diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -10,7 +10,7 @@
   "log_history": [
     {
       "epoch": 0.0010471204188481676,
-      "grad_norm": 1.806400698442994,
+      "grad_norm": 1.806436074491372,
       "learning_rate": 1.7421602787456448e-07,
       "logits/chosen": -11.625,
       "logits/rejected": -11.75,
@@ -25,4750 +25,4750 @@
     },
     {
       "epoch": 0.010471204188481676,
-      "grad_norm": 1.754647374830844,
+      "grad_norm": 1.753031830811777,
       "learning_rate": 1.7421602787456445e-06,
       "logits/chosen": -11.125,
       "logits/rejected": -10.5,
       "logps/chosen": -428.0,
       "logps/rejected": -756.0,
       "loss": 0.6922,
-      "rewards/accuracies": 0.1736111044883728,
-      "rewards/chosen": -0.00021266937255859375,
-      "rewards/margins": 0.0001010894775390625,
-      "rewards/rejected": -0.00031280517578125,
+      "rewards/accuracies": 0.1527777761220932,
+      "rewards/chosen": 0.00015354156494140625,
+      "rewards/margins": 0.000335693359375,
+      "rewards/rejected": -0.00017833709716796875,
       "step": 10
     },
     {
       "epoch": 0.020942408376963352,
-      "grad_norm": 1.6443926478703903,
+      "grad_norm": 1.6525738053961263,
       "learning_rate": 3.484320557491289e-06,
       "logits/chosen": -11.0,
       "logits/rejected": -10.8125,
       "logps/chosen": -402.0,
       "logps/rejected": -576.0,
-      "loss": 0.6886,
-      "rewards/accuracies": 0.39375001192092896,
-      "rewards/chosen": 0.0005645751953125,
-      "rewards/margins": 0.01226806640625,
-      "rewards/rejected": -0.01171875,
+      "loss": 0.6891,
+      "rewards/accuracies": 0.3499999940395355,
+      "rewards/chosen": -0.00055694580078125,
+      "rewards/margins": 0.00958251953125,
+      "rewards/rejected": -0.0101318359375,
       "step": 20
     },
     {
       "epoch": 0.031413612565445025,
-      "grad_norm": 1.6744022302366757,
+      "grad_norm": 1.6752503762489155,
       "learning_rate": 5.226480836236934e-06,
       "logits/chosen": -11.625,
       "logits/rejected": -11.375,
       "logps/chosen": -430.0,
       "logps/rejected": -532.0,
-      "loss": 0.6764,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": -0.0257568359375,
-      "rewards/margins": 0.0263671875,
-      "rewards/rejected": -0.05224609375,
+      "loss": 0.6757,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.025634765625,
+      "rewards/margins": 0.02783203125,
+      "rewards/rejected": -0.053466796875,
       "step": 30
     },
     {
       "epoch": 0.041884816753926704,
-      "grad_norm": 1.9599604304110412,
+      "grad_norm": 1.965171751722186,
       "learning_rate": 6.968641114982578e-06,
       "logits/chosen": -12.4375,
       "logits/rejected": -12.0,
       "logps/chosen": -448.0,
       "logps/rejected": -616.0,
-      "loss": 0.6365,
-      "rewards/accuracies": 0.643750011920929,
+      "loss": 0.6364,
+      "rewards/accuracies": 0.6312500238418579,
       "rewards/chosen": -0.189453125,
-      "rewards/margins": 0.1240234375,
+      "rewards/margins": 0.1220703125,
       "rewards/rejected": -0.3125,
       "step": 40
     },
     {
       "epoch": 0.05235602094240838,
-      "grad_norm": 2.0242389220979042,
+      "grad_norm": 2.0214971099991823,
       "learning_rate": 8.710801393728225e-06,
       "logits/chosen": -12.1875,
       "logits/rejected": -12.0,
       "logps/chosen": -462.0,
       "logps/rejected": -700.0,
       "loss": 0.5849,
-      "rewards/accuracies": 0.6812499761581421,
+      "rewards/accuracies": 0.6625000238418579,
       "rewards/chosen": -0.5078125,
       "rewards/margins": 0.419921875,
-      "rewards/rejected": -0.9296875,
+      "rewards/rejected": -0.92578125,
       "step": 50
     },
     {
       "epoch": 0.06282722513089005,
-      "grad_norm": 1.8508753348060478,
+      "grad_norm": 1.8574919158967238,
       "learning_rate": 1.0452961672473868e-05,
       "logits/chosen": -11.0,
       "logits/rejected": -10.9375,
       "logps/chosen": -398.0,
       "logps/rejected": -692.0,
-      "loss": 0.5395,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -0.50390625,
+      "loss": 0.5398,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -0.5,
       "rewards/margins": 0.76171875,
       "rewards/rejected": -1.265625,
       "step": 60
     },
     {
       "epoch": 0.07329842931937172,
-      "grad_norm": 2.0232007306056765,
+      "grad_norm": 2.0180102172020864,
       "learning_rate": 1.2195121951219513e-05,
       "logits/chosen": -12.0,
       "logits/rejected": -11.6875,
       "logps/chosen": -516.0,
       "logps/rejected": -848.0,
-      "loss": 0.535,
+      "loss": 0.5346,
       "rewards/accuracies": 0.706250011920929,
       "rewards/chosen": -0.81640625,
-      "rewards/margins": 1.0,
+      "rewards/margins": 0.99609375,
       "rewards/rejected": -1.8125,
       "step": 70
     },
     {
       "epoch": 0.08376963350785341,
-      "grad_norm": 2.132580312351081,
+      "grad_norm": 2.118796993161168,
       "learning_rate": 1.3937282229965156e-05,
       "logits/chosen": -12.3125,
       "logits/rejected": -12.1875,
       "logps/chosen": -506.0,
-      "logps/rejected": -924.0,
-      "loss": 0.4983,
-      "rewards/accuracies": 0.768750011920929,
+      "logps/rejected": -920.0,
+      "loss": 0.4982,
+      "rewards/accuracies": 0.7562500238418579,
       "rewards/chosen": -1.0859375,
-      "rewards/margins": 1.53125,
-      "rewards/rejected": -2.625,
+      "rewards/margins": 1.5234375,
+      "rewards/rejected": -2.609375,
       "step": 80
     },
     {
       "epoch": 0.09424083769633508,
-      "grad_norm": 3.3738414136521584,
+      "grad_norm": 3.4881945422416933,
       "learning_rate": 1.56794425087108e-05,
-      "logits/chosen": -12.1875,
-      "logits/rejected": -11.9375,
-      "logps/chosen": -506.0,
-      "logps/rejected": -720.0,
-      "loss": 0.5304,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -510.0,
+      "logps/rejected": -724.0,
+      "loss": 0.5279,
       "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -1.453125,
-      "rewards/margins": 1.0,
-      "rewards/rejected": -2.453125,
+      "rewards/chosen": -1.4921875,
+      "rewards/margins": 0.98828125,
+      "rewards/rejected": -2.484375,
       "step": 90
     },
     {
       "epoch": 0.10471204188481675,
-      "grad_norm": 2.2224399434934434,
+      "grad_norm": 2.152922060914835,
       "learning_rate": 1.742160278745645e-05,
-      "logits/chosen": -11.5,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -656.0,
-      "logps/rejected": -968.0,
-      "loss": 0.5081,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -1.7890625,
-      "rewards/margins": 1.25,
-      "rewards/rejected": -3.046875,
+      "logits/chosen": -11.8125,
+      "logits/rejected": -11.5625,
+      "logps/chosen": -660.0,
+      "logps/rejected": -960.0,
+      "loss": 0.5108,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -1.8203125,
+      "rewards/margins": 1.1640625,
+      "rewards/rejected": -2.984375,
       "step": 100
     },
     {
       "epoch": 0.10471204188481675,
-      "eval_logits/chosen": -10.1875,
-      "eval_logits/rejected": -9.75,
-      "eval_logps/chosen": -432.0,
-      "eval_logps/rejected": -424.0,
-      "eval_loss": 0.6819999814033508,
-      "eval_rewards/accuracies": 0.59765625,
-      "eval_rewards/chosen": -1.140625,
-      "eval_rewards/margins": 0.203125,
-      "eval_rewards/rejected": -1.34375,
-      "eval_runtime": 48.3213,
-      "eval_samples_per_second": 41.39,
-      "eval_steps_per_second": 0.662,
+      "eval_logits/chosen": -10.8125,
+      "eval_logits/rejected": -10.375,
+      "eval_logps/chosen": -442.0,
+      "eval_logps/rejected": -434.0,
+      "eval_loss": 0.6707968711853027,
+      "eval_rewards/accuracies": 0.609375,
+      "eval_rewards/chosen": -1.234375,
+      "eval_rewards/margins": 0.2177734375,
+      "eval_rewards/rejected": -1.453125,
+      "eval_runtime": 49.1287,
+      "eval_samples_per_second": 40.709,
+      "eval_steps_per_second": 0.651,
       "step": 100
     },
     {
       "epoch": 0.11518324607329843,
-      "grad_norm": 2.5536117772295404,
+      "grad_norm": 2.7152160967946757,
       "learning_rate": 1.9163763066202093e-05,
-      "logits/chosen": -9.9375,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -564.0,
-      "logps/rejected": -880.0,
-      "loss": 0.4696,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -1.46875,
-      "rewards/margins": 1.328125,
-      "rewards/rejected": -2.796875,
+      "logits/chosen": -10.1875,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -588.0,
+      "logps/rejected": -912.0,
+      "loss": 0.4537,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -1.703125,
+      "rewards/margins": 1.390625,
+      "rewards/rejected": -3.09375,
       "step": 110
     },
     {
       "epoch": 0.1256544502617801,
-      "grad_norm": 2.6964853754209717,
+      "grad_norm": 2.8078148154437925,
       "learning_rate": 2.0905923344947736e-05,
-      "logits/chosen": -11.0625,
-      "logits/rejected": -11.0,
-      "logps/chosen": -528.0,
-      "logps/rejected": -824.0,
-      "loss": 0.4889,
+      "logits/chosen": -9.375,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -592.0,
+      "logps/rejected": -884.0,
+      "loss": 0.479,
       "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -1.6640625,
-      "rewards/margins": 1.3203125,
-      "rewards/rejected": -2.984375,
+      "rewards/chosen": -2.28125,
+      "rewards/margins": 1.296875,
+      "rewards/rejected": -3.578125,
       "step": 120
     },
     {
       "epoch": 0.13612565445026178,
-      "grad_norm": 2.6235465585012347,
+      "grad_norm": 2.35281267807467,
       "learning_rate": 2.264808362369338e-05,
-      "logits/chosen": -10.0,
-      "logits/rejected": -10.0,
-      "logps/chosen": -584.0,
-      "logps/rejected": -888.0,
-      "loss": 0.474,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -1.75,
-      "rewards/margins": 1.2265625,
-      "rewards/rejected": -2.96875,
+      "logits/chosen": -10.0625,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -572.0,
+      "logps/rejected": -864.0,
+      "loss": 0.4769,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -1.6171875,
+      "rewards/margins": 1.109375,
+      "rewards/rejected": -2.71875,
       "step": 130
     },
     {
       "epoch": 0.14659685863874344,
-      "grad_norm": 3.137523632711562,
+      "grad_norm": 2.972375209292521,
       "learning_rate": 2.4390243902439026e-05,
-      "logits/chosen": -11.4375,
-      "logits/rejected": -11.25,
-      "logps/chosen": -636.0,
-      "logps/rejected": -988.0,
-      "loss": 0.4655,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -1.9140625,
-      "rewards/margins": 1.4453125,
-      "rewards/rejected": -3.359375,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -12.8125,
+      "logps/chosen": -652.0,
+      "logps/rejected": -1004.0,
+      "loss": 0.4693,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -2.046875,
+      "rewards/margins": 1.4609375,
+      "rewards/rejected": -3.515625,
       "step": 140
     },
     {
       "epoch": 0.15706806282722513,
-      "grad_norm": 2.4716487395797584,
+      "grad_norm": 2.7041561232033717,
       "learning_rate": 2.6132404181184672e-05,
-      "logits/chosen": -11.0625,
-      "logits/rejected": -10.875,
-      "logps/chosen": -672.0,
-      "logps/rejected": -924.0,
-      "loss": 0.4827,
-      "rewards/accuracies": 0.793749988079071,
+      "logits/chosen": -12.0,
+      "logits/rejected": -11.75,
+      "logps/chosen": -676.0,
+      "logps/rejected": -928.0,
+      "loss": 0.479,
+      "rewards/accuracies": 0.8187500238418579,
       "rewards/chosen": -2.5,
-      "rewards/margins": 1.1953125,
-      "rewards/rejected": -3.6875,
+      "rewards/margins": 1.25,
+      "rewards/rejected": -3.75,
       "step": 150
     },
     {
       "epoch": 0.16753926701570682,
-      "grad_norm": 2.1480232597076863,
+      "grad_norm": 2.087713233064345,
       "learning_rate": 2.7874564459930312e-05,
-      "logits/chosen": -12.0625,
-      "logits/rejected": -11.5,
-      "logps/chosen": -636.0,
-      "logps/rejected": -904.0,
-      "loss": 0.4591,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -2.453125,
-      "rewards/margins": 1.1875,
-      "rewards/rejected": -3.640625,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -660.0,
+      "logps/rejected": -932.0,
+      "loss": 0.4574,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -2.65625,
+      "rewards/margins": 1.265625,
+      "rewards/rejected": -3.921875,
       "step": 160
     },
     {
       "epoch": 0.17801047120418848,
-      "grad_norm": 2.585566163174424,
+      "grad_norm": 2.381916331427287,
       "learning_rate": 2.9616724738675962e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -9.75,
-      "logps/chosen": -668.0,
-      "logps/rejected": -1104.0,
-      "loss": 0.4211,
-      "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -2.40625,
-      "rewards/margins": 2.140625,
-      "rewards/rejected": -4.53125,
+      "logits/chosen": -9.125,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -724.0,
+      "logps/rejected": -1176.0,
+      "loss": 0.4194,
+      "rewards/accuracies": 0.831250011920929,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 2.28125,
+      "rewards/rejected": -5.28125,
       "step": 170
     },
     {
       "epoch": 0.18848167539267016,
-      "grad_norm": 2.4070574933282933,
+      "grad_norm": 2.527678455386163,
       "learning_rate": 3.13588850174216e-05,
-      "logits/chosen": -10.125,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -712.0,
-      "logps/rejected": -1120.0,
-      "loss": 0.4303,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -2.71875,
+      "logits/chosen": -9.0,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -724.0,
+      "logps/rejected": -1128.0,
+      "loss": 0.4248,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": -2.875,
       "rewards/margins": 2.109375,
-      "rewards/rejected": -4.8125,
+      "rewards/rejected": -4.96875,
       "step": 180
     },
     {
       "epoch": 0.19895287958115182,
-      "grad_norm": 3.015118581025222,
+      "grad_norm": 3.03370138884052,
       "learning_rate": 3.310104529616725e-05,
-      "logits/chosen": -7.96875,
-      "logits/rejected": -7.96875,
-      "logps/chosen": -724.0,
-      "logps/rejected": -1104.0,
-      "loss": 0.4271,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -3.265625,
-      "rewards/margins": 1.8125,
-      "rewards/rejected": -5.0625,
+      "logits/chosen": -9.1875,
+      "logits/rejected": -9.125,
+      "logps/chosen": -760.0,
+      "logps/rejected": -1152.0,
+      "loss": 0.4235,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -3.640625,
+      "rewards/margins": 1.9140625,
+      "rewards/rejected": -5.5625,
       "step": 190
     },
     {
       "epoch": 0.2094240837696335,
-      "grad_norm": 2.379514964244399,
+      "grad_norm": 2.275931638108211,
       "learning_rate": 3.48432055749129e-05,
-      "logits/chosen": -6.84375,
-      "logits/rejected": -6.28125,
-      "logps/chosen": -760.0,
-      "logps/rejected": -1088.0,
-      "loss": 0.4521,
+      "logits/chosen": -10.5,
+      "logits/rejected": -10.125,
+      "logps/chosen": -756.0,
+      "logps/rejected": -1080.0,
+      "loss": 0.4452,
       "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -3.0625,
-      "rewards/margins": 1.5625,
-      "rewards/rejected": -4.625,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 1.4921875,
+      "rewards/rejected": -4.53125,
       "step": 200
     },
     {
       "epoch": 0.2094240837696335,
-      "eval_logits/chosen": -8.0625,
-      "eval_logits/rejected": -7.25,
-      "eval_logps/chosen": -608.0,
-      "eval_logps/rejected": -632.0,
-      "eval_loss": 0.7578906416893005,
-      "eval_rewards/accuracies": 0.646484375,
-      "eval_rewards/chosen": -2.890625,
-      "eval_rewards/margins": 0.53125,
-      "eval_rewards/rejected": -3.421875,
-      "eval_runtime": 46.649,
-      "eval_samples_per_second": 42.873,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -10.1875,
+      "eval_logits/rejected": -9.625,
+      "eval_logps/chosen": -600.0,
+      "eval_logps/rejected": -616.0,
+      "eval_loss": 0.7220468521118164,
+      "eval_rewards/accuracies": 0.6171875,
+      "eval_rewards/chosen": -2.828125,
+      "eval_rewards/margins": 0.443359375,
+      "eval_rewards/rejected": -3.28125,
+      "eval_runtime": 47.0795,
+      "eval_samples_per_second": 42.481,
+      "eval_steps_per_second": 0.68,
       "step": 200
     },
     {
       "epoch": 0.2198952879581152,
-      "grad_norm": 2.4380532238903907,
+      "grad_norm": 2.790099646654149,
       "learning_rate": 3.6585365853658535e-05,
-      "logits/chosen": -10.0,
-      "logits/rejected": -9.625,
-      "logps/chosen": -648.0,
-      "logps/rejected": -900.0,
-      "loss": 0.4556,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -2.8125,
-      "rewards/margins": 1.171875,
-      "rewards/rejected": -3.984375,
+      "logits/chosen": -10.25,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -624.0,
+      "logps/rejected": -880.0,
+      "loss": 0.4422,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -2.609375,
+      "rewards/margins": 1.15625,
+      "rewards/rejected": -3.765625,
       "step": 210
     },
     {
       "epoch": 0.23036649214659685,
-      "grad_norm": 1.7578156817996728,
+      "grad_norm": 2.4445867175858362,
       "learning_rate": 3.8327526132404185e-05,
-      "logits/chosen": -11.9375,
-      "logits/rejected": -11.25,
-      "logps/chosen": -636.0,
-      "logps/rejected": -1012.0,
-      "loss": 0.4034,
-      "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": -2.171875,
-      "rewards/margins": 1.65625,
-      "rewards/rejected": -3.84375,
+      "logits/chosen": -9.8125,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -736.0,
+      "logps/rejected": -1136.0,
+      "loss": 0.4087,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 1.90625,
+      "rewards/rejected": -5.0625,
       "step": 220
     },
     {
       "epoch": 0.24083769633507854,
-      "grad_norm": 2.3073288799435776,
+      "grad_norm": 2.189536210082279,
       "learning_rate": 4.006968641114983e-05,
-      "logits/chosen": -10.75,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -740.0,
-      "logps/rejected": -996.0,
-      "loss": 0.442,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -3.484375,
-      "rewards/margins": 1.5546875,
-      "rewards/rejected": -5.03125,
+      "logits/chosen": -9.6875,
+      "logits/rejected": -9.625,
+      "logps/chosen": -716.0,
+      "logps/rejected": -952.0,
+      "loss": 0.4481,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 1.3515625,
+      "rewards/rejected": -4.625,
       "step": 230
     },
     {
       "epoch": 0.2513089005235602,
-      "grad_norm": 2.9680533124904507,
+      "grad_norm": 4.83195323954807,
       "learning_rate": 4.181184668989547e-05,
-      "logits/chosen": -10.5625,
-      "logits/rejected": -9.875,
-      "logps/chosen": -704.0,
-      "logps/rejected": -1024.0,
-      "loss": 0.4366,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -3.328125,
-      "rewards/margins": 1.625,
-      "rewards/rejected": -4.9375,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -8.375,
+      "logps/chosen": -760.0,
+      "logps/rejected": -1128.0,
+      "loss": 0.4543,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -3.875,
+      "rewards/margins": 2.09375,
+      "rewards/rejected": -6.0,
       "step": 240
     },
     {
       "epoch": 0.2617801047120419,
-      "grad_norm": 3.595072601390531,
+      "grad_norm": 2.9857595735325306,
       "learning_rate": 4.3554006968641115e-05,
-      "logits/chosen": -10.5,
-      "logits/rejected": -9.875,
+      "logits/chosen": -10.9375,
+      "logits/rejected": -10.5,
       "logps/chosen": -728.0,
-      "logps/rejected": -1048.0,
-      "loss": 0.4247,
-      "rewards/accuracies": 0.78125,
+      "logps/rejected": -1032.0,
+      "loss": 0.4357,
+      "rewards/accuracies": 0.8062499761581421,
       "rewards/chosen": -2.75,
-      "rewards/margins": 1.796875,
-      "rewards/rejected": -4.53125,
+      "rewards/margins": 1.609375,
+      "rewards/rejected": -4.34375,
       "step": 250
     },
     {
       "epoch": 0.27225130890052357,
-      "grad_norm": 1.870996228943309,
+      "grad_norm": 3.059164866934553,
       "learning_rate": 4.529616724738676e-05,
-      "logits/chosen": -9.1875,
-      "logits/rejected": -8.5,
-      "logps/chosen": -720.0,
-      "logps/rejected": -1152.0,
-      "loss": 0.4025,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -3.421875,
-      "rewards/margins": 2.40625,
-      "rewards/rejected": -5.84375,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -668.0,
+      "logps/rejected": -1064.0,
+      "loss": 0.4133,
+      "rewards/accuracies": 0.8187500238418579,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 2.046875,
+      "rewards/rejected": -4.9375,
       "step": 260
     },
     {
       "epoch": 0.28272251308900526,
-      "grad_norm": 2.051575033116394,
+      "grad_norm": 1.846300303313416,
       "learning_rate": 4.703832752613241e-05,
-      "logits/chosen": -11.875,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -728.0,
+      "logits/chosen": -10.0,
+      "logits/rejected": -9.625,
+      "logps/chosen": -736.0,
       "logps/rejected": -1136.0,
-      "loss": 0.4484,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 2.03125,
-      "rewards/rejected": -5.5625,
+      "loss": 0.4582,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -3.640625,
+      "rewards/margins": 1.8984375,
+      "rewards/rejected": -5.53125,
       "step": 270
     },
     {
       "epoch": 0.2931937172774869,
-      "grad_norm": 2.4441465903872164,
+      "grad_norm": 2.5014181154607797,
       "learning_rate": 4.878048780487805e-05,
-      "logits/chosen": -12.9375,
-      "logits/rejected": -12.375,
-      "logps/chosen": -724.0,
-      "logps/rejected": -1048.0,
-      "loss": 0.4751,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -3.65625,
-      "rewards/margins": 1.4765625,
-      "rewards/rejected": -5.15625,
+      "logits/chosen": -9.625,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -696.0,
+      "logps/rejected": -1020.0,
+      "loss": 0.4562,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -3.375,
+      "rewards/margins": 1.484375,
+      "rewards/rejected": -4.84375,
       "step": 280
     },
     {
       "epoch": 0.3036649214659686,
-      "grad_norm": 2.626048297305558,
+      "grad_norm": 2.3279926698388334,
       "learning_rate": 4.999983293481418e-05,
-      "logits/chosen": -12.25,
-      "logits/rejected": -12.4375,
-      "logps/chosen": -696.0,
-      "logps/rejected": -1064.0,
-      "loss": 0.4388,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -3.109375,
-      "rewards/margins": 1.640625,
-      "rewards/rejected": -4.75,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -820.0,
+      "logps/rejected": -1184.0,
+      "loss": 0.4449,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -4.34375,
+      "rewards/margins": 1.6015625,
+      "rewards/rejected": -5.96875,
       "step": 290
     },
     {
       "epoch": 0.31413612565445026,
-      "grad_norm": 2.7792476826298977,
+      "grad_norm": 2.8143471281423187,
       "learning_rate": 4.9996862949181595e-05,
-      "logits/chosen": -12.8125,
-      "logits/rejected": -12.8125,
-      "logps/chosen": -692.0,
-      "logps/rejected": -1064.0,
-      "loss": 0.4267,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -2.84375,
-      "rewards/margins": 2.0,
-      "rewards/rejected": -4.84375,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -8.75,
+      "logps/chosen": -760.0,
+      "logps/rejected": -1120.0,
+      "loss": 0.4197,
+      "rewards/accuracies": 0.831250011920929,
+      "rewards/chosen": -3.515625,
+      "rewards/margins": 1.8828125,
+      "rewards/rejected": -5.40625,
       "step": 300
     },
     {
       "epoch": 0.31413612565445026,
-      "eval_logits/chosen": -14.0,
-      "eval_logits/rejected": -13.875,
-      "eval_logps/chosen": -564.0,
-      "eval_logps/rejected": -584.0,
-      "eval_loss": 0.7476093769073486,
-      "eval_rewards/accuracies": 0.5859375,
-      "eval_rewards/chosen": -2.453125,
-      "eval_rewards/margins": 0.515625,
-      "eval_rewards/rejected": -2.96875,
-      "eval_runtime": 46.6053,
-      "eval_samples_per_second": 42.914,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -8.375,
+      "eval_logits/rejected": -7.5,
+      "eval_logps/chosen": -716.0,
+      "eval_logps/rejected": -748.0,
+      "eval_loss": 0.7965937256813049,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -3.984375,
+      "eval_rewards/margins": 0.61328125,
+      "eval_rewards/rejected": -4.59375,
+      "eval_runtime": 47.0746,
+      "eval_samples_per_second": 42.486,
+      "eval_steps_per_second": 0.68,
       "step": 300
     },
     {
       "epoch": 0.32460732984293195,
-      "grad_norm": 2.7275392603478785,
+      "grad_norm": 3.3412549989746747,
       "learning_rate": 4.9990180911526896e-05,
-      "logits/chosen": -14.25,
-      "logits/rejected": -13.8125,
-      "logps/chosen": -648.0,
-      "logps/rejected": -984.0,
-      "loss": 0.4152,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -2.765625,
-      "rewards/margins": 1.6328125,
-      "rewards/rejected": -4.40625,
+      "logits/chosen": -8.75,
+      "logits/rejected": -8.375,
+      "logps/chosen": -868.0,
+      "logps/rejected": -1232.0,
+      "loss": 0.4075,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 1.890625,
+      "rewards/rejected": -6.84375,
       "step": 310
     },
     {
       "epoch": 0.33507853403141363,
-      "grad_norm": 2.2853171512095245,
+      "grad_norm": 2.190128732250711,
       "learning_rate": 4.997978781413745e-05,
-      "logits/chosen": -11.9375,
-      "logits/rejected": -11.125,
-      "logps/chosen": -880.0,
-      "logps/rejected": -1208.0,
-      "loss": 0.4604,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -4.34375,
-      "rewards/margins": 2.15625,
-      "rewards/rejected": -6.5,
+      "logits/chosen": -9.5625,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -920.0,
+      "logps/rejected": -1232.0,
+      "loss": 0.4571,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -4.78125,
+      "rewards/margins": 1.9453125,
+      "rewards/rejected": -6.71875,
       "step": 320
     },
     {
       "epoch": 0.34554973821989526,
-      "grad_norm": 3.537133265761021,
+      "grad_norm": 3.536254358142373,
       "learning_rate": 4.996568520039565e-05,
-      "logits/chosen": -11.8125,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -692.0,
-      "logps/rejected": -1080.0,
-      "loss": 0.4257,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -2.78125,
-      "rewards/margins": 2.03125,
-      "rewards/rejected": -4.8125,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -704.0,
+      "logps/rejected": -1104.0,
+      "loss": 0.4034,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 2.078125,
+      "rewards/rejected": -5.0,
       "step": 330
     },
     {
       "epoch": 0.35602094240837695,
-      "grad_norm": 3.0781821409502292,
+      "grad_norm": 2.893960585920598,
       "learning_rate": 4.994787516454973e-05,
-      "logits/chosen": -10.0,
-      "logits/rejected": -9.5,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -10.875,
       "logps/chosen": -800.0,
-      "logps/rejected": -1120.0,
-      "loss": 0.4442,
-      "rewards/accuracies": 0.8062499761581421,
+      "logps/rejected": -1096.0,
+      "loss": 0.447,
+      "rewards/accuracies": 0.800000011920929,
       "rewards/chosen": -4.03125,
-      "rewards/margins": 1.9609375,
-      "rewards/rejected": -6.0,
+      "rewards/margins": 1.8203125,
+      "rewards/rejected": -5.84375,
       "step": 340
     },
     {
       "epoch": 0.36649214659685864,
-      "grad_norm": 3.330156216164261,
+      "grad_norm": 3.96725204340208,
       "learning_rate": 4.992636035140273e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -12.9375,
-      "logps/chosen": -684.0,
-      "logps/rejected": -1136.0,
-      "loss": 0.4478,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -3.171875,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -12.8125,
+      "logps/chosen": -796.0,
+      "logps/rejected": -1248.0,
+      "loss": 0.4506,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -4.3125,
       "rewards/margins": 2.21875,
-      "rewards/rejected": -5.40625,
+      "rewards/rejected": -6.53125,
       "step": 350
     },
     {
       "epoch": 0.3769633507853403,
-      "grad_norm": 3.0358447044225856,
+      "grad_norm": 2.8513490162823105,
       "learning_rate": 4.99011439559198e-05,
-      "logits/chosen": -12.6875,
-      "logits/rejected": -12.0,
-      "logps/chosen": -756.0,
-      "logps/rejected": -1064.0,
-      "loss": 0.4195,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -3.53125,
-      "rewards/margins": 1.65625,
-      "rewards/rejected": -5.1875,
+      "logits/chosen": -12.5,
+      "logits/rejected": -11.875,
+      "logps/chosen": -792.0,
+      "logps/rejected": -1072.0,
+      "loss": 0.4222,
+      "rewards/accuracies": 0.831250011920929,
+      "rewards/chosen": -3.90625,
+      "rewards/margins": 1.4296875,
+      "rewards/rejected": -5.34375,
       "step": 360
     },
     {
       "epoch": 0.387434554973822,
-      "grad_norm": 3.4866944987496185,
+      "grad_norm": 3.302024051268013,
       "learning_rate": 4.987222972275369e-05,
-      "logits/chosen": -8.9375,
-      "logits/rejected": -7.90625,
-      "logps/chosen": -916.0,
-      "logps/rejected": -1296.0,
-      "loss": 0.3793,
-      "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": -5.25,
