diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -10,12 +10,12 @@
   "log_history": [
     {
       "epoch": 0.0010471204188481676,
-      "grad_norm": 0.005552303844419711,
+      "grad_norm": 5.38308067689226,
       "learning_rate": 1.7421602787456448e-07,
-      "logits/chosen": -11.0,
-      "logits/rejected": -11.0,
-      "logps/chosen": -504.0,
-      "logps/rejected": -504.0,
+      "logits/chosen": -9.75,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -760.0,
+      "logps/rejected": -1992.0,
       "loss": 0.6914,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
@@ -25,4750 +25,4750 @@
     },
     {
       "epoch": 0.010471204188481676,
-      "grad_norm": 0.005930047993848834,
+      "grad_norm": 5.124685883317083,
       "learning_rate": 1.7421602787456445e-06,
-      "logits/chosen": -11.4375,
-      "logits/rejected": -11.4375,
-      "logps/chosen": -560.0,
-      "logps/rejected": -560.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.00250244140625,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.00250244140625,
+      "logits/chosen": -10.8125,
+      "logits/rejected": -8.0,
+      "logps/chosen": -510.0,
+      "logps/rejected": -2080.0,
+      "loss": 0.6905,
+      "rewards/accuracies": 0.0902777761220932,
+      "rewards/chosen": -0.0002346038818359375,
+      "rewards/margins": 0.0002498626708984375,
+      "rewards/rejected": -0.000484466552734375,
       "step": 10
     },
     {
       "epoch": 0.020942408376963352,
-      "grad_norm": 0.005844555689138784,
+      "grad_norm": 5.266345890881455,
       "learning_rate": 3.484320557491289e-06,
-      "logits/chosen": -11.5,
-      "logits/rejected": -11.5,
-      "logps/chosen": -548.0,
-      "logps/rejected": -548.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -4.76837158203125e-06,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -4.76837158203125e-06,
+      "logits/chosen": -10.9375,
+      "logits/rejected": -8.375,
+      "logps/chosen": -486.0,
+      "logps/rejected": -2128.0,
+      "loss": 0.6303,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -0.0067138671875,
+      "rewards/margins": 0.134765625,
+      "rewards/rejected": -0.1416015625,
       "step": 20
     },
     {
       "epoch": 0.031413612565445025,
-      "grad_norm": 0.006235490280701447,
+      "grad_norm": 1.7049065566770096,
       "learning_rate": 5.226480836236934e-06,
-      "logits/chosen": -11.125,
-      "logits/rejected": -11.125,
-      "logps/chosen": -512.0,
-      "logps/rejected": -512.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.0023193359375,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.0023193359375,
+      "logits/chosen": -10.875,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -448.0,
+      "logps/rejected": -2272.0,
+      "loss": 0.3423,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -0.1044921875,
+      "rewards/margins": 1.1640625,
+      "rewards/rejected": -1.265625,
       "step": 30
     },
     {
       "epoch": 0.041884816753926704,
-      "grad_norm": 0.005288278317797941,
+      "grad_norm": 0.5243567019599547,
       "learning_rate": 6.968641114982578e-06,
-      "logits/chosen": -11.0,
-      "logits/rejected": -11.0,
-      "logps/chosen": -532.0,
-      "logps/rejected": -532.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.00811767578125,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.00811767578125,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -576.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.2282,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -0.68359375,
+      "rewards/margins": 3.765625,
+      "rewards/rejected": -4.4375,
       "step": 40
     },
     {
       "epoch": 0.05235602094240838,
-      "grad_norm": 0.0056171109816268985,
+      "grad_norm": 1.8665347307549,
       "learning_rate": 8.710801393728225e-06,
-      "logits/chosen": -10.75,
-      "logits/rejected": -10.75,
-      "logps/chosen": -528.0,
-      "logps/rejected": -528.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.0194091796875,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.0194091796875,
+      "logits/chosen": -12.0,
+      "logits/rejected": -9.875,
+      "logps/chosen": -604.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.2274,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -0.7734375,
+      "rewards/margins": 3.78125,
+      "rewards/rejected": -4.53125,
       "step": 50
     },
     {
       "epoch": 0.06282722513089005,
-      "grad_norm": 0.005804252338228117,
+      "grad_norm": 1.6133377477083573,
       "learning_rate": 1.0452961672473868e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -11.25,
-      "logps/chosen": -504.0,
-      "logps/rejected": -504.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.03955078125,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.03955078125,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -516.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.2086,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -0.5625,
+      "rewards/margins": 3.734375,
+      "rewards/rejected": -4.3125,
       "step": 60
     },
     {
       "epoch": 0.07329842931937172,
-      "grad_norm": 0.006938350621212424,
+      "grad_norm": 2.758300742442637,
       "learning_rate": 1.2195121951219513e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -11.25,
-      "logps/chosen": -524.0,
-      "logps/rejected": -524.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.07275390625,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.07275390625,
+      "logits/chosen": -14.4375,
+      "logits/rejected": -11.0625,
+      "logps/chosen": -532.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.2306,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -0.65234375,
+      "rewards/margins": 3.21875,
+      "rewards/rejected": -3.875,
       "step": 70
     },
     {
       "epoch": 0.08376963350785341,
-      "grad_norm": 0.006083240503299625,
+      "grad_norm": 1.4261185227440032,
       "learning_rate": 1.3937282229965156e-05,
-      "logits/chosen": -10.8125,
-      "logits/rejected": -10.8125,
-      "logps/chosen": -504.0,
-      "logps/rejected": -504.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.12353515625,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.12353515625,
+      "logits/chosen": -14.375,
+      "logits/rejected": -10.75,
+      "logps/chosen": -516.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.2193,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -0.69921875,
+      "rewards/margins": 3.03125,
+      "rewards/rejected": -3.71875,
       "step": 80
     },
     {
       "epoch": 0.09424083769633508,
-      "grad_norm": 0.0058841724445259196,
+      "grad_norm": 1.5121621931310467,
       "learning_rate": 1.56794425087108e-05,
-      "logits/chosen": -10.6875,
-      "logits/rejected": -10.6875,
-      "logps/chosen": -524.0,
-      "logps/rejected": -524.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.21875,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.21875,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -10.875,
+      "logps/chosen": -490.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.1808,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -0.54296875,
+      "rewards/margins": 3.515625,
+      "rewards/rejected": -4.0625,
       "step": 90
     },
     {
       "epoch": 0.10471204188481675,
-      "grad_norm": 0.0055217692214385725,
+      "grad_norm": 1.4750070586077515,
       "learning_rate": 1.742160278745645e-05,
-      "logits/chosen": -10.3125,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -516.0,
-      "logps/rejected": -516.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.291015625,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.291015625,
+      "logits/chosen": -14.25,
+      "logits/rejected": -11.0,
+      "logps/chosen": -500.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.1864,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -0.68359375,
+      "rewards/margins": 3.984375,
+      "rewards/rejected": -4.6875,
       "step": 100
     },
     {
       "epoch": 0.10471204188481675,
-      "eval_logits/chosen": -9.8125,
-      "eval_logits/rejected": -9.4375,
-      "eval_logps/chosen": -352.0,
-      "eval_logps/rejected": -322.0,
-      "eval_loss": 0.6990625262260437,
-      "eval_rewards/accuracies": 0.419921875,
-      "eval_rewards/chosen": -0.345703125,
-      "eval_rewards/margins": -0.005279541015625,
-      "eval_rewards/rejected": -0.33984375,
-      "eval_runtime": 48.861,
-      "eval_samples_per_second": 40.932,
-      "eval_steps_per_second": 0.655,
+      "eval_logits/chosen": -14.6875,
+      "eval_logits/rejected": -14.5,
+      "eval_logps/chosen": -364.0,
+      "eval_logps/rejected": -344.0,
+      "eval_loss": 0.676437497138977,
+      "eval_rewards/accuracies": 0.541015625,
+      "eval_rewards/chosen": -0.447265625,
+      "eval_rewards/margins": 0.1083984375,
+      "eval_rewards/rejected": -0.5546875,
+      "eval_runtime": 48.6011,
+      "eval_samples_per_second": 41.151,
+      "eval_steps_per_second": 0.658,
       "step": 100
     },
     {
       "epoch": 0.11518324607329843,
-      "grad_norm": 0.007178924410337923,
+      "grad_norm": 1.428472636565015,
       "learning_rate": 1.9163763066202093e-05,
-      "logits/chosen": -10.0,
-      "logits/rejected": -10.0,
-      "logps/chosen": -540.0,
-      "logps/rejected": -540.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.39453125,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.39453125,
+      "logits/chosen": -14.0625,
+      "logits/rejected": -11.25,
+      "logps/chosen": -576.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.2104,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -0.8046875,
+      "rewards/margins": 3.0625,
+      "rewards/rejected": -3.875,
       "step": 110
     },
     {
       "epoch": 0.1256544502617801,
-      "grad_norm": 0.0065036855285739315,
+      "grad_norm": 1.136327247749575,
       "learning_rate": 2.0905923344947736e-05,
-      "logits/chosen": -9.75,
-      "logits/rejected": -9.75,
-      "logps/chosen": -584.0,
-      "logps/rejected": -584.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.5078125,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.5078125,
+      "logits/chosen": -12.125,
+      "logits/rejected": -10.125,
+      "logps/chosen": -556.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.1613,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -0.87109375,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -5.40625,
       "step": 120
     },
     {
       "epoch": 0.13612565445026178,
-      "grad_norm": 0.0069986005540183275,
+      "grad_norm": 1.2324544779713202,
       "learning_rate": 2.264808362369338e-05,
-      "logits/chosen": -9.375,
-      "logits/rejected": -9.375,
-      "logps/chosen": -600.0,
-      "logps/rejected": -600.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.69921875,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.69921875,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -512.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.1672,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -0.66796875,
+      "rewards/margins": 4.3125,
+      "rewards/rejected": -4.96875,
       "step": 130
     },
     {
       "epoch": 0.14659685863874344,
-      "grad_norm": 0.007168645297758455,
+      "grad_norm": 1.5984891624397473,
       "learning_rate": 2.4390243902439026e-05,
-      "logits/chosen": -8.5,
-      "logits/rejected": -8.5,
-      "logps/chosen": -540.0,
-      "logps/rejected": -540.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.96484375,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.96484375,
+      "logits/chosen": -11.6875,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -456.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.2005,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -0.5546875,
+      "rewards/margins": 3.78125,
+      "rewards/rejected": -4.34375,
       "step": 140
     },
     {
       "epoch": 0.15706806282722513,
-      "grad_norm": 0.01210750745976283,
+      "grad_norm": 1.2606827932294156,
       "learning_rate": 2.6132404181184672e-05,
-      "logits/chosen": -8.125,
-      "logits/rejected": -8.125,
-      "logps/chosen": -648.0,
-      "logps/rejected": -648.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -1.609375,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -1.609375,
+      "logits/chosen": -11.625,
+      "logits/rejected": -9.125,
+      "logps/chosen": -488.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.1755,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -0.64453125,
+      "rewards/margins": 3.75,
+      "rewards/rejected": -4.40625,
       "step": 150
     },
     {
       "epoch": 0.16753926701570682,
-      "grad_norm": 0.03117049640489613,
+      "grad_norm": 3.093909888170432,
       "learning_rate": 2.7874564459930312e-05,
-      "logits/chosen": -7.0625,
-      "logits/rejected": -7.0625,
-      "logps/chosen": -772.0,
-      "logps/rejected": -772.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -3.015625,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -3.015625,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -10.125,
+      "logps/chosen": -510.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.1924,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.0078125,
+      "rewards/margins": 4.09375,
+      "rewards/rejected": -5.09375,
       "step": 160
     },
     {
       "epoch": 0.17801047120418848,
-      "grad_norm": 0.04003634324569403,
+      "grad_norm": 1.1896686461642088,
       "learning_rate": 2.9616724738675962e-05,
-      "logits/chosen": -7.25,
-      "logits/rejected": -7.25,
-      "logps/chosen": -1080.0,
-      "logps/rejected": -1080.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -5.1875,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -5.1875,
+      "logits/chosen": -13.375,
+      "logits/rejected": -11.0,
+      "logps/chosen": -620.0,
+      "logps/rejected": -2768.0,
+      "loss": 0.1624,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -1.2890625,
+      "rewards/margins": 4.28125,
+      "rewards/rejected": -5.5625,
       "step": 170
     },
     {
       "epoch": 0.18848167539267016,
-      "grad_norm": 0.11743752644574089,
+      "grad_norm": 1.786162172320875,
       "learning_rate": 3.13588850174216e-05,
-      "logits/chosen": -6.5,
-      "logits/rejected": -6.5,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -1272.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -7.6875,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -7.6875,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -540.0,
+      "logps/rejected": -2768.0,
+      "loss": 0.0998,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.0,
+      "rewards/margins": 4.6875,
+      "rewards/rejected": -5.6875,
       "step": 180
     },
     {
       "epoch": 0.19895287958115182,
-      "grad_norm": 0.177519537179685,
+      "grad_norm": 4.125706219478258,
       "learning_rate": 3.310104529616725e-05,
-      "logits/chosen": -5.625,
-      "logits/rejected": -5.625,
-      "logps/chosen": -1632.0,
-      "logps/rejected": -1632.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -11.5,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -11.5,
+      "logits/chosen": -13.0,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -572.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.2826,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -1.296875,
+      "rewards/margins": 5.53125,
+      "rewards/rejected": -6.8125,
       "step": 190
     },
     {
       "epoch": 0.2094240837696335,
-      "grad_norm": 0.7182035026855809,
+      "grad_norm": 1.5835108506389313,
       "learning_rate": 3.48432055749129e-05,
-      "logits/chosen": 2.015625,
-      "logits/rejected": 2.015625,
-      "logps/chosen": -5056.0,
-      "logps/rejected": -5056.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -46.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -46.0,
+      "logits/chosen": -12.75,
+      "logits/rejected": -10.5,
+      "logps/chosen": -494.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.1156,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.83984375,
+      "rewards/margins": 4.5625,
+      "rewards/rejected": -5.40625,
       "step": 200
     },
     {
       "epoch": 0.2094240837696335,
-      "eval_logits/chosen": 4.53125,
-      "eval_logits/rejected": 5.25,
-      "eval_logps/chosen": -6176.0,
-      "eval_logps/rejected": -5408.0,
-      "eval_loss": 17.017187118530273,
