 0.695909214436541,
       "learning_rate": 4.472358744696845e-05,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -700.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.056,
+      "logits/chosen": -15.5,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -760.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.052,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -3.09375,
-      "rewards/margins": 15.5,
-      "rewards/rejected": -18.625,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.75,
       "step": 830
     },
     {
       "epoch": 0.8795811518324608,
-      "grad_norm": 0.33408859873604485,
+      "grad_norm": 0.3605341654613082,
       "learning_rate": 4.4534928112285645e-05,
-      "logits/chosen": -14.4375,
-      "logits/rejected": -7.59375,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4016.0,
-      "loss": 0.0389,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -3.328125,
-      "rewards/margins": 21.625,
-      "rewards/rejected": -24.875,
+      "logits/chosen": -14.625,
+      "logits/rejected": -7.875,
+      "logps/chosen": -768.0,
+      "logps/rejected": -4080.0,
+      "loss": 0.023,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 21.875,
+      "rewards/rejected": -25.625,
       "step": 840
     },
     {
       "epoch": 0.8900523560209425,
-      "grad_norm": 0.5302838139876008,
+      "grad_norm": 0.3873200497688502,
       "learning_rate": 4.43433678268096e-05,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -5.21875,
-      "logps/chosen": -764.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0451,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -3.875,
-      "rewards/margins": 24.625,
-      "rewards/rejected": -28.5,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -7.3125,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0247,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -23.75,
       "step": 850
     },
     {
       "epoch": 0.900523560209424,
-      "grad_norm": 0.23281335650964907,
+      "grad_norm": 0.33704357105780586,
       "learning_rate": 4.414893503737968e-05,
-      "logits/chosen": -14.0625,
-      "logits/rejected": -5.90625,
-      "logps/chosen": -656.0,
-      "logps/rejected": -3504.0,
-      "loss": 0.0305,
+      "logits/chosen": -14.125,
+      "logits/rejected": -7.15625,
+      "logps/chosen": -696.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0381,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -2.921875,
-      "rewards/margins": 18.0,
-      "rewards/rejected": -21.0,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -23.0,
       "step": 860
     },
     {
       "epoch": 0.9109947643979057,
-      "grad_norm": 0.5782393946828994,
+      "grad_norm": 0.7549939316222815,
       "learning_rate": 4.395165861740414e-05,
-      "logits/chosen": -14.0,
-      "logits/rejected": -7.25,
-      "logps/chosen": -652.0,
-      "logps/rejected": -3376.0,
-      "loss": 0.0227,
+      "logits/chosen": -14.0625,
+      "logits/rejected": -7.1875,
+      "logps/chosen": -720.0,
+      "logps/rejected": -4000.0,
+      "loss": 0.0275,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 15.5625,
-      "rewards/rejected": -18.5,
+      "rewards/chosen": -3.578125,
+      "rewards/margins": 21.25,
+      "rewards/rejected": -24.75,
       "step": 870
     },
     {
       "epoch": 0.9214659685863874,
-      "grad_norm": 0.5126565689071665,
+      "grad_norm": 0.8605760336559388,
       "learning_rate": 4.375156786257241e-05,
-      "logits/chosen": -14.0,
-      "logits/rejected": -7.0,
-      "logps/chosen": -732.0,
-      "logps/rejected": -3568.0,
-      "loss": 0.0418,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.140625,
-      "rewards/margins": 17.0,
-      "rewards/rejected": -20.125,
+      "logits/chosen": -14.375,
+      "logits/rejected": -7.40625,
+      "logps/chosen": -844.0,
+      "logps/rejected": -4288.0,
+      "loss": 0.066,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -4.25,
+      "rewards/margins": 23.125,
+      "rewards/rejected": -27.375,
       "step": 880
     },
     {
       "epoch": 0.9319371727748691,
-      "grad_norm": 0.6455665076227766,
+      "grad_norm": 0.7395749818523414,
       "learning_rate": 4.354869248650466e-05,
-      "logits/chosen": -14.4375,
-      "logits/rejected": -8.0,
-      "logps/chosen": -744.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0379,
+      "logits/chosen": -15.0,
+      "logits/rejected": -7.625,
+      "logps/chosen": -788.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.05,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 13.5625,
-      "rewards/rejected": -16.875,
+      "rewards/chosen": -3.78125,
+      "rewards/margins": 12.0,
+      "rewards/rejected": -15.75,
       "step": 890
     },
     {
       "epoch": 0.9424083769633508,
-      "grad_norm": 0.4302318233765888,
+      "grad_norm": 1.532022977641751,
       "learning_rate": 4.334306261633931e-05,
-      "logits/chosen": -14.3125,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -760.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0147,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.46875,
-      "rewards/margins": 13.5625,
-      "rewards/rejected": -17.0,
+      "logits/chosen": -15.0,
+      "logits/rejected": -8.5,
+      "logps/chosen": -744.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.0259,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 10.125,
+      "rewards/rejected": -13.4375,
       "step": 900
     },
     {
       "epoch": 0.9424083769633508,
-      "eval_logits/chosen": -14.1875,
-      "eval_logits/rejected": -13.8125,
+      "eval_logits/chosen": -15.875,
+      "eval_logits/rejected": -15.625,
       "eval_logps/chosen": -680.0,
-      "eval_logps/rejected": -664.0,
-      "eval_loss": 1.1243125200271606,
-      "eval_rewards/accuracies": 0.482421875,
+      "eval_logps/rejected": -668.0,
+      "eval_loss": 1.0868593454360962,
+      "eval_rewards/accuracies": 0.46484375,
       "eval_rewards/chosen": -3.609375,
-      "eval_rewards/margins": 0.1552734375,
-      "eval_rewards/rejected": -3.765625,
-      "eval_runtime": 46.6148,
-      "eval_samples_per_second": 42.905,
-      "eval_steps_per_second": 0.686,
+      "eval_rewards/margins": 0.1630859375,
+      "eval_rewards/rejected": -3.78125,
+      "eval_runtime": 58.4914,
+      "eval_samples_per_second": 34.193,
+      "eval_steps_per_second": 0.547,
       "step": 900
     },
     {
       "epoch": 0.9528795811518325,
-      "grad_norm": 0.02452975154205547,
+      "grad_norm": 0.22783837530173948,
       "learning_rate": 4.313470878825911e-05,
-      "logits/chosen": -14.1875,
-      "logits/rejected": -9.0625,
+      "logits/chosen": -15.5,
+      "logits/rejected": -7.25,
       "logps/chosen": -752.0,
-      "logps/rejected": -3824.0,
-      "loss": 0.0239,
-      "rewards/accuracies": 0.987500011920929,
+      "logps/rejected": -3248.0,
+      "loss": 0.0255,
+      "rewards/accuracies": 0.9937499761581421,
       "rewards/chosen": -3.65625,
-      "rewards/margins": 18.5,
-      "rewards/rejected": -22.125,
+      "rewards/margins": 12.6875,
+      "rewards/rejected": -16.375,
       "step": 910
     },
     {
       "epoch": 0.9633507853403142,
-      "grad_norm": 2.7020826001659572,
+      "grad_norm": 3.204568174262903,
       "learning_rate": 4.292366194295652e-05,
-      "logits/chosen": -14.3125,
-      "logits/rejected": -9.125,
-      "logps/chosen": -736.0,
-      "logps/rejected": -3440.0,
-      "loss": 0.0538,
+      "logits/chosen": -15.3125,
+      "logits/rejected": -7.46875,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3088.0,
+      "loss": 0.055,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 17.0,
-      "rewards/rejected": -20.5,
+      "rewards/chosen": -3.90625,
+      "rewards/margins": 13.125,
+      "rewards/rejected": -17.0,
       "step": 920
     },
     {
       "epoch": 0.9738219895287958,
-      "grad_norm": 0.4395173806647976,
+      "grad_norm": 0.48484340674751086,
       "learning_rate": 4.2709953421038964e-05,
-      "logits/chosen": -14.0,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -728.0,
-      "logps/rejected": -3328.0,
-      "loss": 0.0427,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 12.8125,
-      "rewards/rejected": -16.25,
+      "logits/chosen": -15.5,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.064,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.359375,
+      "rewards/margins": 12.0625,
+      "rewards/rejected": -15.4375,
       "step": 930
     },
     {
       "epoch": 0.9842931937172775,
-      "grad_norm": 0.7638437429048192,
+      "grad_norm": 0.6650139638971434,
       "learning_rate": 4.249361495837478e-05,
-      "logits/chosen": -14.25,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -780.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.0398,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.9375,
-      "rewards/margins": 11.6875,
-      "rewards/rejected": -15.625,
+      "logits/chosen": -15.0,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -672.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.0535,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 9.875,
+      "rewards/rejected": -12.6875,
       "step": 940
     },
     {
       "epoch": 0.9947643979057592,
-      "grad_norm": 0.3251012354208565,
+      "grad_norm": 0.33667618116247766,
       "learning_rate": 4.227467868138035e-05,
-      "logits/chosen": -15.625,
-      "logits/rejected": -10.5,
-      "logps/chosen": -744.0,
-      "logps/rejected": -3280.0,
-      "loss": 0.0393,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 13.6875,
-      "rewards/rejected": -17.25,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -716.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.0315,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.25,
+      "rewards/margins": 10.875,
+      "rewards/rejected": -14.0625,
       "step": 950
     },
     {
       "epoch": 1.0052356020942408,
-      "grad_norm": 0.008119670127014563,
+      "grad_norm": 0.012836104133861779,
       "learning_rate": 4.205317710224938e-05,
-      "logits/chosen": -15.5625,
-      "logits/rejected": -11.375,
-      "logps/chosen": -704.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0166,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -10.0,
+      "logps/chosen": -792.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0997,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -3.03125,
-      "rewards/margins": 14.5625,
-      "rewards/rejected": -17.625,
+      "rewards/chosen": -3.921875,
+      "rewards/margins": 14.0625,
+      "rewards/rejected": -18.0,
       "step": 960
     },
     {
       "epoch": 1.0157068062827226,
-      "grad_norm": 0.059358924457938715,
+      "grad_norm": 0.1326131577640483,
       "learning_rate": 4.1829143114124733e-05,
-      "logits/chosen": -15.9375,
-      "logits/rejected": -12.5,
-      "logps/chosen": -680.0,
-      "logps/rejected": -3488.0,
-      "loss": 0.0021,
+      "logits/chosen": -15.5,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -764.0,
+      "logps/rejected": -3600.0,
+      "loss": 0.0017,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 16.625,
-      "rewards/rejected": -19.5,
+      "rewards/chosen": -3.75,
+      "rewards/margins": 16.875,
+      "rewards/rejected": -20.625,
       "step": 970
     },
     {
       "epoch": 1.0261780104712042,
-      "grad_norm": 0.09333746425797176,
+      "grad_norm": 0.07650473409471704,
       "learning_rate": 4.160260998621387e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -12.1875,
-      "logps/chosen": -740.0,
-      "logps/rejected": -3776.0,
-      "loss": 0.0147,
+      "logits/chosen": -15.4375,
+      "logits/rejected": -9.5,
+      "logps/chosen": -796.0,
+      "logps/rejected": -3568.0,
+      "loss": 0.0041,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 19.625,
-      "rewards/rejected": -22.875,
+      "rewards/chosen": -3.78125,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -20.875,
       "step": 980
     },
     {
       "epoch": 1.036649214659686,
-      "grad_norm": 0.308760639155674,
+      "grad_norm": 0.18375699746984236,
       "learning_rate": 4.137361135884824e-05,
-      "logits/chosen": -15.125,
-      "logits/rejected": -10.4375,
+      "logits/chosen": -14.4375,
+      "logits/rejected": -9.375,
       "logps/chosen": -732.0,
-      "logps/rejected": -3712.0,
-      "loss": 0.0029,
+      "logps/rejected": -3520.0,
+      "loss": 0.0021,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 17.75,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 15.9375,
+      "rewards/rejected": -19.25,
       "step": 990
     },
     {
       "epoch": 1.0471204188481675,
-      "grad_norm": 0.0749640318421138,
+      "grad_norm": 0.16249690509503137,
       "learning_rate": 4.114218123848777e-05,
-      "logits/chosen": -14.5,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -680.0,
-      "logps/rejected": -3616.0,
-      "loss": 0.0038,
+      "logits/chosen": -14.625,
+      "logits/rejected": -9.75,
+      "logps/chosen": -664.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.09375,
-      "rewards/margins": 17.0,
-      "rewards/rejected": -20.125,
+      "rewards/chosen": -2.9375,
+      "rewards/margins": 15.5625,
+      "rewards/rejected": -18.5,
       "step": 1000
     },
     {
       "epoch": 1.0471204188481675,
-      "eval_logits/chosen": -14.125,
-      "eval_logits/rejected": -13.8125,
-      "eval_logps/chosen": -676.0,
-      "eval_logps/rejected": -668.0,
-      "eval_loss": 1.0635312795639038,
-      "eval_rewards/accuracies": 0.494140625,
-      "eval_rewards/chosen": -3.578125,
-      "eval_rewards/margins": 0.2158203125,
-      "eval_rewards/rejected": -3.796875,
-      "eval_runtime": 46.6079,
-      "eval_samples_per_second": 42.911,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -15.25,
+      "eval_logits/rejected": -15.0625,
+      "eval_logps/chosen": -664.0,
+      "eval_logps/rejected": -652.0,
+      "eval_loss": 1.094390630722046,
+      "eval_rewards/accuracies": 0.45703125,
+      "eval_rewards/chosen": -3.4375,
+      "eval_rewards/margins": 0.17578125,
+      "eval_rewards/rejected": -3.625,
+      "eval_runtime": 58.2794,
+      "eval_samples_per_second": 34.317,
+      "eval_steps_per_second": 0.549,
       "step": 1000
     },
     {
       "epoch": 1.057591623036649,
-      "grad_norm": 0.042771840457376135,
+      "grad_norm": 0.07317202583341488,
       "learning_rate": 4.0908353992670824e-05,
-      "logits/chosen": -14.5,
-      "logits/rejected": -9.75,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3648.0,
-      "loss": 0.0029,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -10.125,
+      "logps/chosen": -656.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0014,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 17.875,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -2.859375,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -19.0,
       "step": 1010
     },
     {
       "epoch": 1.068062827225131,
-      "grad_norm": 0.0032161675083494483,
+      "grad_norm": 0.012337831583363917,
       "learning_rate": 4.067216434491059e-05,
-      "logits/chosen": -14.625,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -720.0,
-      "logps/rejected": -3872.0,
-      "loss": 0.0024,
+      "logits/chosen": -15.5,
+      "logits/rejected": -9.75,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0021,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 20.25,
-      "rewards/rejected": -23.625,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 17.5,
+      "rewards/rejected": -20.625,
       "step": 1020
     },
     {
       "epoch": 1.0785340314136125,
-      "grad_norm": 0.013603417056732318,
+      "grad_norm": 0.011577300311291611,
       "learning_rate": 4.043364736953866e-05,
-      "logits/chosen": -14.5,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -840.0,
-      "logps/rejected": -4096.0,
-      "loss": 0.0166,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.921875,
-      "rewards/margins": 21.5,
-      "rewards/rejected": -25.5,
+      "logits/chosen": -15.5625,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -844.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0025,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.9375,
+      "rewards/margins": 18.5,
+      "rewards/rejected": -22.375,
       "step": 1030
     },
     {
       "epoch": 1.0890052356020943,
-      "grad_norm": 0.008062170697864203,
+      "grad_norm": 0.008968050363599173,
       "learning_rate": 4.019283848649643e-05,
-      "logits/chosen": -14.5625,
-      "logits/rejected": -11.0625,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3872.0,
-      "loss": 0.0012,
+      "logits/chosen": -16.0,
+      "logits/rejected": -9.125,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0027,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.1875,
-      "rewards/margins": 19.625,
-      "rewards/rejected": -22.75,
-      "step": 1040
-    },
-    {
-      "epoch": 1.0994764397905759,
-      "grad_norm": 0.13971868572514048,
-      "learning_rate": 3.9949773456075215e-05,
-      "logits/chosen": -14.25,
-      "logits/rejected": -11.1875,
-      "logps/chosen": -696.0,
-      "logps/rejected": -3984.0,
-      "loss": 0.0153,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 20.375,
-      "rewards/rejected": -23.625,
+      "rewards/chosen": -3.625,
+      "rewards/margins": 19.0,
+      "rewards/rejected": -22.625,
+      "step": 1040
+    },
+    {
+      "epoch": 1.0994764397905759,
+      "grad_norm": 0.0025983658707289694,
+      "learning_rate": 3.9949773456075215e-05,
+      "logits/chosen": -16.125,
+      "logits/rejected": -8.75,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0084,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.71875,
+      "rewards/margins": 18.125,
+      "rewards/rejected": -21.875,
       "step": 1050
     },
     {
       "epoch": 1.1099476439790577,
-      "grad_norm": 1.0076852291341252,
+      "grad_norm": 0.08206094206578826,
       "learning_rate": 3.970448837360585e-05,
-      "logits/chosen": -14.5,
-      "logits/rejected": -10.75,
-      "logps/chosen": -796.0,
-      "logps/rejected": -3968.0,
-      "loss": 0.0019,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.59375,
-      "rewards/margins": 21.375,
-      "rewards/rejected": -25.0,
+      "logits/chosen": -16.5,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -816.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0037,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.78125,
+      "rewards/margins": 15.5625,
+      "rewards/rejected": -19.375,
       "step": 1060
     },
     {
       "epoch": 1.1204188481675392,
-      "grad_norm": 0.06158344392086408,
+      "grad_norm": 0.3367673519424652,
       "learning_rate": 3.945701966409846e-05,
-      "logits/chosen": -14.625,
-      "logits/rejected": -11.0625,
-      "logps/chosen": -728.0,
-      "logps/rejected": -3984.0,
-      "loss": 0.0075,