-      "rewards/margins": 2.171875,
-      "rewards/rejected": -7.4375,
+      "logits/chosen": -9.8125,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -956.0,
+      "logps/rejected": -1384.0,
+      "loss": 0.3879,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -5.65625,
+      "rewards/margins": 2.671875,
+      "rewards/rejected": -8.3125,
       "step": 370
     },
     {
       "epoch": 0.39790575916230364,
-      "grad_norm": 2.190025890508139,
+      "grad_norm": 2.1576846738942814,
       "learning_rate": 4.983962194568872e-05,
-      "logits/chosen": -9.3125,
-      "logits/rejected": -7.9375,
-      "logps/chosen": -860.0,
-      "logps/rejected": -1448.0,
-      "loss": 0.4388,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -4.5,
-      "rewards/margins": 2.671875,
-      "rewards/rejected": -7.1875,
+      "logits/chosen": -9.25,
+      "logits/rejected": -8.125,
+      "logps/chosen": -824.0,
+      "logps/rejected": -1408.0,
+      "loss": 0.45,
+      "rewards/accuracies": 0.8187500238418579,
+      "rewards/chosen": -4.125,
+      "rewards/margins": 2.640625,
+      "rewards/rejected": -6.78125,
       "step": 380
     },
     {
       "epoch": 0.4083769633507853,
-      "grad_norm": 2.8930455017260517,
+      "grad_norm": 2.7628662083416833,
       "learning_rate": 4.9803325467003084e-05,
-      "logits/chosen": -10.5,
-      "logits/rejected": -9.0,
-      "logps/chosen": -872.0,
-      "logps/rejected": -1368.0,
-      "loss": 0.4152,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -4.5625,
-      "rewards/margins": 2.421875,
-      "rewards/rejected": -6.96875,
+      "logits/chosen": -11.125,
+      "logits/rejected": -9.875,
+      "logps/chosen": -756.0,
+      "logps/rejected": -1224.0,
+      "loss": 0.4251,
+      "rewards/accuracies": 0.793749988079071,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 2.125,
+      "rewards/rejected": -5.5625,
       "step": 390
     },
     {
       "epoch": 0.418848167539267,
-      "grad_norm": 2.187787713170407,
+      "grad_norm": 2.108113678780066,
       "learning_rate": 4.9763345676749816e-05,
-      "logits/chosen": -12.875,
-      "logits/rejected": -12.0,
-      "logps/chosen": -720.0,
-      "logps/rejected": -1032.0,
-      "loss": 0.4431,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -3.296875,
-      "rewards/margins": 1.6640625,
-      "rewards/rejected": -4.96875,
+      "logits/chosen": -10.875,
+      "logits/rejected": -10.25,
+      "logps/chosen": -772.0,
+      "logps/rejected": -1072.0,
+      "loss": 0.4624,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -3.8125,
+      "rewards/margins": 1.5546875,
+      "rewards/rejected": -5.375,
       "step": 400
     },
     {
       "epoch": 0.418848167539267,
-      "eval_logits/chosen": -13.25,
-      "eval_logits/rejected": -12.5,
-      "eval_logps/chosen": -600.0,
-      "eval_logps/rejected": -616.0,
-      "eval_loss": 0.7253437638282776,
-      "eval_rewards/accuracies": 0.625,
-      "eval_rewards/chosen": -2.8125,
-      "eval_rewards/margins": 0.478515625,
-      "eval_rewards/rejected": -3.28125,
-      "eval_runtime": 46.6362,
-      "eval_samples_per_second": 42.885,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -12.3125,
+      "eval_logits/rejected": -11.3125,
+      "eval_logps/chosen": -676.0,
+      "eval_logps/rejected": -692.0,
+      "eval_loss": 0.7233437299728394,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -3.5625,
+      "eval_rewards/margins": 0.494140625,
+      "eval_rewards/rejected": -4.0625,
+      "eval_runtime": 47.2245,
+      "eval_samples_per_second": 42.351,
+      "eval_steps_per_second": 0.678,
       "step": 400
     },
     {
       "epoch": 0.4293193717277487,
-      "grad_norm": 3.143613213782077,
+      "grad_norm": 2.565615926912782,
       "learning_rate": 4.9719688511956374e-05,
-      "logits/chosen": -11.125,
-      "logits/rejected": -10.0,
-      "logps/chosen": -844.0,
-      "logps/rejected": -1256.0,
-      "loss": 0.4186,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -4.21875,
-      "rewards/margins": 2.171875,
-      "rewards/rejected": -6.375,
+      "logits/chosen": -11.75,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -864.0,
+      "logps/rejected": -1280.0,
+      "loss": 0.4033,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -4.40625,
+      "rewards/margins": 2.234375,
+      "rewards/rejected": -6.65625,
       "step": 410
     },
     {
       "epoch": 0.4397905759162304,
-      "grad_norm": 2.6074737431180868,
+      "grad_norm": 2.5736428925524395,
       "learning_rate": 4.967236045574294e-05,
-      "logits/chosen": -9.3125,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -892.0,
-      "logps/rejected": -1352.0,
-      "loss": 0.4162,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -4.875,
-      "rewards/margins": 2.578125,
-      "rewards/rejected": -7.46875,
+      "logits/chosen": -9.9375,
+      "logits/rejected": -8.5625,
+      "logps/chosen": -912.0,
+      "logps/rejected": -1384.0,
+      "loss": 0.4054,
+      "rewards/accuracies": 0.831250011920929,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 2.640625,
+      "rewards/rejected": -7.75,
       "step": 420
     },
     {
       "epoch": 0.450261780104712,
-      "grad_norm": 2.5927675012838436,
+      "grad_norm": 2.859187192915181,
       "learning_rate": 4.96213685363597e-05,
-      "logits/chosen": -11.125,
-      "logits/rejected": -10.25,
-      "logps/chosen": -664.0,
-      "logps/rejected": -1040.0,
-      "loss": 0.4354,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -2.65625,
-      "rewards/margins": 1.8125,
-      "rewards/rejected": -4.46875,
+      "logits/chosen": -9.75,
+      "logits/rejected": -9.0,
+      "logps/chosen": -780.0,
+      "logps/rejected": -1168.0,
+      "loss": 0.4254,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": -3.84375,
+      "rewards/margins": 1.8828125,
+      "rewards/rejected": -5.75,
       "step": 430
     },
     {
       "epoch": 0.4607329842931937,
-      "grad_norm": 4.075753485824123,
+      "grad_norm": 3.254685782656208,
       "learning_rate": 4.956672032614315e-05,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -10.4375,
-      "logps/chosen": -720.0,
-      "logps/rejected": -1112.0,
-      "loss": 0.4449,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -3.03125,
-      "rewards/margins": 1.6484375,
-      "rewards/rejected": -4.6875,
+      "logits/chosen": -10.3125,
+      "logits/rejected": -9.625,
+      "logps/chosen": -812.0,
+      "logps/rejected": -1232.0,
+      "loss": 0.4422,
+      "rewards/accuracies": 0.8187500238418579,
+      "rewards/chosen": -3.984375,
+      "rewards/margins": 1.8984375,
+      "rewards/rejected": -5.875,
       "step": 440
     },
     {
       "epoch": 0.4712041884816754,
-      "grad_norm": 2.4855696283828825,
+      "grad_norm": 2.575491895393729,
       "learning_rate": 4.950842394039157e-05,
-      "logits/chosen": -11.0625,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -796.0,
-      "logps/rejected": -1248.0,
-      "loss": 0.3663,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -4.09375,
-      "rewards/margins": 2.484375,
-      "rewards/rejected": -6.5625,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -788.0,
+      "logps/rejected": -1256.0,
+      "loss": 0.3665,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": -4.0,
+      "rewards/margins": 2.703125,
+      "rewards/rejected": -6.71875,
       "step": 450
     },
     {
       "epoch": 0.4816753926701571,
-      "grad_norm": 3.5901906362277423,
+      "grad_norm": 3.297298961632602,
       "learning_rate": 4.944648803615994e-05,
-      "logits/chosen": -12.0,
-      "logits/rejected": -10.25,
-      "logps/chosen": -792.0,
-      "logps/rejected": -1376.0,
-      "loss": 0.3751,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -3.84375,
-      "rewards/margins": 3.125,
-      "rewards/rejected": -6.96875,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -900.0,
+      "logps/rejected": -1504.0,
+      "loss": 0.3571,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.90625,
+      "rewards/margins": 3.390625,
+      "rewards/rejected": -8.3125,
       "step": 460
     },
     {
       "epoch": 0.49214659685863876,
-      "grad_norm": 4.229699088239283,
+      "grad_norm": 3.2052642555898516,
       "learning_rate": 4.938092181097433e-05,
-      "logits/chosen": -10.6875,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -968.0,
-      "logps/rejected": -1560.0,
-      "loss": 0.367,
-      "rewards/accuracies": 0.8500000238418579,
-      "rewards/chosen": -5.625,
-      "rewards/margins": 4.09375,
-      "rewards/rejected": -9.75,
+      "logits/chosen": -7.5625,
+      "logits/rejected": -6.84375,
+      "logps/chosen": -1008.0,
+      "logps/rejected": -1576.0,
+      "loss": 0.3558,
+      "rewards/accuracies": 0.862500011920929,
+      "rewards/chosen": -6.0625,
+      "rewards/margins": 3.78125,
+      "rewards/rejected": -9.875,
       "step": 470
     },
     {
       "epoch": 0.5026178010471204,
-      "grad_norm": 4.235572521712871,
+      "grad_norm": 5.923067620926881,
       "learning_rate": 4.931173500146607e-05,
-      "logits/chosen": -11.8125,
-      "logits/rejected": -10.75,
-      "logps/chosen": -876.0,
-      "logps/rejected": -1456.0,
-      "loss": 0.3744,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -5.0,
-      "rewards/margins": 3.265625,
-      "rewards/rejected": -8.25,
+      "logits/chosen": -8.6875,
+      "logits/rejected": -7.78125,
+      "logps/chosen": -940.0,
+      "logps/rejected": -1576.0,
+      "loss": 0.3837,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -5.625,
+      "rewards/margins": 3.875,
+      "rewards/rejected": -9.5,
       "step": 480
     },
     {
       "epoch": 0.5130890052356021,
-      "grad_norm": 2.062653503774281,
+      "grad_norm": 2.1013028895144013,
       "learning_rate": 4.923893788192582e-05,
-      "logits/chosen": -12.5625,
-      "logits/rejected": -11.375,
-      "logps/chosen": -756.0,
-      "logps/rejected": -1176.0,
-      "loss": 0.3873,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 2.28125,
-      "rewards/rejected": -5.84375,
+      "logits/chosen": -8.5625,
+      "logits/rejected": -7.625,
+      "logps/chosen": -920.0,
+      "logps/rejected": -1456.0,
+      "loss": 0.4372,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -5.1875,
+      "rewards/margins": 3.5,
+      "rewards/rejected": -8.6875,
       "step": 490
     },
     {
       "epoch": 0.5235602094240838,
-      "grad_norm": 3.230603689468994,
+      "grad_norm": 2.698247192834292,
       "learning_rate": 4.9162541262777906e-05,
-      "logits/chosen": -8.875,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -928.0,
-      "logps/rejected": -1368.0,
-      "loss": 0.3779,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -5.09375,
-      "rewards/margins": 2.828125,
-      "rewards/rejected": -7.90625,
+      "logits/chosen": -10.125,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -820.0,
+      "logps/rejected": -1216.0,
+      "loss": 0.373,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -4.0,
+      "rewards/margins": 2.34375,
+      "rewards/rejected": -6.34375,
       "step": 500
     },
     {
       "epoch": 0.5235602094240838,
-      "eval_logits/chosen": -6.1875,
-      "eval_logits/rejected": -5.03125,
-      "eval_logps/chosen": -932.0,
-      "eval_logps/rejected": -984.0,
-      "eval_loss": 0.9410781264305115,
-      "eval_rewards/accuracies": 0.611328125,
-      "eval_rewards/chosen": -6.125,
-      "eval_rewards/margins": 0.81640625,
-      "eval_rewards/rejected": -6.9375,
-      "eval_runtime": 46.6393,
-      "eval_samples_per_second": 42.882,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -8.6875,
+      "eval_logits/rejected": -7.3125,
+      "eval_logps/chosen": -768.0,
+      "eval_logps/rejected": -808.0,
+      "eval_loss": 0.8017343878746033,
+      "eval_rewards/accuracies": 0.626953125,
+      "eval_rewards/chosen": -4.5,
+      "eval_rewards/margins": 0.67578125,
+      "eval_rewards/rejected": -5.1875,
+      "eval_runtime": 47.0783,
+      "eval_samples_per_second": 42.482,
+      "eval_steps_per_second": 0.68,
       "step": 500
     },
     {
       "epoch": 0.5340314136125655,
-      "grad_norm": 2.5514470548349597,
+      "grad_norm": 3.41827119232801,
       "learning_rate": 4.90825564889749e-05,
-      "logits/chosen": -9.9375,
-      "logits/rejected": -9.125,
-      "logps/chosen": -936.0,
-      "logps/rejected": -1272.0,
-      "loss": 0.3914,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -4.96875,
-      "rewards/margins": 2.390625,
-      "rewards/rejected": -7.34375,
+      "logits/chosen": -9.125,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -980.0,
+      "logps/rejected": -1288.0,
+      "loss": 0.4038,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -5.40625,
+      "rewards/margins": 2.0625,
+      "rewards/rejected": -7.4375,
       "step": 510
     },
     {
       "epoch": 0.5445026178010471,
-      "grad_norm": 2.188835884226418,
+      "grad_norm": 2.440591801688025,
       "learning_rate": 4.899899543831288e-05,
-      "logits/chosen": -13.375,
-      "logits/rejected": -12.5,
-      "logps/chosen": -892.0,
-      "logps/rejected": -1232.0,
-      "loss": 0.4224,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -4.5625,
-      "rewards/margins": 2.3125,
-      "rewards/rejected": -6.875,
+      "logits/chosen": -10.9375,
+      "logits/rejected": -9.75,
+      "logps/chosen": -952.0,
+      "logps/rejected": -1336.0,
+      "loss": 0.3826,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": -5.15625,
+      "rewards/margins": 2.78125,
+      "rewards/rejected": -7.96875,
       "step": 520
     },
     {
       "epoch": 0.5549738219895288,
-      "grad_norm": 2.7520949733051654,
+      "grad_norm": 3.4988940244923685,
       "learning_rate": 4.891187051966765e-05,
-      "logits/chosen": -13.125,
-      "logits/rejected": -12.6875,
-      "logps/chosen": -964.0,
-      "logps/rejected": -1272.0,
-      "loss": 0.4319,
+      "logits/chosen": -11.75,
+      "logits/rejected": -11.0625,
+      "logps/chosen": -960.0,
+      "logps/rejected": -1312.0,
+      "loss": 0.417,
       "rewards/accuracies": 0.800000011920929,
       "rewards/chosen": -5.34375,
-      "rewards/margins": 1.828125,
-      "rewards/rejected": -7.15625,
+      "rewards/margins": 2.25,
+      "rewards/rejected": -7.59375,
       "step": 530
     },
     {
       "epoch": 0.5654450261780105,
-      "grad_norm": 2.4954303956563937,
+      "grad_norm": 2.7529177257956894,
       "learning_rate": 4.882119467115191e-05,
-      "logits/chosen": -14.5625,
-      "logits/rejected": -13.6875,
-      "logps/chosen": -844.0,
-      "logps/rejected": -1240.0,
-      "loss": 0.3697,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -4.4375,
-      "rewards/margins": 2.234375,
-      "rewards/rejected": -6.6875,
+      "logits/chosen": -12.1875,
+      "logits/rejected": -11.5,
+      "logps/chosen": -940.0,
+      "logps/rejected": -1344.0,
+      "loss": 0.3729,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -5.375,
+      "rewards/margins": 2.3125,
+      "rewards/rejected": -7.6875,
       "step": 540
     },
     {
       "epoch": 0.5759162303664922,
-      "grad_norm": 2.6039466335594126,
+      "grad_norm": 2.9444958471125102,
       "learning_rate": 4.872698135819402e-05,
-      "logits/chosen": -13.5625,
-      "logits/rejected": -11.9375,
-      "logps/chosen": -888.0,
-      "logps/rejected": -1368.0,
-      "loss": 0.3893,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -4.9375,
-      "rewards/margins": 2.453125,
-      "rewards/rejected": -7.375,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -12.125,
+      "logps/chosen": -848.0,
+      "logps/rejected": -1304.0,
+      "loss": 0.384,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": -4.53125,
+      "rewards/margins": 2.171875,
+      "rewards/rejected": -6.71875,
       "step": 550
     },
     {
       "epoch": 0.5863874345549738,
-      "grad_norm": 3.065716722713571,
+      "grad_norm": 2.164423145904123,
       "learning_rate": 4.862924457153831e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -9.625,
-      "logps/chosen": -832.0,
-      "logps/rejected": -1344.0,
-      "loss": 0.3948,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -11.5,
+      "logps/chosen": -772.0,
+      "logps/rejected": -1264.0,
+      "loss": 0.3553,
       "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": -4.90625,
-      "rewards/margins": 2.859375,
-      "rewards/rejected": -7.78125,
+      "rewards/chosen": -4.34375,
+      "rewards/margins": 2.640625,
+      "rewards/rejected": -6.96875,
       "step": 560
     },
     {
       "epoch": 0.5968586387434555,
-      "grad_norm": 2.2845184054681145,
+      "grad_norm": 2.5427492500745785,
       "learning_rate": 4.8527998825167484e-05,
-      "logits/chosen": -10.375,
-      "logits/rejected": -9.0,
-      "logps/chosen": -912.0,
-      "logps/rejected": -1416.0,
-      "loss": 0.35,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -5.375,
-      "rewards/margins": 2.859375,
-      "rewards/rejected": -8.25,
+      "logits/chosen": -9.8125,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -960.0,
+      "logps/rejected": -1448.0,
+      "loss": 0.3727,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -5.8125,
+      "rewards/margins": 2.78125,
+      "rewards/rejected": -8.625,
       "step": 570
     },
     {
       "epoch": 0.6073298429319371,
-      "grad_norm": 4.388822740607571,
+      "grad_norm": 2.8173384255067138,
       "learning_rate": 4.842325915414728e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -8.5625,
-      "logps/chosen": -856.0,
-      "logps/rejected": -1232.0,
-      "loss": 0.3959,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -4.9375,
-      "rewards/margins": 2.375,
-      "rewards/rejected": -7.3125,
+      "logits/chosen": -9.0,
+      "logits/rejected": -7.90625,
+      "logps/chosen": -788.0,
+      "logps/rejected": -1128.0,
+      "loss": 0.3916,
+      "rewards/accuracies": 0.8187500238418579,
+      "rewards/chosen": -4.28125,
+      "rewards/margins": 1.9609375,
+      "rewards/rejected": -6.25,
       "step": 580
     },
     {
       "epoch": 0.6178010471204188,
-      "grad_norm": 3.596658754192997,
+      "grad_norm": 4.092750582522732,
       "learning_rate": 4.831504111239373e-05,
-      "logits/chosen": -11.875,
-      "logits/rejected": -10.4375,
-      "logps/chosen": -956.0,
-      "logps/rejected": -1504.0,
-      "loss": 0.3597,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -5.375,
-      "rewards/margins": 2.796875,
-      "rewards/rejected": -8.1875,
+      "logits/chosen": -9.625,
+      "logits/rejected": -8.5625,
+      "logps/chosen": -976.0,
+      "logps/rejected": -1560.0,
+      "loss": 0.3583,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -5.59375,
+      "rewards/margins": 3.140625,
+      "rewards/rejected": -8.75,
       "step": 590
     },
     {
       "epoch": 0.6282722513089005,
-      "grad_norm": 3.65379678588694,
+      "grad_norm": 2.2019812207382214,
       "learning_rate": 4.8203360770363425e-05,
-      "logits/chosen": -12.625,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -1624.0,
-      "loss": 0.404,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -6.5,
-      "rewards/margins": 3.90625,
-      "rewards/rejected": -10.375,
+      "logits/chosen": -10.25,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -984.0,
+      "logps/rejected": -1456.0,
+      "loss": 0.418,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -5.78125,
+      "rewards/margins": 2.953125,
+      "rewards/rejected": -8.75,
       "step": 600
     },
     {
       "epoch": 0.6282722513089005,
-      "eval_logits/chosen": -12.875,
-      "eval_logits/rejected": -11.6875,
-      "eval_logps/chosen": -840.0,
-      "eval_logps/rejected": -892.0,
-      "eval_loss": 0.8501093983650208,
-      "eval_rewards/accuracies": 0.615234375,
-      "eval_rewards/chosen": -5.21875,
-      "eval_rewards/margins": 0.8359375,
-      "eval_rewards/rejected": -6.0625,
-      "eval_runtime": 46.6125,
-      "eval_samples_per_second": 42.907,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -11.0,
+      "eval_logits/rejected": -9.8125,
+      "eval_logps/chosen": -768.0,
+      "eval_logps/rejected": -816.0,
+      "eval_loss": 0.8380625247955322,
+      "eval_rewards/accuracies": 0.61328125,
+      "eval_rewards/chosen": -4.46875,
+      "eval_rewards/margins": 0.78125,
+      "eval_rewards/rejected": -5.28125,
+      "eval_runtime": 47.079,
+      "eval_samples_per_second": 42.482,
+      "eval_steps_per_second": 0.68,
       "step": 600
     },
     {
       "epoch": 0.6387434554973822,
-      "grad_norm": 3.0781320586208816,
+      "grad_norm": 3.716517958121969,
       "learning_rate": 4.808823471266702e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -12.0625,
-      "logps/chosen": -804.0,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -10.25,
+      "logps/chosen": -816.0,
       "logps/rejected": -1296.0,
-      "loss": 0.3622,
-      "rewards/accuracies": 0.831250011920929,
-      "rewards/chosen": -4.125,
-      "rewards/margins": 2.78125,
-      "rewards/rejected": -6.875,
+      "loss": 0.3654,
+      "rewards/accuracies": 0.875,
+      "rewards/chosen": -4.21875,
+      "rewards/margins": 2.703125,
+      "rewards/rejected": -6.90625,
       "step": 610
     },
     {
       "epoch": 0.6492146596858639,
-      "grad_norm": 3.5301284845661636,
+      "grad_norm": 3.753413187168972,
       "learning_rate": 4.796968003560639e-05,
-      "logits/chosen": -13.0625,
-      "logits/rejected": -11.5,
-      "logps/chosen": -784.0,
-      "logps/rejected": -1240.0,
-      "loss": 0.4145,
+      "logits/chosen": -12.5625,
+      "logits/rejected": -11.0,
+      "logps/chosen": -892.0,
+      "logps/rejected": -1352.0,
+      "loss": 0.452,
       "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -3.78125,
-      "rewards/margins": 2.40625,
-      "rewards/rejected": -6.1875,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 2.53125,
+      "rewards/rejected": -7.375,
       "step": 620
     },
     {
       "epoch": 0.6596858638743456,
-      "grad_norm": 2.047298790167175,
+      "grad_norm": 2.062359612493287,
       "learning_rate": 4.7847714344635864e-05,
-      "logits/chosen": -13.875,
-      "logits/rejected": -12.5,
-      "logps/chosen": -680.0,
-      "logps/rejected": -1032.0,
-      "loss": 0.406,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -2.96875,
-      "rewards/margins": 2.09375,
-      "rewards/rejected": -5.0625,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -13.75,
+      "logps/chosen": -800.0,
+      "logps/rejected": -1160.0,
+      "loss": 0.4173,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -4.15625,
+      "rewards/margins": 2.1875,
+      "rewards/rejected": -6.34375,
       "step": 630
     },
     {
       "epoch": 0.6701570680628273,
-      "grad_norm": 4.0714597435744775,
+      "grad_norm": 3.124870807293616,
       "learning_rate": 4.772235575174776e-05,
-      "logits/chosen": -11.0,
-      "logits/rejected": -9.625,
-      "logps/chosen": -912.0,
-      "logps/rejected": -1328.0,
-      "loss": 0.3977,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -4.78125,
-      "rewards/margins": 2.71875,
-      "rewards/rejected": -7.5,
+      "logits/chosen": -13.625,
+      "logits/rejected": -13.125,
+      "logps/chosen": -972.0,
+      "logps/rejected": -1336.0,
+      "loss": 0.3992,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -5.40625,
+      "rewards/margins": 2.140625,
+      "rewards/rejected": -7.53125,
       "step": 640
     },
     {
       "epoch": 0.680628272251309,
-      "grad_norm": 2.8763490652298307,
+      "grad_norm": 2.499316818622552,
       "learning_rate": 4.7593622872782754e-05,
-      "logits/chosen": -12.1875,
-      "logits/rejected": -11.4375,
-      "logps/chosen": -768.0,
-      "logps/rejected": -1128.0,
-      "loss": 0.4176,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -3.828125,
-      "rewards/margins": 2.0625,
-      "rewards/rejected": -5.90625,
+      "logits/chosen": -12.75,
+      "logits/rejected": -11.875,
+      "logps/chosen": -924.0,
+      "logps/rejected": -1320.0,
+      "loss": 0.4004,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -5.375,
+      "rewards/margins": 2.4375,
+      "rewards/rejected": -7.8125,
       "step": 650
     },
     {
       "epoch": 0.6910994764397905,
-      "grad_norm": 2.8331813022174526,
+      "grad_norm": 3.1708092646053587,
       "learning_rate": 4.746153482466542e-05,
-      "logits/chosen": -14.6875,
-      "logits/rejected": -14.0,
-      "logps/chosen": -608.0,
-      "logps/rejected": -936.0,
-      "loss": 0.4262,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -2.25,
-      "rewards/margins": 1.796875,
-      "rewards/rejected": -4.03125,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -796.0,
+      "logps/rejected": -1160.0,
+      "loss": 0.4005,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": -4.125,
+      "rewards/margins": 2.15625,
+      "rewards/rejected": -6.28125,
       "step": 660
     },
     {
       "epoch": 0.7015706806282722,
-      "grad_norm": 6.667894055354915,
+      "grad_norm": 2.739188543095194,
       "learning_rate": 4.732611122256537e-05,
-      "logits/chosen": -13.625,
-      "logits/rejected": -12.625,
-      "logps/chosen": -840.0,
-      "logps/rejected": -1264.0,
-      "loss": 0.3821,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -4.0625,
-      "rewards/margins": 3.0625,
-      "rewards/rejected": -7.125,
+      "logits/chosen": -8.0625,
+      "logits/rejected": -7.1875,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -1456.0,
+      "loss": 0.3637,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": -6.5,
+      "rewards/margins": 2.546875,
+      "rewards/rejected": -9.0,
       "step": 670
     },
     {
       "epoch": 0.7120418848167539,
-      "grad_norm": 2.290520094156181,
+      "grad_norm": 2.0506422048897694,
       "learning_rate": 4.718737217698435e-05,
-      "logits/chosen": -11.9375,
-      "logits/rejected": -10.8125,
-      "logps/chosen": -784.0,
-      "logps/rejected": -1272.0,
-      "loss": 0.422,
-      "rewards/accuracies": 0.8374999761581421,
-      "rewards/chosen": -4.03125,
-      "rewards/margins": 2.734375,
-      "rewards/rejected": -6.75,
+      "logits/chosen": -8.9375,
+      "logits/rejected": -8.125,
+      "logps/chosen": -868.0,
+      "logps/rejected": -1296.0,
+      "loss": 0.4323,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 2.125,
+      "rewards/rejected": -7.0,
       "step": 680
     },
     {
       "epoch": 0.7225130890052356,
-      "grad_norm": 2.3939624094729393,
+      "grad_norm": 2.2563511617718204,
       "learning_rate": 4.704533829076982e-05,
-      "logits/chosen": -12.375,
-      "logits/rejected": -11.4375,
-      "logps/chosen": -760.0,
-      "logps/rejected": -1184.0,
-      "loss": 0.3652,
-      "rewards/accuracies": 0.8500000238418579,
-      "rewards/chosen": -3.65625,
-      "rewards/margins": 2.140625,
-      "rewards/rejected": -5.78125,
+      "logits/chosen": -11.875,
+      "logits/rejected": -10.875,
+      "logps/chosen": -788.0,
+      "logps/rejected": -1216.0,
+      "loss": 0.3888,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": -3.90625,
+      "rewards/margins": 2.25,
+      "rewards/rejected": -6.15625,
       "step": 690
     },
     {
       "epoch": 0.7329842931937173,
-      "grad_norm": 2.671835002723019,
+      "grad_norm": 2.796314845940636,
       "learning_rate": 4.6900030656055455e-05,
-      "logits/chosen": -11.4375,
-      "logits/rejected": -10.25,
-      "logps/chosen": -788.0,
-      "logps/rejected": -1136.0,
-      "loss": 0.3834,
-      "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -3.796875,
-      "rewards/margins": 2.0625,
-      "rewards/rejected": -5.875,
-      "step": 700
-    },
-    {
+      "logits/chosen": -12.0625,
+      "logits/rejected": -11.0,
+      "logps/chosen": -800.0,
+      "logps/rejected": -1160.0,
+      "loss": 0.4,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -3.90625,
+      "rewards/margins": 2.203125,
+      "rewards/rejected": -6.125,
+      "step": 700
+    },
+    {
       "epoch": 0.7329842931937173,
-      "eval_logits/chosen": -8.625,
-      "eval_logits/rejected": -7.15625,
-      "eval_logps/chosen": -744.0,
-      "eval_logps/rejected": -808.0,
-      "eval_loss": 0.8127187490463257,
-      "eval_rewards/accuracies": 0.626953125,
-      "eval_rewards/chosen": -4.25,
-      "eval_rewards/margins": 0.9609375,
-      "eval_rewards/rejected": -5.21875,
-      "eval_runtime": 46.6384,
-      "eval_samples_per_second": 42.883,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -11.6875,
+      "eval_logits/rejected": -10.75,
+      "eval_logps/chosen": -752.0,
+      "eval_logps/rejected": -800.0,
+      "eval_loss": 0.7899531126022339,
+      "eval_rewards/accuracies": 0.646484375,
+      "eval_rewards/chosen": -4.3125,
+      "eval_rewards/margins": 0.80078125,
+      "eval_rewards/rejected": -5.125,
+      "eval_runtime": 47.0119,
+      "eval_samples_per_second": 42.542,
+      "eval_steps_per_second": 0.681,
       "step": 700
     },
     {
       "epoch": 0.743455497382199,
-      "grad_norm": 3.0168388208465706,
+      "grad_norm": 2.5923435233959697,
       "learning_rate": 4.6751470851128906e-05,
-      "logits/chosen": -10.25,
-      "logits/rejected": -9.0,
-      "logps/chosen": -796.0,
-      "logps/rejected": -1256.0,
-      "loss": 0.399,
+      "logits/chosen": -11.875,
+      "logits/rejected": -10.75,
+      "logps/chosen": -832.0,
+      "logps/rejected": -1272.0,
+      "loss": 0.3732,