-      "eval_rewards/accuracies": 0.423828125,
-      "eval_rewards/chosen": -58.5,
-      "eval_rewards/margins": -7.3125,
-      "eval_rewards/rejected": -51.25,
-      "eval_runtime": 47.1587,
-      "eval_samples_per_second": 42.41,
-      "eval_steps_per_second": 0.679,
+      "eval_logits/chosen": -11.6875,
+      "eval_logits/rejected": -11.4375,
+      "eval_logps/chosen": -398.0,
+      "eval_logps/rejected": -378.0,
+      "eval_loss": 0.6999062299728394,
+      "eval_rewards/accuracies": 0.5234375,
+      "eval_rewards/chosen": -0.80078125,
+      "eval_rewards/margins": 0.09912109375,
+      "eval_rewards/rejected": -0.8984375,
+      "eval_runtime": 46.6479,
+      "eval_samples_per_second": 42.874,
+      "eval_steps_per_second": 0.686,
       "step": 200
     },
     {
       "epoch": 0.2198952879581152,
-      "grad_norm": 0.3156643022440185,
+      "grad_norm": 0.987453416665501,
       "learning_rate": 3.6585365853658535e-05,
-      "logits/chosen": 2.5625,
-      "logits/rejected": 2.5625,
-      "logps/chosen": -16640.0,
-      "logps/rejected": -16640.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -162.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -162.0,
+      "logits/chosen": -11.75,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -524.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.1272,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -1.0234375,
+      "rewards/margins": 5.125,
+      "rewards/rejected": -6.125,
       "step": 210
     },
     {
       "epoch": 0.23036649214659685,
-      "grad_norm": 0.1011855202587046,
+      "grad_norm": 1.0889613168882084,
       "learning_rate": 3.8327526132404185e-05,
-      "logits/chosen": 1.484375,
-      "logits/rejected": 1.484375,
-      "logps/chosen": -30848.0,
-      "logps/rejected": -30848.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -302.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -302.0,
+      "logits/chosen": -12.625,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -532.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.1713,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -0.96875,
+      "rewards/margins": 4.40625,
+      "rewards/rejected": -5.375,
       "step": 220
     },
     {
       "epoch": 0.24083769633507854,
-      "grad_norm": 0.0426157286291694,
+      "grad_norm": 1.6920828920147442,
       "learning_rate": 4.006968641114983e-05,
-      "logits/chosen": 0.7890625,
-      "logits/rejected": 0.7890625,
-      "logps/chosen": -33024.0,
-      "logps/rejected": -33024.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -328.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -328.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -532.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.1088,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.125,
+      "rewards/margins": 5.3125,
+      "rewards/rejected": -6.4375,
       "step": 230
     },
     {
       "epoch": 0.2513089005235602,
-      "grad_norm": 0.02715793589621172,
+      "grad_norm": 0.7898867619854981,
       "learning_rate": 4.181184668989547e-05,
-      "logits/chosen": 0.6875,
-      "logits/rejected": 0.6875,
-      "logps/chosen": -38144.0,
-      "logps/rejected": -38144.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -374.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -374.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -604.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.1776,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.2109375,
+      "rewards/margins": 4.25,
+      "rewards/rejected": -5.46875,
       "step": 240
     },
     {
       "epoch": 0.2617801047120419,
-      "grad_norm": 0.020708774395151855,
+      "grad_norm": 2.123153422041998,
       "learning_rate": 4.3554006968641115e-05,
-      "logits/chosen": 0.34765625,
-      "logits/rejected": 0.34765625,
-      "logps/chosen": -43520.0,
-      "logps/rejected": -43520.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -430.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -430.0,
+      "logits/chosen": -14.125,
+      "logits/rejected": -11.25,
+      "logps/chosen": -696.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.192,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -1.3671875,
+      "rewards/margins": 4.125,
+      "rewards/rejected": -5.5,
       "step": 250
     },
     {
       "epoch": 0.27225130890052357,
-      "grad_norm": 0.014829834370556913,
+      "grad_norm": 1.1472354666715698,
       "learning_rate": 4.529616724738676e-05,
-      "logits/chosen": -0.10693359375,
-      "logits/rejected": -0.10693359375,
-      "logps/chosen": -40192.0,
-      "logps/rejected": -40192.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -396.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -396.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.9375,
+      "logps/chosen": -592.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.1599,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -1.2578125,
+      "rewards/margins": 4.15625,
+      "rewards/rejected": -5.4375,
       "step": 260
     },
     {
       "epoch": 0.28272251308900526,
-      "grad_norm": 0.013335406469528324,
+      "grad_norm": 1.5959333305446395,
       "learning_rate": 4.703832752613241e-05,
-      "logits/chosen": -0.40234375,
-      "logits/rejected": -0.40234375,
-      "logps/chosen": -37888.0,
-      "logps/rejected": -37888.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -374.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -374.0,
+      "logits/chosen": -12.1875,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -560.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.1573,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -1.203125,
+      "rewards/margins": 5.8125,
+      "rewards/rejected": -7.03125,
       "step": 270
     },
     {
       "epoch": 0.2931937172774869,
-      "grad_norm": 0.01326371168621352,
+      "grad_norm": 1.185815344909836,
       "learning_rate": 4.878048780487805e-05,
-      "logits/chosen": -0.578125,
-      "logits/rejected": -0.578125,
-      "logps/chosen": -41984.0,
-      "logps/rejected": -41984.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -414.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -414.0,
+      "logits/chosen": -12.8125,
+      "logits/rejected": -10.25,
+      "logps/chosen": -568.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.2706,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -1.0546875,
+      "rewards/margins": 3.71875,
+      "rewards/rejected": -4.78125,
       "step": 280
     },
     {
       "epoch": 0.3036649214659686,
-      "grad_norm": 0.012534728526770804,
+      "grad_norm": 1.0364545571263137,
       "learning_rate": 4.999983293481418e-05,
-      "logits/chosen": -0.71484375,
-      "logits/rejected": -0.71484375,
-      "logps/chosen": -45056.0,
-      "logps/rejected": -45056.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -446.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -446.0,
+      "logits/chosen": -12.25,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -544.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.2109,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -0.8828125,
+      "rewards/margins": 4.03125,
+      "rewards/rejected": -4.90625,
       "step": 290
     },
     {
       "epoch": 0.31413612565445026,
-      "grad_norm": 0.011334699751743478,
+      "grad_norm": 0.703025557252931,
       "learning_rate": 4.9996862949181595e-05,
-      "logits/chosen": -0.9296875,
-      "logits/rejected": -0.9296875,
-      "logps/chosen": -38400.0,
-      "logps/rejected": -38400.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -380.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -380.0,
+      "logits/chosen": -11.6875,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -556.0,
+      "logps/rejected": -2928.0,
+      "loss": 0.1465,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.2890625,
+      "rewards/margins": 6.09375,
+      "rewards/rejected": -7.375,
       "step": 300
     },
     {
       "epoch": 0.31413612565445026,
-      "eval_logits/chosen": -0.953125,
-      "eval_logits/rejected": -0.9296875,
-      "eval_logps/chosen": -33280.0,
-      "eval_logps/rejected": -28800.0,
-      "eval_loss": 97.35624694824219,
-      "eval_rewards/accuracies": 0.431640625,
-      "eval_rewards/chosen": -330.0,
-      "eval_rewards/margins": -44.75,
-      "eval_rewards/rejected": -286.0,
-      "eval_runtime": 47.1426,
-      "eval_samples_per_second": 42.425,
-      "eval_steps_per_second": 0.679,
+      "eval_logits/chosen": -11.375,
+      "eval_logits/rejected": -10.9375,
+      "eval_logps/chosen": -424.0,
+      "eval_logps/rejected": -406.0,
+      "eval_loss": 0.7544375061988831,
+      "eval_rewards/accuracies": 0.48828125,
+      "eval_rewards/chosen": -1.0546875,
+      "eval_rewards/margins": 0.10791015625,
+      "eval_rewards/rejected": -1.1640625,
+      "eval_runtime": 46.6412,
+      "eval_samples_per_second": 42.881,
+      "eval_steps_per_second": 0.686,
       "step": 300
     },
     {
       "epoch": 0.32460732984293195,
-      "grad_norm": 0.01021976851310903,
+      "grad_norm": 1.3095123966398965,
       "learning_rate": 4.9990180911526896e-05,
-      "logits/chosen": -1.0390625,
-      "logits/rejected": -1.0390625,
-      "logps/chosen": -41728.0,
-      "logps/rejected": -41728.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -412.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -412.0,
+      "logits/chosen": -11.5,
+      "logits/rejected": -9.5,
+      "logps/chosen": -580.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.1313,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -1.3984375,
+      "rewards/margins": 5.875,
+      "rewards/rejected": -7.28125,
       "step": 310
     },
     {
       "epoch": 0.33507853403141363,
-      "grad_norm": 0.00982943677250041,
+      "grad_norm": 2.142337171370044,
       "learning_rate": 4.997978781413745e-05,
-      "logits/chosen": -1.03125,
-      "logits/rejected": -1.03125,
-      "logps/chosen": -40704.0,
-      "logps/rejected": -40704.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -402.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -402.0,
+      "logits/chosen": -12.5,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -664.0,
+      "logps/rejected": -2976.0,
+      "loss": 0.1775,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.171875,
+      "rewards/margins": 6.0625,
+      "rewards/rejected": -8.25,
       "step": 320
     },
     {
       "epoch": 0.34554973821989526,
-      "grad_norm": 0.009719158444227481,
+      "grad_norm": 1.1180308740605138,
       "learning_rate": 4.996568520039565e-05,
-      "logits/chosen": -1.0078125,
-      "logits/rejected": -1.0078125,
-      "logps/chosen": -44544.0,
-      "logps/rejected": -44544.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -440.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -440.0,
+      "logits/chosen": -12.0,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -664.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.1794,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -1.625,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -6.28125,
       "step": 330
     },
     {
       "epoch": 0.35602094240837695,
-      "grad_norm": 0.010076151869135958,
+      "grad_norm": 0.7734569833149393,
       "learning_rate": 4.994787516454973e-05,
-      "logits/chosen": -0.890625,
-      "logits/rejected": -0.890625,
-      "logps/chosen": -42240.0,
-      "logps/rejected": -42240.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -416.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -416.0,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -10.625,
+      "logps/chosen": -624.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.1475,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -1.453125,
+      "rewards/margins": 5.84375,
+      "rewards/rejected": -7.3125,
       "step": 340
     },
     {
       "epoch": 0.36649214659685864,
-      "grad_norm": 0.011238490353511184,
+      "grad_norm": 1.5631652496307882,
       "learning_rate": 4.992636035140273e-05,
-      "logits/chosen": -0.8203125,
-      "logits/rejected": -0.8203125,
-      "logps/chosen": -40448.0,
-      "logps/rejected": -40448.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -400.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -400.0,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -11.0,
+      "logps/chosen": -628.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.2169,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -1.7421875,
+      "rewards/margins": 3.984375,
+      "rewards/rejected": -5.75,
       "step": 350
     },
     {
       "epoch": 0.3769633507853403,
-      "grad_norm": 0.008465062570295539,
+      "grad_norm": 1.9544156123509902,
       "learning_rate": 4.99011439559198e-05,
-      "logits/chosen": -0.640625,
-      "logits/rejected": -0.640625,
-      "logps/chosen": -40960.0,
-      "logps/rejected": -40960.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -404.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -404.0,
+      "logits/chosen": -14.75,
+      "logits/rejected": -11.25,
+      "logps/chosen": -616.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.1645,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -1.65625,
+      "rewards/margins": 5.15625,
+      "rewards/rejected": -6.8125,
       "step": 360
     },
     {
       "epoch": 0.387434554973822,
-      "grad_norm": 0.010933383809386082,
+      "grad_norm": 2.2095468928684725,
       "learning_rate": 4.987222972275369e-05,
-      "logits/chosen": -0.35546875,
-      "logits/rejected": -0.35546875,
-      "logps/chosen": -46336.0,
-      "logps/rejected": -46336.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -458.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -458.0,
+      "logits/chosen": -16.0,
+      "logits/rejected": -12.5625,
+      "logps/chosen": -612.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.1457,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -1.640625,
+      "rewards/margins": 4.78125,
+      "rewards/rejected": -6.4375,
       "step": 370
     },
     {
       "epoch": 0.39790575916230364,
-      "grad_norm": 0.011240717347271166,
+      "grad_norm": 1.57655360572696,
       "learning_rate": 4.983962194568872e-05,
-      "logits/chosen": -0.189453125,
-      "logits/rejected": -0.189453125,
-      "logps/chosen": -42496.0,
-      "logps/rejected": -42496.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -420.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -420.0,
+      "logits/chosen": -16.25,
+      "logits/rejected": -12.875,
+      "logps/chosen": -680.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.1805,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -1.875,
+      "rewards/margins": 5.28125,
+      "rewards/rejected": -7.15625,
       "step": 380
     },
     {
       "epoch": 0.4083769633507853,
-      "grad_norm": 0.01105897511684848,
+      "grad_norm": 2.729095166731873,
       "learning_rate": 4.9803325467003084e-05,
-      "logits/chosen": -0.1611328125,
-      "logits/rejected": -0.1611328125,
-      "logps/chosen": -43008.0,
-      "logps/rejected": -43008.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -424.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -424.0,
+      "logits/chosen": -15.6875,
+      "logits/rejected": -11.625,
+      "logps/chosen": -608.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.1816,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -1.59375,
+      "rewards/margins": 4.0625,
+      "rewards/rejected": -5.6875,
       "step": 390
     },
     {
       "epoch": 0.418848167539267,
-      "grad_norm": 0.009287147078919152,
+      "grad_norm": 1.0787492937668104,
       "learning_rate": 4.9763345676749816e-05,
-      "logits/chosen": -0.048095703125,
-      "logits/rejected": -0.048095703125,
-      "logps/chosen": -44288.0,
-      "logps/rejected": -44288.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -438.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -438.0,
+      "logits/chosen": -15.0,
+      "logits/rejected": -11.75,
+      "logps/chosen": -624.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.1468,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -1.3828125,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -6.125,
       "step": 400
     },
     {
       "epoch": 0.418848167539267,
-      "eval_logits/chosen": 0.29296875,
-      "eval_logits/rejected": 0.41796875,