+      "logits/chosen": -16.5,
+      "logits/rejected": -10.5,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3488.0,
+      "loss": 0.0029,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 20.625,
-      "rewards/rejected": -24.125,
+      "rewards/chosen": -3.25,
+      "rewards/margins": 15.9375,
+      "rewards/rejected": -19.125,
       "step": 1070
     },
     {
       "epoch": 1.130890052356021,
-      "grad_norm": 0.06688023040151427,
+      "grad_norm": 0.006734754661559785,
       "learning_rate": 3.9207404076833374e-05,
-      "logits/chosen": -14.0,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -832.0,
-      "logps/rejected": -4320.0,
-      "loss": 0.0009,
+      "logits/chosen": -16.375,
+      "logits/rejected": -9.5,
+      "logps/chosen": -808.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0025,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.9375,
-      "rewards/margins": 22.625,
-      "rewards/rejected": -26.625,
+      "rewards/chosen": -3.671875,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -21.25,
       "step": 1080
     },
     {
       "epoch": 1.1413612565445026,
-      "grad_norm": 0.00488064881806559,
+      "grad_norm": 0.009584182200972225,
       "learning_rate": 3.895567867990379e-05,
-      "logits/chosen": -14.25,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -832.0,
-      "logps/rejected": -4480.0,
-      "loss": 0.0058,
+      "logits/chosen": -16.375,
+      "logits/rejected": -9.5,
+      "logps/chosen": -808.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0063,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.953125,
-      "rewards/margins": 25.125,
-      "rewards/rejected": -29.125,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -22.625,
       "step": 1090
     },
     {
       "epoch": 1.1518324607329844,
-      "grad_norm": 0.017793955656379326,
+      "grad_norm": 0.007566812119930769,
       "learning_rate": 3.870188085471114e-05,
-      "logits/chosen": -14.4375,
-      "logits/rejected": -9.375,
-      "logps/chosen": -808.0,
-      "logps/rejected": -4512.0,
-      "loss": 0.0192,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.75,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0156,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -4.0625,
-      "rewards/margins": 26.125,
-      "rewards/rejected": -30.125,
+      "rewards/chosen": -3.796875,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -24.0,
       "step": 1100
     },
     {
       "epoch": 1.1518324607329844,
-      "eval_logits/chosen": -14.25,
-      "eval_logits/rejected": -14.0,
-      "eval_logps/chosen": -736.0,
-      "eval_logps/rejected": -744.0,
-      "eval_loss": 1.2316561937332153,
-      "eval_rewards/accuracies": 0.494140625,
-      "eval_rewards/chosen": -4.1875,
-      "eval_rewards/margins": 0.37109375,
-      "eval_rewards/rejected": -4.5625,
-      "eval_runtime": 46.608,
-      "eval_samples_per_second": 42.911,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -16.625,
+      "eval_logits/rejected": -16.5,
+      "eval_logps/chosen": -760.0,
+      "eval_logps/rejected": -748.0,
+      "eval_loss": 1.2452187538146973,
+      "eval_rewards/accuracies": 0.462890625,
+      "eval_rewards/chosen": -4.40625,
+      "eval_rewards/margins": 0.197265625,
+      "eval_rewards/rejected": -4.59375,
+      "eval_runtime": 58.2827,
+      "eval_samples_per_second": 34.316,
+      "eval_steps_per_second": 0.549,
       "step": 1100
     },
     {
       "epoch": 1.162303664921466,
-      "grad_norm": 0.047532137887866706,
+      "grad_norm": 0.02911325880325358,
       "learning_rate": 3.844604829041395e-05,
-      "logits/chosen": -14.3125,
-      "logits/rejected": -9.875,
-      "logps/chosen": -704.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.021,
+      "logits/chosen": -16.375,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -748.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0178,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.3125,
-      "rewards/margins": 19.75,
-      "rewards/rejected": -23.0,
+      "rewards/chosen": -3.734375,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -21.75,
       "step": 1110
     },
     {
       "epoch": 1.1727748691099475,
-      "grad_norm": 0.20795487571604057,
+      "grad_norm": 0.010022198392111673,
       "learning_rate": 3.818821897833095e-05,
-      "logits/chosen": -14.0625,
-      "logits/rejected": -10.125,
-      "logps/chosen": -692.0,
-      "logps/rejected": -3312.0,
-      "loss": 0.0087,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 15.125,
-      "rewards/rejected": -18.0,
+      "logits/chosen": -15.125,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3184.0,
+      "loss": 0.0053,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 13.375,
+      "rewards/rejected": -16.75,
       "step": 1120
     },
     {
       "epoch": 1.1832460732984293,
-      "grad_norm": 0.14485081443392359,
+      "grad_norm": 0.8141745191921965,
       "learning_rate": 3.792843120629935e-05,
-      "logits/chosen": -14.4375,
-      "logits/rejected": -9.25,
-      "logps/chosen": -672.0,
-      "logps/rejected": -3232.0,
-      "loss": 0.005,
+      "logits/chosen": -14.25,
+      "logits/rejected": -8.25,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3056.0,
+      "loss": 0.0052,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 13.875,
-      "rewards/rejected": -16.75,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -15.0,
       "step": 1130
     },
     {
       "epoch": 1.193717277486911,
-      "grad_norm": 0.015638355884919942,
+      "grad_norm": 0.5803312870363045,
       "learning_rate": 3.766672355298907e-05,
-      "logits/chosen": -14.3125,
-      "logits/rejected": -8.875,
-      "logps/chosen": -640.0,
-      "logps/rejected": -3280.0,
-      "loss": 0.007,
+      "logits/chosen": -14.25,
+      "logits/rejected": -8.125,
+      "logps/chosen": -696.0,
+      "logps/rejected": -3120.0,
+      "loss": 0.0069,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.765625,
-      "rewards/margins": 14.9375,
-      "rewards/rejected": -17.75,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -16.25,
       "step": 1140
     },
     {
       "epoch": 1.2041884816753927,
-      "grad_norm": 0.07115231891342906,
+      "grad_norm": 0.017578436805769707,
       "learning_rate": 3.740313488217373e-05,
-      "logits/chosen": -14.375,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -680.0,
-      "logps/rejected": -3296.0,
-      "loss": 0.0051,
+      "logits/chosen": -14.8125,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3168.0,
+      "loss": 0.0013,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.140625,
-      "rewards/margins": 15.875,
-      "rewards/rejected": -19.0,
+      "rewards/chosen": -3.484375,
+      "rewards/margins": 14.1875,
+      "rewards/rejected": -17.75,
       "step": 1150
     },
     {
       "epoch": 1.2146596858638743,
-      "grad_norm": 0.00543138144494391,
+      "grad_norm": 0.0009345576809731534,
       "learning_rate": 3.7137704336959464e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -740.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0019,
+      "logits/chosen": -14.125,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3504.0,
+      "loss": 0.0009,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.78125,
-      "rewards/margins": 17.375,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -3.828125,
+      "rewards/margins": 15.25,
+      "rewards/rejected": -19.125,
       "step": 1160
     },
     {
       "epoch": 1.225130890052356,
-      "grad_norm": 0.12363240002857107,
+      "grad_norm": 0.02404185189101972,
       "learning_rate": 3.687047133397201e-05,
-      "logits/chosen": -14.0,
-      "logits/rejected": -8.875,
-      "logps/chosen": -752.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0015,
+      "logits/chosen": -15.0,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -748.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0043,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.78125,
-      "rewards/margins": 18.5,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -3.75,
+      "rewards/margins": 15.75,
+      "rewards/rejected": -19.5,
       "step": 1170
     },
     {
       "epoch": 1.2356020942408377,
-      "grad_norm": 0.0685303511588328,
+      "grad_norm": 0.03747579944332716,
       "learning_rate": 3.6601475557503415e-05,
-      "logits/chosen": -13.9375,
-      "logits/rejected": -8.5625,
-      "logps/chosen": -788.0,
-      "logps/rejected": -4000.0,
-      "loss": 0.0107,
+      "logits/chosen": -15.625,
+      "logits/rejected": -8.25,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3536.0,
+      "loss": 0.0091,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.9375,
-      "rewards/margins": 20.5,
-      "rewards/rejected": -24.375,
+      "rewards/chosen": -3.1875,
+      "rewards/margins": 16.75,
+      "rewards/rejected": -19.875,
       "step": 1180
     },
     {
       "epoch": 1.2460732984293195,
-      "grad_norm": 0.21537455279483658,
+      "grad_norm": 0.22189544436590225,
       "learning_rate": 3.633075695361882e-05,
-      "logits/chosen": -14.0625,
-      "logits/rejected": -10.0,
-      "logps/chosen": -704.0,
-      "logps/rejected": -4000.0,
-      "loss": 0.0034,
+      "logits/chosen": -15.75,
+      "logits/rejected": -9.125,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0025,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 19.125,
-      "rewards/rejected": -22.5,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 17.375,
+      "rewards/rejected": -20.5,
       "step": 1190
     },
     {
       "epoch": 1.256544502617801,
-      "grad_norm": 0.06871009005234026,
+      "grad_norm": 0.25830477664045604,
       "learning_rate": 3.6058355724224477e-05,
-      "logits/chosen": -14.25,
-      "logits/rejected": -10.375,
-      "logps/chosen": -736.0,
-      "logps/rejected": -3824.0,
-      "loss": 0.0035,
+      "logits/chosen": -15.875,
+      "logits/rejected": -8.875,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0018,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5625,
-      "rewards/margins": 18.75,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -3.359375,
+      "rewards/margins": 17.125,
+      "rewards/rejected": -20.5,
       "step": 1200
     },
     {
       "epoch": 1.256544502617801,
-      "eval_logits/chosen": -14.0625,
-      "eval_logits/rejected": -13.6875,
-      "eval_logps/chosen": -700.0,
-      "eval_logps/rejected": -696.0,
-      "eval_loss": 1.1275312900543213,
-      "eval_rewards/accuracies": 0.51953125,
-      "eval_rewards/chosen": -3.8125,
-      "eval_rewards/margins": 0.267578125,
-      "eval_rewards/rejected": -4.09375,
-      "eval_runtime": 46.639,
-      "eval_samples_per_second": 42.883,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -15.875,
+      "eval_logits/rejected": -15.5625,
+      "eval_logps/chosen": -692.0,
+      "eval_logps/rejected": -680.0,
+      "eval_loss": 1.0496094226837158,
+      "eval_rewards/accuracies": 0.484375,
+      "eval_rewards/chosen": -3.734375,
+      "eval_rewards/margins": 0.1884765625,
+      "eval_rewards/rejected": -3.921875,
+      "eval_runtime": 58.2942,
+      "eval_samples_per_second": 34.309,
+      "eval_steps_per_second": 0.549,
       "step": 1200
     },
     {
       "epoch": 1.2670157068062826,
-      "grad_norm": 0.1746944937990447,
+      "grad_norm": 0.3326734600786279,
       "learning_rate": 3.5784312321097744e-05,
-      "logits/chosen": -14.5,
-      "logits/rejected": -10.5,
-      "logps/chosen": -732.0,
-      "logps/rejected": -3648.0,
-      "loss": 0.0019,
+      "logits/chosen": -16.25,
+      "logits/rejected": -8.5,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3568.0,
+      "loss": 0.0022,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.265625,
-      "rewards/margins": 18.75,
-      "rewards/rejected": -22.0,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -21.25,
       "step": 1210
     },
     {
       "epoch": 1.2774869109947644,
-      "grad_norm": 0.009156384968498658,
+      "grad_norm": 0.03148954765063839,
       "learning_rate": 3.550866743987994e-05,
-      "logits/chosen": -14.5,
-      "logits/rejected": -9.875,
-      "logps/chosen": -780.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.003,
+      "logits/chosen": -15.75,
+      "logits/rejected": -7.875,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0037,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.703125,
-      "rewards/margins": 20.5,
-      "rewards/rejected": -24.125,
+      "rewards/chosen": -3.421875,
+      "rewards/margins": 18.25,
+      "rewards/rejected": -21.625,
       "step": 1220
     },
     {
       "epoch": 1.2879581151832462,
-      "grad_norm": 0.03147888615727965,
+      "grad_norm": 0.01867307854443609,
       "learning_rate": 3.5231462014033024e-05,
-      "logits/chosen": -14.9375,
-      "logits/rejected": -10.625,
-      "logps/chosen": -688.0,
-      "logps/rejected": -3856.0,
-      "loss": 0.0812,
+      "logits/chosen": -16.5,
+      "logits/rejected": -8.75,
+      "logps/chosen": -644.0,
+      "logps/rejected": -3488.0,
+      "loss": 0.0015,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.1875,
-      "rewards/margins": 21.0,
-      "rewards/rejected": -24.25,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 17.875,
+      "rewards/rejected": -20.5,
       "step": 1230
     },
     {
       "epoch": 1.2984293193717278,
-      "grad_norm": 2.9792991133576354,
+      "grad_norm": 3.339824627059515,
       "learning_rate": 3.495273720876095e-05,
-      "logits/chosen": -14.6875,
-      "logits/rejected": -10.4375,
-      "logps/chosen": -736.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0134,
+      "logits/chosen": -15.6875,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0158,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 20.5,
-      "rewards/rejected": -23.875,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 17.875,
+      "rewards/rejected": -21.0,
       "step": 1240
     },
     {
       "epoch": 1.3089005235602094,
-      "grad_norm": 0.009977097170025909,
+      "grad_norm": 0.012621159925290962,
       "learning_rate": 3.4672534414896616e-05,
-      "logits/chosen": -14.9375,
-      "logits/rejected": -10.3125,
+      "logits/chosen": -15.5625,
+      "logits/rejected": -7.9375,
       "logps/chosen": -752.0,
-      "logps/rejected": -3856.0,
-      "loss": 0.0181,
+      "logps/rejected": -3504.0,
+      "loss": 0.002,
       "rewards/accuracies": 1.0,
       "rewards/chosen": -3.28125,
-      "rewards/margins": 20.25,
-      "rewards/rejected": -23.5,
+      "rewards/margins": 16.75,
+      "rewards/rejected": -20.0,
       "step": 1250
     },
     {
       "epoch": 1.3193717277486912,
-      "grad_norm": 0.0038157381375055195,
+      "grad_norm": 0.006486458242010127,
       "learning_rate": 3.439089524275526e-05,
-      "logits/chosen": -14.375,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -688.0,
-      "logps/rejected": -3760.0,
-      "loss": 0.0046,
+      "logits/chosen": -15.0,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3504.0,
+      "loss": 0.0014,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.96875,
-      "rewards/margins": 19.0,
-      "rewards/rejected": -21.875,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -19.25,
       "step": 1260
     },
     {
       "epoch": 1.3298429319371727,
-      "grad_norm": 0.00784451955469286,
+      "grad_norm": 0.008868721539240463,
       "learning_rate": 3.410786151595533e-05,
-      "logits/chosen": -13.8125,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -660.0,
-      "logps/rejected": -3632.0,
-      "loss": 0.0016,
+      "logits/chosen": -15.125,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -636.0,
+      "logps/rejected": -3360.0,
+      "loss": 0.0015,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.84375,
-      "rewards/margins": 18.75,
-      "rewards/rejected": -21.625,
+      "rewards/chosen": -2.578125,
+      "rewards/margins": 16.25,
+      "rewards/rejected": -18.875,
       "step": 1270
     },
     {
       "epoch": 1.3403141361256545,
-      "grad_norm": 0.3186579558758453,
+      "grad_norm": 0.04650516106426507,
       "learning_rate": 3.382347526520765e-05,
-      "logits/chosen": -12.75,
-      "logits/rejected": -6.6875,
-      "logps/chosen": -728.0,
-      "logps/rejected": -3904.0,
-      "loss": 0.0126,
+      "logits/chosen": -15.875,
+      "logits/rejected": -6.59375,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3472.0,
+      "loss": 0.0026,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.296875,
-      "rewards/margins": 20.5,
-      "rewards/rejected": -23.75,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -19.375,
       "step": 1280
     },
     {
       "epoch": 1.350785340314136,
-      "grad_norm": 0.010167443300278703,
+      "grad_norm": 0.006762033774200141,
       "learning_rate": 3.353777872207381e-05,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -6.4375,
-      "logps/chosen": -720.0,
-      "logps/rejected": -3888.0,
-      "loss": 0.0014,
+      "logits/chosen": -15.6875,
+      "logits/rejected": -7.09375,
+      "logps/chosen": -664.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0013,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 20.75,
-      "rewards/rejected": -24.0,
+      "rewards/chosen": -2.640625,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -19.75,
       "step": 1290
     },
     {
       "epoch": 1.3612565445026177,
-      "grad_norm": 0.17740553102425882,
+      "grad_norm": 0.026810506177580448,
       "learning_rate": 3.325081431269475e-05,
-      "logits/chosen": -10.75,
-      "logits/rejected": -6.6875,
-      "logps/chosen": -688.0,
-      "logps/rejected": -4032.0,
-      "loss": 0.0014,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -7.15625,
+      "logps/chosen": -628.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0046,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.328125,
-      "rewards/margins": 21.125,
-      "rewards/rejected": -24.375,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 17.5,
+      "rewards/rejected": -20.25,
       "step": 1300
     },
     {
       "epoch": 1.3612565445026177,
-      "eval_logits/chosen": -11.0,
-      "eval_logits/rejected": -10.3125,
-      "eval_logps/chosen": -700.0,
-      "eval_logps/rejected": -708.0,
-      "eval_loss": 1.1071875095367432,
-      "eval_rewards/accuracies": 0.50390625,
-      "eval_rewards/chosen": -3.828125,
-      "eval_rewards/margins": 0.3671875,
-      "eval_rewards/rejected": -4.1875,
-      "eval_runtime": 46.6275,
-      "eval_samples_per_second": 42.893,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -15.25,
+      "eval_logits/rejected": -14.9375,
+      "eval_logps/chosen": -656.0,
+      "eval_logps/rejected": -648.0,
+      "eval_loss": 1.0483750104904175,
+      "eval_rewards/accuracies": 0.498046875,
+      "eval_rewards/chosen": -3.375,
+      "eval_rewards/margins": 0.240234375,