       "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -3.609375,
-      "rewards/margins": 2.140625,
-      "rewards/rejected": -5.75,
+      "rewards/chosen": -3.96875,
+      "rewards/margins": 1.9375,
+      "rewards/rejected": -5.90625,
       "step": 710
     },
     {
       "epoch": 0.7539267015706806,
-      "grad_norm": 2.9052841560528395,
+      "grad_norm": 4.808763076398256,
       "learning_rate": 4.6599680937227454e-05,
-      "logits/chosen": -11.0625,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -760.0,
-      "logps/rejected": -1232.0,
-      "loss": 0.3828,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -3.359375,
-      "rewards/margins": 2.15625,
-      "rewards/rejected": -5.53125,
+      "logits/chosen": -11.375,
+      "logits/rejected": -9.875,
+      "logps/chosen": -932.0,
+      "logps/rejected": -1432.0,
+      "loss": 0.3776,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 2.390625,
+      "rewards/rejected": -7.5,
       "step": 720
     },
     {
       "epoch": 0.7643979057591623,
-      "grad_norm": 2.5370724301277607,
+      "grad_norm": 3.3056631440336632,
       "learning_rate": 4.644468345526183e-05,
-      "logits/chosen": -12.5625,
-      "logits/rejected": -11.4375,
-      "logps/chosen": -768.0,
-      "logps/rejected": -1264.0,
-      "loss": 0.3865,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -3.546875,
-      "rewards/margins": 2.546875,
-      "rewards/rejected": -6.09375,
+      "logits/chosen": -11.6875,
+      "logits/rejected": -10.625,
+      "logps/chosen": -884.0,
+      "logps/rejected": -1392.0,
+      "loss": 0.3878,
+      "rewards/accuracies": 0.793749988079071,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 2.640625,
+      "rewards/rejected": -7.34375,
       "step": 730
     },
     {
       "epoch": 0.774869109947644,
-      "grad_norm": 3.307569122761129,
+      "grad_norm": 2.1338574011001996,
       "learning_rate": 4.628650142246894e-05,
-      "logits/chosen": -10.5,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -772.0,
-      "logps/rejected": -1328.0,
-      "loss": 0.35,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -3.96875,
-      "rewards/margins": 2.65625,
-      "rewards/rejected": -6.625,
+      "logits/chosen": -7.3125,
+      "logits/rejected": -6.4375,
+      "logps/chosen": -1032.0,
+      "logps/rejected": -1600.0,
+      "loss": 0.3479,
+      "rewards/accuracies": 0.8687499761581421,
+      "rewards/chosen": -6.53125,
+      "rewards/margins": 2.734375,
+      "rewards/rejected": -9.25,
       "step": 740
     },
     {
       "epoch": 0.7853403141361257,
-      "grad_norm": 2.0413598770513546,
+      "grad_norm": 2.3416942555594327,
       "learning_rate": 4.6125158328993765e-05,
-      "logits/chosen": -7.46875,
-      "logits/rejected": -6.28125,
-      "logps/chosen": -832.0,
-      "logps/rejected": -1288.0,
-      "loss": 0.3542,
-      "rewards/accuracies": 0.831250011920929,
-      "rewards/chosen": -4.4375,
-      "rewards/margins": 2.421875,
-      "rewards/rejected": -6.875,
+      "logits/chosen": -7.96875,
+      "logits/rejected": -7.0,
+      "logps/chosen": -948.0,
+      "logps/rejected": -1400.0,
+      "loss": 0.378,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -5.625,
+      "rewards/margins": 2.375,
+      "rewards/rejected": -8.0,
       "step": 750
     },
     {
       "epoch": 0.7958115183246073,
-      "grad_norm": 2.661796929199356,
+      "grad_norm": 2.786491188516209,
       "learning_rate": 4.5960678134401064e-05,
-      "logits/chosen": -7.375,
-      "logits/rejected": -6.59375,
-      "logps/chosen": -828.0,
-      "logps/rejected": -1256.0,
-      "loss": 0.3697,
-      "rewards/accuracies": 0.8500000238418579,
-      "rewards/chosen": -4.25,
-      "rewards/margins": 2.34375,
-      "rewards/rejected": -6.59375,
+      "logits/chosen": -11.125,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -844.0,
+      "logps/rejected": -1264.0,
+      "loss": 0.3708,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -4.40625,
+      "rewards/margins": 2.296875,
+      "rewards/rejected": -6.71875,
       "step": 760
     },
     {
       "epoch": 0.806282722513089,
-      "grad_norm": 2.972096778316046,
+      "grad_norm": 3.7399992316619675,
       "learning_rate": 4.579308526411734e-05,
-      "logits/chosen": -6.3125,
-      "logits/rejected": -5.71875,
-      "logps/chosen": -884.0,
-      "logps/rejected": -1552.0,
-      "loss": 0.3803,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -4.9375,
-      "rewards/margins": 3.40625,
-      "rewards/rejected": -8.3125,
+      "logits/chosen": -11.0,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -932.0,
+      "logps/rejected": -1624.0,
+      "loss": 0.3749,
+      "rewards/accuracies": 0.8500000238418579,
+      "rewards/chosen": -5.40625,
+      "rewards/margins": 3.625,
+      "rewards/rejected": -9.0625,
       "step": 770
     },
     {
       "epoch": 0.8167539267015707,
-      "grad_norm": 3.273343611745989,
+      "grad_norm": 3.419904484617841,
       "learning_rate": 4.562240460580367e-05,
-      "logits/chosen": -8.25,
-      "logits/rejected": -7.46875,
-      "logps/chosen": -840.0,
-      "logps/rejected": -1368.0,
-      "loss": 0.3527,
+      "logits/chosen": -12.125,
+      "logits/rejected": -11.1875,
+      "logps/chosen": -896.0,
+      "logps/rejected": -1408.0,
+      "loss": 0.3598,
       "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -4.5,
-      "rewards/margins": 2.953125,
-      "rewards/rejected": -7.4375,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 2.859375,
+      "rewards/rejected": -7.90625,
       "step": 780
     },
     {
       "epoch": 0.8272251308900523,
-      "grad_norm": 2.3306328811053656,
+      "grad_norm": 2.2317066770296665,
       "learning_rate": 4.544866150565985e-05,
-      "logits/chosen": -7.46875,
-      "logits/rejected": -6.34375,
+      "logits/chosen": -12.125,
+      "logits/rejected": -10.75,
       "logps/chosen": -868.0,
-      "logps/rejected": -1352.0,
-      "loss": 0.3751,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -4.75,
+      "logps/rejected": -1344.0,
+      "loss": 0.3745,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -4.71875,
       "rewards/margins": 2.53125,
       "rewards/rejected": -7.25,
       "step": 790
     },
     {
       "epoch": 0.837696335078534,
-      "grad_norm": 2.6779620111040563,
+      "grad_norm": 2.0984539072740658,
       "learning_rate": 4.527188176466051e-05,
-      "logits/chosen": -6.65625,
-      "logits/rejected": -6.0,
-      "logps/chosen": -928.0,
-      "logps/rejected": -1240.0,
-      "loss": 0.3763,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -5.0,
-      "rewards/margins": 1.9375,
-      "rewards/rejected": -6.9375,
+      "logits/chosen": -11.125,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -904.0,
+      "logps/rejected": -1208.0,
+      "loss": 0.384,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -4.78125,
+      "rewards/margins": 1.8359375,
+      "rewards/rejected": -6.625,
       "step": 800
     },
     {
       "epoch": 0.837696335078534,
-      "eval_logits/chosen": -6.84375,
-      "eval_logits/rejected": -5.15625,
-      "eval_logps/chosen": -772.0,
-      "eval_logps/rejected": -828.0,
-      "eval_loss": 0.8067343831062317,
-      "eval_rewards/accuracies": 0.666015625,
-      "eval_rewards/chosen": -4.5625,
-      "eval_rewards/margins": 0.8515625,
-      "eval_rewards/rejected": -5.40625,
-      "eval_runtime": 46.6335,
-      "eval_samples_per_second": 42.888,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -8.6875,
+      "eval_logits/rejected": -7.21875,
+      "eval_logps/chosen": -844.0,
+      "eval_logps/rejected": -892.0,
+      "eval_loss": 0.8198124766349792,
+      "eval_rewards/accuracies": 0.63671875,
+      "eval_rewards/chosen": -5.25,
+      "eval_rewards/margins": 0.77734375,
+      "eval_rewards/rejected": -6.03125,
+      "eval_runtime": 47.0705,
+      "eval_samples_per_second": 42.489,
+      "eval_steps_per_second": 0.68,
       "step": 800
     },
     {
       "epoch": 0.8481675392670157,
-      "grad_norm": 2.2463263154930906,
+      "grad_norm": 3.617005189722071,
       "learning_rate": 4.509209163472357e-05,
-      "logits/chosen": -11.375,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -828.0,
-      "logps/rejected": -1264.0,
-      "loss": 0.3441,
+      "logits/chosen": -9.0625,
+      "logits/rejected": -7.9375,
+      "logps/chosen": -968.0,
+      "logps/rejected": -1400.0,
+      "loss": 0.3517,
       "rewards/accuracies": 0.8125,
-      "rewards/chosen": -4.15625,
-      "rewards/margins": 2.484375,
-      "rewards/rejected": -6.625,
+      "rewards/chosen": -5.5625,
+      "rewards/margins": 2.5,
+      "rewards/rejected": -8.0625,
       "step": 810
     },
     {
       "epoch": 0.8586387434554974,
-      "grad_norm": 2.3911855853649446,
+      "grad_norm": 2.4091883002327696,
       "learning_rate": 4.4909317814811896e-05,
-      "logits/chosen": -14.0,
-      "logits/rejected": -12.9375,
-      "logps/chosen": -776.0,
-      "logps/rejected": -1272.0,
-      "loss": 0.3974,
+      "logits/chosen": -10.0625,
+      "logits/rejected": -9.0,
+      "logps/chosen": -896.0,
+      "logps/rejected": -1384.0,
+      "loss": 0.421,
       "rewards/accuracies": 0.8125,
-      "rewards/chosen": -3.796875,
-      "rewards/margins": 2.640625,
-      "rewards/rejected": -6.4375,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 2.59375,
+      "rewards/rejected": -7.5625,
       "step": 820
     },
     {
       "epoch": 0.8691099476439791,
-      "grad_norm": 2.153078414771686,
+      "grad_norm": 1.750962791430712,
       "learning_rate": 4.472358744696845e-05,
-      "logits/chosen": -13.8125,
-      "logits/rejected": -12.625,
-      "logps/chosen": -816.0,
-      "logps/rejected": -1256.0,
-      "loss": 0.3508,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -4.15625,
-      "rewards/margins": 2.734375,
-      "rewards/rejected": -6.875,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -772.0,
+      "logps/rejected": -1168.0,
+      "loss": 0.3731,
+      "rewards/accuracies": 0.831250011920929,
+      "rewards/chosen": -3.734375,
+      "rewards/margins": 2.265625,
+      "rewards/rejected": -6.0,
       "step": 830
     },
     {
       "epoch": 0.8795811518324608,
-      "grad_norm": 4.529830971974864,
+      "grad_norm": 4.3978815491579715,
       "learning_rate": 4.4534928112285645e-05,
-      "logits/chosen": -11.375,
-      "logits/rejected": -10.25,
-      "logps/chosen": -1048.0,
+      "logits/chosen": -7.6875,
+      "logits/rejected": -6.625,
+      "logps/chosen": -1040.0,
       "logps/rejected": -1560.0,
-      "loss": 0.3871,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -6.375,
-      "rewards/margins": 2.96875,
+      "loss": 0.3799,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -6.34375,
+      "rewards/margins": 3.015625,
       "rewards/rejected": -9.375,
       "step": 840
     },
     {
       "epoch": 0.8900523560209425,
-      "grad_norm": 2.2531524238058545,
+      "grad_norm": 2.4350082449341772,
       "learning_rate": 4.43433678268096e-05,
-      "logits/chosen": -11.125,
-      "logits/rejected": -10.0,
-      "logps/chosen": -900.0,
-      "logps/rejected": -1376.0,
-      "loss": 0.351,
+      "logits/chosen": -7.0625,
+      "logits/rejected": -6.1875,
+      "logps/chosen": -1020.0,
+      "logps/rejected": -1496.0,
+      "loss": 0.3835,
       "rewards/accuracies": 0.8500000238418579,
-      "rewards/chosen": -5.03125,
-      "rewards/margins": 2.578125,
-      "rewards/rejected": -7.59375,
+      "rewards/chosen": -6.21875,
+      "rewards/margins": 2.59375,
+      "rewards/rejected": -8.8125,
       "step": 850
     },
     {
       "epoch": 0.900523560209424,
-      "grad_norm": 2.4761644745415774,
+      "grad_norm": 3.243850899843546,
       "learning_rate": 4.414893503737968e-05,
-      "logits/chosen": -8.875,
-      "logits/rejected": -7.375,
-      "logps/chosen": -808.0,
-      "logps/rejected": -1296.0,
-      "loss": 0.3436,
-      "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -4.375,
-      "rewards/margins": 2.546875,
-      "rewards/rejected": -6.9375,
+      "logits/chosen": -8.0625,
+      "logits/rejected": -6.53125,
+      "logps/chosen": -852.0,
+      "logps/rejected": -1328.0,
+      "loss": 0.325,
+      "rewards/accuracies": 0.8374999761581421,
+      "rewards/chosen": -4.8125,
+      "rewards/margins": 2.453125,
+      "rewards/rejected": -7.25,
       "step": 860
     },
     {
       "epoch": 0.9109947643979057,
-      "grad_norm": 3.484894698116822,
+      "grad_norm": 3.5841949842500322,
       "learning_rate": 4.395165861740414e-05,
-      "logits/chosen": -6.96875,
-      "logits/rejected": -6.375,
-      "logps/chosen": -956.0,
-      "logps/rejected": -1456.0,
-      "loss": 0.3573,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -5.8125,
-      "rewards/margins": 3.0625,
-      "rewards/rejected": -8.875,
+      "logits/chosen": -7.34375,
+      "logits/rejected": -6.625,
+      "logps/chosen": -1024.0,
+      "logps/rejected": -1560.0,
+      "loss": 0.3574,
+      "rewards/accuracies": 0.8187500238418579,
+      "rewards/chosen": -6.5,
+      "rewards/margins": 3.359375,
+      "rewards/rejected": -9.875,
       "step": 870
     },
     {
       "epoch": 0.9214659685863874,
-      "grad_norm": 2.350506820489632,
+      "grad_norm": 2.1992246953147747,
       "learning_rate": 4.375156786257241e-05,
-      "logits/chosen": -8.375,
-      "logits/rejected": -7.125,
-      "logps/chosen": -976.0,
-      "logps/rejected": -1400.0,
-      "loss": 0.4046,
+      "logits/chosen": -9.875,
+      "logits/rejected": -8.625,
+      "logps/chosen": -1008.0,
+      "logps/rejected": -1432.0,
+      "loss": 0.4129,
       "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -5.40625,
+      "rewards/chosen": -5.71875,
       "rewards/margins": 2.640625,
-      "rewards/rejected": -8.0625,
+      "rewards/rejected": -8.375,
       "step": 880
     },
     {
       "epoch": 0.9319371727748691,
-      "grad_norm": 4.005168156248366,
+      "grad_norm": 3.2427874400780143,
       "learning_rate": 4.354869248650466e-05,
-      "logits/chosen": -8.125,
-      "logits/rejected": -6.75,
-      "logps/chosen": -804.0,
-      "logps/rejected": -1208.0,
-      "loss": 0.3735,
-      "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -3.984375,
-      "rewards/margins": 1.9921875,
-      "rewards/rejected": -5.96875,
+      "logits/chosen": -11.75,
+      "logits/rejected": -10.25,
+      "logps/chosen": -776.0,
+      "logps/rejected": -1168.0,
+      "loss": 0.3742,
+      "rewards/accuracies": 0.831250011920929,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 1.828125,
+      "rewards/rejected": -5.53125,
       "step": 890
     },
     {
       "epoch": 0.9424083769633508,
-      "grad_norm": 4.049235158821143,
+      "grad_norm": 4.560527139654481,
       "learning_rate": 4.334306261633931e-05,
-      "logits/chosen": -6.9375,
-      "logits/rejected": -5.5,
-      "logps/chosen": -884.0,
-      "logps/rejected": -1392.0,
-      "loss": 0.3462,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -4.53125,
-      "rewards/margins": 2.625,
-      "rewards/rejected": -7.15625,
+      "logits/chosen": -9.5625,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -980.0,
+      "logps/rejected": -1536.0,
+      "loss": 0.3594,
+      "rewards/accuracies": 0.793749988079071,
+      "rewards/chosen": -5.5,
+      "rewards/margins": 3.078125,
+      "rewards/rejected": -8.5625,
       "step": 900
     },
     {
       "epoch": 0.9424083769633508,
-      "eval_logits/chosen": -5.4375,
-      "eval_logits/rejected": -3.703125,
-      "eval_logps/chosen": -824.0,
-      "eval_logps/rejected": -892.0,
-      "eval_loss": 0.8711718916893005,
-      "eval_rewards/accuracies": 0.6171875,
-      "eval_rewards/chosen": -5.0625,
-      "eval_rewards/margins": 0.98046875,
-      "eval_rewards/rejected": -6.03125,
-      "eval_runtime": 46.646,
-      "eval_samples_per_second": 42.876,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -7.09375,
+      "eval_logits/rejected": -5.71875,
+      "eval_logps/chosen": -1040.0,
+      "eval_logps/rejected": -1128.0,
+      "eval_loss": 1.0285311937332153,
+      "eval_rewards/accuracies": 0.62890625,
+      "eval_rewards/chosen": -7.25,
+      "eval_rewards/margins": 1.171875,
+      "eval_rewards/rejected": -8.4375,
+      "eval_runtime": 47.0901,
+      "eval_samples_per_second": 42.472,
+      "eval_steps_per_second": 0.68,
       "step": 900
     },
     {
       "epoch": 0.9528795811518325,
-      "grad_norm": 3.2827110650137534,
+      "grad_norm": 2.7133624345526663,
       "learning_rate": 4.313470878825911e-05,
-      "logits/chosen": -7.25,
-      "logits/rejected": -5.8125,
-      "logps/chosen": -920.0,
-      "logps/rejected": -1424.0,
-      "loss": 0.3549,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -5.125,
-      "rewards/margins": 3.25,
-      "rewards/rejected": -8.375,
+      "logits/chosen": -8.75,
+      "logits/rejected": -7.625,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1560.0,
+      "loss": 0.3425,
+      "rewards/accuracies": 0.8687499761581421,
+      "rewards/chosen": -6.375,
+      "rewards/margins": 3.34375,
+      "rewards/rejected": -9.75,
       "step": 910
     },
     {
       "epoch": 0.9633507853403142,
-      "grad_norm": 2.189013520523636,
+      "grad_norm": 2.413433929191455,
       "learning_rate": 4.292366194295652e-05,
-      "logits/chosen": -6.03125,
-      "logits/rejected": -5.40625,
-      "logps/chosen": -880.0,
-      "logps/rejected": -1392.0,
-      "loss": 0.3678,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -4.71875,
-      "rewards/margins": 2.828125,
-      "rewards/rejected": -7.5625,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -944.0,
+      "logps/rejected": -1432.0,
+      "loss": 0.3655,
+      "rewards/accuracies": 0.8187500238418579,
+      "rewards/chosen": -5.40625,
+      "rewards/margins": 2.609375,
+      "rewards/rejected": -8.0,
       "step": 920
     },
     {
       "epoch": 0.9738219895287958,
-      "grad_norm": 2.192842320844609,
+      "grad_norm": 2.408759101192053,
       "learning_rate": 4.2709953421038964e-05,
-      "logits/chosen": -6.15625,
-      "logits/rejected": -5.4375,
-      "logps/chosen": -792.0,
-      "logps/rejected": -1208.0,
-      "loss": 0.3696,
-      "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -4.0625,
-      "rewards/margins": 2.09375,
-      "rewards/rejected": -6.15625,
+      "logits/chosen": -7.6875,
+      "logits/rejected": -6.96875,
+      "logps/chosen": -896.0,
+      "logps/rejected": -1296.0,
+      "loss": 0.3685,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -5.125,
+      "rewards/margins": 1.953125,
+      "rewards/rejected": -7.0625,
       "step": 930
     },
     {
       "epoch": 0.9842931937172775,
-      "grad_norm": 2.5344607332610694,
+      "grad_norm": 3.3791897749286695,
       "learning_rate": 4.249361495837478e-05,
-      "logits/chosen": -8.0625,
-      "logits/rejected": -6.8125,
-      "logps/chosen": -756.0,
-      "logps/rejected": -1160.0,
-      "loss": 0.3828,
+      "logits/chosen": -7.96875,
+      "logits/rejected": -6.5625,
+      "logps/chosen": -868.0,
+      "logps/rejected": -1272.0,
+      "loss": 0.4109,
       "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -3.59375,
+      "rewards/chosen": -4.71875,
       "rewards/margins": 2.078125,
-      "rewards/rejected": -5.6875,
+      "rewards/rejected": -6.8125,
       "step": 940
     },
     {
       "epoch": 0.9947643979057592,
-      "grad_norm": 2.7410134304859026,
+      "grad_norm": 2.4290280048530506,
       "learning_rate": 4.227467868138035e-05,
-      "logits/chosen": -7.8125,
-      "logits/rejected": -6.46875,
-      "logps/chosen": -792.0,
-      "logps/rejected": -1304.0,
-      "loss": 0.3605,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -3.90625,
-      "rewards/margins": 2.640625,
-      "rewards/rejected": -6.53125,
+      "logits/chosen": -7.53125,
+      "logits/rejected": -6.40625,
+      "logps/chosen": -900.0,
+      "logps/rejected": -1432.0,
+      "loss": 0.335,
+      "rewards/accuracies": 0.824999988079071,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 2.8125,
+      "rewards/rejected": -7.78125,
       "step": 950
     },
     {
       "epoch": 1.0052356020942408,
-      "grad_norm": 1.3512640250878485,
+      "grad_norm": 1.688081881995257,
       "learning_rate": 4.205317710224938e-05,
-      "logits/chosen": -7.25,
-      "logits/rejected": -5.96875,
-      "logps/chosen": -808.0,
-      "logps/rejected": -1320.0,
-      "loss": 0.2604,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -3.90625,
-      "rewards/margins": 3.234375,
-      "rewards/rejected": -7.15625,
+      "logits/chosen": -8.5,
+      "logits/rejected": -7.0,
+      "logps/chosen": -912.0,
+      "logps/rejected": -1440.0,
+      "loss": 0.2601,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 3.421875,
+      "rewards/rejected": -8.375,
       "step": 960
     },
     {
       "epoch": 1.0157068062827226,
-      "grad_norm": 0.7991859257916447,
+      "grad_norm": 1.0155288053060518,
       "learning_rate": 4.1829143114124733e-05,
-      "logits/chosen": -5.875,
-      "logits/rejected": -4.4375,
-      "logps/chosen": -1016.0,
-      "logps/rejected": -1616.0,
-      "loss": 0.1285,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -6.03125,
-      "rewards/margins": 4.9375,
-      "rewards/rejected": -10.9375,
+      "logits/chosen": -8.1875,
+      "logits/rejected": -6.5,
+      "logps/chosen": -984.0,
+      "logps/rejected": -1536.0,
+      "loss": 0.1329,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.6875,
+      "rewards/margins": 4.4375,
+      "rewards/rejected": -10.125,
       "step": 970
     },
     {
       "epoch": 1.0261780104712042,
-      "grad_norm": 1.9931558519454577,
+      "grad_norm": 2.518665731566414,
       "learning_rate": 4.160260998621387e-05,
-      "logits/chosen": -8.6875,
-      "logits/rejected": -6.625,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -1752.0,
-      "loss": 0.1106,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.03125,
-      "rewards/margins": 5.6875,
-      "rewards/rejected": -11.6875,
+      "logits/chosen": -7.0625,
+      "logits/rejected": -5.4375,
+      "logps/chosen": -1280.0,
+      "logps/rejected": -2000.0,
+      "loss": 0.1042,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -8.375,
+      "rewards/margins": 5.84375,
+      "rewards/rejected": -14.1875,
       "step": 980
     },
     {
       "epoch": 1.036649214659686,
-      "grad_norm": 2.2795326538678733,
+      "grad_norm": 1.3435112945729693,
       "learning_rate": 4.137361135884824e-05,
-      "logits/chosen": -8.5,
-      "logits/rejected": -7.0,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -1912.0,
-      "loss": 0.0914,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -6.1875,
-      "rewards/margins": 6.5,
-      "rewards/rejected": -12.6875,
-      "step": 990
+      "logits/chosen": -6.96875,
+      "logits/rejected": -5.78125,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -2128.0,
+      "loss": 0.0906,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 6.6875,
+      "rewards/rejected": -14.875,
+      "step": 990
     },
     {
       "epoch": 1.0471204188481675,
-      "grad_norm": 0.8440521735588647,
+      "grad_norm": 0.8826496148665905,
       "learning_rate": 4.114218123848777e-05,
-      "logits/chosen": -6.5625,
-      "logits/rejected": -5.0625,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -2096.0,
-      "loss": 0.0835,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.875,
-      "rewards/margins": 7.25,
-      "rewards/rejected": -15.125,
+      "logits/chosen": -5.625,
+      "logits/rejected": -4.46875,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -2160.0,
+      "loss": 0.0833,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.875,
+      "rewards/margins": 6.9375,
+      "rewards/rejected": -15.75,
       "step": 1000
     },
     {
       "epoch": 1.0471204188481675,
-      "eval_logits/chosen": -5.125,
-      "eval_logits/rejected": -3.21875,
-      "eval_logps/chosen": -1112.0,
-      "eval_logps/rejected": -1216.0,
-      "eval_loss": 1.1522812843322754,
-      "eval_rewards/accuracies": 0.607421875,
-      "eval_rewards/chosen": -7.9375,
-      "eval_rewards/margins": 1.3828125,
-      "eval_rewards/rejected": -9.3125,
-      "eval_runtime": 46.6312,
-      "eval_samples_per_second": 42.89,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -4.09375,
+      "eval_logits/rejected": -2.578125,
+      "eval_logps/chosen": -1240.0,
+      "eval_logps/rejected": -1328.0,
+      "eval_loss": 1.1501250267028809,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -9.25,
+      "eval_rewards/margins": 1.15625,
+      "eval_rewards/rejected": -10.375,
+      "eval_runtime": 47.0762,
+      "eval_samples_per_second": 42.484,
+      "eval_steps_per_second": 0.68,
       "step": 1000
     },
     {
       "epoch": 1.057591623036649,
-      "grad_norm": 1.8284025930315264,
+      "grad_norm": 2.4475079849098083,
       "learning_rate": 4.0908353992670824e-05,
-      "logits/chosen": -7.34375,
-      "logits/rejected": -5.21875,
-      "logps/chosen": -916.0,
-      "logps/rejected": -1584.0,
-      "loss": 0.1184,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -5.25,
-      "rewards/margins": 5.46875,
-      "rewards/rejected": -10.6875,
+      "logits/chosen": -6.84375,
+      "logits/rejected": -5.53125,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -1736.0,
+      "loss": 0.1142,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 5.40625,
+      "rewards/rejected": -12.25,
       "step": 1010
     },
     {
       "epoch": 1.068062827225131,
-      "grad_norm": 2.5792515405223613,
+      "grad_norm": 2.4653747614145973,
       "learning_rate": 4.067216434491059e-05,
-      "logits/chosen": -5.5,
-      "logits/rejected": -4.28125,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -1800.0,
-      "loss": 0.1037,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.40625,
-      "rewards/margins": 5.84375,
-      "rewards/rejected": -12.25,
+      "logits/chosen": -6.75,
+      "logits/rejected": -5.84375,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -1832.0,
+      "loss": 0.1033,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.875,
+      "rewards/margins": 5.65625,
+      "rewards/rejected": -12.5625,
       "step": 1020
     },
     {
       "epoch": 1.0785340314136125,
-      "grad_norm": 1.5895575549253214,
+      "grad_norm": 2.1310452666397706,
       "learning_rate": 4.043364736953866e-05,
-      "logits/chosen": -5.53125,
-      "logits/rejected": -3.8125,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1864.0,
+      "logits/chosen": -6.0,
+      "logits/rejected": -4.53125,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -1944.0,
       "loss": 0.1132,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -6.53125,
-      "rewards/margins": 5.6875,
-      "rewards/rejected": -12.1875,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -7.375,
+      "rewards/margins": 5.65625,
+      "rewards/rejected": -13.0625,
       "step": 1030
     },
     {
       "epoch": 1.0890052356020943,
-      "grad_norm": 1.7488098259352443,
+      "grad_norm": 3.028771262320732,
       "learning_rate": 4.019283848649643e-05,
-      "logits/chosen": -7.0,
-      "logits/rejected": -5.09375,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.1177,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.34375,
-      "rewards/margins": 5.34375,
-      "rewards/rejected": -11.6875,
+      "logits/chosen": -5.65625,
+      "logits/rejected": -4.21875,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -1976.0,
+      "loss": 0.139,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.0,
+      "rewards/margins": 5.84375,
+      "rewards/rejected": -13.8125,
       "step": 1040
     },
     {
       "epoch": 1.0994764397905759,
-      "grad_norm": 1.2073556946963084,
+      "grad_norm": 1.127373422789099,
       "learning_rate": 3.9949773456075215e-05,
-      "logits/chosen": -7.78125,
-      "logits/rejected": -6.0,
-      "logps/chosen": -920.0,
-      "logps/rejected": -1576.0,
-      "loss": 0.0906,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -5.46875,
-      "rewards/margins": 4.5625,
-      "rewards/rejected": -10.0625,
+      "logits/chosen": -5.34375,
+      "logits/rejected": -3.96875,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1704.0,
+      "loss": 0.089,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -6.65625,
+      "rewards/margins": 4.59375,
+      "rewards/rejected": -11.25,
       "step": 1050
     },
     {
       "epoch": 1.1099476439790577,
-      "grad_norm": 1.1159050917303748,
+      "grad_norm": 2.952551522192677,
       "learning_rate": 3.970448837360585e-05,
-      "logits/chosen": -7.3125,
-      "logits/rejected": -5.5625,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -2040.0,
-      "loss": 0.0826,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.375,
-      "rewards/margins": 6.25,
-      "rewards/rejected": -13.625,
+      "logits/chosen": -6.09375,
+      "logits/rejected": -4.71875,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -2176.0,
+      "loss": 0.09,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -9.125,
+      "rewards/margins": 5.875,
+      "rewards/rejected": -15.0,
       "step": 1060
     },
     {
       "epoch": 1.1204188481675392,
-      "grad_norm": 0.7401095721933173,
+      "grad_norm": 1.0412579401484154,
       "learning_rate": 3.945701966409846e-05,
-      "logits/chosen": -5.34375,
-      "logits/rejected": -4.40625,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -2240.0,