-      "eval_logps/chosen": -35328.0,
-      "eval_logps/rejected": -30592.0,
-      "eval_loss": 103.59187316894531,
-      "eval_rewards/accuracies": 0.431640625,
-      "eval_rewards/chosen": -352.0,
-      "eval_rewards/margins": -48.0,
-      "eval_rewards/rejected": -304.0,
-      "eval_runtime": 47.1448,
-      "eval_samples_per_second": 42.422,
-      "eval_steps_per_second": 0.679,
+      "eval_logits/chosen": -15.25,
+      "eval_logits/rejected": -15.1875,
+      "eval_logps/chosen": -428.0,
+      "eval_logps/rejected": -402.0,
+      "eval_loss": 0.7513124942779541,
+      "eval_rewards/accuracies": 0.470703125,
+      "eval_rewards/chosen": -1.1015625,
+      "eval_rewards/margins": 0.0322265625,
+      "eval_rewards/rejected": -1.1328125,
+      "eval_runtime": 46.6407,
+      "eval_samples_per_second": 42.881,
+      "eval_steps_per_second": 0.686,
       "step": 400
     },
     {
       "epoch": 0.4293193717277487,
-      "grad_norm": 0.010277145836894815,
+      "grad_norm": 0.990800383756,
       "learning_rate": 4.9719688511956374e-05,
-      "logits/chosen": 0.30078125,
-      "logits/rejected": 0.30078125,
-      "logps/chosen": -40448.0,
-      "logps/rejected": -40448.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -400.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -400.0,
+      "logits/chosen": -14.75,
+      "logits/rejected": -11.625,
+      "logps/chosen": -664.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.2037,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -1.5625,
+      "rewards/margins": 5.03125,
+      "rewards/rejected": -6.59375,
       "step": 410
     },
     {
       "epoch": 0.4397905759162304,
-      "grad_norm": 0.01039678691436103,
+      "grad_norm": 0.9838504657021493,
       "learning_rate": 4.967236045574294e-05,
-      "logits/chosen": 0.4140625,
-      "logits/rejected": 0.4140625,
-      "logps/chosen": -43008.0,
-      "logps/rejected": -43008.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -424.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -424.0,
+      "logits/chosen": -14.875,
+      "logits/rejected": -11.625,
+      "logps/chosen": -580.0,
+      "logps/rejected": -2656.0,
+      "loss": 0.1242,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -1.5390625,
+      "rewards/margins": 3.796875,
+      "rewards/rejected": -5.34375,
       "step": 420
     },
     {
       "epoch": 0.450261780104712,
-      "grad_norm": 0.012475698792204153,
+      "grad_norm": 2.5137656623349414,
       "learning_rate": 4.96213685363597e-05,
-      "logits/chosen": 0.447265625,
-      "logits/rejected": 0.447265625,
-      "logps/chosen": -44288.0,
-      "logps/rejected": -44288.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -438.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -438.0,
+      "logits/chosen": -15.8125,
+      "logits/rejected": -12.3125,
+      "logps/chosen": -572.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.1511,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -1.5234375,
+      "rewards/margins": 5.90625,
+      "rewards/rejected": -7.4375,
       "step": 430
     },
     {
       "epoch": 0.4607329842931937,
-      "grad_norm": 0.012341985918760179,
+      "grad_norm": 0.9069061490412824,
       "learning_rate": 4.956672032614315e-05,
-      "logits/chosen": 0.302734375,
-      "logits/rejected": 0.302734375,
-      "logps/chosen": -42496.0,
-      "logps/rejected": -42496.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -420.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -420.0,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -11.5625,
+      "logps/chosen": -588.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.1128,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -1.4921875,
+      "rewards/margins": 5.5,
+      "rewards/rejected": -7.0,
       "step": 440
     },
     {
       "epoch": 0.4712041884816754,
-      "grad_norm": 0.01353638227305833,
+      "grad_norm": 1.3568919349746313,
       "learning_rate": 4.950842394039157e-05,
-      "logits/chosen": 0.357421875,
-      "logits/rejected": 0.357421875,
-      "logps/chosen": -45824.0,
-      "logps/rejected": -45824.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -452.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -452.0,
+      "logits/chosen": -14.75,
+      "logits/rejected": -11.375,
+      "logps/chosen": -572.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.1608,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -1.5078125,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -6.03125,
       "step": 450
     },
     {
       "epoch": 0.4816753926701571,
-      "grad_norm": 0.011363332255569828,
+      "grad_norm": 1.1441529193282705,
       "learning_rate": 4.944648803615994e-05,
-      "logits/chosen": 0.5703125,
-      "logits/rejected": 0.5703125,
-      "logps/chosen": -45568.0,
-      "logps/rejected": -45568.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -450.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -450.0,
+      "logits/chosen": -15.0625,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -556.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.1252,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -1.3671875,
+      "rewards/margins": 4.4375,
+      "rewards/rejected": -5.8125,
       "step": 460
     },
     {
       "epoch": 0.49214659685863876,
-      "grad_norm": 0.015690537824906727,
+      "grad_norm": 0.8364076445731514,
       "learning_rate": 4.938092181097433e-05,
-      "logits/chosen": 0.6328125,
-      "logits/rejected": 0.6328125,
-      "logps/chosen": -45824.0,
-      "logps/rejected": -45824.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -454.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -454.0,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -11.1875,
+      "logps/chosen": -640.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.11,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -1.7734375,
+      "rewards/margins": 5.84375,
+      "rewards/rejected": -7.625,
       "step": 470
     },
     {
       "epoch": 0.5026178010471204,
-      "grad_norm": 0.017699737830403618,
+      "grad_norm": 0.7167581392274818,
       "learning_rate": 4.931173500146607e-05,
-      "logits/chosen": 0.337890625,
-      "logits/rejected": 0.337890625,
-      "logps/chosen": -44800.0,
-      "logps/rejected": -44800.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -442.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -442.0,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.625,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.1407,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -2.15625,
+      "rewards/margins": 7.90625,
+      "rewards/rejected": -10.0625,
       "step": 480
     },
     {
       "epoch": 0.5130890052356021,
-      "grad_norm": 0.01649853995955587,
+      "grad_norm": 1.02493568676847,
       "learning_rate": 4.923893788192582e-05,
-      "logits/chosen": 0.28515625,
-      "logits/rejected": 0.28515625,
-      "logps/chosen": -47104.0,
-      "logps/rejected": -47104.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -466.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -466.0,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -676.0,
+      "logps/rejected": -2992.0,
+      "loss": 0.1668,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -2.03125,
+      "rewards/margins": 5.0,
+      "rewards/rejected": -7.03125,
       "step": 490
     },
     {
       "epoch": 0.5235602094240838,
-      "grad_norm": 0.013467813931858657,
+      "grad_norm": 1.2468386390469541,
       "learning_rate": 4.9162541262777906e-05,
-      "logits/chosen": 0.72265625,
-      "logits/rejected": 0.72265625,
-      "logps/chosen": -45312.0,
-      "logps/rejected": -45312.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -448.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -448.0,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -11.0625,
+      "logps/chosen": -616.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.1267,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -1.84375,
+      "rewards/margins": 4.75,
+      "rewards/rejected": -6.59375,
       "step": 500
     },
     {
       "epoch": 0.5235602094240838,
-      "eval_logits/chosen": 1.3046875,
-      "eval_logits/rejected": 1.4765625,
-      "eval_logps/chosen": -37632.0,
-      "eval_logps/rejected": -32512.0,
-      "eval_loss": 109.86737823486328,
-      "eval_rewards/accuracies": 0.43359375,
-      "eval_rewards/chosen": -372.0,
-      "eval_rewards/margins": -50.75,
-      "eval_rewards/rejected": -322.0,
-      "eval_runtime": 47.1552,
-      "eval_samples_per_second": 42.413,
-      "eval_steps_per_second": 0.679,
+      "eval_logits/chosen": -13.375,
+      "eval_logits/rejected": -13.125,
+      "eval_logps/chosen": -460.0,
+      "eval_logps/rejected": -436.0,
+      "eval_loss": 0.7807812690734863,
+      "eval_rewards/accuracies": 0.517578125,
+      "eval_rewards/chosen": -1.421875,
+      "eval_rewards/margins": 0.057373046875,
+      "eval_rewards/rejected": -1.484375,
+      "eval_runtime": 46.6389,
+      "eval_samples_per_second": 42.883,
+      "eval_steps_per_second": 0.686,
       "step": 500
     },
     {
       "epoch": 0.5340314136125655,
-      "grad_norm": 0.012854647834118242,
+      "grad_norm": 0.6155954698882383,
       "learning_rate": 4.90825564889749e-05,
-      "logits/chosen": 0.94140625,
-      "logits/rejected": 0.94140625,
-      "logps/chosen": -45312.0,
-      "logps/rejected": -45312.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -448.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -448.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -11.0625,
+      "logps/chosen": -624.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.1299,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.8125,
+      "rewards/margins": 5.5,
+      "rewards/rejected": -7.3125,
       "step": 510
     },
     {
       "epoch": 0.5445026178010471,
-      "grad_norm": 0.014738155894898568,
+      "grad_norm": 1.8958326659247546,
       "learning_rate": 4.899899543831288e-05,
-      "logits/chosen": 1.015625,
-      "logits/rejected": 1.015625,
-      "logps/chosen": -44800.0,
-      "logps/rejected": -44800.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -444.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -444.0,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -12.0,
+      "logps/chosen": -688.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.2037,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -2.140625,
+      "rewards/margins": 5.34375,
+      "rewards/rejected": -7.5,
       "step": 520
     },
     {
       "epoch": 0.5549738219895288,
-      "grad_norm": 0.032677815601760384,
+      "grad_norm": 1.3822359805479507,
       "learning_rate": 4.891187051966765e-05,
-      "logits/chosen": 1.3125,
-      "logits/rejected": 1.3125,
-      "logps/chosen": -45312.0,
-      "logps/rejected": -45312.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -448.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -448.0,
+      "logits/chosen": -15.375,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -572.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.1362,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.625,
+      "rewards/margins": 4.40625,
+      "rewards/rejected": -6.03125,
       "step": 530
     },
     {
       "epoch": 0.5654450261780105,
-      "grad_norm": 0.014210153031693354,
+      "grad_norm": 0.8236894372265684,
       "learning_rate": 4.882119467115191e-05,
-      "logits/chosen": 1.109375,
-      "logits/rejected": 1.109375,
-      "logps/chosen": -51200.0,
-      "logps/rejected": -51200.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -506.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -506.0,
+      "logits/chosen": -15.25,
+      "logits/rejected": -12.0,
+      "logps/chosen": -688.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.1558,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -1.9921875,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -6.65625,
       "step": 540
     },
     {
       "epoch": 0.5759162303664922,
-      "grad_norm": 0.014048067056600484,
+      "grad_norm": 1.8254673958497956,
       "learning_rate": 4.872698135819402e-05,
-      "logits/chosen": 1.3046875,
-      "logits/rejected": 1.3046875,
-      "logps/chosen": -48896.0,
-      "logps/rejected": -48896.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -484.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -484.0,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -11.0625,
+      "logps/chosen": -580.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.1554,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -1.734375,
+      "rewards/margins": 6.09375,
+      "rewards/rejected": -7.8125,
       "step": 550
     },
     {
       "epoch": 0.5863874345549738,
-      "grad_norm": 0.013864377220155596,
+      "grad_norm": 0.7193719838484375,
       "learning_rate": 4.862924457153831e-05,
-      "logits/chosen": 1.546875,
-      "logits/rejected": 1.546875,
-      "logps/chosen": -42240.0,
-      "logps/rejected": -42240.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -418.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -418.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -608.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.1212,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -1.8359375,
+      "rewards/margins": 5.59375,
+      "rewards/rejected": -7.4375,
       "step": 560
     },
     {
       "epoch": 0.5968586387434555,
-      "grad_norm": 0.0119031380737084,
+      "grad_norm": 1.018114281649203,
       "learning_rate": 4.8527998825167484e-05,
-      "logits/chosen": 1.09375,
-      "logits/rejected": 1.09375,
-      "logps/chosen": -51712.0,
-      "logps/rejected": -51712.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -512.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -512.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3008.0,
+      "loss": 0.1167,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -2.109375,
+      "rewards/margins": 5.75,
+      "rewards/rejected": -7.875,
       "step": 570
     },
     {
       "epoch": 0.6073298429319371,
-      "grad_norm": 0.013039898178854925,
+      "grad_norm": 1.5917670740171264,
       "learning_rate": 4.842325915414728e-05,
-      "logits/chosen": 1.1328125,
-      "logits/rejected": 1.1328125,
-      "logps/chosen": -47872.0,
-      "logps/rejected": -47872.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -472.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -472.0,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -9.625,
+      "logps/chosen": -652.0,
+      "logps/rejected": -2928.0,
+      "loss": 0.1567,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.8125,
+      "rewards/margins": 5.46875,
+      "rewards/rejected": -7.28125,
       "step": 580
     },
     {
       "epoch": 0.6178010471204188,
-      "grad_norm": 0.01752199005002369,
+      "grad_norm": 0.994837983138027,
       "learning_rate": 4.831504111239373e-05,
-      "logits/chosen": 1.125,
-      "logits/rejected": 1.125,
-      "logps/chosen": -49664.0,
-      "logps/rejected": -49664.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -490.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -490.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -632.0,
+      "logps/rejected": -2976.0,
+      "loss": 0.1247,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.9375,
+      "rewards/margins": 5.28125,
+      "rewards/rejected": -7.21875,
       "step": 590
     },
     {
       "epoch": 0.6282722513089005,
-      "grad_norm": 0.014974431039762133,
+      "grad_norm": 1.979701944672432,
       "learning_rate": 4.8203360770363425e-05,
-      "logits/chosen": 1.2109375,
-      "logits/rejected": 1.2109375,
-      "logps/chosen": -49664.0,
-      "logps/rejected": -49664.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -490.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -490.0,
+      "logits/chosen": -14.875,
+      "logits/rejected": -11.625,
+      "logps/chosen": -704.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.1435,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -2.390625,
+      "rewards/margins": 6.53125,
+      "rewards/rejected": -8.9375,
       "step": 600
     },
     {
       "epoch": 0.6282722513089005,
-      "eval_logits/chosen": 1.6640625,
-      "eval_logits/rejected": 1.8828125,
-      "eval_logps/chosen": -39936.0,
-      "eval_logps/rejected": -34560.0,
-      "eval_loss": 116.73625183105469,