+      "eval_rewards/rejected": -3.609375,
+      "eval_runtime": 58.2841,
+      "eval_samples_per_second": 34.315,
+      "eval_steps_per_second": 0.549,
       "step": 1300
     },
     {
       "epoch": 1.3717277486910995,
-      "grad_norm": 0.08233979413254028,
+      "grad_norm": 0.0058493648356022125,
       "learning_rate": 3.296262465149046e-05,
-      "logits/chosen": -12.375,
-      "logits/rejected": -7.5625,
-      "logps/chosen": -756.0,
-      "logps/rejected": -3984.0,
-      "loss": 0.0038,
+      "logits/chosen": -15.125,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0023,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 20.625,
-      "rewards/rejected": -24.125,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -21.875,
       "step": 1310
     },
     {
       "epoch": 1.3821989528795813,
-      "grad_norm": 0.0001292407479547709,
+      "grad_norm": 0.6725534076934339,
       "learning_rate": 3.2673252534831687e-05,
-      "logits/chosen": -12.9375,
-      "logits/rejected": -8.125,
-      "logps/chosen": -716.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.0104,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.3125,
-      "rewards/margins": 20.25,
-      "rewards/rejected": -23.625,
+      "logits/chosen": -15.25,
+      "logits/rejected": -7.875,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0021,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 19.625,
+      "rewards/rejected": -22.5,
       "step": 1320
     },
     {
       "epoch": 1.3926701570680629,
-      "grad_norm": 0.014103322835807182,
+      "grad_norm": 0.0018101752565744089,
       "learning_rate": 3.2382740934684696e-05,
-      "logits/chosen": -12.8125,
-      "logits/rejected": -7.625,
-      "logps/chosen": -724.0,
-      "logps/rejected": -4352.0,
-      "loss": 0.0075,
+      "logits/chosen": -15.0,
+      "logits/rejected": -6.84375,
+      "logps/chosen": -696.0,
+      "logps/rejected": -4032.0,
+      "loss": 0.0136,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 21.125,
+      "rewards/rejected": -24.375,
       "step": 1330
     },
     {
       "epoch": 1.4031413612565444,
-      "grad_norm": 0.08964854484092108,
+      "grad_norm": 0.6280645356588488,
       "learning_rate": 3.209113299222982e-05,
-      "logits/chosen": -12.0,
-      "logits/rejected": -7.03125,
-      "logps/chosen": -688.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0083,
+      "logits/chosen": -14.625,
+      "logits/rejected": -6.59375,
+      "logps/chosen": -628.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0116,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -2.9375,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.25,
       "step": 1340
     },
     {
       "epoch": 1.4136125654450262,
-      "grad_norm": 0.006512173165062479,
+      "grad_norm": 0.005478057676279359,
       "learning_rate": 3.179847201145505e-05,
-      "logits/chosen": -12.125,
-      "logits/rejected": -6.9375,
-      "logps/chosen": -780.0,
-      "logps/rejected": -4352.0,
-      "loss": 0.0016,
+      "logits/chosen": -14.0,
+      "logits/rejected": -6.09375,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.78125,
-      "rewards/margins": 22.75,
-      "rewards/rejected": -26.625,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 18.375,
+      "rewards/rejected": -21.5,
       "step": 1350
     },
     {
       "epoch": 1.4240837696335078,
-      "grad_norm": 0.007417549973581045,
+      "grad_norm": 0.07112654034965081,
       "learning_rate": 3.150480145272528e-05,
-      "logits/chosen": -12.75,
-      "logits/rejected": -7.0,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4064.0,
-      "loss": 0.0199,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -5.78125,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0038,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.6875,
-      "rewards/margins": 21.25,
-      "rewards/rejected": -25.0,
+      "rewards/chosen": -3.359375,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -22.25,
       "step": 1360
     },
     {
       "epoch": 1.4345549738219896,
-      "grad_norm": 0.0024468649289336964,
+      "grad_norm": 0.009229335731704631,
       "learning_rate": 3.121016492632848e-05,
-      "logits/chosen": -13.75,
-      "logits/rejected": -7.59375,
-      "logps/chosen": -724.0,
-      "logps/rejected": -3744.0,
-      "loss": 0.0017,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -5.65625,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0013,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 19.125,
-      "rewards/rejected": -22.375,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 19.5,
+      "rewards/rejected": -22.75,
       "step": 1370
     },
     {
       "epoch": 1.4450261780104712,
-      "grad_norm": 0.008576142450597167,
+      "grad_norm": 0.08762861683253026,
       "learning_rate": 3.091460618599951e-05,
-      "logits/chosen": -14.625,
-      "logits/rejected": -7.8125,
-      "logps/chosen": -752.0,
-      "logps/rejected": -3712.0,
-      "loss": 0.0034,
+      "logits/chosen": -13.75,
+      "logits/rejected": -5.125,
+      "logps/chosen": -696.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0025,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.625,
-      "rewards/margins": 19.875,
-      "rewards/rejected": -23.5,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 19.5,
+      "rewards/rejected": -22.5,
       "step": 1380
     },
     {
       "epoch": 1.4554973821989527,
-      "grad_norm": 0.00437585589580392,
+      "grad_norm": 0.010007788501371465,
       "learning_rate": 3.061816912242265e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.0,
-      "logps/chosen": -768.0,
-      "logps/rejected": -4048.0,
-      "loss": 0.0024,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.640625,
-      "rewards/margins": 21.125,
-      "rewards/rejected": -24.75,
+      "logits/chosen": -14.0625,
+      "logits/rejected": -4.96875,
+      "logps/chosen": -724.0,
+      "logps/rejected": -4064.0,
+      "loss": 0.0019,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 21.75,
+      "rewards/rejected": -24.875,
       "step": 1390
     },
     {
       "epoch": 1.4659685863874345,
-      "grad_norm": 0.002291926262623951,
+      "grad_norm": 0.0002242797151445476,
       "learning_rate": 3.0320897756713785e-05,
-      "logits/chosen": -15.6875,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -760.0,
-      "logps/rejected": -4192.0,
-      "loss": 0.0009,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.609375,
+      "logits/chosen": -14.125,
+      "logits/rejected": -6.28125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -4160.0,
+      "loss": 0.0041,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.0625,
       "rewards/margins": 22.0,
-      "rewards/rejected": -25.625,
+      "rewards/rejected": -25.125,
       "step": 1400
     },
     {
       "epoch": 1.4659685863874345,
-      "eval_logits/chosen": -16.0,
-      "eval_logits/rejected": -15.6875,
-      "eval_logps/chosen": -732.0,
-      "eval_logps/rejected": -748.0,
-      "eval_loss": 1.2157812118530273,
-      "eval_rewards/accuracies": 0.50390625,
-      "eval_rewards/chosen": -4.15625,
-      "eval_rewards/margins": 0.45703125,
-      "eval_rewards/rejected": -4.59375,
-      "eval_runtime": 46.623,
-      "eval_samples_per_second": 42.897,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -14.3125,
+      "eval_logits/rejected": -13.8125,
+      "eval_logps/chosen": -668.0,
+      "eval_logps/rejected": -676.0,
+      "eval_loss": 0.9980156421661377,
+      "eval_rewards/accuracies": 0.513671875,
+      "eval_rewards/chosen": -3.515625,
+      "eval_rewards/margins": 0.337890625,
+      "eval_rewards/rejected": -3.84375,
+      "eval_runtime": 58.3095,
+      "eval_samples_per_second": 34.3,
+      "eval_steps_per_second": 0.549,
       "step": 1400
     },
     {
       "epoch": 1.4764397905759163,
-      "grad_norm": 0.05542532840570852,
+      "grad_norm": 0.016452827145967814,
       "learning_rate": 3.002283623388332e-05,
-      "logits/chosen": -15.625,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -740.0,
-      "logps/rejected": -4000.0,
-      "loss": 0.0136,
+      "logits/chosen": -14.5,
+      "logits/rejected": -6.3125,
+      "logps/chosen": -656.0,
+      "logps/rejected": -3936.0,
+      "loss": 0.0114,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.859375,
-      "rewards/margins": 21.5,
-      "rewards/rejected": -25.375,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 21.625,
+      "rewards/rejected": -24.625,
       "step": 1410
     },
     {
       "epoch": 1.486910994764398,
-      "grad_norm": 1.0152821575564834,
+      "grad_norm": 2.4850971842383194,
       "learning_rate": 2.9724028816280507e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -10.125,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4096.0,
-      "loss": 0.0055,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -6.90625,
+      "logps/chosen": -664.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0158,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.84375,
-      "rewards/margins": 21.875,
-      "rewards/rejected": -25.75,
+      "rewards/chosen": -2.9375,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.375,
       "step": 1420
     },
     {
       "epoch": 1.4973821989528795,
-      "grad_norm": 0.022090154265961406,
+      "grad_norm": 0.021777963803115665,
       "learning_rate": 2.9424519877020524e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -10.0,
-      "logps/chosen": -772.0,
-      "logps/rejected": -4032.0,
-      "loss": 0.0028,
+      "logits/chosen": -15.4375,
+      "logits/rejected": -7.28125,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3520.0,
+      "loss": 0.0036,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 21.375,
-      "rewards/rejected": -25.0,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 17.125,
+      "rewards/rejected": -19.875,
       "step": 1430
     },
     {
-      "epoch": 1.5078534031413613,
-      "grad_norm": 0.025442900489331354,
-      "learning_rate": 2.912435389339504e-05,
-      "logits/chosen": -15.6875,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -732.0,
-      "logps/rejected": -3856.0,
-      "loss": 0.0174,
+      "epoch": 1.5078534031413613,
+      "grad_norm": 0.0022774268344216996,
+      "learning_rate": 2.912435389339504e-05,
+      "logits/chosen": -15.3125,
+      "logits/rejected": -6.625,
+      "logps/chosen": -640.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0049,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.453125,
-      "rewards/margins": 19.75,
-      "rewards/rejected": -23.25,
+      "rewards/chosen": -2.53125,
+      "rewards/margins": 16.75,
+      "rewards/rejected": -19.25,
       "step": 1440
     },
     {
       "epoch": 1.518324607329843,
-      "grad_norm": 0.0020182052090820793,
+      "grad_norm": 0.004488674415128893,
       "learning_rate": 2.8823575440267226e-05,
-      "logits/chosen": -15.625,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0084,
+      "logits/chosen": -15.25,
+      "logits/rejected": -6.375,
+      "logps/chosen": -648.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0024,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 21.375,
-      "rewards/rejected": -24.875,
+      "rewards/chosen": -2.703125,
+      "rewards/margins": 18.25,
+      "rewards/rejected": -21.0,
       "step": 1450
     },
     {
       "epoch": 1.5287958115183247,
-      "grad_norm": 0.06959797078175446,
+      "grad_norm": 0.023025944800861224,
       "learning_rate": 2.8522229183452465e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -10.625,
-      "logps/chosen": -720.0,
-      "logps/rejected": -3712.0,
-      "loss": 0.0358,
+      "logits/chosen": -15.5625,
+      "logits/rejected": -6.75,
+      "logps/chosen": -668.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0369,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.203125,
-      "rewards/margins": 18.125,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -2.703125,
+      "rewards/margins": 17.375,
+      "rewards/rejected": -20.125,
       "step": 1460
     },
     {
       "epoch": 1.5392670157068062,
-      "grad_norm": 0.456908129655633,
+      "grad_norm": 0.05922584640799251,
       "learning_rate": 2.8220359873085373e-05,
-      "logits/chosen": -16.375,
-      "logits/rejected": -11.9375,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3200.0,
-      "loss": 0.0098,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.890625,
-      "rewards/margins": 13.3125,
-      "rewards/rejected": -16.25,
+      "logits/chosen": -15.5625,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -656.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0029,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.609375,
+      "rewards/margins": 14.3125,
+      "rewards/rejected": -17.0,
       "step": 1470
     },
     {
       "epoch": 1.5497382198952878,
-      "grad_norm": 0.16958749551298508,
+      "grad_norm": 0.17754395438600357,
       "learning_rate": 2.7918012336974388e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -11.1875,
-      "logps/chosen": -644.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.006,
+      "logits/chosen": -15.75,
+      "logits/rejected": -7.8125,
+      "logps/chosen": -636.0,
+      "logps/rejected": -3184.0,
+      "loss": 0.0023,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.5625,
-      "rewards/margins": 11.8125,
-      "rewards/rejected": -14.375,
+      "rewards/chosen": -2.4375,
+      "rewards/margins": 14.9375,
+      "rewards/rejected": -17.375,
       "step": 1480
     },
     {
       "epoch": 1.5602094240837696,
-      "grad_norm": 0.018763306908892816,
+      "grad_norm": 0.015737554521410745,
       "learning_rate": 2.7615231473944815e-05,
-      "logits/chosen": -15.75,
-      "logits/rejected": -10.25,
-      "logps/chosen": -648.0,
-      "logps/rejected": -3120.0,
-      "loss": 0.0079,
+      "logits/chosen": -15.0625,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -636.0,
+      "logps/rejected": -3392.0,
+      "loss": 0.0042,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.796875,
-      "rewards/margins": 13.75,
-      "rewards/rejected": -16.5,
+      "rewards/chosen": -2.65625,
+      "rewards/margins": 16.625,
+      "rewards/rejected": -19.25,
       "step": 1490
     },
     {
       "epoch": 1.5706806282722514,
-      "grad_norm": 0.08965357050731275,
+      "grad_norm": 0.06992059734347655,
       "learning_rate": 2.7312062247171327e-05,
-      "logits/chosen": -15.375,
-      "logits/rejected": -10.4375,
-      "logps/chosen": -664.0,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -8.5,
+      "logps/chosen": -644.0,
       "logps/rejected": -3360.0,
-      "loss": 0.0047,
+      "loss": 0.0077,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.03125,
-      "rewards/margins": 16.0,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 16.125,
       "rewards/rejected": -19.0,
       "step": 1500
     },
     {
       "epoch": 1.5706806282722514,
-      "eval_logits/chosen": -15.0625,
-      "eval_logits/rejected": -14.625,
-      "eval_logps/chosen": -660.0,
-      "eval_logps/rejected": -664.0,
-      "eval_loss": 0.9803749918937683,
-      "eval_rewards/accuracies": 0.5,
-      "eval_rewards/chosen": -3.40625,
-      "eval_rewards/margins": 0.3671875,
-      "eval_rewards/rejected": -3.765625,
-      "eval_runtime": 46.6121,
-      "eval_samples_per_second": 42.907,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -14.375,
+      "eval_logits/rejected": -13.875,
+      "eval_logps/chosen": -636.0,
+      "eval_logps/rejected": -640.0,
+      "eval_loss": 1.043375015258789,
+      "eval_rewards/accuracies": 0.490234375,
+      "eval_rewards/chosen": -3.171875,
+      "eval_rewards/margins": 0.353515625,
+      "eval_rewards/rejected": -3.515625,
+      "eval_runtime": 58.3075,
+      "eval_samples_per_second": 34.301,
+      "eval_steps_per_second": 0.549,
       "step": 1500
     },
     {
       "epoch": 1.581151832460733,
-      "grad_norm": 0.06452648807774522,
+      "grad_norm": 0.09946772266223687,
       "learning_rate": 2.7008549677500876e-05,
-      "logits/chosen": -15.4375,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -720.0,
-      "logps/rejected": -3616.0,
-      "loss": 0.0031,
+      "logits/chosen": -15.125,
+      "logits/rejected": -7.875,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0027,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.078125,
-      "rewards/margins": 16.75,
-      "rewards/rejected": -19.875,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 14.75,
+      "rewards/rejected": -17.75,
       "step": 1510
     },
     {
       "epoch": 1.5916230366492146,
-      "grad_norm": 0.006834320487066915,
+      "grad_norm": 0.06393586501249479,
       "learning_rate": 2.670473883676709e-05,
-      "logits/chosen": -15.0,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0015,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.890625,
-      "rewards/margins": 15.75,
-      "rewards/rejected": -18.625,
+      "rewards/chosen": -3.0625,
+      "rewards/margins": 14.625,
+      "rewards/rejected": -17.75,
       "step": 1520
     },
     {
       "epoch": 1.6020942408376964,
-      "grad_norm": 0.06759439088614758,
+      "grad_norm": 0.018838783444432978,
       "learning_rate": 2.6400674841097074e-05,
-      "logits/chosen": -15.6875,
-      "logits/rejected": -10.375,
-      "logps/chosen": -720.0,
-      "logps/rejected": -3856.0,
-      "loss": 0.0009,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0013,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.1875,
-      "rewards/margins": 18.5,
-      "rewards/rejected": -21.75,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 16.75,
+      "rewards/rejected": -20.0,
       "step": 1530
     },
     {
       "epoch": 1.6125654450261782,
-      "grad_norm": 0.2627000696085257,
+      "grad_norm": 1.6226593606979909,
       "learning_rate": 2.609640284421158e-05,
-      "logits/chosen": -15.6875,
-      "logits/rejected": -10.25,
-      "logps/chosen": -656.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0037,
+      "logits/chosen": -15.4375,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -652.0,
+      "logps/rejected": -3488.0,
+      "loss": 0.0054,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.6875,
-      "rewards/margins": 18.75,
-      "rewards/rejected": -21.375,
+      "rewards/chosen": -2.640625,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -19.0,
       "step": 1540
     },
     {
       "epoch": 1.6230366492146597,
-      "grad_norm": 0.0007969231982829138,
+      "grad_norm": 0.003479145576288751,
       "learning_rate": 2.5791968030719726e-05,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.25,
-      "logps/chosen": -732.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.0026,
+      "logits/chosen": -15.5625,
+      "logits/rejected": -7.78125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3472.0,