-      "loss": 0.1085,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 7.625,
-      "rewards/rejected": -15.9375,
+      "logits/chosen": -6.0625,
+      "logits/rejected": -4.625,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -2400.0,
+      "loss": 0.0927,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -9.3125,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -17.5,
       "step": 1070
     },
     {
       "epoch": 1.130890052356021,
-      "grad_norm": 1.1595358397942783,
+      "grad_norm": 1.2484808104674319,
       "learning_rate": 3.9207404076833374e-05,
-      "logits/chosen": -6.15625,
-      "logits/rejected": -4.375,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -2048.0,
-      "loss": 0.0956,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.1875,
-      "rewards/margins": 6.6875,
-      "rewards/rejected": -13.875,
+      "logits/chosen": -7.6875,
+      "logits/rejected": -5.3125,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -2096.0,
+      "loss": 0.1113,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.125,
+      "rewards/margins": 7.25,
+      "rewards/rejected": -14.375,
       "step": 1080
     },
     {
       "epoch": 1.1413612565445026,
-      "grad_norm": 1.2305624655634966,
+      "grad_norm": 1.708984494773098,
       "learning_rate": 3.895567867990379e-05,
-      "logits/chosen": -8.0,
-      "logits/rejected": -5.90625,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -2032.0,
-      "loss": 0.0968,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.0,
-      "rewards/margins": 6.5625,
-      "rewards/rejected": -13.5625,
+      "logits/chosen": -7.9375,
+      "logits/rejected": -5.5625,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -2008.0,
+      "loss": 0.1217,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.65625,
+      "rewards/margins": 6.625,
+      "rewards/rejected": -13.3125,
       "step": 1090
     },
     {
       "epoch": 1.1518324607329844,
-      "grad_norm": 0.9982187213235826,
+      "grad_norm": 1.206510038333789,
       "learning_rate": 3.870188085471114e-05,
-      "logits/chosen": -9.1875,
-      "logits/rejected": -7.0,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -1864.0,
-      "loss": 0.0895,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -6.71875,
-      "rewards/margins": 6.21875,
-      "rewards/rejected": -12.9375,
+      "logits/chosen": -6.0625,
+      "logits/rejected": -4.90625,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -1912.0,
+      "loss": 0.0953,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -7.71875,
+      "rewards/margins": 5.71875,
+      "rewards/rejected": -13.4375,
       "step": 1100
     },
     {
       "epoch": 1.1518324607329844,
-      "eval_logits/chosen": -6.375,
-      "eval_logits/rejected": -4.375,
-      "eval_logps/chosen": -1184.0,
-      "eval_logps/rejected": -1288.0,
-      "eval_loss": 1.358968734741211,
-      "eval_rewards/accuracies": 0.58984375,
-      "eval_rewards/chosen": -8.6875,
-      "eval_rewards/margins": 1.328125,
-      "eval_rewards/rejected": -10.0,
-      "eval_runtime": 46.7701,
-      "eval_samples_per_second": 42.762,
-      "eval_steps_per_second": 0.684,
+      "eval_logits/chosen": -3.59375,
+      "eval_logits/rejected": -2.0,
+      "eval_logps/chosen": -1264.0,
+      "eval_logps/rejected": -1360.0,
+      "eval_loss": 1.1999530792236328,
+      "eval_rewards/accuracies": 0.59375,
+      "eval_rewards/chosen": -9.5,
+      "eval_rewards/margins": 1.2421875,
+      "eval_rewards/rejected": -10.75,
+      "eval_runtime": 47.2586,
+      "eval_samples_per_second": 42.32,
+      "eval_steps_per_second": 0.677,
       "step": 1100
     },
     {
       "epoch": 1.162303664921466,
-      "grad_norm": 1.6241493920690766,
+      "grad_norm": 1.1820958698463258,
       "learning_rate": 3.844604829041395e-05,
-      "logits/chosen": -7.09375,
-      "logits/rejected": -5.0,
+      "logits/chosen": -6.53125,
+      "logits/rejected": -5.03125,
       "logps/chosen": -1056.0,
-      "logps/rejected": -1808.0,
-      "loss": 0.0987,
-      "rewards/accuracies": 0.956250011920929,
+      "logps/rejected": -1768.0,
+      "loss": 0.094,
+      "rewards/accuracies": 0.9624999761581421,
       "rewards/chosen": -6.78125,
-      "rewards/margins": 6.25,
-      "rewards/rejected": -13.0,
+      "rewards/margins": 5.8125,
+      "rewards/rejected": -12.5625,
       "step": 1110
     },
     {
       "epoch": 1.1727748691099475,
-      "grad_norm": 1.7555529505425345,
+      "grad_norm": 1.9618932155541804,
       "learning_rate": 3.818821897833095e-05,
-      "logits/chosen": -4.96875,
-      "logits/rejected": -3.84375,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -1936.0,
-      "loss": 0.1058,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.6875,
-      "rewards/margins": 5.78125,
-      "rewards/rejected": -13.5,
+      "logits/chosen": -6.875,
+      "logits/rejected": -5.46875,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1736.0,
+      "loss": 0.1004,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -6.25,
+      "rewards/margins": 5.25,
+      "rewards/rejected": -11.5,
       "step": 1120
     },
     {
       "epoch": 1.1832460732984293,
-      "grad_norm": 3.041635103170913,
+      "grad_norm": 2.00712723452005,
       "learning_rate": 3.792843120629935e-05,
-      "logits/chosen": -6.21875,
-      "logits/rejected": -4.0,
-      "logps/chosen": -976.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.1042,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -5.84375,
-      "rewards/margins": 5.78125,
-      "rewards/rejected": -11.625,
+      "logits/chosen": -5.90625,
+      "logits/rejected": -4.09375,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -1912.0,
+      "loss": 0.1232,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.03125,
+      "rewards/margins": 5.90625,
+      "rewards/rejected": -12.9375,
       "step": 1130
     },
     {
       "epoch": 1.193717277486911,
-      "grad_norm": 3.218246030672682,
+      "grad_norm": 3.233388775541502,
       "learning_rate": 3.766672355298907e-05,
-      "logits/chosen": -6.8125,
-      "logits/rejected": -4.375,
-      "logps/chosen": -1016.0,
-      "logps/rejected": -1792.0,
-      "loss": 0.0927,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.53125,
-      "rewards/margins": 5.90625,
-      "rewards/rejected": -12.4375,
+      "logits/chosen": -7.3125,
+      "logits/rejected": -5.28125,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1880.0,
+      "loss": 0.1014,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -7.4375,
+      "rewards/margins": 5.84375,
+      "rewards/rejected": -13.25,
       "step": 1140
     },
     {
       "epoch": 1.2041884816753927,
-      "grad_norm": 4.237250020662393,
+      "grad_norm": 4.471216632991261,
       "learning_rate": 3.740313488217373e-05,
-      "logits/chosen": -4.28125,
-      "logits/rejected": -2.578125,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -2176.0,
-      "loss": 0.1099,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -9.125,
-      "rewards/margins": 7.09375,
-      "rewards/rejected": -16.25,
+      "logits/chosen": -7.15625,
+      "logits/rejected": -4.6875,
+      "logps/chosen": -1280.0,
+      "logps/rejected": -2144.0,
+      "loss": 0.11,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -8.9375,
+      "rewards/margins": 7.125,
+      "rewards/rejected": -16.125,
       "step": 1150
     },
     {
       "epoch": 1.2146596858638743,
-      "grad_norm": 2.6838732445212767,
+      "grad_norm": 2.0905299462725346,
       "learning_rate": 3.7137704336959464e-05,
-      "logits/chosen": -2.8125,
-      "logits/rejected": -1.1328125,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.1063,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -6.53125,
-      "rewards/margins": 5.8125,
-      "rewards/rejected": -12.375,
+      "logits/chosen": -4.3125,
+      "logits/rejected": -2.28125,
+      "logps/chosen": -1280.0,
+      "logps/rejected": -2112.0,
+      "loss": 0.1225,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -9.0,
+      "rewards/margins": 6.5625,
+      "rewards/rejected": -15.5,
       "step": 1160
     },
     {
       "epoch": 1.225130890052356,
-      "grad_norm": 2.409271622159364,
+      "grad_norm": 3.168573900259561,
       "learning_rate": 3.687047133397201e-05,
-      "logits/chosen": -4.34375,
-      "logits/rejected": -2.484375,
-      "logps/chosen": -860.0,
-      "logps/rejected": -1672.0,
-      "loss": 0.0987,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -4.75,
-      "rewards/margins": 5.71875,
-      "rewards/rejected": -10.5,
+      "logits/chosen": -5.8125,
+      "logits/rejected": -3.9375,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -1888.0,
+      "loss": 0.1203,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.09375,
+      "rewards/margins": 5.5,
+      "rewards/rejected": -12.5625,
       "step": 1170
     },
     {
       "epoch": 1.2356020942408377,
-      "grad_norm": 1.872892782982811,
+      "grad_norm": 2.331535911720474,
       "learning_rate": 3.6601475557503415e-05,
-      "logits/chosen": -3.5625,
-      "logits/rejected": -1.5078125,
-      "logps/chosen": -1020.0,
-      "logps/rejected": -1696.0,
-      "loss": 0.1018,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -5.96875,
-      "rewards/margins": 5.59375,
-      "rewards/rejected": -11.5625,
+      "logits/chosen": -6.34375,
+      "logits/rejected": -3.734375,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -1864.0,
+      "loss": 0.0932,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.4375,
+      "rewards/margins": 5.8125,
+      "rewards/rejected": -13.25,
       "step": 1180
     },
     {
       "epoch": 1.2460732984293195,
-      "grad_norm": 1.019694109643568,
+      "grad_norm": 2.124068068600138,
       "learning_rate": 3.633075695361882e-05,
-      "logits/chosen": -2.53125,
-      "logits/rejected": -1.4140625,
-      "logps/chosen": -1024.0,
-      "logps/rejected": -1824.0,
-      "loss": 0.0921,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.5,
-      "rewards/margins": 5.78125,
-      "rewards/rejected": -12.25,
+      "logits/chosen": -4.75,
+      "logits/rejected": -3.109375,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -2080.0,
+      "loss": 0.1008,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 6.375,
+      "rewards/rejected": -14.9375,
       "step": 1190
     },
     {
       "epoch": 1.256544502617801,
-      "grad_norm": 3.0964696817150785,
+      "grad_norm": 2.5186377852766966,
       "learning_rate": 3.6058355724224477e-05,
-      "logits/chosen": -3.0625,
-      "logits/rejected": -1.9453125,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -1800.0,
-      "loss": 0.1146,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.4375,
-      "rewards/margins": 5.90625,
-      "rewards/rejected": -12.375,
+      "logits/chosen": -7.0,
+      "logits/rejected": -5.1875,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -1880.0,
+      "loss": 0.1198,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.34375,
+      "rewards/margins": 5.8125,
+      "rewards/rejected": -13.1875,
       "step": 1200
     },
     {
       "epoch": 1.256544502617801,
-      "eval_logits/chosen": -1.0859375,
-      "eval_logits/rejected": 0.70703125,
-      "eval_logps/chosen": -1144.0,
-      "eval_logps/rejected": -1248.0,
-      "eval_loss": 1.284359335899353,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -8.3125,
-      "eval_rewards/margins": 1.2734375,
-      "eval_rewards/rejected": -9.5625,
-      "eval_runtime": 46.632,
-      "eval_samples_per_second": 42.889,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -4.96875,
+      "eval_logits/rejected": -2.9375,
+      "eval_logps/chosen": -1176.0,
+      "eval_logps/rejected": -1264.0,
+      "eval_loss": 1.2178750038146973,
+      "eval_rewards/accuracies": 0.607421875,
+      "eval_rewards/chosen": -8.5625,
+      "eval_rewards/margins": 1.1640625,
+      "eval_rewards/rejected": -9.75,
+      "eval_runtime": 47.1015,
+      "eval_samples_per_second": 42.461,
+      "eval_steps_per_second": 0.679,
       "step": 1200
     },
     {
       "epoch": 1.2670157068062826,
-      "grad_norm": 2.9701757578172727,
+      "grad_norm": 3.4582272711228526,
       "learning_rate": 3.5784312321097744e-05,
-      "logits/chosen": -3.75,
-      "logits/rejected": -2.65625,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -1856.0,
-      "loss": 0.1009,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -6.53125,
-      "rewards/margins": 6.25,
-      "rewards/rejected": -12.8125,
+      "logits/chosen": -6.3125,
+      "logits/rejected": -4.6875,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -1928.0,
+      "loss": 0.1032,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.5625,
+      "rewards/margins": 5.96875,
+      "rewards/rejected": -13.5,
       "step": 1210
     },
     {
       "epoch": 1.2774869109947644,
-      "grad_norm": 2.8993468844463703,
+      "grad_norm": 2.805737307455421,
       "learning_rate": 3.550866743987994e-05,
-      "logits/chosen": -4.5,
-      "logits/rejected": -3.078125,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -1824.0,
-      "loss": 0.1129,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -6.40625,
-      "rewards/margins": 6.03125,
-      "rewards/rejected": -12.4375,
-      "step": 1220
-    },
-    {
-      "epoch": 1.2879581151832462,
-      "grad_norm": 1.145874569139116,
+      "logits/chosen": -7.15625,
+      "logits/rejected": -5.1875,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1832.0,
+      "loss": 0.1246,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 5.6875,
+      "rewards/rejected": -12.5,
+      "step": 1220
+    },
+    {
+      "epoch": 1.2879581151832462,
+      "grad_norm": 2.284438570390368,
       "learning_rate": 3.5231462014033024e-05,
-      "logits/chosen": -5.3125,
-      "logits/rejected": -4.3125,
-      "logps/chosen": -976.0,
-      "logps/rejected": -1864.0,
-      "loss": 0.063,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -5.9375,
-      "rewards/margins": 6.59375,
-      "rewards/rejected": -12.5625,
+      "logits/chosen": -7.96875,
+      "logits/rejected": -6.21875,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1856.0,
+      "loss": 0.0718,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.59375,
+      "rewards/margins": 5.90625,
+      "rewards/rejected": -12.5,
       "step": 1230
     },
     {
       "epoch": 1.2984293193717278,
-      "grad_norm": 1.1193836514507645,
+      "grad_norm": 2.058840468124629,
       "learning_rate": 3.495273720876095e-05,
-      "logits/chosen": -4.5,
-      "logits/rejected": -3.0625,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -1976.0,
-      "loss": 0.0838,
+      "logits/chosen": -5.125,
+      "logits/rejected": -3.421875,
+      "logps/chosen": -1312.0,
+      "logps/rejected": -2176.0,
+      "loss": 0.0799,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.21875,
-      "rewards/margins": 7.03125,
-      "rewards/rejected": -14.25,
+      "rewards/chosen": -9.0,
+      "rewards/margins": 7.21875,
+      "rewards/rejected": -16.25,
       "step": 1240
     },
     {
       "epoch": 1.3089005235602094,
-      "grad_norm": 2.324691765620723,
+      "grad_norm": 0.9713263018281973,
       "learning_rate": 3.4672534414896616e-05,
-      "logits/chosen": -4.65625,
-      "logits/rejected": -3.015625,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -2176.0,
-      "loss": 0.1004,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.75,
-      "rewards/margins": 7.71875,
-      "rewards/rejected": -15.5,
+      "logits/chosen": -5.46875,
+      "logits/rejected": -3.484375,
+      "logps/chosen": -1296.0,
+      "logps/rejected": -2208.0,
+      "loss": 0.0753,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -8.625,
+      "rewards/margins": 7.25,
+      "rewards/rejected": -15.875,
       "step": 1250
     },
     {
       "epoch": 1.3193717277486912,
-      "grad_norm": 2.3590600544525384,
+      "grad_norm": 2.7359457286619695,
       "learning_rate": 3.439089524275526e-05,
-      "logits/chosen": -5.03125,
-      "logits/rejected": -3.109375,
-      "logps/chosen": -964.0,
-      "logps/rejected": -1720.0,
-      "loss": 0.1227,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -5.8125,
-      "rewards/margins": 5.6875,
-      "rewards/rejected": -11.5,
+      "logits/chosen": -5.15625,
+      "logits/rejected": -2.984375,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -1936.0,
+      "loss": 0.0976,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -7.46875,
+      "rewards/margins": 6.1875,
+      "rewards/rejected": -13.6875,
       "step": 1260
     },
     {
       "epoch": 1.3298429319371727,
-      "grad_norm": 1.8159486668552183,
+      "grad_norm": 1.462005068992466,
       "learning_rate": 3.410786151595533e-05,
-      "logits/chosen": -6.21875,
-      "logits/rejected": -3.8125,
-      "logps/chosen": -956.0,
-      "logps/rejected": -1592.0,
-      "loss": 0.0913,
+      "logits/chosen": -5.90625,
+      "logits/rejected": -3.703125,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -1832.0,
+      "loss": 0.0939,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -5.71875,
-      "rewards/margins": 5.5,
-      "rewards/rejected": -11.25,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 6.0,
+      "rewards/rejected": -13.625,
       "step": 1270
     },
     {
       "epoch": 1.3403141361256545,
-      "grad_norm": 1.9280728082989547,
+      "grad_norm": 2.901256884923763,
       "learning_rate": 3.382347526520765e-05,
-      "logits/chosen": -6.28125,
-      "logits/rejected": -3.9375,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -1920.0,
-      "loss": 0.0995,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.3125,
-      "rewards/margins": 6.4375,
-      "rewards/rejected": -13.75,
+      "logits/chosen": -6.25,
+      "logits/rejected": -4.125,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -1992.0,
+      "loss": 0.0974,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 6.1875,
+      "rewards/rejected": -14.5,
       "step": 1280
     },
     {
       "epoch": 1.350785340314136,
-      "grad_norm": 3.125176072583911,
+      "grad_norm": 3.6719217976392375,
       "learning_rate": 3.353777872207381e-05,
-      "logits/chosen": -5.9375,
-      "logits/rejected": -3.609375,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -2016.0,
-      "loss": 0.09,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.71875,
-      "rewards/margins": 6.4375,
-      "rewards/rejected": -14.125,
+      "logits/chosen": -6.875,
+      "logits/rejected": -4.71875,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -2024.0,
+      "loss": 0.0909,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -7.8125,
+      "rewards/margins": 6.375,
+      "rewards/rejected": -14.1875,
       "step": 1290
     },
     {
       "epoch": 1.3612565445026177,
-      "grad_norm": 1.3993462194442678,
+      "grad_norm": 1.8708980452478416,
       "learning_rate": 3.325081431269475e-05,
-      "logits/chosen": -4.65625,
-      "logits/rejected": -3.078125,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -1888.0,
-      "loss": 0.102,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.0,
-      "rewards/margins": 6.0,
-      "rewards/rejected": -13.0,
+      "logits/chosen": -5.1875,
+      "logits/rejected": -3.5625,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -2144.0,
+      "loss": 0.1291,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -8.4375,
+      "rewards/margins": 7.15625,
+      "rewards/rejected": -15.5625,
       "step": 1300
     },
     {
       "epoch": 1.3612565445026177,
-      "eval_logits/chosen": -4.0625,
-      "eval_logits/rejected": -2.09375,
-      "eval_logps/chosen": -1136.0,
-      "eval_logps/rejected": -1232.0,
-      "eval_loss": 1.1325156688690186,
-      "eval_rewards/accuracies": 0.607421875,
-      "eval_rewards/chosen": -8.1875,
-      "eval_rewards/margins": 1.28125,
-      "eval_rewards/rejected": -9.4375,
-      "eval_runtime": 46.6252,
-      "eval_samples_per_second": 42.895,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -4.59375,
+      "eval_logits/rejected": -2.65625,
+      "eval_logps/chosen": -1192.0,
+      "eval_logps/rejected": -1296.0,
+      "eval_loss": 1.165765643119812,
+      "eval_rewards/accuracies": 0.61328125,
+      "eval_rewards/chosen": -8.75,
+      "eval_rewards/margins": 1.296875,
+      "eval_rewards/rejected": -10.0625,
+      "eval_runtime": 47.0911,
+      "eval_samples_per_second": 42.471,
+      "eval_steps_per_second": 0.68,
       "step": 1300
     },
     {
       "epoch": 1.3717277486910995,
-      "grad_norm": 0.818876128439425,
+      "grad_norm": 1.2328616934690808,
       "learning_rate": 3.296262465149046e-05,
-      "logits/chosen": -6.3125,
-      "logits/rejected": -3.796875,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -1800.0,
-      "loss": 0.091,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.40625,
-      "rewards/margins": 5.71875,
-      "rewards/rejected": -12.125,
+      "logits/chosen": -6.46875,
+      "logits/rejected": -3.984375,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -1896.0,
+      "loss": 0.0845,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -6.9375,
+      "rewards/margins": 6.1875,
+      "rewards/rejected": -13.0625,
       "step": 1310
     },
     {
       "epoch": 1.3821989528795813,
-      "grad_norm": 0.5270776446926717,
+      "grad_norm": 1.1147731501344902,
       "learning_rate": 3.2673252534831687e-05,
-      "logits/chosen": -4.59375,
-      "logits/rejected": -2.296875,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1856.0,
-      "loss": 0.0744,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.09375,
-      "rewards/margins": 5.71875,
-      "rewards/rejected": -12.8125,
+      "logits/chosen": -5.28125,
+      "logits/rejected": -2.984375,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -1912.0,
+      "loss": 0.0872,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -7.25,
+      "rewards/margins": 6.0625,
+      "rewards/rejected": -13.3125,
       "step": 1320
     },
     {
       "epoch": 1.3926701570680629,
-      "grad_norm": 4.436881380957505,
+      "grad_norm": 2.9882590329216883,
       "learning_rate": 3.2382740934684696e-05,
-      "logits/chosen": -2.484375,
-      "logits/rejected": -0.9453125,
+      "logits/chosen": -3.21875,
+      "logits/rejected": -1.6875,
       "logps/chosen": -1320.0,
       "logps/rejected": -2272.0,
-      "loss": 0.117,
-      "rewards/accuracies": 0.9312499761581421,
+      "loss": 0.1262,
+      "rewards/accuracies": 0.9437500238418579,
       "rewards/chosen": -9.1875,
-      "rewards/margins": 7.0,
-      "rewards/rejected": -16.25,
+      "rewards/margins": 6.875,
+      "rewards/rejected": -16.125,
       "step": 1330
     },
     {
       "epoch": 1.4031413612565444,
-      "grad_norm": 1.6008848716024413,
+      "grad_norm": 1.2757300715297355,
       "learning_rate": 3.209113299222982e-05,
-      "logits/chosen": -3.5,
-      "logits/rejected": -2.546875,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -1944.0,
-      "loss": 0.0991,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.5,
-      "rewards/margins": 6.3125,
-      "rewards/rejected": -13.8125,
+      "logits/chosen": -3.125,
+      "logits/rejected": -2.734375,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -2048.0,
+      "loss": 0.0969,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -8.625,
+      "rewards/margins": 6.1875,
+      "rewards/rejected": -14.8125,
       "step": 1340
     },
     {
       "epoch": 1.4136125654450262,
-      "grad_norm": 1.7943605925009825,
+      "grad_norm": 2.6096513121300338,
       "learning_rate": 3.179847201145505e-05,
-      "logits/chosen": -5.3125,
-      "logits/rejected": -3.25,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -1928.0,
-      "loss": 0.1004,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -6.625,
-      "rewards/margins": 5.9375,
-      "rewards/rejected": -12.5625,
+      "logits/chosen": -3.671875,
+      "logits/rejected": -2.515625,
+      "logps/chosen": -1280.0,
+      "logps/rejected": -2176.0,
+      "loss": 0.088,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -8.6875,
+      "rewards/margins": 6.4375,
+      "rewards/rejected": -15.125,
       "step": 1350
     },
     {
       "epoch": 1.4240837696335078,
-      "grad_norm": 0.8228436795201145,
+      "grad_norm": 1.0809039561122553,
       "learning_rate": 3.150480145272528e-05,
-      "logits/chosen": -5.28125,
-      "logits/rejected": -3.703125,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -1928.0,
-      "loss": 0.0874,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.65625,
-      "rewards/margins": 5.90625,
-      "rewards/rejected": -13.5625,
+      "logits/chosen": -3.921875,
+      "logits/rejected": -2.984375,
+      "logps/chosen": -1368.0,
+      "logps/rejected": -2192.0,
+      "loss": 0.0968,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -9.6875,
+      "rewards/margins": 6.46875,
+      "rewards/rejected": -16.125,
       "step": 1360
     },
     {
       "epoch": 1.4345549738219896,
-      "grad_norm": 1.301711880847229,
+      "grad_norm": 1.2984827230940517,
       "learning_rate": 3.121016492632848e-05,
-      "logits/chosen": -5.5,
-      "logits/rejected": -3.484375,
-      "logps/chosen": -1216.0,
-      "logps/rejected": -2032.0,
-      "loss": 0.0891,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 6.4375,
-      "rewards/rejected": -14.5625,
+      "logits/chosen": -5.15625,
+      "logits/rejected": -3.53125,
+      "logps/chosen": -1288.0,
+      "logps/rejected": -2112.0,
+      "loss": 0.0998,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -8.8125,
+      "rewards/margins": 6.53125,
+      "rewards/rejected": -15.375,
       "step": 1370
     },
     {
       "epoch": 1.4450261780104712,
-      "grad_norm": 2.352292135411106,
+      "grad_norm": 2.3153852483970154,
       "learning_rate": 3.091460618599951e-05,
-      "logits/chosen": -4.0,
-      "logits/rejected": -2.28125,
-      "logps/chosen": -1288.0,
+      "logits/chosen": -4.625,
+      "logits/rejected": -2.75,
+      "logps/chosen": -1256.0,
       "logps/rejected": -2192.0,
-      "loss": 0.0916,
+      "loss": 0.0897,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 6.875,
-      "rewards/rejected": -15.875,
+      "rewards/chosen": -8.625,
+      "rewards/margins": 7.15625,
+      "rewards/rejected": -15.8125,
       "step": 1380
     },
     {
       "epoch": 1.4554973821989527,
-      "grad_norm": 1.5668802290557498,
+      "grad_norm": 2.568116251434959,
       "learning_rate": 3.061816912242265e-05,
-      "logits/chosen": -4.4375,
-      "logits/rejected": -3.015625,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -2144.0,
-      "loss": 0.101,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 6.21875,
-      "rewards/rejected": -15.0,
+      "logits/chosen": -4.34375,
+      "logits/rejected": -3.03125,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -2112.0,
+      "loss": 0.0867,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 6.3125,
+      "rewards/rejected": -14.625,
       "step": 1390
     },
     {
       "epoch": 1.4659685863874345,
-      "grad_norm": 0.7617513287728593,
+      "grad_norm": 0.820343372273228,
       "learning_rate": 3.0320897756713785e-05,
-      "logits/chosen": -5.25,
-      "logits/rejected": -4.0625,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -2048.0,
-      "loss": 0.0893,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.8125,
-      "rewards/margins": 6.59375,
-      "rewards/rejected": -13.4375,
+      "logits/chosen": -5.125,
+      "logits/rejected": -3.9375,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -2208.0,
+      "loss": 0.0877,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -7.625,
+      "rewards/margins": 7.375,
+      "rewards/rejected": -15.0,
       "step": 1400
     },
     {
       "epoch": 1.4659685863874345,
-      "eval_logits/chosen": -3.734375,
-      "eval_logits/rejected": -1.765625,
-      "eval_logps/chosen": -1096.0,
-      "eval_logps/rejected": -1184.0,
-      "eval_loss": 1.0033906698226929,
-      "eval_rewards/accuracies": 0.630859375,
-      "eval_rewards/chosen": -7.75,
-      "eval_rewards/margins": 1.234375,
-      "eval_rewards/rejected": -9.0,
-      "eval_runtime": 46.638,
-      "eval_samples_per_second": 42.883,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -3.296875,
+      "eval_logits/rejected": -1.3359375,
+      "eval_logps/chosen": -1216.0,
+      "eval_logps/rejected": -1320.0,
+      "eval_loss": 1.1249375343322754,
+      "eval_rewards/accuracies": 0.625,
+      "eval_rewards/chosen": -9.0,
+      "eval_rewards/margins": 1.34375,
+      "eval_rewards/rejected": -10.375,
+      "eval_runtime": 47.0834,
+      "eval_samples_per_second": 42.478,
+      "eval_steps_per_second": 0.68,
       "step": 1400
     },
     {
       "epoch": 1.4764397905759163,
-      "grad_norm": 2.5618002762745338,
+      "grad_norm": 2.1170140180674855,
       "learning_rate": 3.002283623388332e-05,
-      "logits/chosen": -4.8125,
-      "logits/rejected": -3.09375,
-      "logps/chosen": -1032.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.0931,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -6.6875,
-      "rewards/margins": 5.40625,
-      "rewards/rejected": -12.125,
+      "logits/chosen": -4.375,
+      "logits/rejected": -2.828125,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -1928.0,
+      "loss": 0.1084,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -7.8125,
+      "rewards/margins": 5.96875,
+      "rewards/rejected": -13.75,
       "step": 1410