-      "eval_rewards/accuracies": 0.431640625,
-      "eval_rewards/chosen": -396.0,
-      "eval_rewards/margins": -54.0,
-      "eval_rewards/rejected": -342.0,
-      "eval_runtime": 47.1626,
-      "eval_samples_per_second": 42.407,
-      "eval_steps_per_second": 0.679,
+      "eval_logits/chosen": -15.6875,
+      "eval_logits/rejected": -15.625,
+      "eval_logps/chosen": -490.0,
+      "eval_logps/rejected": -462.0,
+      "eval_loss": 0.8511093854904175,
+      "eval_rewards/accuracies": 0.5,
+      "eval_rewards/chosen": -1.7109375,
+      "eval_rewards/margins": 0.03076171875,
+      "eval_rewards/rejected": -1.7421875,
+      "eval_runtime": 46.6343,
+      "eval_samples_per_second": 42.887,
+      "eval_steps_per_second": 0.686,
       "step": 600
     },
     {
       "epoch": 0.6387434554973822,
-      "grad_norm": 0.013560949077688159,
+      "grad_norm": 2.1824077332605794,
       "learning_rate": 4.808823471266702e-05,
-      "logits/chosen": 1.3359375,
-      "logits/rejected": 1.3359375,
-      "logps/chosen": -48128.0,
-      "logps/rejected": -48128.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -478.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -478.0,
+      "logits/chosen": -14.4375,
+      "logits/rejected": -11.3125,
+      "logps/chosen": -724.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.1763,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -2.09375,
+      "rewards/margins": 5.8125,
+      "rewards/rejected": -7.90625,
       "step": 610
     },
     {
       "epoch": 0.6492146596858639,
-      "grad_norm": 0.013539763484461731,
+      "grad_norm": 1.2315438748306793,
       "learning_rate": 4.796968003560639e-05,
-      "logits/chosen": 1.3671875,
-      "logits/rejected": 1.3671875,
-      "logps/chosen": -49664.0,
-      "logps/rejected": -49664.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -490.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -490.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -596.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.1297,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.5859375,
+      "rewards/margins": 4.28125,
+      "rewards/rejected": -5.84375,
       "step": 620
     },
     {
       "epoch": 0.6596858638743456,
-      "grad_norm": 0.014628107511885797,
+      "grad_norm": 0.5749144136274057,
       "learning_rate": 4.7847714344635864e-05,
-      "logits/chosen": 1.9453125,
-      "logits/rejected": 1.9453125,
-      "logps/chosen": -52736.0,
-      "logps/rejected": -52736.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -524.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -524.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -592.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.148,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -1.5703125,
+      "rewards/margins": 6.09375,
+      "rewards/rejected": -7.65625,
       "step": 630
     },
     {
       "epoch": 0.6701570680628273,
-      "grad_norm": 0.015494253389959226,
+      "grad_norm": 0.8964568015584935,
       "learning_rate": 4.772235575174776e-05,
-      "logits/chosen": 2.578125,
-      "logits/rejected": 2.578125,
-      "logps/chosen": -51200.0,
-      "logps/rejected": -51200.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -508.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -508.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -604.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.1514,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.515625,
+      "rewards/margins": 5.0,
+      "rewards/rejected": -6.5,
       "step": 640
     },
     {
       "epoch": 0.680628272251309,
-      "grad_norm": 0.012916907738912421,
+      "grad_norm": 0.7339962590693953,
       "learning_rate": 4.7593622872782754e-05,
-      "logits/chosen": 2.765625,
-      "logits/rejected": 2.765625,
-      "logps/chosen": -49152.0,
-      "logps/rejected": -49152.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -486.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -486.0,
+      "logits/chosen": -15.0,
+      "logits/rejected": -11.0625,
+      "logps/chosen": -536.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.1485,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -1.328125,
+      "rewards/margins": 5.34375,
+      "rewards/rejected": -6.65625,
       "step": 650
     },
     {
       "epoch": 0.6910994764397905,
-      "grad_norm": 0.01178378183083928,
+      "grad_norm": 0.7478148538177102,
       "learning_rate": 4.746153482466542e-05,
-      "logits/chosen": 2.265625,
-      "logits/rejected": 2.265625,
-      "logps/chosen": -51200.0,
-      "logps/rejected": -51200.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -506.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -506.0,
+      "logits/chosen": -15.8125,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -564.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.1465,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -1.40625,
+      "rewards/margins": 5.46875,
+      "rewards/rejected": -6.875,
       "step": 660
     },
     {
       "epoch": 0.7015706806282722,
-      "grad_norm": 0.012734526671173209,
+      "grad_norm": 1.0545978843778105,
       "learning_rate": 4.732611122256537e-05,
-      "logits/chosen": 1.8828125,
-      "logits/rejected": 1.8828125,
-      "logps/chosen": -51200.0,
-      "logps/rejected": -51200.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -506.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -506.0,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -12.3125,
+      "logps/chosen": -632.0,
+      "logps/rejected": -2704.0,
+      "loss": 0.1335,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -1.9375,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -6.625,
       "step": 670
     },
     {
       "epoch": 0.7120418848167539,
-      "grad_norm": 0.013432948723238362,
+      "grad_norm": 2.17361778361498,
       "learning_rate": 4.718737217698435e-05,
-      "logits/chosen": 2.140625,
-      "logits/rejected": 2.140625,
-      "logps/chosen": -53760.0,
-      "logps/rejected": -53760.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -532.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -532.0,
+      "logits/chosen": -14.875,
+      "logits/rejected": -11.875,
+      "logps/chosen": -696.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.132,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.15625,
+      "rewards/margins": 5.40625,
+      "rewards/rejected": -7.5625,
       "step": 680
     },
     {
       "epoch": 0.7225130890052356,
-      "grad_norm": 0.015645285883980287,
+      "grad_norm": 0.8547784490240291,
       "learning_rate": 4.704533829076982e-05,
-      "logits/chosen": 2.34375,
-      "logits/rejected": 2.34375,
-      "logps/chosen": -51712.0,
-      "logps/rejected": -51712.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -512.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -512.0,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -11.5,
+      "logps/chosen": -664.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.1658,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -2.0625,
+      "rewards/margins": 6.125,
+      "rewards/rejected": -8.1875,
       "step": 690
     },
     {
       "epoch": 0.7329842931937173,
-      "grad_norm": 0.014457882611619975,
+      "grad_norm": 0.9289807794147451,
       "learning_rate": 4.6900030656055455e-05,
-      "logits/chosen": 2.28125,
-      "logits/rejected": 2.28125,
-      "logps/chosen": -54272.0,
-      "logps/rejected": -54272.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -536.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -536.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -684.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.1255,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.8828125,
+      "rewards/margins": 4.84375,
+      "rewards/rejected": -6.71875,
       "step": 700
     },
     {
       "epoch": 0.7329842931937173,
-      "eval_logits/chosen": 2.84375,
-      "eval_logits/rejected": 3.140625,
-      "eval_logps/chosen": -42240.0,
-      "eval_logps/rejected": -36608.0,
-      "eval_loss": 123.6395034790039,
-      "eval_rewards/accuracies": 0.43359375,
-      "eval_rewards/chosen": -420.0,
-      "eval_rewards/margins": -57.25,
-      "eval_rewards/rejected": -362.0,
-      "eval_runtime": 47.1629,
-      "eval_samples_per_second": 42.406,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -13.5625,
+      "eval_logits/rejected": -13.125,
+      "eval_logps/chosen": -502.0,
+      "eval_logps/rejected": -492.0,
+      "eval_loss": 0.8534374833106995,
+      "eval_rewards/accuracies": 0.513671875,
+      "eval_rewards/chosen": -1.84375,
+      "eval_rewards/margins": 0.1796875,
+      "eval_rewards/rejected": -2.03125,
+      "eval_runtime": 46.8422,
+      "eval_samples_per_second": 42.697,
+      "eval_steps_per_second": 0.683,
       "step": 700
     },
     {
       "epoch": 0.743455497382199,
-      "grad_norm": 0.015929253433509403,
+      "grad_norm": 1.2303340065121269,
       "learning_rate": 4.6751470851128906e-05,
-      "logits/chosen": 2.078125,
-      "logits/rejected": 2.078125,
-      "logps/chosen": -52224.0,
-      "logps/rejected": -52224.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -520.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -520.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -656.0,
+      "logps/rejected": -3184.0,
+      "loss": 0.1001,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -2.15625,
+      "rewards/margins": 7.6875,
+      "rewards/rejected": -9.875,
       "step": 710
     },
     {
       "epoch": 0.7539267015706806,
-      "grad_norm": 0.017011606591743354,
+      "grad_norm": 0.9622826276528129,
       "learning_rate": 4.6599680937227454e-05,
-      "logits/chosen": 2.015625,
-      "logits/rejected": 2.015625,
-      "logps/chosen": -50432.0,
-      "logps/rejected": -50432.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -500.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -500.0,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -10.375,
+      "logps/chosen": -636.0,
+      "logps/rejected": -2848.0,
+      "loss": 0.158,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -2.1875,
+      "rewards/margins": 5.90625,
+      "rewards/rejected": -8.125,
       "step": 720
     },
     {
       "epoch": 0.7643979057591623,
-      "grad_norm": 0.01612608471514965,
+      "grad_norm": 1.37822733965358,
       "learning_rate": 4.644468345526183e-05,
-      "logits/chosen": 2.1875,
-      "logits/rejected": 2.1875,
-      "logps/chosen": -53760.0,
-      "logps/rejected": -53760.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -532.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -532.0,
+      "logits/chosen": -12.5,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -664.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.1627,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.15625,
+      "rewards/margins": 5.5,
+      "rewards/rejected": -7.6875,
       "step": 730
     },
     {
       "epoch": 0.774869109947644,
-      "grad_norm": 0.019303381763061556,
+      "grad_norm": 1.0983096387223583,
       "learning_rate": 4.628650142246894e-05,
-      "logits/chosen": 2.796875,
-      "logits/rejected": 2.796875,
-      "logps/chosen": -51968.0,
-      "logps/rejected": -51968.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -516.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -516.0,
+      "logits/chosen": -12.3125,
+      "logits/rejected": -10.375,
+      "logps/chosen": -612.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.148,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -1.859375,
+      "rewards/margins": 5.0625,
+      "rewards/rejected": -6.9375,
       "step": 740
     },
     {
       "epoch": 0.7853403141361257,
-      "grad_norm": 0.016461565143255093,
+      "grad_norm": 1.9668590726339368,
       "learning_rate": 4.6125158328993765e-05,
-      "logits/chosen": 2.109375,
-      "logits/rejected": 2.109375,
-      "logps/chosen": -60928.0,
-      "logps/rejected": -60928.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -604.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -604.0,
+      "logits/chosen": -12.4375,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -768.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.1633,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -2.515625,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -7.1875,
       "step": 750
     },
     {
       "epoch": 0.7958115183246073,
-      "grad_norm": 0.0183532260032194,
+      "grad_norm": 1.1683830322642343,
       "learning_rate": 4.5960678134401064e-05,
-      "logits/chosen": 2.78125,
-      "logits/rejected": 2.78125,
-      "logps/chosen": -59648.0,
-      "logps/rejected": -59648.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -592.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -592.0,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -10.0,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3136.0,
+      "loss": 0.1278,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -2.25,
+      "rewards/margins": 6.75,
+      "rewards/rejected": -9.0,
       "step": 760
     },
     {
       "epoch": 0.806282722513089,
-      "grad_norm": 0.015137532173226083,
+      "grad_norm": 1.032006314354136,
       "learning_rate": 4.579308526411734e-05,
-      "logits/chosen": 3.21875,
-      "logits/rejected": 3.21875,
-      "logps/chosen": -49664.0,
-      "logps/rejected": -49664.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -492.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -492.0,
+      "logits/chosen": -11.375,
+      "logits/rejected": -9.625,
+      "logps/chosen": -588.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.1559,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -1.84375,
+      "rewards/margins": 5.28125,
+      "rewards/rejected": -7.125,
       "step": 770
     },
     {
       "epoch": 0.8167539267015707,
-      "grad_norm": 0.015172822265344691,
+      "grad_norm": 1.3437384964915426,
       "learning_rate": 4.562240460580367e-05,
-      "logits/chosen": 2.390625,
-      "logits/rejected": 2.390625,
-      "logps/chosen": -52224.0,
-      "logps/rejected": -52224.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -520.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -520.0,
+      "logits/chosen": -11.4375,
+      "logits/rejected": -9.75,
+      "logps/chosen": -692.0,
+      "logps/rejected": -2720.0,
+      "loss": 0.1536,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -2.09375,
+      "rewards/margins": 4.34375,
+      "rewards/rejected": -6.4375,
       "step": 780
     },
     {
       "epoch": 0.8272251308900523,
-      "grad_norm": 0.016602484645609806,
+      "grad_norm": 0.8735836619606299,
       "learning_rate": 4.544866150565985e-05,
-      "logits/chosen": 2.59375,
-      "logits/rejected": 2.59375,
-      "logps/chosen": -56832.0,
-      "logps/rejected": -56832.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -564.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -564.0,
+      "logits/chosen": -12.125,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -596.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.1346,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.765625,
+      "rewards/margins": 5.0625,
+      "rewards/rejected": -6.84375,
       "step": 790
     },
     {
       "epoch": 0.837696335078534,
-      "grad_norm": 0.013796701726207627,
+      "grad_norm": 0.6435600678567915,
       "learning_rate": 4.527188176466051e-05,
-      "logits/chosen": 2.328125,
-      "logits/rejected": 2.328125,
-      "logps/chosen": -52480.0,
-      "logps/rejected": -52480.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -520.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -520.0,
+      "logits/chosen": -11.8125,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -584.0,
+      "logps/rejected": -2864.0,
+      "loss": 0.1142,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -1.7578125,
+      "rewards/margins": 5.4375,
+      "rewards/rejected": -7.1875,
       "step": 800
     },
     {
       "epoch": 0.837696335078534,
-      "eval_logits/chosen": 3.765625,
-      "eval_logits/rejected": 4.15625,
-      "eval_logps/chosen": -44544.0,
-      "eval_logps/rejected": -38400.0,
-      "eval_loss": 130.5068817138672,
-      "eval_rewards/accuracies": 0.431640625,
-      "eval_rewards/chosen": -442.0,
-      "eval_rewards/margins": -60.25,
-      "eval_rewards/rejected": -382.0,