+      "loss": 0.0048,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.078125,
-      "rewards/margins": 20.75,
-      "rewards/rejected": -23.875,
+      "rewards/chosen": -2.8125,
+      "rewards/margins": 17.375,
+      "rewards/rejected": -20.25,
       "step": 1550
     },
     {
       "epoch": 1.6335078534031413,
-      "grad_norm": 0.08115453310117013,
+      "grad_norm": 0.014043617718315944,
       "learning_rate": 2.548741560940902e-05,
-      "logits/chosen": -15.1875,
-      "logits/rejected": -8.875,
-      "logps/chosen": -700.0,
-      "logps/rejected": -4096.0,
+      "logits/chosen": -14.875,
+      "logits/rejected": -7.34375,
+      "logps/chosen": -668.0,
+      "logps/rejected": -3744.0,
       "loss": 0.0025,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.234375,
-      "rewards/margins": 21.625,
-      "rewards/rejected": -24.875,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 18.375,
+      "rewards/rejected": -21.375,
       "step": 1560
     },
     {
       "epoch": 1.6439790575916229,
-      "grad_norm": 0.12294771473768257,
+      "grad_norm": 0.07679948982186975,
       "learning_rate": 2.5182790806531787e-05,
-      "logits/chosen": -15.375,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -760.0,
-      "logps/rejected": -3824.0,
-      "loss": 0.0048,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 20.5,
-      "rewards/rejected": -24.0,
+      "logits/chosen": -14.125,
+      "logits/rejected": -8.25,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0034,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.421875,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -22.625,
       "step": 1570
     },
     {
       "epoch": 1.6544502617801047,
-      "grad_norm": 0.09859075545652682,
+      "grad_norm": 0.10883293257073702,
       "learning_rate": 2.4878138859089074e-05,
-      "logits/chosen": -15.5,
-      "logits/rejected": -9.375,
-      "logps/chosen": -692.0,
-      "logps/rejected": -4048.0,
-      "loss": 0.005,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.015625,
-      "rewards/margins": 21.875,
-      "rewards/rejected": -24.875,
+      "logits/chosen": -14.0,
+      "logits/rejected": -7.625,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3968.0,
+      "loss": 0.0024,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.359375,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -24.0,
       "step": 1580
     },
     {
       "epoch": 1.6649214659685865,
-      "grad_norm": 0.0059218657013956275,
+      "grad_norm": 0.041327688810381506,
       "learning_rate": 2.4573505008112924e-05,
-      "logits/chosen": -14.0,
-      "logits/rejected": -8.75,
-      "logps/chosen": -724.0,
-      "logps/rejected": -4000.0,
+      "logits/chosen": -14.4375,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3744.0,
       "loss": 0.0006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 20.875,
-      "rewards/rejected": -24.25,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 18.5,
+      "rewards/rejected": -21.75,
       "step": 1590
     },
     {
       "epoch": 1.675392670157068,
-      "grad_norm": 0.012763355039026302,
+      "grad_norm": 0.0032810194722477926,
       "learning_rate": 2.4268934491948032e-05,
-      "logits/chosen": -12.375,
-      "logits/rejected": -8.25,
-      "logps/chosen": -788.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0009,
+      "logits/chosen": -13.375,
+      "logits/rejected": -7.75,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0016,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.875,
-      "rewards/margins": 21.875,
-      "rewards/rejected": -25.75,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -22.5,
       "step": 1600
     },
     {
       "epoch": 1.675392670157068,
-      "eval_logits/chosen": -11.4375,
-      "eval_logits/rejected": -10.6875,
-      "eval_logps/chosen": -724.0,
-      "eval_logps/rejected": -736.0,
-      "eval_loss": 1.0340312719345093,
-      "eval_rewards/accuracies": 0.513671875,
-      "eval_rewards/chosen": -4.03125,
-      "eval_rewards/margins": 0.421875,
-      "eval_rewards/rejected": -4.46875,
-      "eval_runtime": 46.6183,
-      "eval_samples_per_second": 42.902,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -12.9375,
+      "eval_logits/rejected": -12.4375,
+      "eval_logps/chosen": -704.0,
+      "eval_logps/rejected": -716.0,
+      "eval_loss": 1.0881874561309814,
+      "eval_rewards/accuracies": 0.4921875,
+      "eval_rewards/chosen": -3.859375,
+      "eval_rewards/margins": 0.4140625,
+      "eval_rewards/rejected": -4.28125,
+      "eval_runtime": 58.2867,
+      "eval_samples_per_second": 34.313,
+      "eval_steps_per_second": 0.549,
       "step": 1600
     },
     {
       "epoch": 1.6858638743455496,
-      "grad_norm": 0.004842817104352102,
+      "grad_norm": 0.033028521927530974,
       "learning_rate": 2.396447253953385e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -764.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0057,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 22.0,
-      "rewards/rejected": -25.625,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -748.0,
+      "logps/rejected": -4080.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.375,
+      "rewards/margins": 21.25,
+      "rewards/rejected": -24.625,
       "step": 1610
     },
     {
       "epoch": 1.6963350785340314,
-      "grad_norm": 0.6831253458137544,
+      "grad_norm": 0.01041571358803203,
       "learning_rate": 2.3660164363687998e-05,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -7.53125,
-      "logps/chosen": -796.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.0022,
+      "logits/chosen": -14.0,
+      "logits/rejected": -7.3125,
+      "logps/chosen": -772.0,
+      "logps/rejected": -4000.0,
+      "loss": 0.0051,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.6875,
-      "rewards/margins": 18.875,
-      "rewards/rejected": -22.5,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 20.75,
+      "rewards/rejected": -24.25,
       "step": 1620
     },
     {
-      "epoch": 1.7068062827225132,
-      "grad_norm": 0.08770591301485227,
-      "learning_rate": 2.33560551543922e-05,
-      "logits/chosen": -13.3125,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -784.0,
-      "logps/rejected": -3632.0,
-      "loss": 0.001,
+      "epoch": 1.7068062827225132,
+      "grad_norm": 0.05800760554757351,
+      "learning_rate": 2.33560551543922e-05,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -7.59375,
+      "logps/chosen": -764.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0038,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 18.5,
-      "rewards/rejected": -22.0,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 19.625,
+      "rewards/rejected": -22.875,
       "step": 1630
     },
     {
       "epoch": 1.7172774869109948,
-      "grad_norm": 0.06579138395815623,
+      "grad_norm": 0.02780515571834781,
       "learning_rate": 2.3052190072081493e-05,
-      "logits/chosen": -14.25,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -716.0,
-      "logps/rejected": -3648.0,
-      "loss": 0.0008,
+      "logits/chosen": -14.5,
+      "logits/rejected": -8.375,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0014,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 17.875,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 18.5,
+      "rewards/rejected": -21.5,
       "step": 1640
     },
     {
       "epoch": 1.7277486910994764,
-      "grad_norm": 0.026938513311102072,
+      "grad_norm": 0.007195766352180891,
       "learning_rate": 2.2748614240937864e-05,
-      "logits/chosen": -13.6875,
-      "logits/rejected": -9.25,
-      "logps/chosen": -672.0,
-      "logps/rejected": -3632.0,
-      "loss": 0.0048,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -644.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0043,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.09375,
-      "rewards/margins": 18.125,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 18.5,
       "rewards/rejected": -21.25,
       "step": 1650
     },
     {
       "epoch": 1.738219895287958,
-      "grad_norm": 0.0071203481038201915,
+      "grad_norm": 0.01626853234543758,
       "learning_rate": 2.2445372742189334e-05,
-      "logits/chosen": -15.25,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -720.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.0004,
+      "logits/chosen": -14.375,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.203125,
-      "rewards/margins": 18.625,
-      "rewards/rejected": -21.875,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 18.5,
+      "rewards/rejected": -21.375,
       "step": 1660
     },
     {
       "epoch": 1.7486910994764397,
-      "grad_norm": 0.0016386403399205496,
+      "grad_norm": 0.0008831436004743098,
       "learning_rate": 2.214251060741528e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -10.5,
-      "logps/chosen": -764.0,
-      "logps/rejected": -3600.0,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -7.90625,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3520.0,
       "loss": 0.0014,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 18.125,
-      "rewards/rejected": -21.625,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 17.875,
+      "rewards/rejected": -20.875,
       "step": 1670
     },
     {
       "epoch": 1.7591623036649215,
-      "grad_norm": 0.023531711850644822,
+      "grad_norm": 0.5921269634842338,
       "learning_rate": 2.1840072811859302e-05,
-      "logits/chosen": -16.5,
-      "logits/rejected": -10.8125,
-      "logps/chosen": -748.0,
-      "logps/rejected": -3648.0,
-      "loss": 0.0022,
+      "logits/chosen": -15.5,
+      "logits/rejected": -8.125,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3552.0,
+      "loss": 0.0046,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 18.625,
-      "rewards/rejected": -22.0,
+      "rewards/chosen": -2.6875,
+      "rewards/margins": 18.25,
+      "rewards/rejected": -21.0,
       "step": 1680
     },
     {
       "epoch": 1.7696335078534031,
-      "grad_norm": 0.6679285730418462,
+      "grad_norm": 0.07651078377164104,
       "learning_rate": 2.1538104267750287e-05,
-      "logits/chosen": -16.75,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -676.0,
-      "logps/rejected": -3872.0,
-      "loss": 0.0012,
+      "logits/chosen": -15.5,
+      "logits/rejected": -8.5,
+      "logps/chosen": -640.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.078125,
-      "rewards/margins": 19.375,
-      "rewards/rejected": -22.5,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 19.25,
+      "rewards/rejected": -21.875,
       "step": 1690
     },
     {
       "epoch": 1.7801047120418847,
-      "grad_norm": 0.0061073985389830335,
+      "grad_norm": 0.0005663462937260947,
       "learning_rate": 2.1236649817632952e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -11.25,
-      "logps/chosen": -668.0,
+      "logits/chosen": -15.6875,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -644.0,
       "logps/rejected": -3776.0,
-      "loss": 0.0053,
+      "loss": 0.0042,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 19.0,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 19.25,
       "rewards/rejected": -22.25,
       "step": 1700
     },
     {
       "epoch": 1.7801047120418847,
-      "eval_logits/chosen": -16.25,
-      "eval_logits/rejected": -16.125,
-      "eval_logps/chosen": -664.0,
-      "eval_logps/rejected": -672.0,
-      "eval_loss": 0.9808124899864197,
-      "eval_rewards/accuracies": 0.521484375,
-      "eval_rewards/chosen": -3.453125,
-      "eval_rewards/margins": 0.373046875,
-      "eval_rewards/rejected": -3.8125,
-      "eval_runtime": 46.6133,
-      "eval_samples_per_second": 42.906,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -15.9375,
+      "eval_logits/rejected": -15.5,
+      "eval_logps/chosen": -652.0,
+      "eval_logps/rejected": -664.0,
+      "eval_loss": 1.0260624885559082,
+      "eval_rewards/accuracies": 0.494140625,
+      "eval_rewards/chosen": -3.34375,
+      "eval_rewards/margins": 0.423828125,
+      "eval_rewards/rejected": -3.765625,
+      "eval_runtime": 58.2837,
+      "eval_samples_per_second": 34.315,
+      "eval_steps_per_second": 0.549,
       "step": 1700
     },
     {
       "epoch": 1.7905759162303665,
-      "grad_norm": 0.7302477733226329,
+      "grad_norm": 0.10448488283304017,
       "learning_rate": 2.093575422770872e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -10.9375,
-      "logps/chosen": -652.0,
-      "logps/rejected": -3632.0,
-      "loss": 0.0039,
+      "logits/chosen": -15.625,
+      "logits/rejected": -8.125,
+      "logps/chosen": -636.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0036,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.953125,
-      "rewards/margins": 18.5,
-      "rewards/rejected": -21.5,
+      "rewards/chosen": -2.765625,
+      "rewards/margins": 19.625,
+      "rewards/rejected": -22.375,
       "step": 1710
     },
     {
       "epoch": 1.8010471204188483,
-      "grad_norm": 0.051174201301530685,
+      "grad_norm": 0.017697047485813567,
       "learning_rate": 2.0635462181187827e-05,
-      "logits/chosen": -15.875,
-      "logits/rejected": -10.8125,
-      "logps/chosen": -712.0,
+      "logits/chosen": -16.0,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -672.0,
       "logps/rejected": -3744.0,
-      "loss": 0.0061,
+      "loss": 0.0024,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.171875,
-      "rewards/margins": 19.125,
+      "rewards/chosen": -2.78125,
+      "rewards/margins": 19.5,
       "rewards/rejected": -22.25,
       "step": 1720
     },
     {
       "epoch": 1.8115183246073299,
-      "grad_norm": 0.02741825084818544,
+      "grad_norm": 0.036136423955681736,
       "learning_rate": 2.0335818271653917e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -11.0625,
-      "logps/chosen": -720.0,
-      "logps/rejected": -3760.0,
-      "loss": 0.0045,
+      "logits/chosen": -15.8125,
+      "logits/rejected": -8.5625,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0037,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.1875,
-      "rewards/margins": 18.5,
-      "rewards/rejected": -21.625,
+      "rewards/chosen": -2.8125,
+      "rewards/margins": 18.75,
+      "rewards/rejected": -21.5,
       "step": 1730
     },
     {
       "epoch": 1.8219895287958114,
-      "grad_norm": 0.40046550655888324,
+      "grad_norm": 0.34182025065014277,
       "learning_rate": 2.0036866996441815e-05,
-      "logits/chosen": -16.375,
-      "logits/rejected": -11.375,
-      "logps/chosen": -688.0,
-      "logps/rejected": -3872.0,
-      "loss": 0.0079,
+      "logits/chosen": -16.25,
+      "logits/rejected": -8.875,
+      "logps/chosen": -680.0,
+      "logps/rejected": -4016.0,
+      "loss": 0.0028,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.03125,
-      "rewards/margins": 18.75,
-      "rewards/rejected": -21.75,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -23.25,
       "step": 1740
     },
     {
       "epoch": 1.8324607329842932,
-      "grad_norm": 0.027147558716150284,
+      "grad_norm": 0.009434568082569962,
       "learning_rate": 1.9738652750029677e-05,
-      "logits/chosen": -15.875,
-      "logits/rejected": -11.25,
-      "logps/chosen": -736.0,
-      "logps/rejected": -3456.0,
-      "loss": 0.002,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.0021,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.203125,
-      "rewards/margins": 16.75,
-      "rewards/rejected": -19.875,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 19.0,
+      "rewards/rejected": -22.125,
       "step": 1750
     },
     {
       "epoch": 1.8429319371727748,
-      "grad_norm": 0.9747323166327722,
+      "grad_norm": 0.033827301133935075,
       "learning_rate": 1.944121981744631e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -11.25,
-      "logps/chosen": -756.0,
-      "logps/rejected": -3520.0,
-      "loss": 0.003,
+      "logits/chosen": -16.125,
+      "logits/rejected": -8.375,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0064,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.1875,
-      "rewards/margins": 17.25,
-      "rewards/rejected": -20.375,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.25,
       "step": 1760
     },
     {
       "epoch": 1.8534031413612566,
-      "grad_norm": 0.33133657939448796,
+      "grad_norm": 0.6008073276104697,
       "learning_rate": 1.9144612367694887e-05,
-      "logits/chosen": -16.5,
-      "logits/rejected": -11.5,
-      "logps/chosen": -688.0,
-      "logps/rejected": -3520.0,
-      "loss": 0.0024,
+      "logits/chosen": -16.375,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -668.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.0,
-      "rewards/margins": 17.75,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -2.765625,
+      "rewards/margins": 19.5,
+      "rewards/rejected": -22.25,
       "step": 1770
     },
     {
       "epoch": 1.8638743455497382,
-      "grad_norm": 0.035836770744530205,
+      "grad_norm": 0.009516667840302055,
       "learning_rate": 1.8848874447193803e-05,
-      "logits/chosen": -15.9375,
-      "logits/rejected": -11.0,
-      "logps/chosen": -668.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0021,
+      "logits/chosen": -16.25,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -648.0,
+      "logps/rejected": -3872.0,
+      "loss": 0.0011,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.921875,
-      "rewards/margins": 17.625,
-      "rewards/rejected": -20.625,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -22.625,
       "step": 1780
     },
     {
       "epoch": 1.8743455497382198,
-      "grad_norm": 0.004159687050974283,
+      "grad_norm": 0.005503686242518818,
       "learning_rate": 1.855404997323571e-05,
       "logits/chosen": -16.125,
-      "logits/rejected": -11.25,
-      "logps/chosen": -684.0,
-      "logps/rejected": -3600.0,
-      "loss": 0.0011,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0014,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.984375,
-      "rewards/margins": 16.5,
-      "rewards/rejected": -19.5,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 18.625,
+      "rewards/rejected": -21.625,
       "step": 1790
     },
     {
       "epoch": 1.8848167539267016,
-      "grad_norm": 0.021203010570020737,
+      "grad_norm": 0.004920632708374876,
       "learning_rate": 1.82601827274658e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -668.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0006,
+      "logits/chosen": -17.0,
+      "logits/rejected": -8.375,
+      "logps/chosen": -660.0,
+      "logps/rejected": -3600.0,
+      "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.0625,
-      "rewards/margins": 15.8125,
-      "rewards/rejected": -18.875,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 18.375,