     },
     {
       "epoch": 1.486910994764398,
-      "grad_norm": 2.727555469122087,
+      "grad_norm": 2.7336680921377425,
       "learning_rate": 2.9724028816280507e-05,
-      "logits/chosen": -3.84375,
-      "logits/rejected": -2.6875,
+      "logits/chosen": -3.96875,
+      "logits/rejected": -2.890625,
       "logps/chosen": -1168.0,
-      "logps/rejected": -1872.0,
-      "loss": 0.115,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.9375,
-      "rewards/margins": 5.5,
-      "rewards/rejected": -13.4375,
+      "logps/rejected": -1824.0,
+      "loss": 0.1188,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -7.90625,
+      "rewards/margins": 5.09375,
+      "rewards/rejected": -13.0,
       "step": 1420
     },
     {
       "epoch": 1.4973821989528795,
-      "grad_norm": 1.8670853750097924,
+      "grad_norm": 0.89921737457481,
       "learning_rate": 2.9424519877020524e-05,
-      "logits/chosen": -5.46875,
-      "logits/rejected": -3.8125,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -2128.0,
-      "loss": 0.0922,
+      "logits/chosen": -5.0625,
+      "logits/rejected": -3.6875,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -2016.0,
+      "loss": 0.1061,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.0,
-      "rewards/margins": 6.34375,
-      "rewards/rejected": -14.375,
+      "rewards/chosen": -7.875,
+      "rewards/margins": 5.40625,
+      "rewards/rejected": -13.3125,
       "step": 1430
     },
     {
       "epoch": 1.5078534031413613,
-      "grad_norm": 2.082432923035952,
+      "grad_norm": 1.6877993547731167,
       "learning_rate": 2.912435389339504e-05,
-      "logits/chosen": -4.53125,
-      "logits/rejected": -1.890625,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -1944.0,
-      "loss": 0.0963,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.9375,
-      "rewards/margins": 5.875,
-      "rewards/rejected": -13.8125,
+      "logits/chosen": -2.578125,
+      "logits/rejected": -0.2177734375,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -1976.0,
+      "loss": 0.0907,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.6875,
+      "rewards/margins": 5.46875,
+      "rewards/rejected": -14.125,
       "step": 1440
     },
     {
       "epoch": 1.518324607329843,
-      "grad_norm": 2.0307989334818037,
+      "grad_norm": 2.999790836072498,
       "learning_rate": 2.8823575440267226e-05,
-      "logits/chosen": -5.90625,
-      "logits/rejected": -4.15625,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1936.0,
-      "loss": 0.095,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -7.03125,
-      "rewards/margins": 5.9375,
-      "rewards/rejected": -13.0,
+      "logits/chosen": -3.578125,
+      "logits/rejected": -2.46875,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -2040.0,
+      "loss": 0.1018,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 5.78125,
+      "rewards/rejected": -14.0625,
       "step": 1450
     },
     {
       "epoch": 1.5287958115183247,
-      "grad_norm": 2.0862438932717584,
+      "grad_norm": 2.332152261924408,
       "learning_rate": 2.8522229183452465e-05,
-      "logits/chosen": -5.53125,
-      "logits/rejected": -3.171875,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -1768.0,
-      "loss": 0.1032,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.5,
-      "rewards/margins": 5.78125,
-      "rewards/rejected": -12.25,
+      "logits/chosen": -4.15625,
+      "logits/rejected": -1.703125,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -1912.0,
+      "loss": 0.1236,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.53125,
+      "rewards/margins": 6.21875,
+      "rewards/rejected": -13.75,
       "step": 1460
     },
     {
       "epoch": 1.5392670157068062,
-      "grad_norm": 1.792237151103703,
+      "grad_norm": 2.398247558300056,
       "learning_rate": 2.8220359873085373e-05,
-      "logits/chosen": -4.65625,
-      "logits/rejected": -2.90625,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -1920.0,
-      "loss": 0.0987,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -6.96875,
-      "rewards/margins": 6.0625,
-      "rewards/rejected": -13.0625,
+      "logits/chosen": -4.21875,
+      "logits/rejected": -2.3125,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -1952.0,
+      "loss": 0.0829,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.21875,
+      "rewards/margins": 6.125,
+      "rewards/rejected": -13.3125,
       "step": 1470
     },
     {
       "epoch": 1.5497382198952878,
-      "grad_norm": 1.0375757121861808,
+      "grad_norm": 1.8227549611865144,
       "learning_rate": 2.7918012336974388e-05,
-      "logits/chosen": -4.53125,
-      "logits/rejected": -3.359375,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -1848.0,
-      "loss": 0.0893,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -6.78125,
-      "rewards/margins": 6.09375,
-      "rewards/rejected": -12.875,
+      "logits/chosen": -3.46875,
+      "logits/rejected": -2.359375,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -1936.0,
+      "loss": 0.1023,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.6875,
+      "rewards/margins": 6.125,
+      "rewards/rejected": -13.8125,
       "step": 1480
     },
     {
       "epoch": 1.5602094240837696,
-      "grad_norm": 1.6441271000213438,
+      "grad_norm": 3.5892433035479483,
       "learning_rate": 2.7615231473944815e-05,
-      "logits/chosen": -4.1875,
-      "logits/rejected": -2.453125,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -1832.0,
-      "loss": 0.0834,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.21875,
-      "rewards/margins": 5.71875,
-      "rewards/rejected": -12.9375,
+      "logits/chosen": -2.125,
+      "logits/rejected": -0.921875,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -2048.0,
+      "loss": 0.0891,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -8.875,
+      "rewards/margins": 6.21875,
+      "rewards/rejected": -15.0625,
       "step": 1490
     },
     {
       "epoch": 1.5706806282722514,
-      "grad_norm": 4.096341983067086,
+      "grad_norm": 2.7830723360928036,
       "learning_rate": 2.7312062247171327e-05,
-      "logits/chosen": -5.1875,
-      "logits/rejected": -3.59375,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -1976.0,
-      "loss": 0.0968,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.5625,
-      "rewards/margins": 6.0625,
-      "rewards/rejected": -13.625,
+      "logits/chosen": -3.609375,
+      "logits/rejected": -2.390625,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -2080.0,
+      "loss": 0.1044,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -8.4375,
+      "rewards/margins": 6.15625,
+      "rewards/rejected": -14.625,
       "step": 1500
     },
     {
       "epoch": 1.5706806282722514,
-      "eval_logits/chosen": -3.0625,
-      "eval_logits/rejected": -1.1015625,
-      "eval_logps/chosen": -1280.0,
-      "eval_logps/rejected": -1416.0,
-      "eval_loss": 1.221203088760376,
-      "eval_rewards/accuracies": 0.595703125,
-      "eval_rewards/chosen": -9.625,
-      "eval_rewards/margins": 1.625,
-      "eval_rewards/rejected": -11.25,
-      "eval_runtime": 46.6286,
-      "eval_samples_per_second": 42.892,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -1.53125,
+      "eval_logits/rejected": 0.318359375,
+      "eval_logps/chosen": -1296.0,
+      "eval_logps/rejected": -1424.0,
+      "eval_loss": 1.152273416519165,
+      "eval_rewards/accuracies": 0.6484375,
+      "eval_rewards/chosen": -9.75,
+      "eval_rewards/margins": 1.6328125,
+      "eval_rewards/rejected": -11.375,
+      "eval_runtime": 47.0923,
+      "eval_samples_per_second": 42.47,
+      "eval_steps_per_second": 0.68,
       "step": 1500
     },
     {
       "epoch": 1.581151832460733,
-      "grad_norm": 1.3761788744023733,
+      "grad_norm": 0.542876855072402,
       "learning_rate": 2.7008549677500876e-05,
-      "logits/chosen": -6.28125,
-      "logits/rejected": -4.4375,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1952.0,
-      "loss": 0.0978,
+      "logits/chosen": -4.9375,
+      "logits/rejected": -3.328125,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -1888.0,
+      "loss": 0.0931,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.6875,
-      "rewards/margins": 5.9375,
-      "rewards/rejected": -13.625,
+      "rewards/chosen": -7.5,
+      "rewards/margins": 5.5625,
+      "rewards/rejected": -13.0625,
       "step": 1510
     },
     {
       "epoch": 1.5916230366492146,
-      "grad_norm": 1.7673584569409686,
+      "grad_norm": 1.021372172518936,
       "learning_rate": 2.670473883676709e-05,
-      "logits/chosen": -5.625,
-      "logits/rejected": -3.765625,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -2176.0,
-      "loss": 0.076,
+      "logits/chosen": -3.421875,
+      "logits/rejected": -2.265625,
+      "logps/chosen": -1224.0,
+      "logps/rejected": -2160.0,
+      "loss": 0.0753,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.8125,
-      "rewards/margins": 6.625,
-      "rewards/rejected": -14.4375,
+      "rewards/chosen": -8.125,
+      "rewards/margins": 6.25,
+      "rewards/rejected": -14.375,
       "step": 1520
     },
     {
       "epoch": 1.6020942408376964,
-      "grad_norm": 2.8473280798246337,
+      "grad_norm": 2.0816524921421324,
       "learning_rate": 2.6400674841097074e-05,
-      "logits/chosen": -6.34375,
-      "logits/rejected": -4.4375,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -1984.0,
-      "loss": 0.0908,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -8.5625,
-      "rewards/margins": 5.78125,
-      "rewards/rejected": -14.375,
+      "logits/chosen": -4.3125,
+      "logits/rejected": -3.0,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -2096.0,
+      "loss": 0.0919,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -9.5,
+      "rewards/margins": 5.9375,
+      "rewards/rejected": -15.4375,
       "step": 1530
     },
     {
       "epoch": 1.6125654450261782,
-      "grad_norm": 1.4334704714919722,
+      "grad_norm": 1.925179478757633,
       "learning_rate": 2.609640284421158e-05,
-      "logits/chosen": -6.4375,
-      "logits/rejected": -4.40625,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -1928.0,
-      "loss": 0.105,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 6.125,
-      "rewards/rejected": -13.75,
+      "logits/chosen": -5.25,
+      "logits/rejected": -3.5625,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -2048.0,
+      "loss": 0.1042,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -8.5625,
+      "rewards/margins": 6.46875,
+      "rewards/rejected": -15.0625,
       "step": 1540
     },
     {
       "epoch": 1.6230366492146597,
-      "grad_norm": 1.7976112099630002,
+      "grad_norm": 1.3879869247573788,
       "learning_rate": 2.5791968030719726e-05,
-      "logits/chosen": -6.78125,
-      "logits/rejected": -4.375,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1896.0,
-      "loss": 0.0956,
+      "logits/chosen": -6.875,
+      "logits/rejected": -4.65625,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -2016.0,
+      "loss": 0.088,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -6.5,
-      "rewards/margins": 6.09375,
-      "rewards/rejected": -12.625,
+      "rewards/chosen": -7.4375,
+      "rewards/margins": 6.375,
+      "rewards/rejected": -13.8125,
       "step": 1550
     },
     {
       "epoch": 1.6335078534031413,
-      "grad_norm": 2.344835879610543,
+      "grad_norm": 1.9339185977439384,
       "learning_rate": 2.548741560940902e-05,
-      "logits/chosen": -6.5,
-      "logits/rejected": -4.5625,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -1928.0,
-      "loss": 0.0998,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.71875,
-      "rewards/margins": 6.0,
-      "rewards/rejected": -12.75,
+      "logits/chosen": -6.71875,
+      "logits/rejected": -4.9375,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -2016.0,
+      "loss": 0.0984,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.125,
+      "rewards/margins": 6.53125,
+      "rewards/rejected": -13.625,
       "step": 1560
     },
     {
       "epoch": 1.6439790575916229,
-      "grad_norm": 1.6914663154842435,
+      "grad_norm": 2.7758278606408666,
       "learning_rate": 2.5182790806531787e-05,
-      "logits/chosen": -4.125,
-      "logits/rejected": -3.109375,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -2080.0,
-      "loss": 0.1025,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.0,
-      "rewards/margins": 6.65625,
-      "rewards/rejected": -14.6875,
+      "logits/chosen": -6.0,
+      "logits/rejected": -4.75,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -2048.0,
+      "loss": 0.113,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -7.5625,
+      "rewards/margins": 6.71875,
+      "rewards/rejected": -14.3125,
       "step": 1570
     },
     {
       "epoch": 1.6544502617801047,
-      "grad_norm": 2.7577901809452747,
+      "grad_norm": 2.015381760096074,
       "learning_rate": 2.4878138859089074e-05,
-      "logits/chosen": -4.03125,
-      "logits/rejected": -2.546875,
-      "logps/chosen": -1168.0,
+      "logits/chosen": -6.1875,
+      "logits/rejected": -4.28125,
+      "logps/chosen": -1160.0,
       "logps/rejected": -2048.0,
-      "loss": 0.0836,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.65625,
-      "rewards/margins": 6.84375,
-      "rewards/rejected": -14.5,
+      "loss": 0.0796,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 6.96875,
+      "rewards/rejected": -14.5625,
       "step": 1580
     },
     {
       "epoch": 1.6649214659685865,
-      "grad_norm": 1.5159383239881494,
+      "grad_norm": 1.1340180389635373,
       "learning_rate": 2.4573505008112924e-05,
-      "logits/chosen": -3.828125,
-      "logits/rejected": -2.15625,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1888.0,
-      "loss": 0.074,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.90625,
-      "rewards/margins": 5.8125,
-      "rewards/rejected": -13.75,
+      "logits/chosen": -5.6875,
+      "logits/rejected": -3.859375,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -1968.0,
+      "loss": 0.066,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -8.375,
+      "rewards/margins": 6.1875,
+      "rewards/rejected": -14.5,
       "step": 1590
     },
     {
       "epoch": 1.675392670157068,
-      "grad_norm": 1.6817330614934756,
+      "grad_norm": 1.4426622442082426,
       "learning_rate": 2.4268934491948032e-05,
-      "logits/chosen": -3.375,
-      "logits/rejected": -2.40625,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -2000.0,
-      "loss": 0.082,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 6.09375,
-      "rewards/rejected": -14.5625,
+      "logits/chosen": -5.0625,
+      "logits/rejected": -3.875,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -2064.0,
+      "loss": 0.0798,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -9.25,
+      "rewards/margins": 5.9375,
+      "rewards/rejected": -15.1875,
       "step": 1600
     },
     {
       "epoch": 1.675392670157068,
-      "eval_logits/chosen": -0.287109375,
-      "eval_logits/rejected": 1.4609375,
-      "eval_logps/chosen": -1408.0,
-      "eval_logps/rejected": -1528.0,
-      "eval_loss": 1.2544218301773071,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -10.875,
-      "eval_rewards/margins": 1.515625,
-      "eval_rewards/rejected": -12.375,
-      "eval_runtime": 46.6215,
-      "eval_samples_per_second": 42.899,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -2.234375,
+      "eval_logits/rejected": -0.361328125,
+      "eval_logps/chosen": -1488.0,
+      "eval_logps/rejected": -1624.0,
+      "eval_loss": 1.3625078201293945,
+      "eval_rewards/accuracies": 0.6171875,
+      "eval_rewards/chosen": -11.6875,
+      "eval_rewards/margins": 1.65625,
+      "eval_rewards/rejected": -13.375,
+      "eval_runtime": 47.0885,
+      "eval_samples_per_second": 42.473,
+      "eval_steps_per_second": 0.68,
       "step": 1600
     },
     {
       "epoch": 1.6858638743455496,
-      "grad_norm": 0.43405874807942346,
+      "grad_norm": 1.0310987175899597,
       "learning_rate": 2.396447253953385e-05,
-      "logits/chosen": -3.734375,
-      "logits/rejected": -2.546875,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -2160.0,
-      "loss": 0.0693,
+      "logits/chosen": -5.75,
+      "logits/rejected": -4.15625,
+      "logps/chosen": -1352.0,
+      "logps/rejected": -2240.0,
+      "loss": 0.0718,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -8.75,
-      "rewards/margins": 6.96875,
-      "rewards/rejected": -15.75,
+      "rewards/chosen": -9.125,
+      "rewards/margins": 7.3125,
+      "rewards/rejected": -16.5,
       "step": 1610
     },
     {
       "epoch": 1.6963350785340314,
-      "grad_norm": 1.8682621956809984,
+      "grad_norm": 1.5346354875413641,
       "learning_rate": 2.3660164363687998e-05,
-      "logits/chosen": -3.859375,
-      "logits/rejected": -2.234375,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -2112.0,
-      "loss": 0.0701,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 6.65625,
-      "rewards/rejected": -15.0,
+      "logits/chosen": -6.09375,
+      "logits/rejected": -4.34375,
+      "logps/chosen": -1304.0,
+      "logps/rejected": -2096.0,
+      "loss": 0.0764,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -8.5625,
+      "rewards/margins": 6.3125,
+      "rewards/rejected": -14.8125,
       "step": 1620
     },
     {
       "epoch": 1.7068062827225132,
-      "grad_norm": 2.3638432631185684,
+      "grad_norm": 2.616419576418859,
       "learning_rate": 2.33560551543922e-05,
-      "logits/chosen": -4.09375,
-      "logits/rejected": -2.515625,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -1936.0,
-      "loss": 0.0815,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.28125,
-      "rewards/margins": 6.0,
-      "rewards/rejected": -13.3125,
+      "logits/chosen": -6.09375,
+      "logits/rejected": -3.953125,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -1992.0,
+      "loss": 0.073,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -7.78125,
+      "rewards/margins": 6.09375,
+      "rewards/rejected": -13.875,
       "step": 1630
     },
     {
       "epoch": 1.7172774869109948,
-      "grad_norm": 1.8312961050539147,
+      "grad_norm": 1.745924264164314,
       "learning_rate": 2.3052190072081493e-05,
-      "logits/chosen": -3.734375,
-      "logits/rejected": -1.796875,
-      "logps/chosen": -1152.0,
-      "logps/rejected": -1936.0,
-      "loss": 0.0768,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.40625,
-      "rewards/margins": 6.59375,
-      "rewards/rejected": -14.0,
+      "logits/chosen": -5.78125,
+      "logits/rejected": -3.21875,
+      "logps/chosen": -1224.0,
+      "logps/rejected": -1992.0,
+      "loss": 0.0798,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 6.40625,
+      "rewards/rejected": -14.5625,
       "step": 1640
     },
     {
       "epoch": 1.7277486910994764,
-      "grad_norm": 0.9024961976467272,
+      "grad_norm": 1.0855532344640153,
       "learning_rate": 2.2748614240937864e-05,
-      "logits/chosen": -3.640625,
-      "logits/rejected": -2.265625,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -2008.0,
-      "loss": 0.0631,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.0625,
-      "rewards/margins": 6.8125,
-      "rewards/rejected": -13.875,
+      "logits/chosen": -4.59375,
+      "logits/rejected": -3.21875,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -2208.0,
+      "loss": 0.0601,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 7.3125,
+      "rewards/rejected": -15.8125,
       "step": 1650
     },
     {
       "epoch": 1.738219895287958,
-      "grad_norm": 1.9173753677799552,
+      "grad_norm": 2.2131216719549776,
       "learning_rate": 2.2445372742189334e-05,
-      "logits/chosen": -3.109375,
-      "logits/rejected": -1.4375,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -2240.0,
-      "loss": 0.0891,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 7.28125,
-      "rewards/rejected": -16.0,
+      "logits/chosen": -4.21875,
+      "logits/rejected": -2.203125,
+      "logps/chosen": -1352.0,
+      "logps/rejected": -2336.0,
+      "loss": 0.1005,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -9.3125,
+      "rewards/margins": 7.5625,
+      "rewards/rejected": -16.875,
       "step": 1660
     },
     {
       "epoch": 1.7486910994764397,
-      "grad_norm": 2.0859799153453613,
+      "grad_norm": 1.532888727305927,
       "learning_rate": 2.214251060741528e-05,
-      "logits/chosen": -3.03125,
-      "logits/rejected": -1.7890625,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -2336.0,
-      "loss": 0.0693,
+      "logits/chosen": -5.1875,
+      "logits/rejected": -3.28125,
+      "logps/chosen": -1296.0,
+      "logps/rejected": -2176.0,
+      "loss": 0.0749,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 7.4375,
-      "rewards/rejected": -17.125,
+      "rewards/chosen": -8.6875,
+      "rewards/margins": 6.78125,
+      "rewards/rejected": -15.4375,
       "step": 1670
     },
     {
       "epoch": 1.7591623036649215,
-      "grad_norm": 1.8184345910733002,
+      "grad_norm": 2.7468523088707806,
       "learning_rate": 2.1840072811859302e-05,
-      "logits/chosen": -3.078125,
-      "logits/rejected": -1.5703125,
-      "logps/chosen": -1368.0,
-      "logps/rejected": -2288.0,
-      "loss": 0.0778,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.25,
-      "rewards/margins": 7.34375,
-      "rewards/rejected": -16.625,
+      "logits/chosen": -5.0625,
+      "logits/rejected": -3.0625,
+      "logps/chosen": -1296.0,
+      "logps/rejected": -2160.0,
+      "loss": 0.0872,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 6.8125,
+      "rewards/rejected": -15.3125,
       "step": 1680
     },
     {
       "epoch": 1.7696335078534031,
-      "grad_norm": 1.2891188477486866,
+      "grad_norm": 1.88322783183091,
       "learning_rate": 2.1538104267750287e-05,
-      "logits/chosen": -3.09375,
-      "logits/rejected": -1.734375,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -1872.0,
-      "loss": 0.1045,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.4375,
-      "rewards/margins": 5.71875,
-      "rewards/rejected": -13.1875,
+      "logits/chosen": -4.0625,
+      "logits/rejected": -2.46875,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -1992.0,
+      "loss": 0.1035,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -8.4375,
+      "rewards/margins": 5.90625,
+      "rewards/rejected": -14.3125,
       "step": 1690
     },
     {
       "epoch": 1.7801047120418847,
-      "grad_norm": 1.3581495655114637,
+      "grad_norm": 1.6324407923143298,
       "learning_rate": 2.1236649817632952e-05,
-      "logits/chosen": -4.0,
-      "logits/rejected": -3.140625,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -1808.0,
-      "loss": 0.0972,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -6.90625,
+      "logits/chosen": -3.9375,
+      "logits/rejected": -3.0625,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -2032.0,
+      "loss": 0.0847,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -9.125,
       "rewards/margins": 5.6875,
-      "rewards/rejected": -12.5625,
+      "rewards/rejected": -14.8125,
       "step": 1700
     },
     {
       "epoch": 1.7801047120418847,
-      "eval_logits/chosen": -1.984375,
-      "eval_logits/rejected": -0.294921875,
-      "eval_logps/chosen": -1208.0,
-      "eval_logps/rejected": -1320.0,
-      "eval_loss": 1.1152499914169312,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -8.875,
-      "eval_rewards/margins": 1.421875,
-      "eval_rewards/rejected": -10.3125,
-      "eval_runtime": 46.6349,
-      "eval_samples_per_second": 42.886,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -1.09375,
+      "eval_logits/rejected": 0.578125,
+      "eval_logps/chosen": -1544.0,
+      "eval_logps/rejected": -1656.0,
+      "eval_loss": 1.307437539100647,
+      "eval_rewards/accuracies": 0.62109375,
+      "eval_rewards/chosen": -12.25,
+      "eval_rewards/margins": 1.4375,
+      "eval_rewards/rejected": -13.6875,
+      "eval_runtime": 47.1105,
+      "eval_samples_per_second": 42.453,
+      "eval_steps_per_second": 0.679,
       "step": 1700
     },
     {
       "epoch": 1.7905759162303665,
-      "grad_norm": 3.045548713113102,
+      "grad_norm": 3.4698621331944133,
       "learning_rate": 2.093575422770872e-05,
-      "logits/chosen": -3.671875,
-      "logits/rejected": -2.484375,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.0938,
+      "logits/chosen": -2.5,
+      "logits/rejected": -1.546875,
+      "logps/chosen": -1408.0,
+      "logps/rejected": -2192.0,
+      "loss": 0.08,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.875,
-      "rewards/margins": 5.65625,
-      "rewards/rejected": -12.5,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 6.34375,
+      "rewards/rejected": -16.625,
       "step": 1710
     },
     {
       "epoch": 1.8010471204188483,
-      "grad_norm": 2.074792223244018,
+      "grad_norm": 3.244021486396508,
       "learning_rate": 2.0635462181187827e-05,
-      "logits/chosen": -4.5625,
-      "logits/rejected": -3.359375,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -2112.0,
-      "loss": 0.0862,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 6.59375,
-      "rewards/rejected": -14.8125,
+      "logits/chosen": -3.4375,
+      "logits/rejected": -2.5625,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.0912,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -10.9375,
+      "rewards/margins": 6.96875,
+      "rewards/rejected": -17.875,
       "step": 1720
     },
     {
       "epoch": 1.8115183246073299,
-      "grad_norm": 2.0771716318505757,
+      "grad_norm": 2.602033619168437,
       "learning_rate": 2.0335818271653917e-05,
-      "logits/chosen": -4.6875,
-      "logits/rejected": -3.5,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1936.0,
-      "loss": 0.073,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.9375,
-      "rewards/margins": 6.28125,
-      "rewards/rejected": -13.1875,
+      "logits/chosen": -4.15625,
+      "logits/rejected": -3.015625,
+      "logps/chosen": -1312.0,
+      "logps/rejected": -2160.0,
+      "loss": 0.0834,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -9.125,
+      "rewards/margins": 6.34375,
+      "rewards/rejected": -15.4375,
       "step": 1730
     },
     {
       "epoch": 1.8219895287958114,
-      "grad_norm": 1.6183678632381786,
+      "grad_norm": 2.7601551424860826,
       "learning_rate": 2.0036866996441815e-05,
-      "logits/chosen": -4.71875,
-      "logits/rejected": -2.4375,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -2016.0,
-      "loss": 0.0751,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.71875,
-      "rewards/margins": 6.6875,
-      "rewards/rejected": -14.375,
+      "logits/chosen": -4.40625,
+      "logits/rejected": -2.375,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -2096.0,
+      "loss": 0.0875,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -8.75,
+      "rewards/margins": 6.375,
+      "rewards/rejected": -15.125,
       "step": 1740
     },
     {
       "epoch": 1.8324607329842932,
-      "grad_norm": 1.96383796631754,
+      "grad_norm": 3.55242376181993,
       "learning_rate": 1.9738652750029677e-05,
-      "logits/chosen": -3.90625,
-      "logits/rejected": -2.359375,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -2176.0,
-      "loss": 0.0727,
+      "logits/chosen": -3.453125,
+      "logits/rejected": -2.296875,
+      "logps/chosen": -1352.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.0833,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -8.375,
-      "rewards/margins": 7.40625,
-      "rewards/rejected": -15.75,
+      "rewards/chosen": -9.25,
+      "rewards/margins": 7.0,
+      "rewards/rejected": -16.25,
       "step": 1750
     },
     {
       "epoch": 1.8429319371727748,
-      "grad_norm": 1.8600399859166736,
+      "grad_norm": 2.4991335839252526,
       "learning_rate": 1.944121981744631e-05,
-      "logits/chosen": -4.8125,
-      "logits/rejected": -2.953125,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -2288.0,
-      "loss": 0.0823,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -8.4375,
-      "rewards/margins": 7.625,
-      "rewards/rejected": -16.0,
+      "logits/chosen": -5.21875,
+      "logits/rejected": -3.484375,
+      "logps/chosen": -1352.0,
+      "logps/rejected": -2240.0,
+      "loss": 0.0849,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -9.0,
+      "rewards/margins": 6.5625,
+      "rewards/rejected": -15.625,
       "step": 1760
     },
     {
       "epoch": 1.8534031413612566,
-      "grad_norm": 1.4254977139340186,
+      "grad_norm": 1.0867713650815005,
       "learning_rate": 1.9144612367694887e-05,
-      "logits/chosen": -3.984375,
-      "logits/rejected": -2.078125,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -1928.0,
-      "loss": 0.0785,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.25,
-      "rewards/margins": 6.6875,
-      "rewards/rejected": -13.9375,
+      "logits/chosen": -5.125,
+      "logits/rejected": -3.3125,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -2008.0,
+      "loss": 0.0667,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -8.375,
+      "rewards/margins": 6.3125,
+      "rewards/rejected": -14.6875,
       "step": 1770
     },
     {
       "epoch": 1.8638743455497382,
-      "grad_norm": 3.5029641552411337,
+      "grad_norm": 1.5280300607044794,
       "learning_rate": 1.8848874447193803e-05,
-      "logits/chosen": -2.9375,