-      "eval_runtime": 47.1719,
-      "eval_samples_per_second": 42.398,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -12.375,
+      "eval_logits/rejected": -11.875,
+      "eval_logps/chosen": -478.0,
+      "eval_logps/rejected": -452.0,
+      "eval_loss": 0.8559062480926514,
+      "eval_rewards/accuracies": 0.4921875,
+      "eval_rewards/chosen": -1.59375,
+      "eval_rewards/margins": 0.047119140625,
+      "eval_rewards/rejected": -1.640625,
+      "eval_runtime": 46.6154,
+      "eval_samples_per_second": 42.904,
+      "eval_steps_per_second": 0.686,
       "step": 800
     },
     {
       "epoch": 0.8481675392670157,
-      "grad_norm": 0.012556283303654143,
+      "grad_norm": 1.3268621315118674,
       "learning_rate": 4.509209163472357e-05,
-      "logits/chosen": 2.453125,
-      "logits/rejected": 2.453125,
-      "logps/chosen": -58112.0,
-      "logps/rejected": -58112.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -576.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -576.0,
+      "logits/chosen": -12.625,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -676.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.1462,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.96875,
+      "rewards/margins": 5.625,
+      "rewards/rejected": -7.59375,
       "step": 810
     },
     {
       "epoch": 0.8586387434554974,
-      "grad_norm": 0.014792958195181644,
+      "grad_norm": 1.1138973929256597,
       "learning_rate": 4.4909317814811896e-05,
-      "logits/chosen": 3.265625,
-      "logits/rejected": 3.265625,
-      "logps/chosen": -54784.0,
-      "logps/rejected": -54784.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -544.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -544.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.875,
+      "logps/chosen": -608.0,
+      "logps/rejected": -3008.0,
+      "loss": 0.131,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -1.8359375,
+      "rewards/margins": 6.21875,
+      "rewards/rejected": -8.0625,
       "step": 820
     },
     {
       "epoch": 0.8691099476439791,
-      "grad_norm": 0.014966336050440345,
+      "grad_norm": 1.0913026710199536,
       "learning_rate": 4.472358744696845e-05,
-      "logits/chosen": 3.1875,
-      "logits/rejected": 3.1875,
-      "logps/chosen": -55808.0,
-      "logps/rejected": -55808.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -556.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -556.0,
+      "logits/chosen": -12.875,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -640.0,
+      "logps/rejected": -2784.0,
+      "loss": 0.1549,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -2.03125,
+      "rewards/margins": 4.78125,
+      "rewards/rejected": -6.8125,
       "step": 830
     },
     {
       "epoch": 0.8795811518324608,
-      "grad_norm": 0.01395931430052463,
+      "grad_norm": 1.0701708659411528,
       "learning_rate": 4.4534928112285645e-05,
-      "logits/chosen": 3.203125,
-      "logits/rejected": 3.203125,
-      "logps/chosen": -55040.0,
-      "logps/rejected": -55040.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -548.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -548.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.75,
+      "logps/chosen": -640.0,
+      "logps/rejected": -2752.0,
+      "loss": 0.1353,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -1.890625,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -6.59375,
       "step": 840
     },
     {
       "epoch": 0.8900523560209425,
-      "grad_norm": 0.013877008900216554,
+      "grad_norm": 1.0935244825784507,
       "learning_rate": 4.43433678268096e-05,
-      "logits/chosen": 3.609375,
-      "logits/rejected": 3.609375,
-      "logps/chosen": -55040.0,
-      "logps/rejected": -55040.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -544.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -544.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -656.0,
+      "logps/rejected": -2880.0,
+      "loss": 0.1409,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -2.078125,
+      "rewards/margins": 5.375,
+      "rewards/rejected": -7.4375,
       "step": 850
     },
     {
       "epoch": 0.900523560209424,
-      "grad_norm": 0.014232927929297166,
+      "grad_norm": 0.8458258396301148,
       "learning_rate": 4.414893503737968e-05,
-      "logits/chosen": 3.375,
-      "logits/rejected": 3.375,
-      "logps/chosen": -58112.0,
-      "logps/rejected": -58112.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -576.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -576.0,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -10.125,
+      "logps/chosen": -612.0,
+      "logps/rejected": -2912.0,
+      "loss": 0.1224,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -2.03125,
+      "rewards/margins": 5.4375,
+      "rewards/rejected": -7.4375,
       "step": 860
     },
     {
       "epoch": 0.9109947643979057,
-      "grad_norm": 0.014837219024986017,
+      "grad_norm": 1.223796863993444,
       "learning_rate": 4.395165861740414e-05,
-      "logits/chosen": 3.203125,
-      "logits/rejected": 3.203125,
-      "logps/chosen": -57856.0,
-      "logps/rejected": -57856.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -576.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -576.0,
+      "logits/chosen": -12.8125,
+      "logits/rejected": -10.625,
+      "logps/chosen": -632.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.1037,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.09375,
+      "rewards/margins": 6.8125,
+      "rewards/rejected": -8.875,
       "step": 870
     },
     {
       "epoch": 0.9214659685863874,
-      "grad_norm": 0.014119052390831888,
+      "grad_norm": 1.6368484057945356,
       "learning_rate": 4.375156786257241e-05,
-      "logits/chosen": 2.890625,
-      "logits/rejected": 2.890625,
-      "logps/chosen": -58624.0,
-      "logps/rejected": -58624.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -580.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -580.0,
+      "logits/chosen": -12.625,
+      "logits/rejected": -10.5,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3184.0,
+      "loss": 0.12,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -2.203125,
+      "rewards/margins": 7.28125,
+      "rewards/rejected": -9.5,
       "step": 880
     },
     {
       "epoch": 0.9319371727748691,
-      "grad_norm": 0.01550542768542966,
+      "grad_norm": 1.7310358088279334,
       "learning_rate": 4.354869248650466e-05,
-      "logits/chosen": 2.953125,
-      "logits/rejected": 2.953125,
-      "logps/chosen": -62720.0,
-      "logps/rejected": -62720.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -624.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -624.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.875,
+      "logps/chosen": -728.0,
+      "logps/rejected": -2944.0,
+      "loss": 0.1523,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -2.5,
+      "rewards/margins": 5.8125,
+      "rewards/rejected": -8.3125,
       "step": 890
     },
     {
       "epoch": 0.9424083769633508,
-      "grad_norm": 0.01609534664970966,
+      "grad_norm": 0.7170892614715749,
       "learning_rate": 4.334306261633931e-05,
-      "logits/chosen": 3.109375,
-      "logits/rejected": 3.109375,
-      "logps/chosen": -54784.0,
-      "logps/rejected": -54784.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -540.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -540.0,
+      "logits/chosen": -14.5,
+      "logits/rejected": -10.9375,
+      "logps/chosen": -644.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.1596,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -2.234375,
+      "rewards/margins": 4.96875,
+      "rewards/rejected": -7.21875,
       "step": 900
     },
     {
       "epoch": 0.9424083769633508,
-      "eval_logits/chosen": 3.859375,
-      "eval_logits/rejected": 4.25,
-      "eval_logps/chosen": -46848.0,
-      "eval_logps/rejected": -40448.0,
-      "eval_loss": 137.39688110351562,
-      "eval_rewards/accuracies": 0.427734375,
-      "eval_rewards/chosen": -466.0,
-      "eval_rewards/margins": -63.5,
-      "eval_rewards/rejected": -402.0,
-      "eval_runtime": 47.1714,
-      "eval_samples_per_second": 42.399,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -15.0625,
+      "eval_logits/rejected": -14.875,
+      "eval_logps/chosen": -524.0,
+      "eval_logps/rejected": -498.0,
+      "eval_loss": 0.8357968926429749,
+      "eval_rewards/accuracies": 0.4921875,
+      "eval_rewards/chosen": -2.046875,
+      "eval_rewards/margins": 0.051025390625,
+      "eval_rewards/rejected": -2.09375,
+      "eval_runtime": 46.6162,
+      "eval_samples_per_second": 42.904,
+      "eval_steps_per_second": 0.686,
       "step": 900
     },
     {
       "epoch": 0.9528795811518325,
-      "grad_norm": 0.016656400979311477,
+      "grad_norm": 1.0608482317231847,
       "learning_rate": 4.313470878825911e-05,
-      "logits/chosen": 3.546875,
-      "logits/rejected": 3.546875,
-      "logps/chosen": -55552.0,
-      "logps/rejected": -55552.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -552.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -552.0,
+      "logits/chosen": -14.625,
+      "logits/rejected": -11.625,
+      "logps/chosen": -668.0,
+      "logps/rejected": -2832.0,
+      "loss": 0.1358,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -2.109375,
+      "rewards/margins": 4.78125,
+      "rewards/rejected": -6.875,
       "step": 910
     },
     {
       "epoch": 0.9633507853403142,
-      "grad_norm": 0.014307183359898043,
+      "grad_norm": 1.8533029509980772,
       "learning_rate": 4.292366194295652e-05,
-      "logits/chosen": 3.140625,
-      "logits/rejected": 3.140625,
-      "logps/chosen": -53504.0,
-      "logps/rejected": -53504.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -528.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -528.0,
+      "logits/chosen": -13.8125,
+      "logits/rejected": -10.625,
+      "logps/chosen": -696.0,
+      "logps/rejected": -2896.0,
+      "loss": 0.1337,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -2.015625,
+      "rewards/margins": 5.0625,
+      "rewards/rejected": -7.09375,
       "step": 920
     },
     {
       "epoch": 0.9738219895287958,
-      "grad_norm": 0.01600346042310868,
+      "grad_norm": 1.5361849089317439,
       "learning_rate": 4.2709953421038964e-05,
-      "logits/chosen": 3.3125,
-      "logits/rejected": 3.3125,
-      "logps/chosen": -52992.0,
-      "logps/rejected": -52992.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -524.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -524.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.375,
+      "logps/chosen": -576.0,
+      "logps/rejected": -3008.0,
+      "loss": 0.1116,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.8203125,
+      "rewards/margins": 5.8125,
+      "rewards/rejected": -7.625,
       "step": 930
     },
     {
       "epoch": 0.9842931937172775,
-      "grad_norm": 0.014052092139330804,
+      "grad_norm": 0.8184270360275369,
       "learning_rate": 4.249361495837478e-05,
-      "logits/chosen": 2.75,
-      "logits/rejected": 2.75,
-      "logps/chosen": -54272.0,
-      "logps/rejected": -54272.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -536.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -536.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.375,
+      "logps/chosen": -632.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.1171,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -2.296875,
+      "rewards/margins": 6.59375,
+      "rewards/rejected": -8.875,
       "step": 940
     },
     {
       "epoch": 0.9947643979057592,
-      "grad_norm": 0.01622838138211811,
+      "grad_norm": 0.5850180924962325,
       "learning_rate": 4.227467868138035e-05,
-      "logits/chosen": 2.46875,
-      "logits/rejected": 2.46875,
-      "logps/chosen": -58368.0,
-      "logps/rejected": -58368.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -580.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -580.0,
+      "logits/chosen": -13.25,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.1354,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -2.125,
+      "rewards/margins": 5.65625,
+      "rewards/rejected": -7.8125,
       "step": 950
     },
     {
       "epoch": 1.0052356020942408,
-      "grad_norm": 0.017041307443443135,
+      "grad_norm": 0.9212669460316035,
       "learning_rate": 4.205317710224938e-05,
-      "logits/chosen": 3.296875,
-      "logits/rejected": 3.296875,
-      "logps/chosen": -55808.0,
-      "logps/rejected": -55808.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -556.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -556.0,
+      "logits/chosen": -12.4375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -568.0,
+      "logps/rejected": -2800.0,
+      "loss": 0.0981,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -1.5546875,
+      "rewards/margins": 5.21875,
+      "rewards/rejected": -6.78125,
       "step": 960
     },
     {
       "epoch": 1.0157068062827226,
-      "grad_norm": 0.015739739418284014,
+      "grad_norm": 0.6838241096137042,
       "learning_rate": 4.1829143114124733e-05,
-      "logits/chosen": 3.3125,
-      "logits/rejected": 3.3125,
-      "logps/chosen": -59904.0,
-      "logps/rejected": -59904.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -592.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -592.0,
+      "logits/chosen": -12.375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -620.0,
+      "logps/rejected": -3024.0,
+      "loss": 0.0565,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -1.7890625,
+      "rewards/margins": 6.0625,
+      "rewards/rejected": -7.84375,
       "step": 970
     },
     {
       "epoch": 1.0261780104712042,
-      "grad_norm": 0.014865298097133097,
+      "grad_norm": 0.17737987455409518,
       "learning_rate": 4.160260998621387e-05,
-      "logits/chosen": 3.453125,
-      "logits/rejected": 3.453125,
-      "logps/chosen": -56064.0,
-      "logps/rejected": -56064.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -556.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -556.0,
+      "logits/chosen": -12.125,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -624.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.041,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -1.875,
+      "rewards/margins": 7.09375,
+      "rewards/rejected": -8.9375,
       "step": 980
     },
     {
       "epoch": 1.036649214659686,
-      "grad_norm": 0.016487518811756783,
+      "grad_norm": 0.2550137332359494,
       "learning_rate": 4.137361135884824e-05,
-      "logits/chosen": 3.46875,
-      "logits/rejected": 3.46875,
-      "logps/chosen": -60672.0,
-      "logps/rejected": -60672.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -600.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -600.0,
+      "logits/chosen": -12.5,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -616.0,
+      "logps/rejected": -3088.0,
+      "loss": 0.0626,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -1.8671875,
+      "rewards/margins": 7.34375,
+      "rewards/rejected": -9.1875,
       "step": 990
     },
     {
       "epoch": 1.0471204188481675,
-      "grad_norm": 0.014560493385555024,
+      "grad_norm": 0.1161136630664759,
       "learning_rate": 4.114218123848777e-05,
-      "logits/chosen": 3.828125,
-      "logits/rejected": 3.828125,
-      "logps/chosen": -64000.0,
-      "logps/rejected": -64000.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -636.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -636.0,
+      "logits/chosen": -12.8125,
+      "logits/rejected": -10.25,
+      "logps/chosen": -628.0,
+      "logps/rejected": -3136.0,
+      "loss": 0.0231,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -1.9296875,
+      "rewards/margins": 6.53125,
+      "rewards/rejected": -8.5,
       "step": 1000
     },
     {
       "epoch": 1.0471204188481675,
-      "eval_logits/chosen": 5.15625,
-      "eval_logits/rejected": 5.6875,
-      "eval_logps/chosen": -49152.0,
-      "eval_logps/rejected": -42496.0,
-      "eval_loss": 143.70375061035156,
-      "eval_rewards/accuracies": 0.4296875,
-      "eval_rewards/chosen": -488.0,
-      "eval_rewards/margins": -66.5,
-      "eval_rewards/rejected": -422.0,
-      "eval_runtime": 47.1561,
-      "eval_samples_per_second": 42.412,
-      "eval_steps_per_second": 0.679,