+      "rewards/rejected": -21.375,
       "step": 1800
     },
     {
       "epoch": 1.8848167539267016,
-      "eval_logits/chosen": -16.25,
-      "eval_logits/rejected": -16.125,
+      "eval_logits/chosen": -17.0,
+      "eval_logits/rejected": -16.625,
       "eval_logps/chosen": -644.0,
-      "eval_logps/rejected": -640.0,
-      "eval_loss": 0.9780625104904175,
-      "eval_rewards/accuracies": 0.509765625,
-      "eval_rewards/chosen": -3.28125,
-      "eval_rewards/margins": 0.2578125,
-      "eval_rewards/rejected": -3.53125,
-      "eval_runtime": 46.6291,
-      "eval_samples_per_second": 42.892,
-      "eval_steps_per_second": 0.686,
+      "eval_logps/rejected": -648.0,
+      "eval_loss": 1.0536249876022339,
+      "eval_rewards/accuracies": 0.49609375,
+      "eval_rewards/chosen": -3.234375,
+      "eval_rewards/margins": 0.35546875,
+      "eval_rewards/rejected": -3.59375,
+      "eval_runtime": 58.2334,
+      "eval_samples_per_second": 34.345,
+      "eval_steps_per_second": 0.55,
       "step": 1800
     },
     {
       "epoch": 1.8952879581151834,
-      "grad_norm": 0.058687191398231996,
+      "grad_norm": 0.066175417672082,
       "learning_rate": 1.7967316349380213e-05,
-      "logits/chosen": -16.375,
-      "logits/rejected": -11.1875,
-      "logps/chosen": -660.0,
-      "logps/rejected": -3344.0,
-      "loss": 0.0014,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.953125,
-      "rewards/margins": 16.375,
-      "rewards/rejected": -19.375,
+      "logits/chosen": -17.125,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -640.0,
+      "logps/rejected": -3536.0,
+      "loss": 0.0015,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 18.5,
+      "rewards/rejected": -21.25,
       "step": 1810
     },
     {
       "epoch": 1.905759162303665,
-      "grad_norm": 0.016867326143500675,
+      "grad_norm": 0.04774849680002709,
       "learning_rate": 1.7675494329845516e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -11.0625,
-      "logps/chosen": -712.0,
-      "logps/rejected": -3680.0,
-      "loss": 0.0014,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.296875,
-      "rewards/margins": 17.375,
-      "rewards/rejected": -20.625,
+      "logits/chosen": -17.0,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0076,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 18.75,
+      "rewards/rejected": -21.75,
       "step": 1820
     },
     {
       "epoch": 1.9162303664921465,
-      "grad_norm": 0.7720075659731835,
+      "grad_norm": 0.7753027409098596,
       "learning_rate": 1.738476000464026e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -10.625,
-      "logps/chosen": -696.0,
-      "logps/rejected": -3728.0,
-      "loss": 0.0042,
+      "logits/chosen": -17.125,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0029,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.171875,
-      "rewards/margins": 18.25,
-      "rewards/rejected": -21.5,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 18.75,
+      "rewards/rejected": -21.75,
       "step": 1830
     },
     {
       "epoch": 1.9267015706806283,
-      "grad_norm": 0.08525421786180916,
+      "grad_norm": 0.23528590314663267,
       "learning_rate": 1.7095156548019646e-05,
-      "logits/chosen": -16.5,
-      "logits/rejected": -11.0,
-      "logps/chosen": -692.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.0021,
+      "logits/chosen": -17.5,
+      "logits/rejected": -9.0,
+      "logps/chosen": -660.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0022,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.203125,
-      "rewards/margins": 18.875,
-      "rewards/rejected": -22.125,
+      "rewards/chosen": -2.875,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.25,
       "step": 1840
     },
     {
       "epoch": 1.93717277486911,
-      "grad_norm": 0.7631845497930444,
+      "grad_norm": 1.7641936894504935,
       "learning_rate": 1.6806726966304064e-05,
-      "logits/chosen": -16.375,
-      "logits/rejected": -10.5,
-      "logps/chosen": -728.0,
-      "logps/rejected": -3824.0,
-      "loss": 0.003,
+      "logits/chosen": -17.125,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0041,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.234375,
-      "rewards/margins": 19.5,
-      "rewards/rejected": -22.75,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.125,
       "step": 1850
     },
     {
       "epoch": 1.9476439790575917,
-      "grad_norm": 0.0006169662897405676,
+      "grad_norm": 0.0008543202453597533,
       "learning_rate": 1.6519514091492625e-05,
-      "logits/chosen": -16.25,
-      "logits/rejected": -11.25,
-      "logps/chosen": -672.0,
-      "logps/rejected": -3840.0,
-      "loss": 0.0042,
+      "logits/chosen": -17.0,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -636.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0046,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 20.0,
-      "rewards/rejected": -23.25,
+      "rewards/chosen": -2.859375,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -22.75,
       "step": 1860
     },
     {
       "epoch": 1.9581151832460733,
-      "grad_norm": 0.08044453253738494,
+      "grad_norm": 0.015423756626784501,
       "learning_rate": 1.6233560574902627e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -11.125,
-      "logps/chosen": -632.0,
-      "logps/rejected": -4016.0,
-      "loss": 0.0127,
+      "logits/chosen": -17.25,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -596.0,
+      "logps/rejected": -4000.0,
+      "loss": 0.0202,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -2.90625,
-      "rewards/margins": 20.0,
-      "rewards/rejected": -23.0,
+      "rewards/chosen": -2.53125,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -22.75,
       "step": 1870
     },
     {
       "epoch": 1.9685863874345548,
-      "grad_norm": 0.5919665698163148,
+      "grad_norm": 0.3843444092451444,
       "learning_rate": 1.5948908880835753e-05,
-      "logits/chosen": -16.625,
-      "logits/rejected": -10.25,
-      "logps/chosen": -704.0,
-      "logps/rejected": -3888.0,
-      "loss": 0.0059,
+      "logits/chosen": -17.125,
+      "logits/rejected": -7.3125,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0017,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 20.375,
-      "rewards/rejected": -23.5,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -22.875,
       "step": 1880
     },
     {
       "epoch": 1.9790575916230366,
-      "grad_norm": 0.12358338730824676,
+      "grad_norm": 5.491501940440811,
       "learning_rate": 1.5665601280272124e-05,
-      "logits/chosen": -15.875,
-      "logits/rejected": -9.875,
-      "logps/chosen": -696.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0019,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.125,
+      "logps/chosen": -652.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0076,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 22.375,
-      "rewards/rejected": -25.75,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 19.5,
+      "rewards/rejected": -22.5,
       "step": 1890
     },
     {
       "epoch": 1.9895287958115184,
-      "grad_norm": 0.033466910356039985,
+      "grad_norm": 0.03840122246563417,
       "learning_rate": 1.538367984459298e-05,
-      "logits/chosen": -15.3125,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -768.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0086,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0083,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -4.0,
-      "rewards/margins": 22.625,
-      "rewards/rejected": -26.625,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -22.25,
       "step": 1900
     },
     {
       "epoch": 1.9895287958115184,
-      "eval_logits/chosen": -15.6875,
-      "eval_logits/rejected": -15.4375,
-      "eval_logps/chosen": -732.0,
-      "eval_logps/rejected": -756.0,
-      "eval_loss": 1.175937533378601,
-      "eval_rewards/accuracies": 0.501953125,
-      "eval_rewards/chosen": -4.15625,
-      "eval_rewards/margins": 0.51953125,
-      "eval_rewards/rejected": -4.65625,
-      "eval_runtime": 46.4755,
-      "eval_samples_per_second": 43.033,
-      "eval_steps_per_second": 0.689,
+      "eval_logits/chosen": -16.625,
+      "eval_logits/rejected": -16.25,
+      "eval_logps/chosen": -668.0,
+      "eval_logps/rejected": -672.0,
+      "eval_loss": 1.1038906574249268,
+      "eval_rewards/accuracies": 0.48828125,
+      "eval_rewards/chosen": -3.484375,
+      "eval_rewards/margins": 0.32421875,
+      "eval_rewards/rejected": -3.8125,
+      "eval_runtime": 58.2697,
+      "eval_samples_per_second": 34.323,
+      "eval_steps_per_second": 0.549,
       "step": 1900
     },
     {
       "epoch": 2.0,
-      "grad_norm": 0.0028641918145212233,
+      "grad_norm": 0.0014401745187546163,
       "learning_rate": 1.5103186439333133e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -712.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0084,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.8125,
+      "logps/chosen": -652.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0075,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 22.75,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 19.0,
+      "rewards/rejected": -22.0,
       "step": 1910
     },
     {
       "epoch": 2.0104712041884816,
-      "grad_norm": 0.02255346416045209,
+      "grad_norm": 0.019644502273423865,
       "learning_rate": 1.4824162717963829e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.375,
-      "logps/chosen": -708.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0004,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.71875,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 23.5,
-      "rewards/rejected": -26.875,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -22.75,
       "step": 1920
     },
     {
       "epoch": 2.020942408376963,
-      "grad_norm": 0.008187559730452492,
+      "grad_norm": 0.01052187415361049,
       "learning_rate": 1.4546650115707192e-05,
-      "logits/chosen": -15.9375,
-      "logits/rejected": -9.25,
-      "logps/chosen": -668.0,
-      "logps/rejected": -3984.0,
-      "loss": 0.0006,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.28125,
+      "logps/chosen": -656.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.046875,
-      "rewards/margins": 22.125,
-      "rewards/rejected": -25.25,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -21.875,
       "step": 1930
     },
     {
       "epoch": 2.031413612565445,
-      "grad_norm": 0.19527172330719228,
+      "grad_norm": 0.007561785708249631,
       "learning_rate": 1.4270689843383111e-05,
-      "logits/chosen": -15.6875,
-      "logits/rejected": -9.0,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.09375,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3968.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 22.75,
-      "rewards/rejected": -26.125,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -23.0,
       "step": 1940
     },
     {
       "epoch": 2.0418848167539267,
-      "grad_norm": 0.008195639586709077,
+      "grad_norm": 0.009392603808878408,
       "learning_rate": 1.3996322881289347e-05,
-      "logits/chosen": -15.875,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -684.0,
-      "logps/rejected": -4064.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.5,
+      "logps/chosen": -628.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 21.5,
-      "rewards/rejected": -24.75,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.125,
       "step": 1950
     },
     {
       "epoch": 2.0523560209424083,
-      "grad_norm": 0.0052374601160772105,
+      "grad_norm": 0.052833199836017995,
       "learning_rate": 1.3723589973115963e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -9.625,
-      "logps/chosen": -692.0,
-      "logps/rejected": -4128.0,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.78125,
+      "logps/chosen": -656.0,
+      "logps/rejected": -3824.0,
       "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.296875,
-      "rewards/margins": 22.0,
-      "rewards/rejected": -25.375,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.375,
       "step": 1960
     },
     {
       "epoch": 2.06282722513089,
-      "grad_norm": 0.006729745159134903,
+      "grad_norm": 0.007857491781331962,
       "learning_rate": 1.3452531619894837e-05,
-      "logits/chosen": -15.4375,
-      "logits/rejected": -9.125,
-      "logps/chosen": -724.0,
-      "logps/rejected": -4080.0,
-      "loss": 0.0002,
+      "logits/chosen": -15.875,
+      "logits/rejected": -7.375,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 22.25,
-      "rewards/rejected": -25.75,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 19.0,
+      "rewards/rejected": -22.25,
       "step": 1970
     },
     {
       "epoch": 2.073298429319372,
-      "grad_norm": 0.007408094853921691,
+      "grad_norm": 0.012024377845656697,
       "learning_rate": 1.3183188073985245e-05,
-      "logits/chosen": -15.875,
-      "logits/rejected": -9.75,
-      "logps/chosen": -704.0,
-      "logps/rejected": -4032.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.875,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 22.125,
-      "rewards/rejected": -25.5,
+      "rewards/chosen": -3.0625,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -22.875,
       "step": 1980
     },
     {
       "epoch": 2.0837696335078535,
-      "grad_norm": 0.00016992488910610567,
+      "grad_norm": 3.816047971919616e-05,
       "learning_rate": 1.2915599333096351e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -736.0,
-      "logps/rejected": -4080.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.71875,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 22.125,
-      "rewards/rejected": -25.5,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.375,
       "step": 1990
     },
     {
       "epoch": 2.094240837696335,
-      "grad_norm": 0.0005705127481438246,
+      "grad_norm": 0.031216607652373246,
       "learning_rate": 1.264980513434752e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -764.0,
-      "logps/rejected": -4080.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.75,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 22.0,
-      "rewards/rejected": -25.5,
+      "rewards/chosen": -3.375,
+      "rewards/margins": 18.625,
+      "rewards/rejected": -22.0,
       "step": 2000
     },
-    {
-      "epoch": 2.094240837696335,
-      "eval_logits/chosen": -15.6875,
-      "eval_logits/rejected": -15.4375,
-      "eval_logps/chosen": -704.0,
-      "eval_logps/rejected": -720.0,
-      "eval_loss": 1.1181249618530273,
-      "eval_rewards/accuracies": 0.5,
-      "eval_rewards/chosen": -3.859375,
-      "eval_rewards/margins": 0.447265625,
-      "eval_rewards/rejected": -4.3125,
-      "eval_runtime": 46.5637,
-      "eval_samples_per_second": 42.952,
-      "eval_steps_per_second": 0.687,
+    {
+      "epoch": 2.094240837696335,
+      "eval_logits/chosen": -16.625,
+      "eval_logits/rejected": -16.125,
+      "eval_logps/chosen": -672.0,
+      "eval_logps/rejected": -672.0,
+      "eval_loss": 1.1159218549728394,
+      "eval_rewards/accuracies": 0.4921875,
+      "eval_rewards/chosen": -3.515625,
+      "eval_rewards/margins": 0.330078125,
+      "eval_rewards/rejected": -3.84375,
+      "eval_runtime": 58.2928,
+      "eval_samples_per_second": 34.31,
+      "eval_steps_per_second": 0.549,
       "step": 2000
     },
     {
       "epoch": 2.1047120418848166,
-      "grad_norm": 0.11613537072612715,
+      "grad_norm": 0.0040657528991151935,
       "learning_rate": 1.2385844948367324e-05,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.625,
-      "logps/chosen": -692.0,
-      "logps/rejected": -4384.0,
-      "loss": 0.0008,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.71875,
+      "logps/chosen": -668.0,
+      "logps/rejected": -4032.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -3.25,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -23.5,
       "step": 2010
     },
     {
       "epoch": 2.115183246073298,
-      "grad_norm": 0.002797304095375945,
+      "grad_norm": 0.0002459355111861258,
       "learning_rate": 1.2123757973432114e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.375,
-      "logps/chosen": -768.0,
-      "logps/rejected": -4224.0,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.25,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3952.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 22.0,
-      "rewards/rejected": -25.625,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 19.5,
+      "rewards/rejected": -22.75,
       "step": 2020
     },
     {
       "epoch": 2.1256544502617802,
-      "grad_norm": 0.05288470633614975,
+      "grad_norm": 0.002058151572530592,
       "learning_rate": 1.1863583129645106e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.25,
-      "logps/chosen": -720.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.8125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 23.375,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -23.125,
       "step": 2030
     },
     {
       "epoch": 2.136125654450262,
-      "grad_norm": 0.00028503993547607416,
+      "grad_norm": 0.2396628989311444,
       "learning_rate": 1.1605359053156607e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -9.25,
-      "logps/chosen": -728.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0014,
+      "logits/chosen": -16.625,
+      "logits/rejected": -6.96875,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -2.875,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.0,
       "step": 2040
     },
     {
       "epoch": 2.1465968586387434,
-      "grad_norm": 0.0005365799312041201,
+      "grad_norm": 2.8359130495461864,
       "learning_rate": 1.1349124090426644e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.5,
-      "logps/chosen": -720.0,
-      "logps/rejected": -4064.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.625,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0025,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 22.5,
-      "rewards/rejected": -25.875,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.375,
       "step": 2050
     },
     {
       "epoch": 2.157068062827225,
-      "grad_norm": 0.012308329920858068,
+      "grad_norm": 0.007543450291516159,
       "learning_rate": 1.1094916292530406e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -10.0,
-      "logps/chosen": -728.0,
-      "logps/rejected": -4192.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.96875,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0059,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.3125,
-      "rewards/margins": 23.0,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 19.625,
+      "rewards/rejected": -22.625,
       "step": 2060
     },
     {
       "epoch": 2.167539267015707,
-      "grad_norm": 0.027286484413604088,
+      "grad_norm": 0.025906161458217353,