-      "logits/rejected": -1.2578125,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -1944.0,
-      "loss": 0.0917,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.53125,
-      "rewards/margins": 6.65625,
-      "rewards/rejected": -14.1875,
+      "logits/chosen": -4.1875,
+      "logits/rejected": -2.4375,
+      "logps/chosen": -1304.0,
+      "logps/rejected": -2128.0,
+      "loss": 0.0814,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -9.125,
+      "rewards/margins": 6.90625,
+      "rewards/rejected": -16.0,
       "step": 1780
     },
     {
       "epoch": 1.8743455497382198,
-      "grad_norm": 2.6217409118658677,
+      "grad_norm": 2.4360778138595505,
       "learning_rate": 1.855404997323571e-05,
-      "logits/chosen": -3.625,
-      "logits/rejected": -2.40625,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -1984.0,
-      "loss": 0.0938,
+      "logits/chosen": -5.125,
+      "logits/rejected": -3.671875,
+      "logps/chosen": -1336.0,
+      "logps/rejected": -2176.0,
+      "loss": 0.0884,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.75,
-      "rewards/margins": 6.53125,
-      "rewards/rejected": -14.3125,
+      "rewards/chosen": -9.25,
+      "rewards/margins": 6.96875,
+      "rewards/rejected": -16.25,
       "step": 1790
     },
     {
       "epoch": 1.8848167539267016,
-      "grad_norm": 1.2265373653162488,
+      "grad_norm": 1.3496661620645678,
       "learning_rate": 1.82601827274658e-05,
-      "logits/chosen": -3.84375,
-      "logits/rejected": -2.046875,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.1053,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.78125,
-      "rewards/margins": 5.9375,
-      "rewards/rejected": -12.75,
+      "logits/chosen": -5.5625,
+      "logits/rejected": -3.703125,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -1952.0,
+      "loss": 0.1018,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -8.375,
+      "rewards/margins": 6.0,
+      "rewards/rejected": -14.375,
       "step": 1800
     },
     {
       "epoch": 1.8848167539267016,
-      "eval_logits/chosen": -1.265625,
-      "eval_logits/rejected": 0.60546875,
-      "eval_logps/chosen": -1088.0,
-      "eval_logps/rejected": -1200.0,
-      "eval_loss": 1.0633906126022339,
-      "eval_rewards/accuracies": 0.623046875,
-      "eval_rewards/chosen": -7.6875,
-      "eval_rewards/margins": 1.4375,
-      "eval_rewards/rejected": -9.125,
-      "eval_runtime": 46.632,
-      "eval_samples_per_second": 42.889,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -3.3125,
+      "eval_logits/rejected": -1.421875,
+      "eval_logps/chosen": -1264.0,
+      "eval_logps/rejected": -1376.0,
+      "eval_loss": 1.1159921884536743,
+      "eval_rewards/accuracies": 0.638671875,
+      "eval_rewards/chosen": -9.4375,
+      "eval_rewards/margins": 1.421875,
+      "eval_rewards/rejected": -10.8125,
+      "eval_runtime": 47.1075,
+      "eval_samples_per_second": 42.456,
+      "eval_steps_per_second": 0.679,
       "step": 1800
     },
     {
       "epoch": 1.8952879581151834,
-      "grad_norm": 3.698381135634359,
+      "grad_norm": 2.506337636253115,
       "learning_rate": 1.7967316349380213e-05,
-      "logits/chosen": -3.296875,
-      "logits/rejected": -2.03125,
-      "logps/chosen": -996.0,
-      "logps/rejected": -1728.0,
-      "loss": 0.0923,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -6.25,
-      "rewards/margins": 5.84375,
-      "rewards/rejected": -12.0625,
+      "logits/chosen": -5.03125,
+      "logits/rejected": -3.71875,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -1936.0,
+      "loss": 0.0892,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -8.0625,
+      "rewards/margins": 6.09375,
+      "rewards/rejected": -14.125,
       "step": 1810
     },
     {
       "epoch": 1.905759162303665,
-      "grad_norm": 1.2925052376816188,
+      "grad_norm": 1.9237111956589796,
       "learning_rate": 1.7675494329845516e-05,
-      "logits/chosen": -2.90625,
-      "logits/rejected": -1.453125,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -1960.0,
-      "loss": 0.0777,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.59375,
-      "rewards/margins": 6.28125,
-      "rewards/rejected": -13.875,
+      "logits/chosen": -4.75,
+      "logits/rejected": -3.171875,
+      "logps/chosen": -1312.0,
+      "logps/rejected": -2112.0,
+      "loss": 0.0655,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -9.1875,
+      "rewards/margins": 6.21875,
+      "rewards/rejected": -15.4375,
       "step": 1820
     },
     {
       "epoch": 1.9162303664921465,
-      "grad_norm": 0.8622030234952285,
+      "grad_norm": 0.582904913258408,
       "learning_rate": 1.738476000464026e-05,
-      "logits/chosen": -3.390625,
-      "logits/rejected": -1.6640625,
-      "logps/chosen": -1072.0,
-      "logps/rejected": -1888.0,
-      "loss": 0.086,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -6.8125,
+      "logits/chosen": -5.25,
+      "logits/rejected": -3.5,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -2064.0,
+      "loss": 0.0875,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -8.5625,
       "rewards/margins": 6.5625,
-      "rewards/rejected": -13.375,
+      "rewards/rejected": -15.125,
       "step": 1830
     },
     {
-      "epoch": 1.9267015706806283,
-      "grad_norm": 1.7739326531475843,
-      "learning_rate": 1.7095156548019646e-05,
-      "logits/chosen": -4.75,
-      "logits/rejected": -3.015625,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -1928.0,
-      "loss": 0.0803,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.65625,
-      "rewards/margins": 6.71875,
-      "rewards/rejected": -13.375,
+      "epoch": 1.9267015706806283,
+      "grad_norm": 3.7238681440342476,
+      "learning_rate": 1.7095156548019646e-05,
+      "logits/chosen": -6.75,
+      "logits/rejected": -4.8125,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -2112.0,
+      "loss": 0.0851,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -8.5,
+      "rewards/margins": 6.8125,
+      "rewards/rejected": -15.3125,
       "step": 1840
     },
     {
       "epoch": 1.93717277486911,
-      "grad_norm": 3.366028477915385,
+      "grad_norm": 3.2782252173379,
       "learning_rate": 1.6806726966304064e-05,
-      "logits/chosen": -4.65625,
-      "logits/rejected": -2.671875,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -1912.0,
-      "loss": 0.102,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.15625,
+      "logits/chosen": -6.125,
+      "logits/rejected": -4.25,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -2112.0,
+      "loss": 0.1105,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -9.1875,
       "rewards/margins": 6.03125,
-      "rewards/rejected": -13.1875,
+      "rewards/rejected": -15.1875,
       "step": 1850
     },
     {
       "epoch": 1.9476439790575917,
-      "grad_norm": 1.5697122318440628,
+      "grad_norm": 1.6699418962967145,
       "learning_rate": 1.6519514091492625e-05,
-      "logits/chosen": -4.03125,
-      "logits/rejected": -2.578125,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -1864.0,
-      "loss": 0.0829,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -4.03125,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -2040.0,
+      "loss": 0.0809,
       "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.0,
-      "rewards/margins": 6.5,
-      "rewards/rejected": -13.5,
+      "rewards/chosen": -8.6875,
+      "rewards/margins": 6.53125,
+      "rewards/rejected": -15.25,
       "step": 1860
     },
     {
       "epoch": 1.9581151832460733,
-      "grad_norm": 2.1968085541085003,
+      "grad_norm": 1.3679248396395456,
       "learning_rate": 1.6233560574902627e-05,
-      "logits/chosen": -3.265625,
-      "logits/rejected": -1.640625,
-      "logps/chosen": -1064.0,
-      "logps/rejected": -1936.0,
-      "loss": 0.1028,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.09375,
+      "logits/chosen": -5.21875,
+      "logits/rejected": -3.453125,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -2080.0,
+      "loss": 0.0936,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -8.4375,
       "rewards/margins": 6.4375,
-      "rewards/rejected": -13.5,
+      "rewards/rejected": -14.875,
       "step": 1870
     },
     {
       "epoch": 1.9685863874345548,
-      "grad_norm": 1.238671179667089,
+      "grad_norm": 2.4854442522097564,
       "learning_rate": 1.5948908880835753e-05,
-      "logits/chosen": -3.453125,
-      "logits/rejected": -1.65625,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -1848.0,
-      "loss": 0.0764,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.0,
-      "rewards/margins": 6.21875,
-      "rewards/rejected": -13.25,
+      "logits/chosen": -5.625,
+      "logits/rejected": -3.3125,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -2048.0,
+      "loss": 0.081,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -8.8125,
+      "rewards/margins": 6.4375,
+      "rewards/rejected": -15.25,
       "step": 1880
     },
     {
       "epoch": 1.9790575916230366,
-      "grad_norm": 1.2524311881941284,
+      "grad_norm": 1.01399499181541,
       "learning_rate": 1.5665601280272124e-05,
-      "logits/chosen": -3.796875,
-      "logits/rejected": -2.4375,
-      "logps/chosen": -1120.0,
-      "logps/rejected": -1952.0,
-      "loss": 0.0894,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -7.375,
-      "rewards/margins": 6.09375,
-      "rewards/rejected": -13.5,
+      "logits/chosen": -6.15625,
+      "logits/rejected": -4.5,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -2064.0,
+      "loss": 0.0831,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -8.3125,
+      "rewards/margins": 6.28125,
+      "rewards/rejected": -14.5625,
       "step": 1890
     },
     {
       "epoch": 1.9895287958115184,
-      "grad_norm": 1.8642022626087018,
+      "grad_norm": 2.2420767620448676,
       "learning_rate": 1.538367984459298e-05,
-      "logits/chosen": -3.171875,
-      "logits/rejected": -1.53125,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1984.0,
-      "loss": 0.0543,
+      "logits/chosen": -5.71875,
+      "logits/rejected": -3.46875,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -2024.0,
+      "loss": 0.0649,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -7.28125,
-      "rewards/margins": 6.84375,
-      "rewards/rejected": -14.125,
+      "rewards/chosen": -7.90625,
+      "rewards/margins": 6.59375,
+      "rewards/rejected": -14.5,
       "step": 1900
     },
     {
       "epoch": 1.9895287958115184,
-      "eval_logits/chosen": -0.5390625,
-      "eval_logits/rejected": 1.1953125,
-      "eval_logps/chosen": -1384.0,
-      "eval_logps/rejected": -1512.0,
-      "eval_loss": 1.3391094207763672,
-      "eval_rewards/accuracies": 0.6328125,
-      "eval_rewards/chosen": -10.625,
-      "eval_rewards/margins": 1.625,
-      "eval_rewards/rejected": -12.25,
-      "eval_runtime": 46.6172,
-      "eval_samples_per_second": 42.903,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -3.046875,
+      "eval_logits/rejected": -1.03125,
+      "eval_logps/chosen": -1392.0,
+      "eval_logps/rejected": -1536.0,
+      "eval_loss": 1.314226508140564,
+      "eval_rewards/accuracies": 0.62109375,
+      "eval_rewards/chosen": -10.75,
+      "eval_rewards/margins": 1.6796875,
+      "eval_rewards/rejected": -12.4375,
+      "eval_runtime": 47.0816,
+      "eval_samples_per_second": 42.479,
+      "eval_steps_per_second": 0.68,
       "step": 1900
     },
     {
       "epoch": 2.0,
-      "grad_norm": 3.152387361974524,
+      "grad_norm": 2.039933962862527,
       "learning_rate": 1.5103186439333133e-05,
-      "logits/chosen": -2.53125,
-      "logits/rejected": -1.7109375,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -2160.0,
-      "loss": 0.1001,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.84375,
-      "rewards/margins": 7.96875,
-      "rewards/rejected": -15.8125,
+      "logits/chosen": -4.4375,
+      "logits/rejected": -3.3125,
+      "logps/chosen": -1256.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.1064,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -8.75,
+      "rewards/margins": 7.78125,
+      "rewards/rejected": -16.5,
       "step": 1910
     },
     {
       "epoch": 2.0104712041884816,
-      "grad_norm": 0.36179553406631054,
+      "grad_norm": 0.5977599378677944,
       "learning_rate": 1.4824162717963829e-05,
-      "logits/chosen": -3.234375,
-      "logits/rejected": -0.66796875,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -2080.0,
-      "loss": 0.0155,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -7.4375,
-      "rewards/margins": 8.0,
-      "rewards/rejected": -15.4375,
+      "logits/chosen": -5.1875,
+      "logits/rejected": -2.125,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -2192.0,
+      "loss": 0.0153,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -8.375,
+      "rewards/margins": 8.0625,
+      "rewards/rejected": -16.375,
       "step": 1920
     },
     {
       "epoch": 2.020942408376963,
-      "grad_norm": 0.12208535086886389,
+      "grad_norm": 0.36787439417115725,
       "learning_rate": 1.4546650115707192e-05,
-      "logits/chosen": -3.578125,
-      "logits/rejected": -1.9296875,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -2040.0,
-      "loss": 0.0133,
+      "logits/chosen": -5.53125,
+      "logits/rejected": -3.484375,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -2160.0,
+      "loss": 0.0127,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -7.625,
-      "rewards/margins": 7.96875,
-      "rewards/rejected": -15.625,
+      "rewards/chosen": -8.6875,
+      "rewards/margins": 8.0625,
+      "rewards/rejected": -16.75,
       "step": 1930
     },
     {
       "epoch": 2.031413612565445,
-      "grad_norm": 0.1708741890155528,
+      "grad_norm": 0.061516779366227535,
       "learning_rate": 1.4270689843383111e-05,
-      "logits/chosen": -1.5546875,
-      "logits/rejected": 0.16015625,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -2336.0,
-      "loss": 0.0164,
+      "logits/chosen": -3.765625,
+      "logits/rejected": -1.625,
+      "logps/chosen": -1384.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0133,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.5625,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -17.875,
+      "rewards/chosen": -9.8125,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -19.0,
       "step": 1940
     },
     {
       "epoch": 2.0418848167539267,
-      "grad_norm": 0.32979174746088114,
+      "grad_norm": 0.3655557372419591,
       "learning_rate": 1.3996322881289347e-05,
-      "logits/chosen": -2.578125,
-      "logits/rejected": -0.7421875,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -2352.0,
-      "loss": 0.0174,
+      "logits/chosen": -4.5,
+      "logits/rejected": -2.375,
+      "logps/chosen": -1376.0,
+      "logps/rejected": -2496.0,
+      "loss": 0.0212,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -18.0,
+      "rewards/chosen": -10.0625,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -19.375,
       "step": 1950
     },
     {
       "epoch": 2.0523560209424083,
-      "grad_norm": 0.10692400680838644,
+      "grad_norm": 0.3483109003084252,
       "learning_rate": 1.3723589973115963e-05,
-      "logits/chosen": -2.234375,
-      "logits/rejected": -0.482421875,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0095,
+      "logits/chosen": -4.125,
+      "logits/rejected": -2.09375,
+      "logps/chosen": -1296.0,
+      "logps/rejected": -2576.0,
+      "loss": 0.0116,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.5625,
-      "rewards/margins": 10.5625,
-      "rewards/rejected": -19.125,
+      "rewards/chosen": -9.3125,
+      "rewards/margins": 10.3125,
+      "rewards/rejected": -19.625,
       "step": 1960
     },
     {
       "epoch": 2.06282722513089,
-      "grad_norm": 0.059475266864551825,
+      "grad_norm": 0.2220265302563937,
       "learning_rate": 1.3452531619894837e-05,
-      "logits/chosen": -2.21875,
-      "logits/rejected": -0.55859375,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.011,
+      "logits/chosen": -3.859375,
+      "logits/rejected": -1.9765625,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0088,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.3125,
-      "rewards/margins": 10.375,
-      "rewards/rejected": -19.75,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 10.1875,
+      "rewards/rejected": -20.5,
       "step": 1970
     },
     {
       "epoch": 2.073298429319372,
-      "grad_norm": 0.8110130799488728,
+      "grad_norm": 0.2494338602476271,
       "learning_rate": 1.3183188073985245e-05,
-      "logits/chosen": -2.921875,
-      "logits/rejected": -1.1171875,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -2416.0,
-      "loss": 0.0152,
+      "logits/chosen": -4.125,
+      "logits/rejected": -2.3125,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0184,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 9.625,
-      "rewards/rejected": -18.625,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 9.4375,
+      "rewards/rejected": -19.875,
       "step": 1980
     },
     {
       "epoch": 2.0837696335078535,
-      "grad_norm": 0.21455635138761006,
+      "grad_norm": 0.1354327516775415,
       "learning_rate": 1.2915599333096351e-05,
-      "logits/chosen": -3.234375,
-      "logits/rejected": -1.5390625,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0085,
+      "logits/chosen": -4.5,
+      "logits/rejected": -2.65625,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.0084,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.8125,
-      "rewards/margins": 10.5,
-      "rewards/rejected": -20.375,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 10.3125,
+      "rewards/rejected": -21.375,
       "step": 1990
     },
     {
       "epoch": 2.094240837696335,
-      "grad_norm": 0.04917032024327522,
+      "grad_norm": 0.059306827743589294,
       "learning_rate": 1.264980513434752e-05,
-      "logits/chosen": -3.5,
-      "logits/rejected": -0.5234375,
-      "logps/chosen": -1384.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0139,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 9.875,
-      "rewards/rejected": -19.375,
+      "logits/chosen": -4.375,
+      "logits/rejected": -1.71875,
+      "logps/chosen": -1560.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.0155,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -21.0,
       "step": 2000
     },
     {
       "epoch": 2.094240837696335,
-      "eval_logits/chosen": -0.134765625,
-      "eval_logits/rejected": 1.65625,
-      "eval_logps/chosen": -1600.0,
-      "eval_logps/rejected": -1768.0,
-      "eval_loss": 1.6983593702316284,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -12.8125,
-      "eval_rewards/margins": 1.9921875,
-      "eval_rewards/rejected": -14.8125,
-      "eval_runtime": 46.6332,
-      "eval_samples_per_second": 42.888,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -1.1640625,
+      "eval_logits/rejected": 0.66015625,
+      "eval_logps/chosen": -1768.0,
+      "eval_logps/rejected": -1944.0,
+      "eval_loss": 1.7396718263626099,
+      "eval_rewards/accuracies": 0.630859375,
+      "eval_rewards/chosen": -14.4375,
+      "eval_rewards/margins": 2.078125,
+      "eval_rewards/rejected": -16.5,
+      "eval_runtime": 47.1041,
+      "eval_samples_per_second": 42.459,
+      "eval_steps_per_second": 0.679,
       "step": 2000
     },
     {
       "epoch": 2.1047120418848166,
-      "grad_norm": 0.15367757661322679,
+      "grad_norm": 0.13848395497179677,
       "learning_rate": 1.2385844948367324e-05,
-      "logits/chosen": -3.109375,
-      "logits/rejected": -1.2109375,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2400.0,
-      "loss": 0.0185,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -19.0,
+      "logits/chosen": -4.28125,
+      "logits/rejected": -2.359375,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0217,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.6875,
+      "rewards/margins": 9.75,
+      "rewards/rejected": -20.5,
       "step": 2010
     },
     {
       "epoch": 2.115183246073298,
-      "grad_norm": 0.29405073863564757,
+      "grad_norm": 0.22370134033723152,
       "learning_rate": 1.2123757973432114e-05,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -1.2578125,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -2400.0,
-      "loss": 0.0147,
+      "logits/chosen": -4.8125,
+      "logits/rejected": -2.734375,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0173,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -19.0,
+      "rewards/chosen": -10.625,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -19.625,
       "step": 2020
     },
     {
       "epoch": 2.1256544502617802,
-      "grad_norm": 0.28363610183538124,
+      "grad_norm": 0.1993368306799259,
       "learning_rate": 1.1863583129645106e-05,
-      "logits/chosen": -2.890625,
-      "logits/rejected": -1.609375,
-      "logps/chosen": -1328.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0169,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.375,
-      "rewards/margins": 10.875,
-      "rewards/rejected": -20.25,
+      "logits/chosen": -4.15625,
+      "logits/rejected": -2.9375,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0201,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -10.375,
+      "rewards/margins": 9.75,
+      "rewards/rejected": -20.125,
       "step": 2030
     },
     {
       "epoch": 2.136125654450262,
-      "grad_norm": 0.26104422167727537,
+      "grad_norm": 0.22485964142272405,
       "learning_rate": 1.1605359053156607e-05,
-      "logits/chosen": -3.609375,
-      "logits/rejected": -0.99609375,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0145,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.3125,
-      "rewards/margins": 10.0,
-      "rewards/rejected": -19.375,
+      "logits/chosen": -4.4375,
+      "logits/rejected": -1.96875,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0118,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 9.6875,
+      "rewards/rejected": -20.125,
       "step": 2040
     },
     {
       "epoch": 2.1465968586387434,
-      "grad_norm": 0.11786274213524123,
+      "grad_norm": 0.08852618802289816,
       "learning_rate": 1.1349124090426644e-05,
-      "logits/chosen": -2.5,
-      "logits/rejected": -0.361328125,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0155,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 10.4375,
-      "rewards/rejected": -19.375,
+      "logits/chosen": -3.515625,
+      "logits/rejected": -1.5234375,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.0134,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 10.5625,
+      "rewards/rejected": -20.875,
       "step": 2050
     },
     {
       "epoch": 2.157068062827225,
-      "grad_norm": 0.06610681893085839,
+      "grad_norm": 0.47479179910766867,
       "learning_rate": 1.1094916292530406e-05,
-      "logits/chosen": -2.21875,
-      "logits/rejected": -0.87890625,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -2400.0,
-      "loss": 0.0149,
+      "logits/chosen": -3.34375,
+      "logits/rejected": -1.9609375,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0154,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -8.5625,
-      "rewards/margins": 9.5625,
-      "rewards/rejected": -18.125,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 9.25,
+      "rewards/rejected": -19.5,
       "step": 2060
     },
     {
       "epoch": 2.167539267015707,
-      "grad_norm": 0.4208410650885062,
+      "grad_norm": 0.09578626557974271,
       "learning_rate": 1.0842773409507623e-05,
-      "logits/chosen": -2.265625,
-      "logits/rejected": -1.09375,
-      "logps/chosen": -1216.0,
-      "logps/rejected": -2432.0,
-      "loss": 0.0129,
+      "logits/chosen": -3.3125,
+      "logits/rejected": -2.09375,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.018,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 9.8125,
-      "rewards/rejected": -18.25,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 10.0,
+      "rewards/rejected": -20.5,
       "step": 2070
     },
     {
       "epoch": 2.1780104712041886,
-      "grad_norm": 0.21079815364900423,
+      "grad_norm": 0.23092446662860747,
       "learning_rate": 1.0592732884756754e-05,
-      "logits/chosen": -2.78125,
-      "logits/rejected": -1.3828125,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0109,
+      "logits/chosen": -4.3125,
+      "logits/rejected": -2.734375,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.0128,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -9.3125,
-      "rewards/margins": 10.6875,
-      "rewards/rejected": -20.0,
+      "rewards/chosen": -10.8125,
+      "rewards/margins": 10.8125,
+      "rewards/rejected": -21.625,
       "step": 2080
     },
     {
       "epoch": 2.18848167539267,
-      "grad_norm": 0.25869132184156307,
+      "grad_norm": 0.08950959724147078,
       "learning_rate": 1.0344831849474507e-05,
-      "logits/chosen": -2.25,
-      "logits/rejected": -0.640625,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0078,
+      "logits/chosen": -3.9375,
+      "logits/rejected": -2.0,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2768.0,
+      "loss": 0.01,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.75,
+      "rewards/chosen": -11.1875,
       "rewards/margins": 10.3125,
-      "rewards/rejected": -20.125,
+      "rewards/rejected": -21.5,
       "step": 2090
     },
     {
       "epoch": 2.1989528795811517,
-      "grad_norm": 0.1278014009680594,
+      "grad_norm": 0.14829506514767885,
       "learning_rate": 1.0099107117141879e-05,
-      "logits/chosen": -1.890625,
-      "logits/rejected": 0.03564453125,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0268,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.4375,
-      "rewards/margins": 10.3125,
-      "rewards/rejected": -19.75,
+      "logits/chosen": -3.40625,
+      "logits/rejected": -1.390625,
+      "logps/chosen": -1464.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.0091,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.875,
+      "rewards/margins": 10.125,
+      "rewards/rejected": -21.0,
       "step": 2100
     },
     {
       "epoch": 2.1989528795811517,
-      "eval_logits/chosen": 0.427734375,
-      "eval_logits/rejected": 2.09375,
-      "eval_logps/chosen": -1600.0,
-      "eval_logps/rejected": -1776.0,
-      "eval_loss": 1.6732343435287476,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -12.8125,
-      "eval_rewards/margins": 2.046875,
-      "eval_rewards/rejected": -14.875,
-      "eval_runtime": 46.6294,
-      "eval_samples_per_second": 42.891,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -1.3671875,
+      "eval_logits/rejected": 0.50390625,
+      "eval_logps/chosen": -1720.0,
+      "eval_logps/rejected": -1904.0,
+      "eval_loss": 1.7226641178131104,
+      "eval_rewards/accuracies": 0.625,
+      "eval_rewards/chosen": -14.0,
+      "eval_rewards/margins": 2.15625,
+      "eval_rewards/rejected": -16.125,
+      "eval_runtime": 47.0855,
+      "eval_samples_per_second": 42.476,
+      "eval_steps_per_second": 0.68,
       "step": 2100
     },
     {
       "epoch": 2.2094240837696333,
-      "grad_norm": 0.2677220605255813,
+      "grad_norm": 0.14880506586335354,
       "learning_rate": 9.855595178057334e-06,
-      "logits/chosen": -2.421875,
-      "logits/rejected": -0.83984375,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0095,
+      "logits/chosen": -3.84375,
+      "logits/rejected": -2.171875,
+      "logps/chosen": -1560.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0099,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.0,
-      "rewards/margins": 10.75,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -11.5,
+      "rewards/margins": 10.625,
+      "rewards/rejected": -22.125,
       "step": 2110
     },
     {
       "epoch": 2.2198952879581153,
-      "grad_norm": 0.11255096447913922,
+      "grad_norm": 0.11619690142811516,
       "learning_rate": 9.614332193917949e-06,
-      "logits/chosen": -1.921875,
-      "logits/rejected": -0.0137939453125,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0209,
+      "logits/chosen": -3.671875,
+      "logits/rejected": -1.453125,
+      "logps/chosen": -1568.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.0204,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 11.0,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -11.375,
+      "rewards/margins": 10.625,
+      "rewards/rejected": -22.0,
       "step": 2120
     },
     {
       "epoch": 2.230366492146597,
-      "grad_norm": 0.8018291593808143,
+      "grad_norm": 0.6474828677254899,
       "learning_rate": 9.375353992449385e-06,
-      "logits/chosen": -2.84375,
-      "logits/rejected": -1.671875,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0099,
+      "logits/chosen": -4.5,
+      "logits/rejected": -2.953125,
+      "logps/chosen": -1568.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.0077,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.0625,
-      "rewards/margins": 10.5625,
-      "rewards/rejected": -20.625,
+      "rewards/chosen": -11.75,
+      "rewards/margins": 11.0,
+      "rewards/rejected": -22.75,
       "step": 2130
     },
     {
       "epoch": 2.2408376963350785,
-      "grad_norm": 0.22908228744504044,
+      "grad_norm": 1.2532545067430394,
       "learning_rate": 9.138696062085442e-06,
-      "logits/chosen": -2.515625,
-      "logits/rejected": -1.09375,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0126,
+      "logits/chosen": -4.21875,