+      "eval_logits/chosen": -12.5,
+      "eval_logits/rejected": -12.0625,
+      "eval_logps/chosen": -506.0,
+      "eval_logps/rejected": -492.0,
+      "eval_loss": 0.8379687666893005,
+      "eval_rewards/accuracies": 0.5078125,
+      "eval_rewards/chosen": -1.8828125,
+      "eval_rewards/margins": 0.140625,
+      "eval_rewards/rejected": -2.03125,
+      "eval_runtime": 46.599,
+      "eval_samples_per_second": 42.919,
+      "eval_steps_per_second": 0.687,
       "step": 1000
     },
     {
       "epoch": 1.057591623036649,
-      "grad_norm": 0.015789621433931752,
+      "grad_norm": 0.6095935446682996,
       "learning_rate": 4.0908353992670824e-05,
-      "logits/chosen": 4.0625,
-      "logits/rejected": 4.0625,
-      "logps/chosen": -57600.0,
-      "logps/rejected": -57600.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -572.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -572.0,
+      "logits/chosen": -12.5,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -604.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0274,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -1.7265625,
+      "rewards/margins": 7.96875,
+      "rewards/rejected": -9.6875,
       "step": 1010
     },
     {
       "epoch": 1.068062827225131,
-      "grad_norm": 0.014095089315044398,
+      "grad_norm": 1.2134087825889897,
       "learning_rate": 4.067216434491059e-05,
-      "logits/chosen": 4.0,
-      "logits/rejected": 4.0,
-      "logps/chosen": -60160.0,
-      "logps/rejected": -60160.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -596.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -596.0,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -10.375,
+      "logps/chosen": -640.0,
+      "logps/rejected": -3120.0,
+      "loss": 0.0548,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -1.9921875,
+      "rewards/margins": 7.1875,
+      "rewards/rejected": -9.1875,
       "step": 1020
     },
     {
       "epoch": 1.0785340314136125,
-      "grad_norm": 0.013785580987969824,
+      "grad_norm": 0.4360694049816956,
       "learning_rate": 4.043364736953866e-05,
-      "logits/chosen": 3.890625,
-      "logits/rejected": 3.890625,
-      "logps/chosen": -61440.0,
-      "logps/rejected": -61440.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -608.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -608.0,
+      "logits/chosen": -12.25,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3008.0,
+      "loss": 0.022,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.1875,
+      "rewards/margins": 7.34375,
+      "rewards/rejected": -9.5,
       "step": 1030
     },
     {
       "epoch": 1.0890052356020943,
-      "grad_norm": 0.01264166895842069,
+      "grad_norm": 0.8980156145363332,
       "learning_rate": 4.019283848649643e-05,
-      "logits/chosen": 4.0625,
-      "logits/rejected": 4.0625,
-      "logps/chosen": -65024.0,
-      "logps/rejected": -65024.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -644.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -644.0,
+      "logits/chosen": -12.9375,
+      "logits/rejected": -10.625,
+      "logps/chosen": -696.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0351,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.375,
+      "rewards/margins": 7.75,
+      "rewards/rejected": -10.125,
       "step": 1040
     },
     {
       "epoch": 1.0994764397905759,
-      "grad_norm": 0.014835435035743876,
+      "grad_norm": 0.9939934691456076,
       "learning_rate": 3.9949773456075215e-05,
-      "logits/chosen": 3.6875,
-      "logits/rejected": 3.6875,
-      "logps/chosen": -62976.0,
-      "logps/rejected": -62976.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -624.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -624.0,
+      "logits/chosen": -12.875,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -764.0,
+      "logps/rejected": -3120.0,
+      "loss": 0.0218,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.34375,
+      "rewards/margins": 7.1875,
+      "rewards/rejected": -9.5,
       "step": 1050
     },
     {
       "epoch": 1.1099476439790577,
-      "grad_norm": 0.015406953613750432,
+      "grad_norm": 0.6896214407859734,
       "learning_rate": 3.970448837360585e-05,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "logps/chosen": -62976.0,
-      "logps/rejected": -62976.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -624.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -624.0,
+      "logits/chosen": -12.625,
+      "logits/rejected": -10.625,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3168.0,
+      "loss": 0.0232,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.296875,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -10.5,
       "step": 1060
     },
     {
       "epoch": 1.1204188481675392,
-      "grad_norm": 0.01615223213177818,
+      "grad_norm": 1.3521974326991437,
       "learning_rate": 3.945701966409846e-05,
-      "logits/chosen": 4.0625,
-      "logits/rejected": 4.0625,
-      "logps/chosen": -61952.0,
-      "logps/rejected": -61952.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -616.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -616.0,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.0346,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.734375,
+      "rewards/margins": 8.0625,
+      "rewards/rejected": -10.75,
       "step": 1070
     },
     {
       "epoch": 1.130890052356021,
-      "grad_norm": 0.015028142073122393,
+      "grad_norm": 0.18880758895126787,
       "learning_rate": 3.9207404076833374e-05,
-      "logits/chosen": 4.375,
-      "logits/rejected": 4.375,
-      "logps/chosen": -60416.0,
-      "logps/rejected": -60416.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -600.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -600.0,
+      "logits/chosen": -12.0,
+      "logits/rejected": -9.75,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0297,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.796875,
+      "rewards/margins": 7.40625,
+      "rewards/rejected": -10.1875,
       "step": 1080
     },
     {
       "epoch": 1.1413612565445026,
-      "grad_norm": 0.015445120326342732,
+      "grad_norm": 0.6157827170820976,
       "learning_rate": 3.895567867990379e-05,
-      "logits/chosen": 4.75,
-      "logits/rejected": 4.75,
-      "logps/chosen": -61952.0,
-      "logps/rejected": -61952.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -616.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -616.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -10.625,
+      "logps/chosen": -660.0,
+      "logps/rejected": -3168.0,
+      "loss": 0.0192,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.375,
+      "rewards/margins": 7.21875,
+      "rewards/rejected": -9.625,
       "step": 1090
     },
     {
       "epoch": 1.1518324607329844,
-      "grad_norm": 0.01695017149386086,
+      "grad_norm": 0.8362615091972094,
       "learning_rate": 3.870188085471114e-05,
-      "logits/chosen": 3.78125,
-      "logits/rejected": 3.78125,
-      "logps/chosen": -66048.0,
-      "logps/rejected": -66048.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -656.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -656.0,
+      "logits/chosen": -12.5625,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0293,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.390625,
+      "rewards/margins": 7.9375,
+      "rewards/rejected": -10.3125,
       "step": 1100
     },
     {
       "epoch": 1.1518324607329844,
-      "eval_logits/chosen": 6.0,
-      "eval_logits/rejected": 6.59375,
-      "eval_logps/chosen": -51200.0,
-      "eval_logps/rejected": -44288.0,
-      "eval_loss": 150.1531219482422,
-      "eval_rewards/accuracies": 0.4296875,
-      "eval_rewards/chosen": -510.0,
-      "eval_rewards/margins": -69.5,
-      "eval_rewards/rejected": -440.0,
-      "eval_runtime": 47.1765,
-      "eval_samples_per_second": 42.394,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -12.875,
+      "eval_logits/rejected": -12.4375,
+      "eval_logps/chosen": -544.0,
+      "eval_logps/rejected": -528.0,
+      "eval_loss": 0.8317187428474426,
+      "eval_rewards/accuracies": 0.548828125,
+      "eval_rewards/chosen": -2.265625,
+      "eval_rewards/margins": 0.1328125,
+      "eval_rewards/rejected": -2.390625,
+      "eval_runtime": 46.599,
+      "eval_samples_per_second": 42.919,
+      "eval_steps_per_second": 0.687,
       "step": 1100
     },
     {
       "epoch": 1.162303664921466,
-      "grad_norm": 0.01429499584037187,
+      "grad_norm": 0.09931936518209031,
       "learning_rate": 3.844604829041395e-05,
-      "logits/chosen": 4.46875,
-      "logits/rejected": 4.46875,
-      "logps/chosen": -65024.0,
-      "logps/rejected": -65024.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -648.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -648.0,
+      "logits/chosen": -12.875,
+      "logits/rejected": -10.75,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3152.0,
+      "loss": 0.026,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.421875,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -10.6875,
       "step": 1110
     },
     {
       "epoch": 1.1727748691099475,
-      "grad_norm": 0.016099040436650766,
+      "grad_norm": 0.18941580523761176,
       "learning_rate": 3.818821897833095e-05,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "logps/chosen": -70656.0,
-      "logps/rejected": -70656.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -700.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -700.0,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -10.375,
+      "logps/chosen": -768.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.029,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.625,
+      "rewards/margins": 7.59375,
+      "rewards/rejected": -10.1875,
       "step": 1120
     },
     {
       "epoch": 1.1832460732984293,
-      "grad_norm": 0.01730415250377742,
+      "grad_norm": 0.531097510461554,
       "learning_rate": 3.792843120629935e-05,
-      "logits/chosen": 4.625,
-      "logits/rejected": 4.625,
-      "logps/chosen": -62464.0,
-      "logps/rejected": -62464.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -620.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -620.0,
+      "logits/chosen": -12.5,
+      "logits/rejected": -10.375,
+      "logps/chosen": -600.0,
+      "logps/rejected": -3072.0,
+      "loss": 0.0324,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.125,
+      "rewards/margins": 7.625,
+      "rewards/rejected": -9.75,
       "step": 1130
     },
     {
       "epoch": 1.193717277486911,
-      "grad_norm": 0.017426177917885866,
+      "grad_norm": 0.12176198767356863,
       "learning_rate": 3.766672355298907e-05,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "logps/chosen": -61696.0,
-      "logps/rejected": -61696.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -612.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -612.0,
+      "logits/chosen": -12.375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -588.0,
+      "logps/rejected": -3120.0,
+      "loss": 0.0484,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -1.890625,
+      "rewards/margins": 7.375,
+      "rewards/rejected": -9.25,
       "step": 1140
     },
     {
       "epoch": 1.2041884816753927,
-      "grad_norm": 0.015604032330743407,
+      "grad_norm": 0.3992045723599849,
       "learning_rate": 3.740313488217373e-05,
-      "logits/chosen": 4.4375,
-      "logits/rejected": 4.4375,
-      "logps/chosen": -66048.0,
-      "logps/rejected": -66048.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -656.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -656.0,
+      "logits/chosen": -12.375,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -676.0,
+      "logps/rejected": -2960.0,
+      "loss": 0.0301,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.03125,
+      "rewards/margins": 6.4375,
+      "rewards/rejected": -8.5,
       "step": 1150
     },
     {
       "epoch": 1.2146596858638743,
-      "grad_norm": 0.01647191456010102,
+      "grad_norm": 0.19131973731463828,
       "learning_rate": 3.7137704336959464e-05,
-      "logits/chosen": 3.609375,
-      "logits/rejected": 3.609375,
-      "logps/chosen": -73216.0,
-      "logps/rejected": -73216.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -724.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -724.0,
+      "logits/chosen": -12.8125,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.0167,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.078125,
+      "rewards/margins": 6.75,
+      "rewards/rejected": -8.8125,
       "step": 1160
     },
     {
       "epoch": 1.225130890052356,
-      "grad_norm": 0.016264863841572302,
+      "grad_norm": 0.11470142938828308,
       "learning_rate": 3.687047133397201e-05,
-      "logits/chosen": 3.84375,
-      "logits/rejected": 3.84375,
-      "logps/chosen": -64512.0,
-      "logps/rejected": -64512.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -640.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -640.0,
+      "logits/chosen": -12.4375,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3088.0,
+      "loss": 0.0234,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.5,
+      "rewards/margins": 7.59375,
+      "rewards/rejected": -10.125,
       "step": 1170
     },
     {
       "epoch": 1.2356020942408377,
-      "grad_norm": 0.015478269279548113,
+      "grad_norm": 0.08832261340319174,
       "learning_rate": 3.6601475557503415e-05,
-      "logits/chosen": 3.53125,
-      "logits/rejected": 3.53125,
-      "logps/chosen": -63232.0,
-      "logps/rejected": -63232.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -628.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -628.0,
+      "logits/chosen": -12.5,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.0239,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.515625,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -10.6875,
       "step": 1180
     },
     {
       "epoch": 1.2460732984293195,
-      "grad_norm": 0.016117248575659635,
+      "grad_norm": 0.3449797385252149,
       "learning_rate": 3.633075695361882e-05,
-      "logits/chosen": 4.3125,
-      "logits/rejected": 4.3125,
-      "logps/chosen": -63488.0,
-      "logps/rejected": -63488.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -632.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -632.0,
+      "logits/chosen": -12.9375,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3072.0,
+      "loss": 0.0346,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.359375,
+      "rewards/margins": 7.5,
+      "rewards/rejected": -9.875,
       "step": 1190
     },
     {
       "epoch": 1.256544502617801,
-      "grad_norm": 0.015796628746679735,
+      "grad_norm": 0.06528359676207203,
       "learning_rate": 3.6058355724224477e-05,
-      "logits/chosen": 4.09375,
-      "logits/rejected": 4.09375,
-      "logps/chosen": -66048.0,
-      "logps/rejected": -66048.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -656.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -656.0,
+      "logits/chosen": -12.5625,
+      "logits/rejected": -9.875,
+      "logps/chosen": -632.0,
+      "logps/rejected": -3264.0,
+      "loss": 0.0171,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.234375,
+      "rewards/margins": 7.875,
+      "rewards/rejected": -10.125,
       "step": 1200
     },
     {
       "epoch": 1.256544502617801,
-      "eval_logits/chosen": 5.15625,
-      "eval_logits/rejected": 5.6875,
-      "eval_logps/chosen": -53504.0,
-      "eval_logps/rejected": -46336.0,
-      "eval_loss": 156.70574951171875,
-      "eval_rewards/accuracies": 0.4296875,
-      "eval_rewards/chosen": -532.0,
-      "eval_rewards/margins": -72.5,
-      "eval_rewards/rejected": -460.0,
-      "eval_runtime": 47.1867,
-      "eval_samples_per_second": 42.385,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -12.625,
+      "eval_logits/rejected": -12.1875,
+      "eval_logps/chosen": -556.0,
+      "eval_logps/rejected": -540.0,
+      "eval_loss": 0.8362187743186951,
+      "eval_rewards/accuracies": 0.533203125,
+      "eval_rewards/chosen": -2.375,
+      "eval_rewards/margins": 0.13671875,
+      "eval_rewards/rejected": -2.515625,
+      "eval_runtime": 46.5864,
+      "eval_samples_per_second": 42.931,
+      "eval_steps_per_second": 0.687,
       "step": 1200
     },
     {
       "epoch": 1.2670157068062826,
-      "grad_norm": 0.01477064851560785,