       "learning_rate": 1.0842773409507623e-05,
-      "logits/chosen": -16.0,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -672.0,
-      "logps/rejected": -4352.0,
-      "loss": 0.0005,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.9375,
+      "logps/chosen": -648.0,
+      "logps/rejected": -3984.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.21875,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.0,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -23.125,
       "step": 2070
     },
     {
       "epoch": 2.1780104712041886,
-      "grad_norm": 0.0005312850621567488,
+      "grad_norm": 0.0021617160254147508,
       "learning_rate": 1.0592732884756754e-05,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -8.9375,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.40625,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0016,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 23.125,
-      "rewards/rejected": -26.625,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 19.5,
+      "rewards/rejected": -22.625,
       "step": 2080
     },
     {
       "epoch": 2.18848167539267,
-      "grad_norm": 0.004439815026913123,
+      "grad_norm": 0.024913904537620553,
       "learning_rate": 1.0344831849474507e-05,
-      "logits/chosen": -15.875,
-      "logits/rejected": -9.625,
-      "logps/chosen": -744.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0005,
+      "logits/chosen": -16.5,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -696.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.125,
       "step": 2090
     },
     {
       "epoch": 2.1989528795811517,
-      "grad_norm": 0.3963294943253363,
+      "grad_norm": 0.013588028916770528,
       "learning_rate": 1.0099107117141879e-05,
-      "logits/chosen": -16.125,
-      "logits/rejected": -9.5,
-      "logps/chosen": -696.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0145,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.46875,
+      "logps/chosen": -668.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0027,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 23.875,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -3.0625,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -23.0,
       "step": 2100
     },
     {
       "epoch": 2.1989528795811517,
-      "eval_logits/chosen": -15.875,
-      "eval_logits/rejected": -15.625,
-      "eval_logps/chosen": -720.0,
-      "eval_logps/rejected": -740.0,
-      "eval_loss": 1.1573437452316284,
+      "eval_logits/chosen": -16.625,
+      "eval_logits/rejected": -16.125,
+      "eval_logps/chosen": -680.0,
+      "eval_logps/rejected": -688.0,
+      "eval_loss": 1.153546929359436,
       "eval_rewards/accuracies": 0.498046875,
-      "eval_rewards/chosen": -4.03125,
-      "eval_rewards/margins": 0.494140625,
-      "eval_rewards/rejected": -4.53125,
-      "eval_runtime": 46.6315,
-      "eval_samples_per_second": 42.889,
-      "eval_steps_per_second": 0.686,
+      "eval_rewards/chosen": -3.59375,
+      "eval_rewards/margins": 0.404296875,
+      "eval_rewards/rejected": -4.0,
+      "eval_runtime": 58.2702,
+      "eval_samples_per_second": 34.323,
+      "eval_steps_per_second": 0.549,
       "step": 2100
     },
     {
       "epoch": 2.2094240837696333,
-      "grad_norm": 0.03869382246328654,
+      "grad_norm": 0.04130571683561593,
       "learning_rate": 9.855595178057334e-06,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.125,
-      "logps/chosen": -764.0,
-      "logps/rejected": -4384.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.125,
+      "logits/rejected": -7.0625,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.578125,
-      "rewards/margins": 25.25,
-      "rewards/rejected": -28.75,
+      "rewards/chosen": -3.1875,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.25,
       "step": 2110
     },
     {
       "epoch": 2.2198952879581153,
-      "grad_norm": 0.013988600771111089,
+      "grad_norm": 0.0010595350766436481,
       "learning_rate": 9.614332193917949e-06,
-      "logits/chosen": -16.0,
-      "logits/rejected": -9.25,
-      "logps/chosen": -808.0,
-      "logps/rejected": -4640.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.3125,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3936.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.875,
-      "rewards/margins": 26.25,
-      "rewards/rejected": -30.125,
+      "rewards/chosen": -3.375,
+      "rewards/margins": 19.625,
+      "rewards/rejected": -23.0,
       "step": 2120
     },
     {
       "epoch": 2.230366492146597,
-      "grad_norm": 0.00046560201671487993,
+      "grad_norm": 0.00039288425775861873,
       "learning_rate": 9.375353992449385e-06,
-      "logits/chosen": -16.125,
-      "logits/rejected": -9.5,
-      "logps/chosen": -752.0,
-      "logps/rejected": -4544.0,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.5,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3856.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.640625,
-      "rewards/margins": 27.0,
-      "rewards/rejected": -30.625,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -23.75,
       "step": 2130
     },
     {
       "epoch": 2.2408376963350785,
-      "grad_norm": 0.006318341470712007,
+      "grad_norm": 0.011138562502434618,
       "learning_rate": 9.138696062085442e-06,
-      "logits/chosen": -16.125,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -700.0,
-      "logps/rejected": -4448.0,
-      "loss": 0.0015,
+      "logits/chosen": -16.75,
+      "logits/rejected": -7.90625,
+      "logps/chosen": -656.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.265625,
-      "rewards/margins": 25.375,
-      "rewards/rejected": -28.625,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -23.125,
       "step": 2140
     },
     {
       "epoch": 2.25130890052356,
-      "grad_norm": 0.0017725362204231568,
+      "grad_norm": 0.0013893341612525078,
       "learning_rate": 8.904393546698007e-06,
-      "logits/chosen": -16.25,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -760.0,
-      "logps/rejected": -4544.0,
-      "loss": 0.0008,
+      "logits/chosen": -16.75,
+      "logits/rejected": -7.09375,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3968.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 25.75,
-      "rewards/rejected": -29.25,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -23.75,
       "step": 2150
     },
     {
       "epoch": 2.261780104712042,
-      "grad_norm": 0.003614554588243917,
+      "grad_norm": 0.007195232686183097,
       "learning_rate": 8.672481240378142e-06,
-      "logits/chosen": -15.875,
-      "logits/rejected": -9.625,
-      "logps/chosen": -660.0,
-      "logps/rejected": -4192.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.96875,
+      "logps/chosen": -628.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.171875,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -2.828125,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -22.625,
       "step": 2160
     },
     {
       "epoch": 2.2722513089005236,
-      "grad_norm": 0.008315848170885314,
+      "grad_norm": 0.0036460062092003812,
       "learning_rate": 8.44299358226919e-06,
-      "logits/chosen": -15.375,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -708.0,
-      "logps/rejected": -4384.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.46875,
+      "logps/chosen": -664.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.46875,
-      "rewards/margins": 24.875,
-      "rewards/rejected": -28.375,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -23.5,
       "step": 2170
     },
     {
       "epoch": 2.282722513089005,
-      "grad_norm": 0.004981971886133196,
+      "grad_norm": 0.003258206509873447,
       "learning_rate": 8.215964651452457e-06,
-      "logits/chosen": -15.25,
-      "logits/rejected": -9.0,
-      "logps/chosen": -752.0,
-      "logps/rejected": -4416.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.40625,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0015,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.71875,
-      "rewards/margins": 25.125,
-      "rewards/rejected": -28.875,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.375,
       "step": 2180
     },
     {
       "epoch": 2.2931937172774868,
-      "grad_norm": 0.0031426449534539664,
+      "grad_norm": 0.006920860918525047,
       "learning_rate": 7.991428161886502e-06,
-      "logits/chosen": -15.3125,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -752.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.578125,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -3.140625,
+      "rewards/margins": 19.5,
+      "rewards/rejected": -22.625,
       "step": 2190
     },
     {
       "epoch": 2.303664921465969,
-      "grad_norm": 0.0037110116382837077,
+      "grad_norm": 0.025010564776443638,
       "learning_rate": 7.76941745740061e-06,
-      "logits/chosen": -15.375,
-      "logits/rejected": -8.25,
-      "logps/chosen": -804.0,
-      "logps/rejected": -4352.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.375,
+      "logits/rejected": -6.65625,
+      "logps/chosen": -748.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.921875,
-      "rewards/margins": 24.5,
-      "rewards/rejected": -28.5,
+      "rewards/chosen": -3.375,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.75,
       "step": 2200
     },
     {
       "epoch": 2.303664921465969,
-      "eval_logits/chosen": -15.4375,
-      "eval_logits/rejected": -15.0625,
-      "eval_logps/chosen": -740.0,
-      "eval_logps/rejected": -760.0,
-      "eval_loss": 1.1923437118530273,
-      "eval_rewards/accuracies": 0.49609375,
-      "eval_rewards/chosen": -4.21875,
-      "eval_rewards/margins": 0.5234375,
-      "eval_rewards/rejected": -4.71875,
-      "eval_runtime": 46.6192,
-      "eval_samples_per_second": 42.901,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -16.625,
+      "eval_logits/rejected": -16.25,
+      "eval_logps/chosen": -676.0,
+      "eval_logps/rejected": -688.0,
+      "eval_loss": 1.150484323501587,
+      "eval_rewards/accuracies": 0.490234375,
+      "eval_rewards/chosen": -3.578125,
+      "eval_rewards/margins": 0.40625,
+      "eval_rewards/rejected": -3.984375,
+      "eval_runtime": 58.1557,
+      "eval_samples_per_second": 34.39,
+      "eval_steps_per_second": 0.55,
       "step": 2200
     },
     {
-      "epoch": 2.3141361256544504,
-      "grad_norm": 0.0021420210284135524,
-      "learning_rate": 7.549965506743123e-06,
-      "logits/chosen": -15.875,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -720.0,
-      "logps/rejected": -4416.0,
-      "loss": 0.0027,
+      "epoch": 2.3141361256544504,
+      "grad_norm": 0.017206480910386915,
+      "learning_rate": 7.549965506743123e-06,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.375,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.46875,
-      "rewards/margins": 25.5,
-      "rewards/rejected": -29.0,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -23.25,
       "step": 2210
     },
     {
       "epoch": 2.324607329842932,
-      "grad_norm": 0.006628126378196115,
+      "grad_norm": 0.006938246554670781,
       "learning_rate": 7.333104898685631e-06,
-      "logits/chosen": -15.6875,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -720.0,
-      "logps/rejected": -4096.0,
+      "logits/chosen": -16.125,
+      "logits/rejected": -7.90625,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3728.0,
       "loss": 0.0005,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 23.125,
-      "rewards/rejected": -26.5,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -22.75,
       "step": 2220
     },
     {
       "epoch": 2.3350785340314135,
-      "grad_norm": 0.003120591763705076,
+      "grad_norm": 0.006668859986802676,
       "learning_rate": 7.118867837183518e-06,
-      "logits/chosen": -15.875,
-      "logits/rejected": -9.125,
-      "logps/chosen": -812.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.34375,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3936.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.640625,
-      "rewards/margins": 23.0,
-      "rewards/rejected": -26.625,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -23.25,
       "step": 2230
     },
     {
       "epoch": 2.345549738219895,
-      "grad_norm": 0.0016742405990676577,
+      "grad_norm": 0.004008993128219139,
       "learning_rate": 6.9072861365936045e-06,
-      "logits/chosen": -16.125,
-      "logits/rejected": -9.25,
-      "logps/chosen": -800.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.34375,
+      "logps/chosen": -764.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.578125,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -23.125,
       "step": 2240
     },
     {
       "epoch": 2.356020942408377,
-      "grad_norm": 0.008628441618630276,
+      "grad_norm": 0.007663924676482636,
       "learning_rate": 6.6983912169497005e-06,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -752.0,
-      "logps/rejected": -4080.0,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.28125,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3776.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.296875,
-      "rewards/margins": 22.5,
-      "rewards/rejected": -25.75,
+      "rewards/chosen": -2.875,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -22.75,
       "step": 2250
     },
     {
       "epoch": 2.3664921465968587,
-      "grad_norm": 0.009387104873476765,
+      "grad_norm": 0.01791500012532374,
       "learning_rate": 6.49221409929677e-06,
-      "logits/chosen": -15.625,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -676.0,
-      "logps/rejected": -4448.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.9375,
+      "logps/chosen": -644.0,
+      "logps/rejected": -4080.0,
+      "loss": 0.0018,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 24.375,
-      "rewards/rejected": -27.75,
+      "rewards/chosen": -2.9375,
+      "rewards/margins": 21.25,
+      "rewards/rejected": -24.125,
       "step": 2260
     },
     {
       "epoch": 2.3769633507853403,
-      "grad_norm": 0.0018548983537483204,
+      "grad_norm": 0.001348586125923643,
       "learning_rate": 6.2887854010842075e-06,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.375,
-      "logps/chosen": -704.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.46875,
+      "logps/chosen": -664.0,
+      "logps/rejected": -3872.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.328125,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -23.5,
       "step": 2270
     },
     {
       "epoch": 2.387434554973822,
-      "grad_norm": 0.0053515289057978616,
+      "grad_norm": 0.011837733461653115,
       "learning_rate": 6.0881353316191385e-06,
-      "logits/chosen": -15.4375,
-      "logits/rejected": -9.5,
-      "logps/chosen": -772.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.25,
+      "logits/rejected": -8.0,
+      "logps/chosen": -720.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.609375,
-      "rewards/margins": 23.875,
-      "rewards/rejected": -27.375,
+      "rewards/chosen": -3.078125,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -23.0,
       "step": 2280
     },
     {
       "epoch": 2.3979057591623034,
-      "grad_norm": 0.020287290523149492,
+      "grad_norm": 0.001281725369752337,
       "learning_rate": 5.890293687580381e-06,
-      "logits/chosen": -15.9375,
-      "logits/rejected": -10.0,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0004,
+      "logits/chosen": -16.5,
+      "logits/rejected": -8.125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5625,
-      "rewards/margins": 22.75,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -3.0625,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.5,
       "step": 2290
     },
     {
       "epoch": 2.4083769633507854,
-      "grad_norm": 0.16211807808816686,
+      "grad_norm": 0.010432225996481727,
       "learning_rate": 5.695289848593532e-06,
-      "logits/chosen": -15.625,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -720.0,
-      "logps/rejected": -3952.0,
-      "loss": 0.0005,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.78125,
+      "logps/chosen": -664.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.0006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 22.375,
-      "rewards/rejected": -25.75,
+      "rewards/chosen": -2.859375,
+      "rewards/margins": 19.625,
+      "rewards/rejected": -22.5,
       "step": 2300
     },
     {
       "epoch": 2.4083769633507854,
-      "eval_logits/chosen": -15.5625,
-      "eval_logits/rejected": -15.3125,
-      "eval_logps/chosen": -720.0,
-      "eval_logps/rejected": -736.0,
-      "eval_loss": 1.1497187614440918,
+      "eval_logits/chosen": -16.75,
+      "eval_logits/rejected": -16.25,
+      "eval_logps/chosen": -672.0,
+      "eval_logps/rejected": -684.0,
+      "eval_loss": 1.153499960899353,
       "eval_rewards/accuracies": 0.490234375,
-      "eval_rewards/chosen": -4.0,
-      "eval_rewards/margins": 0.482421875,
-      "eval_rewards/rejected": -4.5,
-      "eval_runtime": 46.6235,
-      "eval_samples_per_second": 42.897,
-      "eval_steps_per_second": 0.686,
+      "eval_rewards/chosen": -3.546875,
+      "eval_rewards/margins": 0.40234375,
+      "eval_rewards/rejected": -3.953125,
+      "eval_runtime": 58.2152,
+      "eval_samples_per_second": 34.355,
+      "eval_steps_per_second": 0.55,
       "step": 2300
     },
     {
       "epoch": 2.418848167539267,
-      "grad_norm": 0.0005933024775404446,
+      "grad_norm": 0.012993559894127019,
       "learning_rate": 5.50315277286812e-06,
-      "logits/chosen": -16.125,
-      "logits/rejected": -9.5,
-      "logps/chosen": -692.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.875,
+      "logits/rejected": -7.75,
+      "logps/chosen": -648.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 20.375,
+      "rewards/rejected": -23.375,
       "step": 2310
     },
     {
       "epoch": 2.4293193717277486,
-      "grad_norm": 0.0015257956912693446,
+      "grad_norm": 0.0007165363106639997,
       "learning_rate": 5.313910992897281e-06,
-      "logits/chosen": -15.375,
-      "logits/rejected": -9.125,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4192.0,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.28125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3904.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.453125,
-      "rewards/margins": 23.125,
-      "rewards/rejected": -26.5,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -23.625,
       "step": 2320
     },
     {
       "epoch": 2.4397905759162306,
-      "grad_norm": 0.0021306748692230867,
+      "grad_norm": 0.0046072441922562655,
       "learning_rate": 5.127592611220658e-06,
-      "logits/chosen": -15.5625,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -752.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.9375,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -3.0625,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -23.0,