+      "logits/rejected": -2.171875,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.0113,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.1875,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -20.625,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -23.5,
       "step": 2140
     },
     {
       "epoch": 2.25130890052356,
-      "grad_norm": 0.0796469932739297,
+      "grad_norm": 0.17305712795534955,
       "learning_rate": 8.904393546698007e-06,
-      "logits/chosen": -2.78125,
-      "logits/rejected": -0.99609375,
-      "logps/chosen": -1368.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0123,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 10.875,
-      "rewards/rejected": -20.5,
+      "logits/chosen": -4.40625,
+      "logits/rejected": -1.9375,
+      "logps/chosen": -1576.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.0151,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -11.75,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -23.25,
       "step": 2150
     },
     {
       "epoch": 2.261780104712042,
-      "grad_norm": 1.5020595186353467,
+      "grad_norm": 0.19183191663432336,
       "learning_rate": 8.672481240378142e-06,
-      "logits/chosen": -1.6015625,
-      "logits/rejected": -0.40234375,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0192,
+      "logits/chosen": -3.4375,
+      "logits/rejected": -1.4296875,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.0199,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.5625,
-      "rewards/margins": 10.6875,
-      "rewards/rejected": -20.25,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 10.75,
+      "rewards/rejected": -21.875,
       "step": 2160
     },
     {
       "epoch": 2.2722513089005236,
-      "grad_norm": 0.181865615798319,
+      "grad_norm": 1.0618485036326681,
       "learning_rate": 8.44299358226919e-06,
-      "logits/chosen": -1.4765625,
-      "logits/rejected": -0.2470703125,
-      "logps/chosen": -1416.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.0168,
+      "logits/chosen": -3.78125,
+      "logits/rejected": -1.6953125,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.0174,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.1875,
-      "rewards/margins": 10.8125,
-      "rewards/rejected": -21.0,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 10.875,
+      "rewards/rejected": -22.125,
       "step": 2170
     },
     {
       "epoch": 2.282722513089005,
-      "grad_norm": 0.2701043696604385,
+      "grad_norm": 0.276997394354002,
       "learning_rate": 8.215964651452457e-06,
-      "logits/chosen": -1.1015625,
-      "logits/rejected": 0.53125,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.0094,
+      "logits/chosen": -2.984375,
+      "logits/rejected": -0.828125,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.0074,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.9375,
-      "rewards/margins": 11.1875,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -11.3125,
+      "rewards/margins": 10.9375,
+      "rewards/rejected": -22.25,
       "step": 2180
     },
     {
       "epoch": 2.2931937172774868,
-      "grad_norm": 0.13281907178711774,
+      "grad_norm": 0.23799460846539638,
       "learning_rate": 7.991428161886502e-06,
-      "logits/chosen": -2.109375,
-      "logits/rejected": -0.51171875,
-      "logps/chosen": -1384.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0078,
+      "logits/chosen": -3.9375,
+      "logits/rejected": -1.90625,
+      "logps/chosen": -1504.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0072,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.875,
-      "rewards/margins": 11.125,
-      "rewards/rejected": -21.0,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 10.25,
+      "rewards/rejected": -21.375,
       "step": 2190
     },
     {
       "epoch": 2.303664921465969,
-      "grad_norm": 0.10187667199373601,
+      "grad_norm": 0.18076747142854013,
       "learning_rate": 7.76941745740061e-06,
-      "logits/chosen": -1.7890625,
-      "logits/rejected": -0.69140625,
-      "logps/chosen": -1488.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0119,
+      "logits/chosen": -3.75,
+      "logits/rejected": -1.8671875,
+      "logps/chosen": -1624.0,
+      "logps/rejected": -2928.0,
+      "loss": 0.0101,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -10.625,
-      "rewards/margins": 11.25,
-      "rewards/rejected": -21.875,
+      "rewards/chosen": -11.9375,
+      "rewards/margins": 11.3125,
+      "rewards/rejected": -23.25,
       "step": 2200
     },
     {
       "epoch": 2.303664921465969,
-      "eval_logits/chosen": 1.1171875,
-      "eval_logits/rejected": 2.6875,
-      "eval_logps/chosen": -1712.0,
-      "eval_logps/rejected": -1904.0,
-      "eval_loss": 1.7791718244552612,
-      "eval_rewards/accuracies": 0.619140625,
-      "eval_rewards/chosen": -13.875,
-      "eval_rewards/margins": 2.265625,
-      "eval_rewards/rejected": -16.125,
-      "eval_runtime": 46.6202,
-      "eval_samples_per_second": 42.9,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -0.8359375,
+      "eval_logits/rejected": 0.984375,
+      "eval_logps/chosen": -1824.0,
+      "eval_logps/rejected": -2016.0,
+      "eval_loss": 1.8446015119552612,
+      "eval_rewards/accuracies": 0.640625,
+      "eval_rewards/chosen": -15.0625,
+      "eval_rewards/margins": 2.234375,
+      "eval_rewards/rejected": -17.25,
+      "eval_runtime": 47.0945,
+      "eval_samples_per_second": 42.468,
+      "eval_steps_per_second": 0.679,
       "step": 2200
     },
     {
       "epoch": 2.3141361256544504,
-      "grad_norm": 0.27596116717149305,
+      "grad_norm": 0.22755235241110316,
       "learning_rate": 7.549965506743123e-06,
-      "logits/chosen": -1.3046875,
-      "logits/rejected": -0.51953125,
-      "logps/chosen": -1456.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0089,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.75,
-      "rewards/margins": 10.6875,
-      "rewards/rejected": -21.375,
+      "logits/chosen": -3.546875,
+      "logits/rejected": -2.015625,
+      "logps/chosen": -1576.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.0102,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -11.9375,
+      "rewards/margins": 10.8125,
+      "rewards/rejected": -22.75,
       "step": 2210
     },
     {
       "epoch": 2.324607329842932,
-      "grad_norm": 0.07432946666753426,
+      "grad_norm": 0.04908119155051022,
       "learning_rate": 7.333104898685631e-06,
-      "logits/chosen": -1.3046875,
-      "logits/rejected": -0.408203125,
-      "logps/chosen": -1472.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0113,
+      "logits/chosen": -3.453125,
+      "logits/rejected": -1.921875,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0108,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.8125,
-      "rewards/margins": 10.8125,
-      "rewards/rejected": -21.625,
+      "rewards/chosen": -11.5,
+      "rewards/margins": 11.0,
+      "rewards/rejected": -22.5,
       "step": 2220
     },
     {
       "epoch": 2.3350785340314135,
-      "grad_norm": 0.09573332662972851,
+      "grad_norm": 0.08915612839685524,
       "learning_rate": 7.118867837183518e-06,
-      "logits/chosen": -2.34375,
-      "logits/rejected": -0.36328125,
-      "logps/chosen": -1560.0,
-      "logps/rejected": -2928.0,
-      "loss": 0.0087,
+      "logits/chosen": -4.625,
+      "logits/rejected": -1.875,
+      "logps/chosen": -1616.0,
+      "logps/rejected": -2976.0,
+      "loss": 0.0202,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.0,
-      "rewards/margins": 12.1875,
-      "rewards/rejected": -23.125,
+      "rewards/chosen": -11.5,
+      "rewards/margins": 12.125,
+      "rewards/rejected": -23.625,
       "step": 2230
     },
     {
       "epoch": 2.345549738219895,
-      "grad_norm": 0.05351664263350839,
+      "grad_norm": 0.12457909704050821,
       "learning_rate": 6.9072861365936045e-06,
-      "logits/chosen": -2.140625,
-      "logits/rejected": -1.015625,
-      "logps/chosen": -1648.0,
-      "logps/rejected": -3104.0,
-      "loss": 0.0072,
+      "logits/chosen": -4.4375,
+      "logits/rejected": -2.484375,
+      "logps/chosen": -1640.0,
+      "logps/rejected": -3024.0,
+      "loss": 0.0078,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.9375,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -24.25,
+      "rewards/chosen": -11.8125,
+      "rewards/margins": 11.6875,
+      "rewards/rejected": -23.5,
       "step": 2240
     },
     {
       "epoch": 2.356020942408377,
-      "grad_norm": 0.28640353272680796,
+      "grad_norm": 0.2286841613244401,
       "learning_rate": 6.6983912169497005e-06,
-      "logits/chosen": -1.6640625,
-      "logits/rejected": -0.059326171875,
-      "logps/chosen": -1544.0,
-      "logps/rejected": -2912.0,
-      "loss": 0.0151,
+      "logits/chosen": -3.71875,
+      "logits/rejected": -1.6328125,
+      "logps/chosen": -1560.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.0135,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 11.5,
-      "rewards/rejected": -22.625,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 10.9375,
+      "rewards/rejected": -22.125,
       "step": 2250
     },
     {
       "epoch": 2.3664921465968587,
-      "grad_norm": 0.11254126554289635,
+      "grad_norm": 0.042937808849063154,
       "learning_rate": 6.49221409929677e-06,
-      "logits/chosen": -0.78125,
-      "logits/rejected": 0.73046875,
-      "logps/chosen": -1432.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0118,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.5,
-      "rewards/margins": 11.375,
-      "rewards/rejected": -21.875,
+      "logits/chosen": -2.71875,
+      "logits/rejected": -0.54296875,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.0097,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 10.9375,
+      "rewards/rejected": -22.0,
       "step": 2260
     },
     {
       "epoch": 2.3769633507853403,
-      "grad_norm": 0.07749707545628833,
+      "grad_norm": 0.35863060407012703,
       "learning_rate": 6.2887854010842075e-06,
-      "logits/chosen": -0.91015625,
-      "logits/rejected": -0.28515625,
-      "logps/chosen": -1472.0,
-      "logps/rejected": -2816.0,
-      "loss": 0.0112,
+      "logits/chosen": -2.5,
+      "logits/rejected": -1.4609375,
+      "logps/chosen": -1616.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.0124,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -11.0,
-      "rewards/margins": 11.5625,
-      "rewards/rejected": -22.5,
+      "rewards/chosen": -12.4375,
+      "rewards/margins": 10.9375,
+      "rewards/rejected": -23.375,
       "step": 2270
     },
     {
       "epoch": 2.387434554973822,
-      "grad_norm": 0.18337239522093765,
+      "grad_norm": 1.6091554416518914,
       "learning_rate": 6.0881353316191385e-06,
-      "logits/chosen": -1.671875,
-      "logits/rejected": -0.796875,
-      "logps/chosen": -1472.0,
-      "logps/rejected": -2816.0,
-      "loss": 0.0086,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.5625,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -22.0,
+      "logits/chosen": -3.828125,
+      "logits/rejected": -2.265625,
+      "logps/chosen": -1576.0,
+      "logps/rejected": -2928.0,
+      "loss": 0.0125,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -11.625,
+      "rewards/margins": 11.625,
+      "rewards/rejected": -23.25,
       "step": 2280
     },
     {
       "epoch": 2.3979057591623034,
-      "grad_norm": 0.17683438834216056,
+      "grad_norm": 0.1263364201350342,
       "learning_rate": 5.890293687580381e-06,
-      "logits/chosen": -0.875,
-      "logits/rejected": 0.40234375,
-      "logps/chosen": -1528.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0115,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.1875,
-      "rewards/margins": 11.1875,
-      "rewards/rejected": -22.375,
+      "logits/chosen": -2.671875,
+      "logits/rejected": -0.83984375,
+      "logps/chosen": -1600.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.0122,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -11.875,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -23.25,
       "step": 2290
     },
     {
       "epoch": 2.4083769633507854,
-      "grad_norm": 0.15107603947868276,
+      "grad_norm": 0.6813179391799218,
       "learning_rate": 5.695289848593532e-06,
-      "logits/chosen": -2.34375,
-      "logits/rejected": -1.5625,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0139,
+      "logits/chosen": -4.0,
+      "logits/rejected": -2.71875,
+      "logps/chosen": -1480.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.0145,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -10.125,
-      "rewards/margins": 10.5625,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 10.625,
+      "rewards/rejected": -21.625,
       "step": 2300
     },
     {
       "epoch": 2.4083769633507854,
-      "eval_logits/chosen": 1.0625,
-      "eval_logits/rejected": 2.671875,
-      "eval_logps/chosen": -1672.0,
-      "eval_logps/rejected": -1864.0,
-      "eval_loss": 1.7628437280654907,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -13.5,
-      "eval_rewards/margins": 2.21875,
-      "eval_rewards/rejected": -15.75,
-      "eval_runtime": 46.6098,
-      "eval_samples_per_second": 42.909,
-      "eval_steps_per_second": 0.687,
+      "eval_logits/chosen": -0.66796875,
+      "eval_logits/rejected": 1.171875,
+      "eval_logps/chosen": -1768.0,
+      "eval_logps/rejected": -1968.0,
+      "eval_loss": 1.791085958480835,
+      "eval_rewards/accuracies": 0.62890625,
+      "eval_rewards/chosen": -14.5,
+      "eval_rewards/margins": 2.265625,
+      "eval_rewards/rejected": -16.75,
+      "eval_runtime": 47.0854,
+      "eval_samples_per_second": 42.476,
+      "eval_steps_per_second": 0.68,
       "step": 2300
     },
     {
       "epoch": 2.418848167539267,
-      "grad_norm": 0.13239508388772323,
+      "grad_norm": 0.3636623403280782,
       "learning_rate": 5.50315277286812e-06,
-      "logits/chosen": -1.8359375,
-      "logits/rejected": -0.5625,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.0134,
+      "logits/chosen": -3.8125,
+      "logits/rejected": -2.078125,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.0128,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.0625,
-      "rewards/margins": 11.5,
-      "rewards/rejected": -21.625,
+      "rewards/chosen": -11.125,
+      "rewards/margins": 11.125,
+      "rewards/rejected": -22.25,
       "step": 2310
     },
     {
       "epoch": 2.4293193717277486,
-      "grad_norm": 0.2719996458718832,
+      "grad_norm": 0.19419140933440066,
       "learning_rate": 5.313910992897281e-06,
-      "logits/chosen": -1.296875,
-      "logits/rejected": 0.0654296875,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.0087,
+      "logits/chosen": -3.296875,
+      "logits/rejected": -1.25,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.0079,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.5625,
-      "rewards/margins": 11.8125,
-      "rewards/rejected": -21.375,
+      "rewards/chosen": -10.8125,
+      "rewards/margins": 11.625,
+      "rewards/rejected": -22.375,
       "step": 2320
     },
     {
       "epoch": 2.4397905759162306,
-      "grad_norm": 0.261447212888783,
+      "grad_norm": 0.3216749125997076,
       "learning_rate": 5.127592611220658e-06,
-      "logits/chosen": -1.5546875,
-      "logits/rejected": -0.353515625,
-      "logps/chosen": -1464.0,
-      "logps/rejected": -2848.0,
-      "loss": 0.0175,
+      "logits/chosen": -3.515625,
+      "logits/rejected": -1.7734375,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.0205,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -10.5625,
-      "rewards/margins": 11.5625,
-      "rewards/rejected": -22.125,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 11.3125,
+      "rewards/rejected": -22.625,
       "step": 2330
     },
     {
       "epoch": 2.450261780104712,
-      "grad_norm": 0.5563041707090628,
+      "grad_norm": 0.20579957150346426,
       "learning_rate": 4.944225296251159e-06,
-      "logits/chosen": -1.59375,
-      "logits/rejected": -1.015625,
-      "logps/chosen": -1368.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0087,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.0625,
-      "rewards/margins": 11.5,
-      "rewards/rejected": -21.625,
+      "logits/chosen": -3.609375,
+      "logits/rejected": -2.421875,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.0106,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.875,
+      "rewards/margins": 11.0,
+      "rewards/rejected": -21.875,
       "step": 2340
     },
     {
       "epoch": 2.4607329842931938,
-      "grad_norm": 0.5071113225650288,
+      "grad_norm": 0.11492631505791115,
       "learning_rate": 4.763836278166172e-06,
-      "logits/chosen": -1.5859375,
-      "logits/rejected": 0.0079345703125,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0122,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.875,
-      "rewards/margins": 10.6875,
-      "rewards/rejected": -20.625,
+      "logits/chosen": -4.125,
+      "logits/rejected": -1.8359375,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0092,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.375,
+      "rewards/margins": 10.4375,
+      "rewards/rejected": -20.875,
       "step": 2350
     },
     {
       "epoch": 2.4712041884816753,
-      "grad_norm": 0.06716960269111422,
+      "grad_norm": 0.05241783911536356,
       "learning_rate": 4.586452344863876e-06,
-      "logits/chosen": -1.875,
-      "logits/rejected": -0.64453125,
-      "logps/chosen": -1440.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0111,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -10.3125,
-      "rewards/margins": 10.875,
-      "rewards/rejected": -21.25,
+      "logits/chosen": -3.90625,
+      "logits/rejected": -1.9921875,
+      "logps/chosen": -1512.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0124,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -11.0,
+      "rewards/margins": 11.0,
+      "rewards/rejected": -22.0,
       "step": 2360
     },
     {
       "epoch": 2.481675392670157,
-      "grad_norm": 0.07782024350468128,
+      "grad_norm": 0.03555405707770405,
       "learning_rate": 4.412099837985192e-06,
-      "logits/chosen": -1.46875,
-      "logits/rejected": 0.1142578125,
-      "logps/chosen": -1448.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.0116,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.4375,
-      "rewards/margins": 12.0,
-      "rewards/rejected": -22.375,
+      "logits/chosen": -3.46875,
+      "logits/rejected": -1.25,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2928.0,
+      "loss": 0.0117,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 11.9375,
+      "rewards/rejected": -23.0,
       "step": 2370
     },
     {
       "epoch": 2.492146596858639,
-      "grad_norm": 0.7681663409564508,
+      "grad_norm": 0.33592489701696837,
       "learning_rate": 4.240804649002089e-06,
-      "logits/chosen": -2.03125,
-      "logits/rejected": -0.1630859375,
-      "logps/chosen": -1352.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0142,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -9.9375,
-      "rewards/margins": 10.9375,
-      "rewards/rejected": -20.875,
+      "logits/chosen": -3.828125,
+      "logits/rejected": -1.5078125,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0184,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -10.75,
+      "rewards/margins": 10.6875,
+      "rewards/rejected": -21.375,
       "step": 2380
     },
     {
       "epoch": 2.5026178010471205,
-      "grad_norm": 0.6032809041735147,
+      "grad_norm": 1.128534879252504,
       "learning_rate": 4.072592215372598e-06,
-      "logits/chosen": -2.765625,
-      "logits/rejected": -1.0859375,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2800.0,
-      "loss": 0.0136,
+      "logits/chosen": -4.40625,
+      "logits/rejected": -2.34375,
+      "logps/chosen": -1560.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.0122,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.875,
-      "rewards/margins": 11.5625,
-      "rewards/rejected": -22.375,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 11.6875,
+      "rewards/rejected": -23.0,
       "step": 2390
     },
     {
       "epoch": 2.513089005235602,
-      "grad_norm": 0.47210989221013594,
+      "grad_norm": 0.415737848268598,
       "learning_rate": 3.907487516763389e-06,
-      "logits/chosen": -1.0390625,
-      "logits/rejected": -0.171875,
-      "logps/chosen": -1496.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.0106,
+      "logits/chosen": -3.109375,
+      "logits/rejected": -1.6796875,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.0114,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.0625,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -22.5,
+      "rewards/chosen": -11.3125,
+      "rewards/margins": 11.3125,
+      "rewards/rejected": -22.625,
       "step": 2400
     },
     {
       "epoch": 2.513089005235602,
-      "eval_logits/chosen": 1.265625,
-      "eval_logits/rejected": 2.84375,
-      "eval_logps/chosen": -1776.0,
-      "eval_logps/rejected": -1976.0,
-      "eval_loss": 1.8903906345367432,
-      "eval_rewards/accuracies": 0.603515625,
-      "eval_rewards/chosen": -14.5625,
-      "eval_rewards/margins": 2.328125,
-      "eval_rewards/rejected": -16.875,
-      "eval_runtime": 46.6325,
-      "eval_samples_per_second": 42.889,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -0.54296875,
+      "eval_logits/rejected": 1.234375,
+      "eval_logps/chosen": -1792.0,
+      "eval_logps/rejected": -1992.0,
+      "eval_loss": 1.7978359460830688,
+      "eval_rewards/accuracies": 0.626953125,
+      "eval_rewards/chosen": -14.75,
+      "eval_rewards/margins": 2.28125,
+      "eval_rewards/rejected": -17.0,
+      "eval_runtime": 47.0926,
+      "eval_samples_per_second": 42.469,
+      "eval_steps_per_second": 0.68,
       "step": 2400
     },
     {
       "epoch": 2.5235602094240837,
-      "grad_norm": 0.31620391243647683,
+      "grad_norm": 0.28650815039070604,
       "learning_rate": 3.745515071340286e-06,
-      "logits/chosen": -1.1796875,
-      "logits/rejected": 0.04833984375,
-      "logps/chosen": -1528.0,
-      "logps/rejected": -2944.0,
-      "loss": 0.0115,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 12.3125,
-      "rewards/rejected": -23.375,
+      "logits/chosen": -3.0,
+      "logits/rejected": -1.2734375,
+      "logps/chosen": -1584.0,
+      "logps/rejected": -2928.0,
+      "loss": 0.0125,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -11.6875,
+      "rewards/margins": 11.625,
+      "rewards/rejected": -23.25,
       "step": 2410
     },
     {
       "epoch": 2.5340314136125652,
-      "grad_norm": 0.7299929267611917,
+      "grad_norm": 0.050769491764024226,
       "learning_rate": 3.586698932127236e-06,
-      "logits/chosen": -2.21875,
-      "logits/rejected": -0.6171875,
-      "logps/chosen": -1464.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0139,
+      "logits/chosen": -4.0,
+      "logits/rejected": -1.8828125,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2768.0,
+      "loss": 0.0127,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.6875,
-      "rewards/margins": 11.625,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 10.9375,
       "rewards/rejected": -22.25,
       "step": 2420
     },
     {
       "epoch": 2.5445026178010473,
-      "grad_norm": 0.19208808324836454,
+      "grad_norm": 0.6324762557871049,
       "learning_rate": 3.4310626834344744e-06,
-      "logits/chosen": -2.234375,
-      "logits/rejected": -1.109375,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2880.0,
-      "loss": 0.0106,
+      "logits/chosen": -4.0625,
+      "logits/rejected": -2.640625,
+      "logps/chosen": -1608.0,
+      "logps/rejected": -2912.0,
+      "loss": 0.0119,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.0,
-      "rewards/margins": 11.625,
-      "rewards/rejected": -22.625,
+      "rewards/chosen": -11.875,
+      "rewards/margins": 11.1875,
+      "rewards/rejected": -23.0,
       "step": 2430
     },
     {
       "epoch": 2.554973821989529,
-      "grad_norm": 0.03654100891360164,
+      "grad_norm": 0.0572509741374605,
       "learning_rate": 3.2786294373562343e-06,
-      "logits/chosen": -1.4296875,
-      "logits/rejected": 0.1787109375,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2992.0,
-      "loss": 0.0066,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.125,
+      "logits/chosen": -3.359375,
+      "logits/rejected": -1.296875,
+      "logps/chosen": -1552.0,
+      "logps/rejected": -3008.0,
+      "loss": 0.0103,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -11.375,
       "rewards/margins": 12.3125,
-      "rewards/rejected": -23.375,
+      "rewards/rejected": -23.75,
       "step": 2440
     },
     {
       "epoch": 2.5654450261780104,
-      "grad_norm": 0.8723319894976825,
+      "grad_norm": 0.41395884185200127,
       "learning_rate": 3.1294218303385526e-06,
-      "logits/chosen": -1.9296875,
-      "logits/rejected": -0.78125,
-      "logps/chosen": -1560.0,
-      "logps/rejected": -3120.0,
-      "loss": 0.0112,
+      "logits/chosen": -3.875,
+      "logits/rejected": -2.234375,
+      "logps/chosen": -1624.0,
+      "logps/rejected": -3088.0,
+      "loss": 0.008,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.1875,
-      "rewards/margins": 12.75,
-      "rewards/rejected": -24.0,
+      "rewards/chosen": -11.875,
+      "rewards/margins": 11.875,
+      "rewards/rejected": -23.75,
       "step": 2450
     },
     {
       "epoch": 2.5759162303664924,
-      "grad_norm": 0.27318352974306526,
+      "grad_norm": 0.45416520058418547,
       "learning_rate": 2.9834620198177483e-06,
-      "logits/chosen": -1.6875,
-      "logits/rejected": -0.255859375,
-      "logps/chosen": -1552.0,
+      "logits/chosen": -3.515625,
+      "logits/rejected": -1.6640625,
+      "logps/chosen": -1584.0,
       "logps/rejected": -2864.0,
-      "loss": 0.0182,
+      "loss": 0.0141,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.4375,
-      "rewards/margins": 11.625,
+      "rewards/chosen": -11.75,
+      "rewards/margins": 11.3125,
       "rewards/rejected": -23.125,
       "step": 2460
     },
     {
       "epoch": 2.5863874345549736,
-      "grad_norm": 0.033093854612854724,
+      "grad_norm": 0.06555993289453634,
       "learning_rate": 2.840771680930068e-06,
-      "logits/chosen": -1.703125,
-      "logits/rejected": -0.51171875,
-      "logps/chosen": -1632.0,
-      "logps/rejected": -3184.0,
-      "loss": 0.0087,
+      "logits/chosen": -3.6875,
+      "logits/rejected": -2.015625,
+      "logps/chosen": -1656.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0108,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.0,
-      "rewards/margins": 12.875,
-      "rewards/rejected": -24.875,
+      "rewards/chosen": -12.3125,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -25.0,
       "step": 2470
     },
     {
       "epoch": 2.5968586387434556,
-      "grad_norm": 0.2906893301107401,
+      "grad_norm": 0.14730237647348368,
       "learning_rate": 2.701372003292868e-06,
-      "logits/chosen": -0.53515625,
-      "logits/rejected": 0.6640625,
-      "logps/chosen": -1432.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0147,
+      "logits/chosen": -2.546875,
+      "logits/rejected": -0.80859375,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.013,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.5625,
-      "rewards/margins": 11.5,
-      "rewards/rejected": -22.0,
+      "rewards/chosen": -11.375,
+      "rewards/margins": 10.9375,
+      "rewards/rejected": -22.25,
       "step": 2480
     },
     {
       "epoch": 2.607329842931937,
-      "grad_norm": 0.30978344608724984,
+      "grad_norm": 0.19443148530885196,
       "learning_rate": 2.56528368785795e-06,
-      "logits/chosen": -0.64453125,
-      "logits/rejected": 0.6640625,
-      "logps/chosen": -1456.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0153,
+      "logits/chosen": -2.78125,
+      "logits/rejected": -1.03125,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.0154,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.4375,
-      "rewards/margins": 11.0625,
-      "rewards/rejected": -21.5,
+      "rewards/chosen": -11.125,
+      "rewards/margins": 10.5625,
+      "rewards/rejected": -21.625,
       "step": 2490
     },
     {
       "epoch": 2.6178010471204187,
-      "grad_norm": 3.1669374993550483,
+      "grad_norm": 0.10815148657918729,
       "learning_rate": 2.432526943837471e-06,
-      "logits/chosen": -0.96875,
-      "logits/rejected": 0.443359375,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -2816.0,
-      "loss": 0.0133,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.3125,
-      "rewards/margins": 11.75,
-      "rewards/rejected": -22.125,
+      "logits/chosen": -3.046875,
+      "logits/rejected": -0.859375,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.012,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -11.125,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -22.625,
       "step": 2500
     },
     {
       "epoch": 2.6178010471204187,
-      "eval_logits/chosen": 1.34375,
-      "eval_logits/rejected": 2.90625,
-      "eval_logps/chosen": -1728.0,
-      "eval_logps/rejected": -1920.0,
-      "eval_loss": 1.7944843769073486,
-      "eval_rewards/accuracies": 0.6171875,
-      "eval_rewards/chosen": -14.125,
-      "eval_rewards/margins": 2.21875,
-      "eval_rewards/rejected": -16.375,
-      "eval_runtime": 46.6216,
-      "eval_samples_per_second": 42.899,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -0.5234375,