+      "grad_norm": 0.13409407316838193,
       "learning_rate": 3.5784312321097744e-05,
-      "logits/chosen": 3.1875,
-      "logits/rejected": 3.1875,
-      "logps/chosen": -71168.0,
-      "logps/rejected": -71168.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -704.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -704.0,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0197,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.53125,
+      "rewards/margins": 7.5625,
+      "rewards/rejected": -10.0625,
       "step": 1210
     },
     {
       "epoch": 1.2774869109947644,
-      "grad_norm": 0.01632438181472591,
+      "grad_norm": 0.2086770903320873,
       "learning_rate": 3.550866743987994e-05,
-      "logits/chosen": 4.03125,
-      "logits/rejected": 4.03125,
-      "logps/chosen": -66560.0,
-      "logps/rejected": -66560.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -660.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -660.0,
+      "logits/chosen": -12.3125,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -764.0,
+      "logps/rejected": -3184.0,
+      "loss": 0.0256,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -11.0625,
       "step": 1220
     },
     {
       "epoch": 1.2879581151832462,
-      "grad_norm": 0.016982026516076273,
+      "grad_norm": 4.179359658448476,
       "learning_rate": 3.5231462014033024e-05,
-      "logits/chosen": 3.59375,
-      "logits/rejected": 3.59375,
-      "logps/chosen": -73728.0,
-      "logps/rejected": -73728.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -732.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -732.0,
+      "logits/chosen": -12.375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -812.0,
+      "logps/rejected": -3104.0,
+      "loss": 0.0614,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 7.875,
+      "rewards/rejected": -11.0,
       "step": 1230
     },
     {
       "epoch": 1.2984293193717278,
-      "grad_norm": 0.01528994518254844,
+      "grad_norm": 0.3993866482932888,
       "learning_rate": 3.495273720876095e-05,
-      "logits/chosen": 4.03125,
-      "logits/rejected": 4.03125,
-      "logps/chosen": -66048.0,
-      "logps/rejected": -66048.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -656.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -656.0,
+      "logits/chosen": -12.1875,
+      "logits/rejected": -9.375,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.019,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.75,
+      "rewards/margins": 7.96875,
+      "rewards/rejected": -10.6875,
       "step": 1240
     },
     {
       "epoch": 1.3089005235602094,
-      "grad_norm": 0.014619185496747218,
+      "grad_norm": 1.3227606871810966,
       "learning_rate": 3.4672534414896616e-05,
-      "logits/chosen": 3.984375,
-      "logits/rejected": 3.984375,
-      "logps/chosen": -68096.0,
-      "logps/rejected": -68096.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -676.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -676.0,
+      "logits/chosen": -12.75,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -768.0,
+      "logps/rejected": -3168.0,
+      "loss": 0.0182,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -11.0,
       "step": 1250
     },
     {
       "epoch": 1.3193717277486912,
-      "grad_norm": 0.013580391295527268,
+      "grad_norm": 0.09648882989672453,
       "learning_rate": 3.439089524275526e-05,
-      "logits/chosen": 3.96875,
-      "logits/rejected": 3.96875,
-      "logps/chosen": -68096.0,
-      "logps/rejected": -68096.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -676.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -676.0,
+      "logits/chosen": -12.625,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.0256,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.65625,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -11.1875,
       "step": 1260
     },
     {
       "epoch": 1.3298429319371727,
-      "grad_norm": 0.0143591434363842,
+      "grad_norm": 0.8207102260091197,
       "learning_rate": 3.410786151595533e-05,
-      "logits/chosen": 3.5,
-      "logits/rejected": 3.5,
-      "logps/chosen": -69632.0,
-      "logps/rejected": -69632.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -688.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -688.0,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -10.0,
+      "logps/chosen": -748.0,
+      "logps/rejected": -3168.0,
+      "loss": 0.0352,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.765625,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -10.9375,
       "step": 1270
     },
     {
-      "epoch": 1.3403141361256545,
-      "grad_norm": 0.014786942512265854,
-      "learning_rate": 3.382347526520765e-05,
-      "logits/chosen": 3.75,
-      "logits/rejected": 3.75,
-      "logps/chosen": -64768.0,
-      "logps/rejected": -64768.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -640.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -640.0,
+      "epoch": 1.3403141361256545,
+      "grad_norm": 0.49631322875828615,
+      "learning_rate": 3.382347526520765e-05,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.0183,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 7.8125,
+      "rewards/rejected": -10.5625,
       "step": 1280
     },
     {
       "epoch": 1.350785340314136,
-      "grad_norm": 0.014772759159727642,
+      "grad_norm": 0.4694601062335547,
       "learning_rate": 3.353777872207381e-05,
-      "logits/chosen": 3.640625,
-      "logits/rejected": 3.640625,
-      "logps/chosen": -73728.0,
-      "logps/rejected": -73728.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -732.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -732.0,
+      "logits/chosen": -12.125,
+      "logits/rejected": -9.75,
+      "logps/chosen": -792.0,
+      "logps/rejected": -3264.0,
+      "loss": 0.0274,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.859375,
+      "rewards/margins": 7.90625,
+      "rewards/rejected": -10.75,
       "step": 1290
     },
     {
       "epoch": 1.3612565445026177,
-      "grad_norm": 0.016222674307981223,
+      "grad_norm": 0.4906816240186705,
       "learning_rate": 3.325081431269475e-05,
-      "logits/chosen": 4.03125,
-      "logits/rejected": 4.03125,
-      "logps/chosen": -64512.0,
-      "logps/rejected": -64512.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -640.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -640.0,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -9.875,
+      "logps/chosen": -700.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0174,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.546875,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -10.6875,
       "step": 1300
     },
     {
       "epoch": 1.3612565445026177,
-      "eval_logits/chosen": 5.15625,
-      "eval_logits/rejected": 5.59375,
-      "eval_logps/chosen": -55552.0,
-      "eval_logps/rejected": -48128.0,
-      "eval_loss": 162.70562744140625,
-      "eval_rewards/accuracies": 0.43359375,
-      "eval_rewards/chosen": -552.0,
-      "eval_rewards/margins": -75.0,
-      "eval_rewards/rejected": -476.0,
-      "eval_runtime": 47.1801,
-      "eval_samples_per_second": 42.391,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -12.25,
+      "eval_logits/rejected": -11.75,
+      "eval_logps/chosen": -612.0,
+      "eval_logps/rejected": -600.0,
+      "eval_loss": 0.8684218525886536,
+      "eval_rewards/accuracies": 0.51953125,
+      "eval_rewards/chosen": -2.9375,
+      "eval_rewards/margins": 0.166015625,
+      "eval_rewards/rejected": -3.09375,
+      "eval_runtime": 46.6049,
+      "eval_samples_per_second": 42.914,
+      "eval_steps_per_second": 0.687,
       "step": 1300
     },
     {
       "epoch": 1.3717277486910995,
-      "grad_norm": 0.012275173240251153,
+      "grad_norm": 0.5658517944851932,
       "learning_rate": 3.296262465149046e-05,
-      "logits/chosen": 4.5625,
-      "logits/rejected": 4.5625,
-      "logps/chosen": -64256.0,
-      "logps/rejected": -64256.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -636.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -636.0,
+      "logits/chosen": -12.1875,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -776.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0334,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.8125,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -11.125,
       "step": 1310
     },
     {
       "epoch": 1.3821989528795813,
-      "grad_norm": 0.015385598510831457,
+      "grad_norm": 0.37353285828126176,
       "learning_rate": 3.2673252534831687e-05,
-      "logits/chosen": 4.46875,
-      "logits/rejected": 4.46875,
-      "logps/chosen": -68096.0,
-      "logps/rejected": -68096.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -676.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -676.0,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0359,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.78125,
+      "rewards/margins": 7.65625,
+      "rewards/rejected": -10.4375,
       "step": 1320
     },
     {
       "epoch": 1.3926701570680629,
-      "grad_norm": 0.01683644997658926,
+      "grad_norm": 0.17740342198350953,
       "learning_rate": 3.2382740934684696e-05,
-      "logits/chosen": 3.875,
-      "logits/rejected": 3.875,
-      "logps/chosen": -67072.0,
-      "logps/rejected": -67072.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -664.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -664.0,
+      "logits/chosen": -11.6875,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3264.0,
+      "loss": 0.0213,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 8.125,
+      "rewards/rejected": -11.0625,
       "step": 1330
     },
     {
       "epoch": 1.4031413612565444,
-      "grad_norm": 0.014943856848419524,
+      "grad_norm": 1.2932760724936367,
       "learning_rate": 3.209113299222982e-05,
-      "logits/chosen": 4.09375,
-      "logits/rejected": 4.09375,
-      "logps/chosen": -69120.0,
-      "logps/rejected": -69120.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -688.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -688.0,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3264.0,
+      "loss": 0.0406,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -11.875,
       "step": 1340
     },
     {
       "epoch": 1.4136125654450262,
-      "grad_norm": 0.015026821742916792,
+      "grad_norm": 0.677287622358092,
       "learning_rate": 3.179847201145505e-05,
-      "logits/chosen": 3.484375,
-      "logits/rejected": 3.484375,
-      "logps/chosen": -74240.0,
-      "logps/rejected": -74240.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -736.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -736.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.25,
+      "logps/chosen": -720.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0371,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.625,
+      "rewards/margins": 7.84375,
+      "rewards/rejected": -10.5,
       "step": 1350
     },
     {
       "epoch": 1.4240837696335078,
-      "grad_norm": 0.013242095025126363,
+      "grad_norm": 0.16190054322062475,
       "learning_rate": 3.150480145272528e-05,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "logps/chosen": -65536.0,
-      "logps/rejected": -65536.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -648.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -648.0,
+      "logits/chosen": -12.625,
+      "logits/rejected": -10.125,
+      "logps/chosen": -760.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0159,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.609375,
+      "rewards/margins": 6.65625,
+      "rewards/rejected": -9.25,
       "step": 1360
     },
     {
       "epoch": 1.4345549738219896,
-      "grad_norm": 0.015337059510861211,
+      "grad_norm": 0.2308589290080698,
       "learning_rate": 3.121016492632848e-05,
-      "logits/chosen": 3.5625,
-      "logits/rejected": 3.5625,
-      "logps/chosen": -72192.0,
-      "logps/rejected": -72192.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -716.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -716.0,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -10.0,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3136.0,
+      "loss": 0.0173,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.28125,
+      "rewards/margins": 7.71875,
+      "rewards/rejected": -10.0,
       "step": 1370
     },
     {
       "epoch": 1.4450261780104712,
-      "grad_norm": 0.012741842436650665,
+      "grad_norm": 0.22275028531554653,
       "learning_rate": 3.091460618599951e-05,
-      "logits/chosen": 3.8125,
-      "logits/rejected": 3.8125,
-      "logps/chosen": -71680.0,
-      "logps/rejected": -71680.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -712.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -712.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -9.75,
+      "logps/chosen": -688.0,
+      "logps/rejected": -3360.0,
+      "loss": 0.0199,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.5,
+      "rewards/margins": 8.75,
+      "rewards/rejected": -11.25,
       "step": 1380
     },
     {
       "epoch": 1.4554973821989527,
-      "grad_norm": 0.01577790430062567,
+      "grad_norm": 0.4557083934840219,
       "learning_rate": 3.061816912242265e-05,
-      "logits/chosen": 3.90625,
-      "logits/rejected": 3.90625,
-      "logps/chosen": -68096.0,
-      "logps/rejected": -68096.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -676.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -676.0,
+      "logits/chosen": -12.625,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0196,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.4375,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -11.9375,
       "step": 1390
     },
     {
       "epoch": 1.4659685863874345,
-      "grad_norm": 0.011971722796064059,
+      "grad_norm": 1.0326552236672093,
       "learning_rate": 3.0320897756713785e-05,
-      "logits/chosen": 3.84375,
-      "logits/rejected": 3.84375,
-      "logps/chosen": -74240.0,
-      "logps/rejected": -74240.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -740.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -740.0,
+      "logits/chosen": -12.4375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3328.0,
+      "loss": 0.0191,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.5,
+      "rewards/margins": 8.9375,
+      "rewards/rejected": -11.4375,
       "step": 1400
     },
     {
       "epoch": 1.4659685863874345,
-      "eval_logits/chosen": 5.34375,
-      "eval_logits/rejected": 5.84375,
-      "eval_logps/chosen": -57600.0,
-      "eval_logps/rejected": -49664.0,
-      "eval_loss": 168.47438049316406,
-      "eval_rewards/accuracies": 0.431640625,
-      "eval_rewards/chosen": -572.0,
-      "eval_rewards/margins": -77.5,
-      "eval_rewards/rejected": -494.0,
-      "eval_runtime": 47.1746,
-      "eval_samples_per_second": 42.396,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -12.0,
+      "eval_logits/rejected": -11.5625,
+      "eval_logps/chosen": -588.0,
+      "eval_logps/rejected": -572.0,
+      "eval_loss": 0.8702499866485596,
+      "eval_rewards/accuracies": 0.515625,
+      "eval_rewards/chosen": -2.6875,
+      "eval_rewards/margins": 0.150390625,
+      "eval_rewards/rejected": -2.84375,
+      "eval_runtime": 46.5939,
+      "eval_samples_per_second": 42.924,
+      "eval_steps_per_second": 0.687,
       "step": 1400
     },
     {
       "epoch": 1.4764397905759163,
-      "grad_norm": 0.013137939151115312,
+      "grad_norm": 0.6840147426394543,
       "learning_rate": 3.002283623388332e-05,
-      "logits/chosen": 4.21875,
-      "logits/rejected": 4.21875,
-      "logps/chosen": -67072.0,
-      "logps/rejected": -67072.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -664.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -664.0,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0255,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.59375,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -11.4375,
       "step": 1410
     },
     {
       "epoch": 1.486910994764398,
-      "grad_norm": 0.013606904924170415,
+      "grad_norm": 0.09885297804480472,
       "learning_rate": 2.9724028816280507e-05,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "logps/chosen": -68096.0,
-      "logps/rejected": -68096.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -676.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -676.0,