       "step": 2330
     },
     {
       "epoch": 2.450261780104712,
-      "grad_norm": 0.0023837467762273293,
+      "grad_norm": 0.002823160595406029,
       "learning_rate": 4.944225296251159e-06,
-      "logits/chosen": -15.6875,
-      "logits/rejected": -9.5,
-      "logps/chosen": -680.0,
-      "logps/rejected": -4256.0,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.84375,
+      "logps/chosen": -632.0,
+      "logps/rejected": -3904.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.265625,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -2.8125,
+      "rewards/margins": 20.75,
+      "rewards/rejected": -23.625,
       "step": 2340
     },
     {
       "epoch": 2.4607329842931938,
-      "grad_norm": 0.0028220932839165924,
+      "grad_norm": 0.00910844472276119,
       "learning_rate": 4.763836278166172e-06,
-      "logits/chosen": -15.4375,
-      "logits/rejected": -9.25,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0004,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -696.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.6875,
-      "rewards/margins": 23.125,
-      "rewards/rejected": -26.875,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.5,
       "step": 2350
     },
     {
       "epoch": 2.4712041884816753,
-      "grad_norm": 0.012780597797202939,
+      "grad_norm": 0.0012900917126641776,
       "learning_rate": 4.586452344863876e-06,
-      "logits/chosen": -15.9375,
-      "logits/rejected": -9.375,
-      "logps/chosen": -744.0,
-      "logps/rejected": -4064.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.75,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 22.875,
-      "rewards/rejected": -26.375,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 19.5,
+      "rewards/rejected": -22.375,
       "step": 2360
     },
     {
       "epoch": 2.481675392670157,
-      "grad_norm": 0.0037221110127764515,
+      "grad_norm": 0.005811356538760154,
       "learning_rate": 4.412099837985192e-06,
-      "logits/chosen": -16.0,
-      "logits/rejected": -9.125,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4384.0,
+      "logits/chosen": -16.875,
+      "logits/rejected": -7.15625,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3968.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 24.375,
-      "rewards/rejected": -27.875,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -23.5,
       "step": 2370
     },
     {
       "epoch": 2.492146596858639,
-      "grad_norm": 0.0023870278735102614,
+      "grad_norm": 0.0008920808467065192,
       "learning_rate": 4.240804649002089e-06,
-      "logits/chosen": -15.875,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -688.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -652.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 23.875,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -3.078125,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.125,
       "step": 2380
     },
     {
       "epoch": 2.5026178010471205,
-      "grad_norm": 0.004447776994587039,
+      "grad_norm": 0.23992775780446185,
       "learning_rate": 4.072592215372598e-06,
-      "logits/chosen": -15.9375,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -792.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0003,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.53125,
+      "logps/chosen": -760.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0017,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.71875,
-      "rewards/margins": 23.375,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -3.421875,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -23.125,
       "step": 2390
     },
     {
       "epoch": 2.513089005235602,
-      "grad_norm": 0.003882461986638262,
+      "grad_norm": 0.0015010265535231868,
       "learning_rate": 3.907487516763389e-06,
-      "logits/chosen": -16.0,
-      "logits/rejected": -9.75,
-      "logps/chosen": -712.0,
-      "logps/rejected": -4160.0,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.9375,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3760.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.3125,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.0,
       "step": 2400
     },
     {
       "epoch": 2.513089005235602,
-      "eval_logits/chosen": -15.6875,
-      "eval_logits/rejected": -15.375,
-      "eval_logps/chosen": -720.0,
-      "eval_logps/rejected": -740.0,
-      "eval_loss": 1.1574687957763672,
-      "eval_rewards/accuracies": 0.49609375,
-      "eval_rewards/chosen": -4.03125,
-      "eval_rewards/margins": 0.490234375,
-      "eval_rewards/rejected": -4.53125,
-      "eval_runtime": 46.637,
-      "eval_samples_per_second": 42.884,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -16.625,
+      "eval_logits/rejected": -16.25,
+      "eval_logps/chosen": -676.0,
+      "eval_logps/rejected": -688.0,
+      "eval_loss": 1.1581406593322754,
+      "eval_rewards/accuracies": 0.4921875,
+      "eval_rewards/chosen": -3.578125,
+      "eval_rewards/margins": 0.408203125,
+      "eval_rewards/rejected": -3.984375,
+      "eval_runtime": 58.2647,
+      "eval_samples_per_second": 34.326,
+      "eval_steps_per_second": 0.549,
       "step": 2400
     },
     {
       "epoch": 2.5235602094240837,
-      "grad_norm": 0.0012252707805558566,
+      "grad_norm": 0.004019092072233536,
       "learning_rate": 3.745515071340286e-06,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.375,
-      "logps/chosen": -768.0,
-      "logps/rejected": -4288.0,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3840.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.65625,
-      "rewards/margins": 24.125,
-      "rewards/rejected": -27.75,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -23.25,
       "step": 2410
     },
     {
       "epoch": 2.5340314136125652,
-      "grad_norm": 0.0017846656108641288,
-      "learning_rate": 3.586698932127236e-06,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.5,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4160.0,
+      "grad_norm": 0.002196560886759549,
+      "learning_rate": 3.586698932127236e-06,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.71875,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3792.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 23.0,
-      "rewards/rejected": -26.5,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 19.625,
+      "rewards/rejected": -22.875,
       "step": 2420
     },
     {
       "epoch": 2.5445026178010473,
-      "grad_norm": 0.0014629997433764392,
+      "grad_norm": 0.37930967230265117,
       "learning_rate": 3.4310626834344744e-06,
-      "logits/chosen": -16.125,
-      "logits/rejected": -9.5,
-      "logps/chosen": -740.0,
-      "logps/rejected": -4048.0,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3616.0,
       "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 23.125,
-      "rewards/rejected": -26.5,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -22.25,
       "step": 2430
     },
     {
       "epoch": 2.554973821989529,
-      "grad_norm": 0.01275322917738364,
+      "grad_norm": 0.01008287960866285,
       "learning_rate": 3.2786294373562343e-06,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.875,
-      "logps/chosen": -744.0,
-      "logps/rejected": -4384.0,
+      "logits/chosen": -16.25,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3952.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 24.5,
-      "rewards/rejected": -28.0,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -23.625,
       "step": 2440
     },
     {
       "epoch": 2.5654450261780104,
-      "grad_norm": 0.0010671485432906243,
+      "grad_norm": 0.005355173605038295,
       "learning_rate": 3.1294218303385526e-06,
-      "logits/chosen": -15.4375,
-      "logits/rejected": -9.375,
-      "logps/chosen": -788.0,
-      "logps/rejected": -4352.0,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3936.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.75,
-      "rewards/margins": 23.875,
-      "rewards/rejected": -27.625,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 20.375,
+      "rewards/rejected": -23.5,
       "step": 2450
     },
     {
       "epoch": 2.5759162303664924,
-      "grad_norm": 0.12088700867936222,
+      "grad_norm": 0.009225437912353517,
       "learning_rate": 2.9834620198177483e-06,
-      "logits/chosen": -15.375,
-      "logits/rejected": -9.5,
-      "logps/chosen": -740.0,
-      "logps/rejected": -4192.0,
-      "loss": 0.0006,
+      "logits/chosen": -16.0,
+      "logits/rejected": -7.75,
+      "logps/chosen": -696.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.453125,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.375,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -23.0,
       "step": 2460
     },
     {
       "epoch": 2.5863874345549736,
-      "grad_norm": 0.0006003360864240354,
+      "grad_norm": 0.0037874363962814212,
       "learning_rate": 2.840771680930068e-06,
-      "logits/chosen": -16.125,
-      "logits/rejected": -9.5,
-      "logps/chosen": -772.0,
-      "logps/rejected": -4288.0,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.8125,
+      "logps/chosen": -720.0,
+      "logps/rejected": -3840.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.59375,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -23.125,
       "step": 2470
     },
     {
       "epoch": 2.5968586387434556,
-      "grad_norm": 0.0026496517415497303,
+      "grad_norm": 0.001155281353524686,
       "learning_rate": 2.701372003292868e-06,
-      "logits/chosen": -15.5625,
-      "logits/rejected": -9.25,
-      "logps/chosen": -716.0,
-      "logps/rejected": -4352.0,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.53125,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3920.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.46875,
-      "rewards/margins": 24.75,
-      "rewards/rejected": -28.125,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 20.875,
+      "rewards/rejected": -23.875,
       "step": 2480
     },
     {
       "epoch": 2.607329842931937,
-      "grad_norm": 0.00590808534468086,
+      "grad_norm": 0.005693439718259309,
       "learning_rate": 2.56528368785795e-06,
-      "logits/chosen": -16.0,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -760.0,
-      "logps/rejected": -4576.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.875,
+      "logits/rejected": -7.15625,
+      "logps/chosen": -720.0,
+      "logps/rejected": -4032.0,
+      "loss": 0.0012,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 25.625,
-      "rewards/rejected": -29.125,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 20.75,
+      "rewards/rejected": -23.875,
       "step": 2490
     },
     {
       "epoch": 2.6178010471204187,
-      "grad_norm": 0.0033884560301643664,
+      "grad_norm": 0.002800450158938199,
       "learning_rate": 2.432526943837471e-06,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4448.0,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -688.0,
+      "logps/rejected": -4032.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 24.875,
-      "rewards/rejected": -28.375,
+      "rewards/chosen": -3.0625,
+      "rewards/margins": 21.0,
+      "rewards/rejected": -24.125,
       "step": 2500
     },
     {
       "epoch": 2.6178010471204187,
-      "eval_logits/chosen": -15.5625,
-      "eval_logits/rejected": -15.25,
-      "eval_logps/chosen": -728.0,
-      "eval_logps/rejected": -748.0,
-      "eval_loss": 1.167562484741211,
-      "eval_rewards/accuracies": 0.4921875,
-      "eval_rewards/chosen": -4.09375,
-      "eval_rewards/margins": 0.50390625,
-      "eval_rewards/rejected": -4.59375,
-      "eval_runtime": 46.6275,
-      "eval_samples_per_second": 42.893,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -16.75,
+      "eval_logits/rejected": -16.375,
+      "eval_logps/chosen": -672.0,
+      "eval_logps/rejected": -684.0,
+      "eval_loss": 1.1608593463897705,
+      "eval_rewards/accuracies": 0.49609375,
+      "eval_rewards/chosen": -3.5625,
+      "eval_rewards/margins": 0.408203125,
+      "eval_rewards/rejected": -3.96875,
+      "eval_runtime": 58.2861,
+      "eval_samples_per_second": 34.314,
+      "eval_steps_per_second": 0.549,
       "step": 2500
     },
     {
       "epoch": 2.6282722513089007,
-      "grad_norm": 0.04083566818181006,
+      "grad_norm": 0.0018387213756474478,
       "learning_rate": 2.3031214857028446e-06,
-      "logits/chosen": -15.625,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -764.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0065,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.625,
-      "rewards/margins": 23.875,
-      "rewards/rejected": -27.5,
+      "logits/chosen": -16.25,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 19.625,
+      "rewards/rejected": -22.75,
       "step": 2510
     },
     {
       "epoch": 2.6387434554973823,
-      "grad_norm": 0.25925832899674267,
+      "grad_norm": 0.036156161947461754,
       "learning_rate": 2.177086530257139e-06,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -716.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0005,
+      "logits/chosen": -16.625,
+      "logits/rejected": -8.375,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3936.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 24.125,
-      "rewards/rejected": -27.625,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 20.875,
+      "rewards/rejected": -23.875,
       "step": 2520
     },
     {
       "epoch": 2.649214659685864,
-      "grad_norm": 0.0004399400679548829,
+      "grad_norm": 0.0016827659304670189,
       "learning_rate": 2.054440793781354e-06,
-      "logits/chosen": -15.6875,
-      "logits/rejected": -10.0,
-      "logps/chosen": -712.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.375,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3920.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.34375,
-      "rewards/margins": 23.125,
-      "rewards/rejected": -26.5,
+      "rewards/chosen": -2.9375,
+      "rewards/margins": 20.375,
+      "rewards/rejected": -23.375,
       "step": 2530
     },
     {
       "epoch": 2.6596858638743455,
-      "grad_norm": 0.0015963132819992276,
+      "grad_norm": 0.017451898496348647,
       "learning_rate": 1.9352024892550587e-06,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.625,
-      "logps/chosen": -728.0,
-      "logps/rejected": -4288.0,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.75,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3936.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.0,
+      "rewards/chosen": -3.078125,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -23.75,
       "step": 2540
     },
     {
       "epoch": 2.670157068062827,
-      "grad_norm": 0.004838724667207601,
+      "grad_norm": 0.001515934461611262,
       "learning_rate": 1.8193893236517379e-06,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -748.0,
-      "logps/rejected": -4384.0,
+      "logits/chosen": -16.125,
+      "logits/rejected": -8.0,
+      "logps/chosen": -700.0,
+      "logps/rejected": -4000.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 24.375,
-      "rewards/rejected": -28.0,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 21.25,
+      "rewards/rejected": -24.375,
       "step": 2550
     },
     {
       "epoch": 2.680628272251309,
-      "grad_norm": 0.003960961926870279,
+      "grad_norm": 0.0029832776797642543,
       "learning_rate": 1.7070184953092956e-06,
-      "logits/chosen": -15.625,
-      "logits/rejected": -9.375,
-      "logps/chosen": -744.0,
-      "logps/rejected": -4192.0,
-      "loss": 0.0006,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0004,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 23.25,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -23.625,
       "step": 2560
     },
     {
       "epoch": 2.6910994764397906,
-      "grad_norm": 0.007326074819022257,
+      "grad_norm": 0.0032817218662073975,
       "learning_rate": 1.5981066913760918e-06,
-      "logits/chosen": -15.4375,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -716.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.375,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -664.0,
+      "logps/rejected": -3872.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 23.25,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -23.25,
       "step": 2570
     },
     {
       "epoch": 2.701570680628272,
-      "grad_norm": 0.02450402087761141,
+      "grad_norm": 0.019063697471025442,
       "learning_rate": 1.492670085332884e-06,
-      "logits/chosen": -15.9375,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -748.0,
-      "logps/rejected": -4352.0,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -708.0,
+      "logps/rejected": -4016.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5625,
-      "rewards/margins": 23.25,
-      "rewards/rejected": -26.875,
+      "rewards/chosen": -3.140625,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -23.625,
       "step": 2580
     },
     {
       "epoch": 2.712041884816754,
-      "grad_norm": 0.006134961075889198,
+      "grad_norm": 0.0054267972711354836,
       "learning_rate": 1.3907243345910787e-06,
-      "logits/chosen": -15.875,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -744.0,
-      "logps/rejected": -4160.0,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.0625,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3872.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.296875,
-      "rewards/margins": 22.25,
-      "rewards/rejected": -25.5,
+      "rewards/chosen": -2.9375,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -22.75,
       "step": 2590
     },
     {
       "epoch": 2.7225130890052354,
-      "grad_norm": 0.012201214496547234,
+      "grad_norm": 0.012953630820954048,
       "learning_rate": 1.292284578167538e-06,
-      "logits/chosen": -15.5,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -712.0,
-      "logps/rejected": -4032.0,
-      "loss": 0.0014,
+      "logits/chosen": -16.125,
+      "logits/rejected": -7.5,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 22.875,
-      "rewards/rejected": -26.25,
+      "rewards/chosen": -2.9375,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -23.0,
       "step": 2600
     },
     {
       "epoch": 2.7225130890052354,
-      "eval_logits/chosen": -15.625,
-      "eval_logits/rejected": -15.3125,
-      "eval_logps/chosen": -720.0,
-      "eval_logps/rejected": -740.0,
-      "eval_loss": 1.1490000486373901,
-      "eval_rewards/accuracies": 0.501953125,
-      "eval_rewards/chosen": -4.03125,
-      "eval_rewards/margins": 0.478515625,
-      "eval_rewards/rejected": -4.5,
-      "eval_runtime": 46.6332,
-      "eval_samples_per_second": 42.888,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -16.75,
+      "eval_logits/rejected": -16.375,
+      "eval_logps/chosen": -676.0,