+      "eval_logits/rejected": 1.265625,
+      "eval_logps/chosen": -1784.0,
+      "eval_logps/rejected": -1984.0,
+      "eval_loss": 1.7939844131469727,
+      "eval_rewards/accuracies": 0.630859375,
+      "eval_rewards/chosen": -14.6875,
+      "eval_rewards/margins": 2.28125,
+      "eval_rewards/rejected": -17.0,
+      "eval_runtime": 47.0982,
+      "eval_samples_per_second": 42.464,
+      "eval_steps_per_second": 0.679,
       "step": 2500
     },
     {
       "epoch": 2.6282722513089007,
-      "grad_norm": 0.25114900523762806,
+      "grad_norm": 2.0052364520965056,
       "learning_rate": 2.3031214857028446e-06,
-      "logits/chosen": -1.0703125,
-      "logits/rejected": -0.10400390625,
-      "logps/chosen": -1464.0,
-      "logps/rejected": -2832.0,
-      "loss": 0.01,
+      "logits/chosen": -3.078125,
+      "logits/rejected": -1.609375,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2912.0,
+      "loss": 0.0118,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.4375,
-      "rewards/margins": 11.9375,
-      "rewards/rejected": -22.375,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 12.0,
+      "rewards/rejected": -23.125,
       "step": 2510
     },
     {
       "epoch": 2.6387434554973823,
-      "grad_norm": 0.18518869248809075,
+      "grad_norm": 0.09847657007482878,
       "learning_rate": 2.177086530257139e-06,
-      "logits/chosen": -1.6171875,
-      "logits/rejected": -0.255859375,
-      "logps/chosen": -1488.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0122,
+      "logits/chosen": -3.578125,
+      "logits/rejected": -1.671875,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.0146,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.0,
-      "rewards/margins": 10.5625,
-      "rewards/rejected": -21.5,
+      "rewards/chosen": -11.4375,
+      "rewards/margins": 10.9375,
+      "rewards/rejected": -22.375,
       "step": 2520
     },
     {
       "epoch": 2.649214659685864,
-      "grad_norm": 0.3677767520124152,
+      "grad_norm": 0.16869525008885042,
       "learning_rate": 2.054440793781354e-06,
-      "logits/chosen": -1.8046875,
-      "logits/rejected": -0.0262451171875,
-      "logps/chosen": -1448.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.0141,
+      "logits/chosen": -3.609375,
+      "logits/rejected": -1.265625,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0124,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -10.4375,
+      "rewards/chosen": -11.375,
       "rewards/margins": 11.125,
-      "rewards/rejected": -21.625,
+      "rewards/rejected": -22.5,
       "step": 2530
     },
     {
       "epoch": 2.6596858638743455,
-      "grad_norm": 0.08071873284564589,
+      "grad_norm": 0.10456975971332139,
       "learning_rate": 1.9352024892550587e-06,
-      "logits/chosen": -1.421875,
-      "logits/rejected": -0.049072265625,
-      "logps/chosen": -1424.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.011,
+      "logits/chosen": -3.046875,
+      "logits/rejected": -1.1953125,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.012,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.25,
-      "rewards/margins": 10.9375,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -11.3125,
+      "rewards/margins": 10.8125,
+      "rewards/rejected": -22.125,
       "step": 2540
     },
     {
       "epoch": 2.670157068062827,
-      "grad_norm": 0.22694537948423107,
+      "grad_norm": 0.23987183672477652,
       "learning_rate": 1.8193893236517379e-06,
-      "logits/chosen": -2.03125,
-      "logits/rejected": -0.50390625,
-      "logps/chosen": -1472.0,
-      "logps/rejected": -2800.0,
-      "loss": 0.013,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.5,
+      "logits/chosen": -3.703125,
+      "logits/rejected": -1.734375,
+      "logps/chosen": -1592.0,
+      "logps/rejected": -2912.0,
+      "loss": 0.0177,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -11.6875,
       "rewards/margins": 11.25,
-      "rewards/rejected": -21.75,
+      "rewards/rejected": -23.0,
       "step": 2550
     },
     {
       "epoch": 2.680628272251309,
-      "grad_norm": 0.17563079209313892,
+      "grad_norm": 0.28280240621718694,
       "learning_rate": 1.7070184953092956e-06,
-      "logits/chosen": -1.8671875,
-      "logits/rejected": 0.1845703125,
-      "logps/chosen": -1368.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0105,
+      "logits/chosen": -3.6875,
+      "logits/rejected": -0.94921875,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.0117,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.625,
-      "rewards/margins": 11.375,
-      "rewards/rejected": -21.0,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 11.25,
+      "rewards/rejected": -21.75,
       "step": 2560
     },
     {
       "epoch": 2.6910994764397906,
-      "grad_norm": 0.2427374378493851,
+      "grad_norm": 0.17868788890856102,
       "learning_rate": 1.5981066913760918e-06,
-      "logits/chosen": -0.7890625,
-      "logits/rejected": 0.78125,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.006,
+      "logits/chosen": -2.59375,
+      "logits/rejected": -0.53125,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.0076,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 10.6875,
-      "rewards/rejected": -20.5,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 10.25,
+      "rewards/rejected": -20.875,
       "step": 2570
     },
     {
       "epoch": 2.701570680628272,
-      "grad_norm": 0.2822726814219828,
+      "grad_norm": 0.15309507992376212,
       "learning_rate": 1.492670085332884e-06,
-      "logits/chosen": -2.375,
-      "logits/rejected": -0.80859375,
-      "logps/chosen": -1440.0,
-      "logps/rejected": -2880.0,
-      "loss": 0.0102,
+      "logits/chosen": -3.921875,
+      "logits/rejected": -1.8671875,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.0093,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.4375,
-      "rewards/margins": 11.8125,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -11.4375,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -23.0,
       "step": 2580
     },
     {
       "epoch": 2.712041884816754,
-      "grad_norm": 0.031176925309414637,
+      "grad_norm": 0.11211247878950366,
       "learning_rate": 1.3907243345910787e-06,
-      "logits/chosen": -1.6171875,
-      "logits/rejected": -0.1767578125,
-      "logps/chosen": -1496.0,
-      "logps/rejected": -2800.0,
-      "loss": 0.0085,
+      "logits/chosen": -3.640625,
+      "logits/rejected": -1.375,
+      "logps/chosen": -1592.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.0134,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.625,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -22.0,
+      "rewards/chosen": -11.5625,
+      "rewards/margins": 11.1875,
+      "rewards/rejected": -22.75,
       "step": 2590
     },
     {
       "epoch": 2.7225130890052354,
-      "grad_norm": 0.06744671485238259,
+      "grad_norm": 0.0878471738085003,
       "learning_rate": 1.292284578167538e-06,
-      "logits/chosen": -1.4921875,
-      "logits/rejected": -0.2431640625,
-      "logps/chosen": -1448.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0065,
+      "logits/chosen": -3.140625,
+      "logits/rejected": -1.3671875,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.0075,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.5625,
-      "rewards/margins": 11.25,
-      "rewards/rejected": -21.75,
+      "rewards/chosen": -11.375,
+      "rewards/margins": 11.1875,
+      "rewards/rejected": -22.625,
       "step": 2600
     },
     {
       "epoch": 2.7225130890052354,
-      "eval_logits/chosen": 1.2578125,
-      "eval_logits/rejected": 2.8125,
-      "eval_logps/chosen": -1728.0,
-      "eval_logps/rejected": -1928.0,
-      "eval_loss": 1.7784218788146973,
-      "eval_rewards/accuracies": 0.615234375,
-      "eval_rewards/chosen": -14.125,
-      "eval_rewards/margins": 2.234375,
-      "eval_rewards/rejected": -16.375,
-      "eval_runtime": 46.6277,
-      "eval_samples_per_second": 42.893,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -0.40625,
+      "eval_logits/rejected": 1.359375,
+      "eval_logps/chosen": -1792.0,
+      "eval_logps/rejected": -1992.0,
+      "eval_loss": 1.8008984327316284,
+      "eval_rewards/accuracies": 0.630859375,
+      "eval_rewards/chosen": -14.75,
+      "eval_rewards/margins": 2.265625,
+      "eval_rewards/rejected": -17.0,
+      "eval_runtime": 47.0873,
+      "eval_samples_per_second": 42.474,
+      "eval_steps_per_second": 0.68,
       "step": 2600
     },
     {
       "epoch": 2.7329842931937174,
-      "grad_norm": 0.16676870803629498,
+      "grad_norm": 0.07370534954517018,
       "learning_rate": 1.1973654344364925e-06,
-      "logits/chosen": -1.5546875,
-      "logits/rejected": -0.6953125,
-      "logps/chosen": -1448.0,
-      "logps/rejected": -2800.0,
-      "loss": 0.0081,
+      "logits/chosen": -3.171875,
+      "logits/rejected": -1.953125,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.0086,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.3125,
-      "rewards/margins": 11.3125,
-      "rewards/rejected": -21.625,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -22.5,
       "step": 2610
     },
     {
       "epoch": 2.743455497382199,
-      "grad_norm": 0.12928274926533845,
+      "grad_norm": 0.3365025689074477,
       "learning_rate": 1.1059809989586583e-06,
-      "logits/chosen": -1.25,
-      "logits/rejected": 0.1982421875,
-      "logps/chosen": -1464.0,
-      "logps/rejected": -2928.0,
-      "loss": 0.0073,
+      "logits/chosen": -3.046875,
+      "logits/rejected": -0.94921875,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -3008.0,
+      "loss": 0.0093,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.625,
-      "rewards/margins": 12.0625,
-      "rewards/rejected": -22.625,
+      "rewards/chosen": -11.375,
+      "rewards/margins": 12.125,
+      "rewards/rejected": -23.5,
       "step": 2620
     },
     {
       "epoch": 2.7539267015706805,
-      "grad_norm": 0.11458867598959759,
+      "grad_norm": 0.08829671354669028,
       "learning_rate": 1.018144842388044e-06,
-      "logits/chosen": -0.671875,
-      "logits/rejected": 0.90234375,
-      "logps/chosen": -1440.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.0078,
+      "logits/chosen": -2.46875,
+      "logits/rejected": -0.255859375,
+      "logps/chosen": -1528.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.0105,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.125,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -21.5,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -22.375,
       "step": 2630
     },
     {
       "epoch": 2.7643979057591626,
-      "grad_norm": 0.16453060460125196,
+      "grad_norm": 0.1732226483399812,
       "learning_rate": 9.338700084567109e-07,
-      "logits/chosen": -1.578125,
-      "logits/rejected": -0.36328125,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -2816.0,
-      "loss": 0.0093,
+      "logits/chosen": -3.109375,
+      "logits/rejected": -1.4765625,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.0078,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.1875,
-      "rewards/margins": 12.0625,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 11.6875,
+      "rewards/rejected": -22.75,
       "step": 2640
     },
     {
       "epoch": 2.774869109947644,
-      "grad_norm": 0.4688204673145244,
+      "grad_norm": 0.21408267013956153,
       "learning_rate": 8.531690120377606e-07,
-      "logits/chosen": -1.5078125,
-      "logits/rejected": -0.341796875,
-      "logps/chosen": -1592.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.0077,
+      "logits/chosen": -2.96875,
+      "logits/rejected": -1.453125,
+      "logps/chosen": -1728.0,
+      "logps/rejected": -3024.0,
+      "loss": 0.0122,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.5,
+      "rewards/chosen": -12.875,
       "rewards/margins": 11.375,
-      "rewards/rejected": -22.875,
+      "rewards/rejected": -24.25,
       "step": 2650
     },
     {
       "epoch": 2.7853403141361257,
-      "grad_norm": 0.226890892587328,
+      "grad_norm": 0.12461493186298428,
       "learning_rate": 7.760538372868636e-07,
-      "logits/chosen": -1.6953125,
-      "logits/rejected": -0.34375,
-      "logps/chosen": -1432.0,
-      "logps/rejected": -2720.0,
+      "logits/chosen": -3.3125,
+      "logits/rejected": -1.4453125,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2832.0,
       "loss": 0.0137,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.3125,
-      "rewards/margins": 11.375,
-      "rewards/rejected": -21.625,
+      "rewards/chosen": -11.375,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -22.875,
       "step": 2660
     },
     {
       "epoch": 2.7958115183246073,
-      "grad_norm": 0.0332686297158602,
+      "grad_norm": 0.08539447462187891,
       "learning_rate": 7.025359358626166e-07,
-      "logits/chosen": -1.703125,
-      "logits/rejected": 0.333984375,
-      "logps/chosen": -1528.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0086,
+      "logits/chosen": -3.34375,
+      "logits/rejected": -0.84765625,
+      "logps/chosen": -1648.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.0076,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.8125,
-      "rewards/margins": 11.375,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -12.0,
+      "rewards/margins": 11.75,
+      "rewards/rejected": -23.75,
       "step": 2670
     },
     {
       "epoch": 2.806282722513089,
-      "grad_norm": 0.2974033538076973,
+      "grad_norm": 0.03546327245127398,
       "learning_rate": 6.326262252259463e-07,
-      "logits/chosen": -1.8046875,
-      "logits/rejected": 0.169921875,
-      "logps/chosen": -1488.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.0108,
+      "logits/chosen": -3.15625,
+      "logits/rejected": -0.90234375,
+      "logps/chosen": -1592.0,
+      "logps/rejected": -3008.0,
+      "loss": 0.0098,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.625,
+      "rewards/chosen": -11.6875,
       "rewards/margins": 12.125,
-      "rewards/rejected": -22.75,
+      "rewards/rejected": -23.875,
       "step": 2680
     },
     {
       "epoch": 2.816753926701571,
-      "grad_norm": 0.26037676087946277,
+      "grad_norm": 0.22716375235408476,
       "learning_rate": 5.66335087018871e-07,
-      "logits/chosen": -1.6796875,
-      "logits/rejected": -0.52734375,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2944.0,
-      "loss": 0.0058,
+      "logits/chosen": -3.25,
+      "logits/rejected": -1.6484375,
+      "logps/chosen": -1584.0,
+      "logps/rejected": -3024.0,
+      "loss": 0.0076,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.9375,
-      "rewards/margins": 11.625,
-      "rewards/rejected": -22.5,
+      "rewards/chosen": -11.625,
+      "rewards/margins": 11.6875,
+      "rewards/rejected": -23.25,
       "step": 2690
     },
     {
       "epoch": 2.8272251308900525,
-      "grad_norm": 0.731496490192096,
+      "grad_norm": 0.5299507901285828,
       "learning_rate": 5.036723655228226e-07,
-      "logits/chosen": -1.140625,
-      "logits/rejected": -0.1796875,
-      "logps/chosen": -1432.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0126,
+      "logits/chosen": -2.765625,
+      "logits/rejected": -1.3046875,
+      "logps/chosen": -1512.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.0142,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.625,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -22.0,
+      "rewards/chosen": -11.4375,
+      "rewards/margins": 11.625,
+      "rewards/rejected": -23.0,
       "step": 2700
     },
     {
       "epoch": 2.8272251308900525,
-      "eval_logits/chosen": 1.2734375,
-      "eval_logits/rejected": 2.828125,
-      "eval_logps/chosen": -1752.0,
-      "eval_logps/rejected": -1944.0,
-      "eval_loss": 1.7994999885559082,
-      "eval_rewards/accuracies": 0.609375,
-      "eval_rewards/chosen": -14.3125,
-      "eval_rewards/margins": 2.25,
-      "eval_rewards/rejected": -16.625,
-      "eval_runtime": 46.6289,
-      "eval_samples_per_second": 42.892,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -0.302734375,
+      "eval_logits/rejected": 1.4453125,
+      "eval_logps/chosen": -1816.0,
+      "eval_logps/rejected": -2016.0,
+      "eval_loss": 1.8148125410079956,
+      "eval_rewards/accuracies": 0.630859375,
+      "eval_rewards/chosen": -14.9375,
+      "eval_rewards/margins": 2.296875,
+      "eval_rewards/rejected": -17.25,
+      "eval_runtime": 47.0841,
+      "eval_samples_per_second": 42.477,
+      "eval_steps_per_second": 0.68,
       "step": 2700
     },
     {
       "epoch": 2.837696335078534,
-      "grad_norm": 0.7198056574728556,
+      "grad_norm": 1.7591547555964455,
       "learning_rate": 4.446473661967432e-07,
-      "logits/chosen": -1.9921875,
-      "logits/rejected": -0.59375,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.0097,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 11.0625,
-      "rewards/rejected": -22.125,
+      "logits/chosen": -3.625,
+      "logits/rejected": -1.8671875,
+      "logps/chosen": -1592.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0123,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -11.9375,
+      "rewards/margins": 10.6875,
+      "rewards/rejected": -22.625,
       "step": 2710
     },
     {
       "epoch": 2.8481675392670156,
-      "grad_norm": 0.4753855724493522,
+      "grad_norm": 1.0528192304485011,
       "learning_rate": 3.892688542952411e-07,
-      "logits/chosen": -0.69140625,
-      "logits/rejected": 0.625,
-      "logps/chosen": -1408.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.0094,
+      "logits/chosen": -2.5,
+      "logits/rejected": -0.67578125,
+      "logps/chosen": -1464.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0128,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.25,
-      "rewards/margins": 11.625,
-      "rewards/rejected": -21.875,
+      "rewards/chosen": -10.8125,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -22.25,
       "step": 2720
     },
     {
       "epoch": 2.858638743455497,
-      "grad_norm": 0.2906802488180948,
+      "grad_norm": 0.21803424026895357,
       "learning_rate": 3.3754505356694e-07,
-      "logits/chosen": -2.3125,
-      "logits/rejected": -0.71875,
-      "logps/chosen": -1544.0,
+      "logits/chosen": -3.984375,
+      "logits/rejected": -1.90625,
+      "logps/chosen": -1608.0,
       "logps/rejected": -2800.0,
-      "loss": 0.019,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.25,
-      "rewards/margins": 10.9375,
-      "rewards/rejected": -22.125,
+      "loss": 0.0136,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -11.875,
+      "rewards/margins": 10.4375,
+      "rewards/rejected": -22.25,
       "step": 2730
     },
     {
       "epoch": 2.869109947643979,
-      "grad_norm": 0.265663073141491,
+      "grad_norm": 0.1258121861832878,
       "learning_rate": 2.8948364503322276e-07,
-      "logits/chosen": -1.1484375,
-      "logits/rejected": 0.38671875,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0122,
+      "logits/chosen": -2.75,
+      "logits/rejected": -0.8359375,
+      "logps/chosen": -1504.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.0123,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.3125,
-      "rewards/margins": 11.4375,
-      "rewards/rejected": -21.75,
+      "rewards/chosen": -11.3125,
+      "rewards/margins": 11.125,
+      "rewards/rejected": -22.375,
       "step": 2740
     },
     {
       "epoch": 2.8795811518324608,
-      "grad_norm": 0.19920546039714782,
+      "grad_norm": 0.5767318390725228,
       "learning_rate": 2.4509176584762706e-07,
-      "logits/chosen": -1.390625,
-      "logits/rejected": 0.0634765625,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0107,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.125,
-      "rewards/margins": 10.875,
-      "rewards/rejected": -21.0,
+      "logits/chosen": -2.984375,
+      "logits/rejected": -0.98046875,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.0097,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -11.0,
+      "rewards/margins": 10.625,
+      "rewards/rejected": -21.5,
       "step": 2750
     },
     {
       "epoch": 2.8900523560209423,
-      "grad_norm": 0.5103962747661847,
+      "grad_norm": 0.09425982935238154,
       "learning_rate": 2.0437600823595694e-07,
-      "logits/chosen": -1.6640625,
-      "logits/rejected": -0.2099609375,
-      "logps/chosen": -1496.0,
-      "logps/rejected": -2784.0,
-      "loss": 0.0113,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.5625,
-      "rewards/margins": 11.125,
-      "rewards/rejected": -21.625,
+      "logits/chosen": -3.484375,
+      "logits/rejected": -1.40625,
+      "logps/chosen": -1592.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.0105,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -11.5,
+      "rewards/margins": 11.0,
+      "rewards/rejected": -22.5,
       "step": 2760
     },
     {
       "epoch": 2.900523560209424,
-      "grad_norm": 0.3945394222875742,
+      "grad_norm": 0.314688452433757,
       "learning_rate": 1.6734241851733e-07,
-      "logits/chosen": -1.8671875,
-      "logits/rejected": -0.462890625,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.0088,
+      "logits/chosen": -3.390625,
+      "logits/rejected": -1.53125,
+      "logps/chosen": -1672.0,
+      "logps/rejected": -2992.0,
+      "loss": 0.0092,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.375,
-      "rewards/margins": 11.125,
-      "rewards/rejected": -22.5,
+      "rewards/chosen": -12.375,
+      "rewards/margins": 11.4375,
+      "rewards/rejected": -23.75,
       "step": 2770
     },
     {
       "epoch": 2.9109947643979055,
-      "grad_norm": 0.21459014215233083,
+      "grad_norm": 1.533065818931154,
       "learning_rate": 1.339964962062984e-07,
-      "logits/chosen": -1.7578125,
-      "logits/rejected": -0.83984375,
-      "logps/chosen": -1440.0,
-      "logps/rejected": -2736.0,
-      "loss": 0.0051,
+      "logits/chosen": -3.28125,
+      "logits/rejected": -2.015625,
+      "logps/chosen": -1512.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0067,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.625,
+      "rewards/chosen": -11.4375,
       "rewards/margins": 11.1875,
-      "rewards/rejected": -21.875,
+      "rewards/rejected": -22.625,
       "step": 2780
     },
     {
       "epoch": 2.9214659685863875,
-      "grad_norm": 0.26252937123483133,
+      "grad_norm": 0.5216332775391516,
       "learning_rate": 1.0434319319617136e-07,
-      "logits/chosen": -1.671875,
-      "logits/rejected": 0.1298828125,
-      "logps/chosen": -1456.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0176,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.625,
+      "logits/chosen": -3.234375,
+      "logits/rejected": -0.93359375,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.01,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -11.4375,
       "rewards/margins": 10.75,
-      "rewards/rejected": -21.375,
+      "rewards/rejected": -22.125,
       "step": 2790
     },
     {
       "epoch": 2.931937172774869,
-      "grad_norm": 0.6790062750047958,
+      "grad_norm": 0.341009303562548,
       "learning_rate": 7.838691302365363e-08,
-      "logits/chosen": -1.171875,
-      "logits/rejected": 0.150390625,
-      "logps/chosen": -1464.0,
-      "logps/rejected": -2880.0,
-      "loss": 0.0068,
+      "logits/chosen": -2.8125,
+      "logits/rejected": -1.0078125,
+      "logps/chosen": -1552.0,
+      "logps/rejected": -2912.0,
+      "loss": 0.0102,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.625,
-      "rewards/margins": 11.9375,
-      "rewards/rejected": -22.5,
+      "rewards/chosen": -11.4375,
+      "rewards/margins": 11.5,
+      "rewards/rejected": -22.875,
       "step": 2800
     },
     {
       "epoch": 2.931937172774869,
-      "eval_logits/chosen": 1.2578125,
-      "eval_logits/rejected": 2.828125,
-      "eval_logps/chosen": -1752.0,
-      "eval_logps/rejected": -1952.0,
-      "eval_loss": 1.8052500486373901,
-      "eval_rewards/accuracies": 0.609375,
-      "eval_rewards/chosen": -14.375,
-      "eval_rewards/margins": 2.265625,
-      "eval_rewards/rejected": -16.625,
-      "eval_runtime": 46.6314,
-      "eval_samples_per_second": 42.89,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -0.3125,
+      "eval_logits/rejected": 1.4375,
+      "eval_logps/chosen": -1816.0,
+      "eval_logps/rejected": -2016.0,
+      "eval_loss": 1.8180078268051147,
+      "eval_rewards/accuracies": 0.626953125,
+      "eval_rewards/chosen": -14.9375,
+      "eval_rewards/margins": 2.3125,
+      "eval_rewards/rejected": -17.25,
+      "eval_runtime": 47.0939,
+      "eval_samples_per_second": 42.468,
+      "eval_steps_per_second": 0.679,
       "step": 2800
     },
     {
       "epoch": 2.9424083769633507,
-      "grad_norm": 0.31425102267712873,
+      "grad_norm": 0.1707304319835883,
       "learning_rate": 5.613151021490459e-08,
-      "logits/chosen": -1.015625,
-      "logits/rejected": -0.296875,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -2800.0,
-      "loss": 0.0054,
+      "logits/chosen": -2.640625,
+      "logits/rejected": -1.5234375,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.0058,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.625,
-      "rewards/margins": 11.6875,
-      "rewards/rejected": -22.25,
+      "rewards/chosen": -11.3125,
+      "rewards/margins": 11.75,
+      "rewards/rejected": -23.0,
       "step": 2810
     },
     {
       "epoch": 2.9528795811518327,
-      "grad_norm": 0.09801027577800256,
+      "grad_norm": 0.5595562848919824,
       "learning_rate": 3.7580289713159965e-08,
-      "logits/chosen": -1.3359375,
-      "logits/rejected": 0.46484375,
-      "logps/chosen": -1496.0,
-      "logps/rejected": -2864.0,
-      "loss": 0.0074,
+      "logits/chosen": -2.921875,
+      "logits/rejected": -0.72265625,
+      "logps/chosen": -1600.0,
+      "logps/rejected": -2976.0,
+      "loss": 0.0094,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.8125,
+      "rewards/chosen": -11.8125,
       "rewards/margins": 11.875,
-      "rewards/rejected": -22.625,
+      "rewards/rejected": -23.75,
       "step": 2820
     },
     {
       "epoch": 2.9633507853403143,
-      "grad_norm": 0.3159087299773979,
+      "grad_norm": 0.34613763016441956,
       "learning_rate": 2.2736006387927277e-08,
-      "logits/chosen": -1.7734375,
-      "logits/rejected": 0.1474609375,
-      "logps/chosen": -1472.0,
-      "logps/rejected": -2688.0,
-      "loss": 0.0086,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.5625,
-      "rewards/margins": 11.0,
-      "rewards/rejected": -21.625,
+      "logits/chosen": -3.234375,
+      "logits/rejected": -0.93359375,
+      "logps/chosen": -1576.0,
+      "logps/rejected": -2768.0,
+      "loss": 0.012,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -11.625,
+      "rewards/margins": 10.75,
+      "rewards/rejected": -22.375,
       "step": 2830
     },
     {
       "epoch": 2.973821989528796,
-      "grad_norm": 0.1583739041052338,
+      "grad_norm": 1.0870109941132398,
       "learning_rate": 1.1600864625893604e-08,
-      "logits/chosen": -1.734375,
-      "logits/rejected": -0.490234375,
-      "logps/chosen": -1448.0,
-      "logps/rejected": -2896.0,
-      "loss": 0.0167,
+      "logits/chosen": -3.15625,
+      "logits/rejected": -1.6796875,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2976.0,
+      "loss": 0.0113,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.4375,
-      "rewards/margins": 12.0625,
-      "rewards/rejected": -22.5,
+      "rewards/chosen": -11.375,
+      "rewards/margins": 12.0,
+      "rewards/rejected": -23.375,
       "step": 2840
     },
     {
       "epoch": 2.9842931937172774,
-      "grad_norm": 0.08503265376484782,
+      "grad_norm": 0.017941993915583238,
       "learning_rate": 4.176518003573549e-09,
-      "logits/chosen": -2.046875,
-      "logits/rejected": -0.3671875,
-      "logps/chosen": -1456.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.0207,
+      "logits/chosen": -3.703125,
+      "logits/rejected": -1.53125,
+      "logps/chosen": -1560.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.016,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.625,
-      "rewards/margins": 11.5,
-      "rewards/rejected": -22.125,
+      "rewards/chosen": -11.625,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -23.0,
       "step": 2850
     },
     {
       "epoch": 2.994764397905759,
-      "grad_norm": 0.6178102711626517,
+      "grad_norm": 0.11366211393856238,
       "learning_rate": 4.640690417528903e-10,
-      "logits/chosen": -1.4296875,
-      "logits/rejected": -0.1806640625,
-      "logps/chosen": -1416.0,
-      "logps/rejected": -2672.0,
-      "loss": 0.014,
+      "logits/chosen": -2.984375,
+      "logits/rejected": -1.484375,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.0163,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -10.375,
-      "rewards/margins": 11.375,
-      "rewards/rejected": -21.75,
+      "rewards/chosen": -11.4375,
+      "rewards/margins": 10.8125,
+      "rewards/rejected": -22.25,
       "step": 2860
     },
     {
       "epoch": 3.0,
       "step": 2865,
       "total_flos": 0.0,
-      "train_loss": 0.17591854184865952,
-      "train_runtime": 12314.0377,
-      "train_samples_per_second": 14.89,
-      "train_steps_per_second": 0.233
+      "train_loss": 0.17683555641411486,
+      "train_runtime": 12375.7398,
+      "train_samples_per_second": 14.816,
+      "train_steps_per_second": 0.232
     }
   ],
   "logging_steps": 10,