+      "eval_logps/rejected": -688.0,
+      "eval_loss": 1.166812539100647,
+      "eval_rewards/accuracies": 0.4921875,
+      "eval_rewards/chosen": -3.59375,
+      "eval_rewards/margins": 0.412109375,
+      "eval_rewards/rejected": -4.0,
+      "eval_runtime": 58.3395,
+      "eval_samples_per_second": 34.282,
+      "eval_steps_per_second": 0.549,
       "step": 2600
     },
     {
       "epoch": 2.7329842931937174,
-      "grad_norm": 0.001116568577486857,
+      "grad_norm": 0.0005930920535898555,
       "learning_rate": 1.1973654344364925e-06,
-      "logits/chosen": -15.625,
-      "logits/rejected": -8.3125,
-      "logps/chosen": -760.0,
-      "logps/rejected": -4192.0,
+      "logits/chosen": -16.625,
+      "logits/rejected": -6.53125,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3856.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 23.5,
-      "rewards/rejected": -27.0,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 20.375,
+      "rewards/rejected": -23.625,
       "step": 2610
     },
     {
       "epoch": 2.743455497382199,
-      "grad_norm": 0.07212731011897386,
+      "grad_norm": 0.06878874178980637,
       "learning_rate": 1.1059809989586583e-06,
-      "logits/chosen": -16.125,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4288.0,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3952.0,
       "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.390625,
-      "rewards/margins": 23.375,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -23.5,
       "step": 2620
     },
     {
       "epoch": 2.7539267015706805,
-      "grad_norm": 0.009638352141423667,
+      "grad_norm": 0.013777314780696511,
       "learning_rate": 1.018144842388044e-06,
-      "logits/chosen": -15.5,
-      "logits/rejected": -9.125,
-      "logps/chosen": -796.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0009,
+      "logits/chosen": -16.75,
+      "logits/rejected": -7.34375,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0013,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.71875,
-      "rewards/margins": 23.25,
-      "rewards/rejected": -27.0,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -23.625,
       "step": 2630
     },
     {
       "epoch": 2.7643979057591626,
-      "grad_norm": 0.010691617149800768,
+      "grad_norm": 0.03199188541916619,
       "learning_rate": 9.338700084567109e-07,
-      "logits/chosen": -15.6875,
-      "logits/rejected": -10.25,
-      "logps/chosen": -728.0,
-      "logps/rejected": -4288.0,
+      "logits/chosen": -16.625,
+      "logits/rejected": -8.5625,
+      "logps/chosen": -680.0,
+      "logps/rejected": -4016.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 23.25,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 21.125,
+      "rewards/rejected": -24.125,
       "step": 2640
     },
     {
       "epoch": 2.774869109947644,
-      "grad_norm": 0.002358334883783129,
+      "grad_norm": 0.0063123311946760305,
       "learning_rate": 8.531690120377606e-07,
-      "logits/chosen": -15.875,
-      "logits/rejected": -9.0,
-      "logps/chosen": -784.0,
-      "logps/rejected": -4096.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.21875,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.59375,
-      "rewards/margins": 23.5,
-      "rewards/rejected": -27.0,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 20.375,
+      "rewards/rejected": -23.5,
       "step": 2650
     },
     {
       "epoch": 2.7853403141361257,
-      "grad_norm": 0.0012789627743359042,
+      "grad_norm": 0.0030481750710455013,
       "learning_rate": 7.760538372868636e-07,
-      "logits/chosen": -16.0,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -708.0,
-      "logps/rejected": -4160.0,
+      "logits/chosen": -16.875,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -668.0,
+      "logps/rejected": -3920.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.328125,
-      "rewards/margins": 21.75,
-      "rewards/rejected": -25.125,
+      "rewards/chosen": -2.9375,
+      "rewards/margins": 19.625,
+      "rewards/rejected": -22.5,
       "step": 2660
     },
     {
       "epoch": 2.7958115183246073,
-      "grad_norm": 0.0044279669844624146,
+      "grad_norm": 0.005162417109600259,
       "learning_rate": 7.025359358626166e-07,
-      "logits/chosen": -15.9375,
-      "logits/rejected": -9.375,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4352.0,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.46875,
+      "logps/chosen": -712.0,
+      "logps/rejected": -4000.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 24.25,
-      "rewards/rejected": -27.75,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 21.125,
+      "rewards/rejected": -24.125,
       "step": 2670
     },
     {
       "epoch": 2.806282722513089,
-      "grad_norm": 0.005464768231011858,
+      "grad_norm": 0.005196174682409069,
       "learning_rate": 6.326262252259463e-07,
-      "logits/chosen": -15.5,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -768.0,
-      "logps/rejected": -4256.0,
-      "loss": 0.0005,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3920.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.609375,
-      "rewards/margins": 23.125,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.25,
       "step": 2680
     },
     {
       "epoch": 2.816753926701571,
-      "grad_norm": 0.004072117142116979,
+      "grad_norm": 0.005880559633775991,
       "learning_rate": 5.66335087018871e-07,
-      "logits/chosen": -15.9375,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -756.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0008,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.875,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3872.0,
+      "loss": 0.0007,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.5,
-      "rewards/margins": 22.625,
-      "rewards/rejected": -26.125,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -23.25,
       "step": 2690
     },
     {
       "epoch": 2.8272251308900525,
-      "grad_norm": 0.000725402238058513,
+      "grad_norm": 0.0006464919664875174,
       "learning_rate": 5.036723655228226e-07,
-      "logits/chosen": -15.875,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -720.0,
-      "logps/rejected": -4320.0,
+      "logits/chosen": -16.75,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3920.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -23.25,
       "step": 2700
     },
     {
       "epoch": 2.8272251308900525,
-      "eval_logits/chosen": -15.5625,
-      "eval_logits/rejected": -15.25,
-      "eval_logps/chosen": -720.0,
-      "eval_logps/rejected": -740.0,
-      "eval_loss": 1.1504687070846558,
-      "eval_rewards/accuracies": 0.49609375,
-      "eval_rewards/chosen": -4.03125,
-      "eval_rewards/margins": 0.482421875,
-      "eval_rewards/rejected": -4.53125,
-      "eval_runtime": 46.6248,
-      "eval_samples_per_second": 42.896,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -16.75,
+      "eval_logits/rejected": -16.375,
+      "eval_logps/chosen": -676.0,
+      "eval_logps/rejected": -688.0,
+      "eval_loss": 1.1668281555175781,
+      "eval_rewards/accuracies": 0.490234375,
+      "eval_rewards/chosen": -3.59375,
+      "eval_rewards/margins": 0.412109375,
+      "eval_rewards/rejected": -4.0,
+      "eval_runtime": 58.2645,
+      "eval_samples_per_second": 34.326,
+      "eval_steps_per_second": 0.549,
       "step": 2700
     },
     {
       "epoch": 2.837696335078534,
-      "grad_norm": 0.0006535721522905963,
+      "grad_norm": 0.0006720959085331416,
       "learning_rate": 4.446473661967432e-07,
-      "logits/chosen": -16.0,
-      "logits/rejected": -9.125,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4064.0,
+      "logits/chosen": -17.125,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3760.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 22.625,
-      "rewards/rejected": -26.125,
+      "rewards/chosen": -2.875,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -22.875,
       "step": 2710
     },
     {
       "epoch": 2.8481675392670156,
-      "grad_norm": 0.013383438899806253,
+      "grad_norm": 0.007680763860389793,
       "learning_rate": 3.892688542952411e-07,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -8.9375,
-      "logps/chosen": -716.0,
-      "logps/rejected": -4128.0,
-      "loss": 0.0002,
+      "logits/chosen": -17.0,
+      "logits/rejected": -7.3125,
+      "logps/chosen": -656.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 23.125,
-      "rewards/rejected": -26.625,
+      "rewards/chosen": -2.8125,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.0,
       "step": 2720
     },
     {
       "epoch": 2.858638743455497,
-      "grad_norm": 0.010721294233034109,
+      "grad_norm": 0.001718671925712482,
       "learning_rate": 3.3754505356694e-07,
-      "logits/chosen": -16.125,
-      "logits/rejected": -9.25,
-      "logps/chosen": -744.0,
-      "logps/rejected": -4320.0,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.75,
+      "logps/chosen": -692.0,
+      "logps/rejected": -4000.0,
       "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 21.0,
+      "rewards/rejected": -24.0,
       "step": 2730
     },
     {
       "epoch": 2.869109947643979,
-      "grad_norm": 0.004703826451267692,
+      "grad_norm": 0.004639820582322925,
       "learning_rate": 2.8948364503322276e-07,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -720.0,
-      "logps/rejected": -4160.0,
-      "loss": 0.0031,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -27.125,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0006,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 20.875,
+      "rewards/rejected": -23.875,
       "step": 2740
     },
     {
       "epoch": 2.8795811518324608,
-      "grad_norm": 0.013951572871782138,
+      "grad_norm": 0.00678097350711074,
       "learning_rate": 2.4509176584762706e-07,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -8.5,
-      "logps/chosen": -732.0,
-      "logps/rejected": -4224.0,
+      "logits/chosen": -16.75,
+      "logits/rejected": -6.8125,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3872.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 23.25,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -23.25,
       "step": 2750
     },
     {
       "epoch": 2.8900523560209423,
-      "grad_norm": 0.0013388301522194548,
+      "grad_norm": 0.0017287559570423386,
       "learning_rate": 2.0437600823595694e-07,
-      "logits/chosen": -15.5,
-      "logits/rejected": -9.25,
-      "logps/chosen": -788.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.125,
+      "logits/rejected": -7.625,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3920.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.703125,
-      "rewards/margins": 23.125,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -23.125,
       "step": 2760
     },
     {
       "epoch": 2.900523560209424,
-      "grad_norm": 0.03748902951038762,
+      "grad_norm": 0.056861200483102226,
       "learning_rate": 1.6734241851733e-07,
-      "logits/chosen": -15.875,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -800.0,
-      "logps/rejected": -4352.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.5,
+      "logits/rejected": -7.15625,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3952.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.75,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.75,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -23.75,
       "step": 2770
     },
     {
       "epoch": 2.9109947643979055,
-      "grad_norm": 7.792196761086201e-05,
+      "grad_norm": 0.000185478066581428,
       "learning_rate": 1.339964962062984e-07,
-      "logits/chosen": -16.0,
-      "logits/rejected": -10.0,
-      "logps/chosen": -700.0,
-      "logps/rejected": -4384.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.625,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -640.0,
+      "logps/rejected": -4016.0,
+      "loss": 0.0008,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 24.0,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -2.8125,
+      "rewards/margins": 21.0,
+      "rewards/rejected": -23.75,
       "step": 2780
     },
     {
       "epoch": 2.9214659685863875,
-      "grad_norm": 0.001844729047164741,
+      "grad_norm": 0.01764894970149795,
       "learning_rate": 1.0434319319617136e-07,
-      "logits/chosen": -15.875,
-      "logits/rejected": -10.0,
-      "logps/chosen": -712.0,
-      "logps/rejected": -4416.0,
-      "loss": 0.0004,
+      "logits/chosen": -16.625,
+      "logits/rejected": -8.25,
+      "logps/chosen": -676.0,
+      "logps/rejected": -4048.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 24.5,
-      "rewards/rejected": -28.0,
+      "rewards/chosen": -3.0625,
+      "rewards/margins": 21.25,
+      "rewards/rejected": -24.375,
       "step": 2790
     },
     {
       "epoch": 2.931937172774869,
-      "grad_norm": 0.0014877818919606682,
+      "grad_norm": 0.0006298773319941659,
       "learning_rate": 7.838691302365363e-08,
-      "logits/chosen": -15.8125,
-      "logits/rejected": -9.5,
-      "logps/chosen": -728.0,
-      "logps/rejected": -4224.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.75,
+      "logits/rejected": -7.84375,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3920.0,
+      "loss": 0.0003,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 23.375,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 20.75,
+      "rewards/rejected": -23.625,
       "step": 2800
     },
     {
       "epoch": 2.931937172774869,
-      "eval_logits/chosen": -15.5625,
-      "eval_logits/rejected": -15.25,
-      "eval_logps/chosen": -724.0,
-      "eval_logps/rejected": -744.0,
-      "eval_loss": 1.151843786239624,
-      "eval_rewards/accuracies": 0.498046875,
-      "eval_rewards/chosen": -4.0625,
-      "eval_rewards/margins": 0.482421875,
-      "eval_rewards/rejected": -4.53125,
-      "eval_runtime": 46.626,
-      "eval_samples_per_second": 42.895,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -16.875,
+      "eval_logits/rejected": -16.375,
+      "eval_logps/chosen": -676.0,
+      "eval_logps/rejected": -688.0,
+      "eval_loss": 1.166765570640564,
+      "eval_rewards/accuracies": 0.490234375,
+      "eval_rewards/chosen": -3.59375,
+      "eval_rewards/margins": 0.412109375,
+      "eval_rewards/rejected": -4.0,
+      "eval_runtime": 58.2876,
+      "eval_samples_per_second": 34.313,
+      "eval_steps_per_second": 0.549,
       "step": 2800
     },
     {
       "epoch": 2.9424083769633507,
-      "grad_norm": 0.009167364388714315,
+      "grad_norm": 0.0040975119117121605,
       "learning_rate": 5.613151021490459e-08,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.4375,
-      "logps/chosen": -644.0,
-      "logps/rejected": -4352.0,
-      "loss": 0.0002,
+      "logits/chosen": -16.875,
+      "logits/rejected": -7.75,
+      "logps/chosen": -608.0,
+      "logps/rejected": -4048.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.171875,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -26.75,
+      "rewards/chosen": -2.8125,
+      "rewards/margins": 20.875,
+      "rewards/rejected": -23.625,
       "step": 2810
     },
     {
       "epoch": 2.9528795811518327,
-      "grad_norm": 0.0045128825185065755,
+      "grad_norm": 0.0065534249330385565,
       "learning_rate": 3.7580289713159965e-08,
-      "logits/chosen": -15.625,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -764.0,
-      "logps/rejected": -3920.0,
+      "logits/chosen": -16.75,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -696.0,
+      "logps/rejected": -3632.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.578125,
-      "rewards/margins": 21.75,
-      "rewards/rejected": -25.375,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 19.625,
+      "rewards/rejected": -22.5,
       "step": 2820
     },
     {
       "epoch": 2.9633507853403143,
-      "grad_norm": 0.0070913888976011415,
+      "grad_norm": 0.002657069864231234,
       "learning_rate": 2.2736006387927277e-08,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.75,
-      "logps/chosen": -760.0,
-      "logps/rejected": -4352.0,
+      "logits/chosen": -16.75,
+      "logits/rejected": -8.25,
+      "logps/chosen": -716.0,
+      "logps/rejected": -4000.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.609375,
-      "rewards/margins": 23.875,
-      "rewards/rejected": -27.5,
+      "rewards/chosen": -3.140625,
+      "rewards/margins": 21.0,
+      "rewards/rejected": -24.125,
       "step": 2830
     },
     {
       "epoch": 2.973821989528796,
-      "grad_norm": 0.004162508187154523,
+      "grad_norm": 0.0023835430771258844,
       "learning_rate": 1.1600864625893604e-08,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.0625,
-      "logps/chosen": -724.0,
-      "logps/rejected": -4288.0,
-      "loss": 0.0001,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3936.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.40625,
-      "rewards/margins": 23.875,
-      "rewards/rejected": -27.25,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 20.875,
+      "rewards/rejected": -23.75,
       "step": 2840
     },
     {
       "epoch": 2.9842931937172774,
-      "grad_norm": 0.005228912394016512,
+      "grad_norm": 0.002712137820076971,
       "learning_rate": 4.176518003573549e-09,
-      "logits/chosen": -15.875,
-      "logits/rejected": -9.0,
-      "logps/chosen": -736.0,
-      "logps/rejected": -4384.0,
-      "loss": 0.0004,
+      "logits/chosen": -16.625,
+      "logits/rejected": -7.40625,
+      "logps/chosen": -696.0,
+      "logps/rejected": -4016.0,
+      "loss": 0.0006,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 24.25,
-      "rewards/rejected": -27.875,
+      "rewards/chosen": -3.140625,
+      "rewards/margins": 20.875,
+      "rewards/rejected": -24.0,
       "step": 2850
     },
     {
       "epoch": 2.994764397905759,
-      "grad_norm": 0.0019148515385177618,
+      "grad_norm": 0.007304194259673967,
       "learning_rate": 4.640690417528903e-10,
-      "logits/chosen": -15.75,
-      "logits/rejected": -9.375,
-      "logps/chosen": -716.0,
-      "logps/rejected": -4320.0,
+      "logits/chosen": -16.25,
+      "logits/rejected": -7.71875,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3936.0,
       "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 24.5,
-      "rewards/rejected": -28.0,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 21.125,
+      "rewards/rejected": -24.25,
       "step": 2860
     },
     {
       "epoch": 3.0,
       "step": 2865,
       "total_flos": 0.0,
-      "train_loss": 0.02262802565294394,
-      "train_runtime": 12206.3813,
-      "train_samples_per_second": 15.021,
-      "train_steps_per_second": 0.235
+      "train_loss": 0.02151861389916891,
+      "train_runtime": 13053.1122,
+      "train_samples_per_second": 14.047,
+      "train_steps_per_second": 0.219
     }
   ],
   "logging_steps": 10,