+      "logits/chosen": -11.6875,
+      "logits/rejected": -9.625,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3392.0,
+      "loss": 0.0139,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.5625,
+      "rewards/margins": 9.4375,
+      "rewards/rejected": -12.0,
       "step": 1420
     },
     {
       "epoch": 1.4973821989528795,
-      "grad_norm": 0.01259196291056396,
+      "grad_norm": 0.5447267084878693,
       "learning_rate": 2.9424519877020524e-05,
-      "logits/chosen": 4.0625,
-      "logits/rejected": 4.0625,
-      "logps/chosen": -67072.0,
-      "logps/rejected": -67072.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -664.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -664.0,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -9.875,
+      "logps/chosen": -664.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0351,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.515625,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -11.375,
       "step": 1430
     },
     {
       "epoch": 1.5078534031413613,
-      "grad_norm": 0.011134596774090143,
+      "grad_norm": 1.7612852542771738,
       "learning_rate": 2.912435389339504e-05,
-      "logits/chosen": 4.09375,
-      "logits/rejected": 4.09375,
-      "logps/chosen": -73728.0,
-      "logps/rejected": -73728.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -732.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -732.0,
+      "logits/chosen": -12.0,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -636.0,
+      "logps/rejected": -3360.0,
+      "loss": 0.033,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.546875,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -11.75,
       "step": 1440
     },
     {
       "epoch": 1.518324607329843,
-      "grad_norm": 0.012428707001269383,
+      "grad_norm": 0.21797196230306862,
       "learning_rate": 2.8823575440267226e-05,
-      "logits/chosen": 4.03125,
-      "logits/rejected": 4.03125,
-      "logps/chosen": -69632.0,
-      "logps/rejected": -69632.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -692.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -692.0,
+      "logits/chosen": -12.5,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.019,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.359375,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -10.625,
       "step": 1450
     },
     {
       "epoch": 1.5287958115183247,
-      "grad_norm": 0.013061485166396302,
+      "grad_norm": 0.1398275390233516,
       "learning_rate": 2.8522229183452465e-05,
-      "logits/chosen": 4.0625,
-      "logits/rejected": 4.0625,
-      "logps/chosen": -69632.0,
-      "logps/rejected": -69632.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -688.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -688.0,
+      "logits/chosen": -12.4375,
+      "logits/rejected": -9.5,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0184,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.359375,
+      "rewards/margins": 8.4375,
+      "rewards/rejected": -10.75,
       "step": 1460
     },
     {
       "epoch": 1.5392670157068062,
-      "grad_norm": 0.014320515428634304,
+      "grad_norm": 0.07480740506899823,
       "learning_rate": 2.8220359873085373e-05,
-      "logits/chosen": 3.546875,
-      "logits/rejected": 3.546875,
-      "logps/chosen": -75776.0,
-      "logps/rejected": -75776.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -752.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -752.0,
+      "logits/chosen": -12.0,
+      "logits/rejected": -9.75,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0397,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.78125,
+      "rewards/margins": 7.90625,
+      "rewards/rejected": -10.6875,
       "step": 1470
     },
     {
       "epoch": 1.5497382198952878,
-      "grad_norm": 0.013196013818110143,
+      "grad_norm": 0.242912155955285,
       "learning_rate": 2.7918012336974388e-05,
-      "logits/chosen": 3.84375,
-      "logits/rejected": 3.84375,
-      "logps/chosen": -75776.0,
-      "logps/rejected": -75776.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -752.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -752.0,
+      "logits/chosen": -11.25,
+      "logits/rejected": -9.125,
+      "logps/chosen": -816.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0257,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.75,
+      "rewards/margins": 7.875,
+      "rewards/rejected": -10.625,
       "step": 1480
     },
     {
       "epoch": 1.5602094240837696,
-      "grad_norm": 0.012857422444597541,
+      "grad_norm": 0.7594533324298329,
       "learning_rate": 2.7615231473944815e-05,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "logps/chosen": -69120.0,
-      "logps/rejected": -69120.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -684.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -684.0,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -684.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0282,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.484375,
+      "rewards/margins": 7.96875,
+      "rewards/rejected": -10.4375,
       "step": 1490
     },
     {
       "epoch": 1.5706806282722514,
-      "grad_norm": 0.01140621673323182,
+      "grad_norm": 1.6055276260680513,
       "learning_rate": 2.7312062247171327e-05,
-      "logits/chosen": 4.40625,
-      "logits/rejected": 4.40625,
-      "logps/chosen": -73728.0,
-      "logps/rejected": -73728.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -732.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -732.0,
+      "logits/chosen": -11.6875,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.036,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.734375,
+      "rewards/margins": 7.4375,
+      "rewards/rejected": -10.125,
       "step": 1500
     },
     {
       "epoch": 1.5706806282722514,
-      "eval_logits/chosen": 5.53125,
-      "eval_logits/rejected": 6.03125,
-      "eval_logps/chosen": -59392.0,
-      "eval_logps/rejected": -51200.0,
-      "eval_loss": 173.848876953125,
-      "eval_rewards/accuracies": 0.4296875,
-      "eval_rewards/chosen": -588.0,
-      "eval_rewards/margins": -80.0,
-      "eval_rewards/rejected": -510.0,
-      "eval_runtime": 47.1795,
-      "eval_samples_per_second": 42.391,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -11.8125,
+      "eval_logits/rejected": -11.3125,
+      "eval_logps/chosen": -576.0,
+      "eval_logps/rejected": -560.0,
+      "eval_loss": 0.9228906035423279,
+      "eval_rewards/accuracies": 0.509765625,
+      "eval_rewards/chosen": -2.578125,
+      "eval_rewards/margins": 0.134765625,
+      "eval_rewards/rejected": -2.71875,
+      "eval_runtime": 46.5961,
+      "eval_samples_per_second": 42.922,
+      "eval_steps_per_second": 0.687,
       "step": 1500
     },
     {
       "epoch": 1.581151832460733,
-      "grad_norm": 0.012860782489733298,
+      "grad_norm": 0.44937458978019706,
       "learning_rate": 2.7008549677500876e-05,
-      "logits/chosen": 3.484375,
-      "logits/rejected": 3.484375,
-      "logps/chosen": -82944.0,
-      "logps/rejected": -82944.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -824.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -824.0,
+      "logits/chosen": -12.4375,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0131,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 7.625,
+      "rewards/rejected": -10.5,
       "step": 1510
     },
     {
       "epoch": 1.5916230366492146,
-      "grad_norm": 0.012427973848508278,
+      "grad_norm": 0.13225735658196988,
       "learning_rate": 2.670473883676709e-05,
-      "logits/chosen": 4.09375,
-      "logits/rejected": 4.09375,
-      "logps/chosen": -80384.0,
-      "logps/rejected": -80384.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -800.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -800.0,
+      "logits/chosen": -12.25,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3264.0,
+      "loss": 0.0257,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.734375,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -11.0,
       "step": 1520
     },
     {
       "epoch": 1.6020942408376964,
-      "grad_norm": 0.013442569573618494,
+      "grad_norm": 0.3470754497933429,
       "learning_rate": 2.6400674841097074e-05,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "logps/chosen": -74240.0,
-      "logps/rejected": -74240.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -736.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -736.0,
+      "logits/chosen": -12.0,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0232,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.75,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -11.0,
       "step": 1530
     },
     {
       "epoch": 1.6125654450261782,
-      "grad_norm": 0.014009662433481458,
+      "grad_norm": 0.31904725186480803,
       "learning_rate": 2.609640284421158e-05,
-      "logits/chosen": 4.4375,
-      "logits/rejected": 4.4375,
-      "logps/chosen": -78848.0,
-      "logps/rejected": -78848.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -784.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -784.0,
+      "logits/chosen": -11.75,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.0205,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 7.6875,
+      "rewards/rejected": -10.625,
       "step": 1540
     },
     {
       "epoch": 1.6230366492146597,
-      "grad_norm": 0.010657449204283307,
+      "grad_norm": 0.3882924340654773,
       "learning_rate": 2.5791968030719726e-05,
-      "logits/chosen": 3.921875,
-      "logits/rejected": 3.921875,
-      "logps/chosen": -75776.0,
-      "logps/rejected": -75776.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -752.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -752.0,
+      "logits/chosen": -11.25,
+      "logits/rejected": -9.375,
+      "logps/chosen": -768.0,
+      "logps/rejected": -3200.0,
+      "loss": 0.0295,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 8.0,
+      "rewards/rejected": -11.1875,
       "step": 1550
     },
     {
       "epoch": 1.6335078534031413,
-      "grad_norm": 0.011906504359558066,
+      "grad_norm": 0.1078678678208724,
       "learning_rate": 2.548741560940902e-05,
-      "logits/chosen": 4.21875,
-      "logits/rejected": 4.21875,
-      "logps/chosen": -74752.0,
-      "logps/rejected": -74752.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -740.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -740.0,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -856.0,
+      "logps/rejected": -3264.0,
+      "loss": 0.0299,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.25,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -11.5625,
       "step": 1560
     },
     {
       "epoch": 1.6439790575916229,
-      "grad_norm": 0.01300327047369842,
+      "grad_norm": 1.6353071104324566,
       "learning_rate": 2.5182790806531787e-05,
-      "logits/chosen": 4.40625,
-      "logits/rejected": 4.40625,
-      "logps/chosen": -80896.0,
-      "logps/rejected": -80896.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -804.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -804.0,
+      "logits/chosen": -12.3125,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -768.0,
+      "logps/rejected": -3392.0,
+      "loss": 0.0166,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -11.6875,
       "step": 1570
     },
     {
       "epoch": 1.6544502617801047,
-      "grad_norm": 0.010903433666418192,
+      "grad_norm": 0.5004458601183499,
       "learning_rate": 2.4878138859089074e-05,
-      "logits/chosen": 4.34375,
-      "logits/rejected": 4.34375,
-      "logps/chosen": -69120.0,
-      "logps/rejected": -69120.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -684.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -684.0,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -9.125,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0272,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -11.25,
       "step": 1580
     },
     {
       "epoch": 1.6649214659685865,
-      "grad_norm": 0.013386936611589588,
+      "grad_norm": 0.2045399227444437,
       "learning_rate": 2.4573505008112924e-05,
-      "logits/chosen": 3.828125,
-      "logits/rejected": 3.828125,
-      "logps/chosen": -75776.0,
-      "logps/rejected": -75776.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -752.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -752.0,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -8.625,
+      "logps/chosen": -764.0,
+      "logps/rejected": -3408.0,
+      "loss": 0.0183,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 8.4375,
+      "rewards/rejected": -11.4375,
       "step": 1590
     },
     {
       "epoch": 1.675392670157068,
-      "grad_norm": 0.012631366931526493,
+      "grad_norm": 0.44927815319300485,
       "learning_rate": 2.4268934491948032e-05,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "logps/chosen": -76800.0,
-      "logps/rejected": -76800.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -764.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -764.0,
+      "logits/chosen": -12.1875,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3088.0,
+      "loss": 0.018,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.6875,
+      "rewards/margins": 8.0,
+      "rewards/rejected": -10.6875,
       "step": 1600
     },
     {
       "epoch": 1.675392670157068,
-      "eval_logits/chosen": 5.375,
-      "eval_logits/rejected": 5.84375,
-      "eval_logps/chosen": -60928.0,
-      "eval_logps/rejected": -52736.0,
-      "eval_loss": 178.62261962890625,
-      "eval_rewards/accuracies": 0.43359375,
-      "eval_rewards/chosen": -608.0,
-      "eval_rewards/margins": -82.5,
-      "eval_rewards/rejected": -524.0,
-      "eval_runtime": 47.1632,
-      "eval_samples_per_second": 42.406,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -12.875,
+      "eval_logits/rejected": -12.4375,
+      "eval_logps/chosen": -600.0,
+      "eval_logps/rejected": -588.0,
+      "eval_loss": 0.9153749942779541,
+      "eval_rewards/accuracies": 0.505859375,
+      "eval_rewards/chosen": -2.828125,
+      "eval_rewards/margins": 0.15234375,
+      "eval_rewards/rejected": -2.984375,
+      "eval_runtime": 46.5864,
+      "eval_samples_per_second": 42.931,
+      "eval_steps_per_second": 0.687,
       "step": 1600
     },
     {
       "epoch": 1.6858638743455496,
-      "grad_norm": 0.010101258098591628,
+      "grad_norm": 0.4102090929634209,
       "learning_rate": 2.396447253953385e-05,
-      "logits/chosen": 4.34375,
-      "logits/rejected": 4.34375,
-      "logps/chosen": -74240.0,
-      "logps/rejected": -74240.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -740.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -740.0,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3328.0,
+      "loss": 0.0129,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.875,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -11.5625,
       "step": 1610
     },
     {
       "epoch": 1.6963350785340314,
-      "grad_norm": 0.011847815275036772,
+      "grad_norm": 0.03945129756196896,
       "learning_rate": 2.3660164363687998e-05,
-      "logits/chosen": 3.578125,
-      "logits/rejected": 3.578125,
-      "logps/chosen": -80896.0,
-      "logps/rejected": -80896.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -804.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -804.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -800.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.0575,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -11.5,
       "step": 1620
     },
     {
       "epoch": 1.7068062827225132,
-      "grad_norm": 0.011194593716480038,
+      "grad_norm": 0.9283966646541102,
       "learning_rate": 2.33560551543922e-05,
-      "logits/chosen": 3.84375,
-      "logits/rejected": 3.84375,
-      "logps/chosen": -75264.0,
-      "logps/rejected": -75264.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -748.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -748.0,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -10.125,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.0496,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.703125,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -11.5,
       "step": 1630
     },
     {
       "epoch": 1.7172774869109948,
-      "grad_norm": 0.010636784457652712,
+      "grad_norm": 1.3535331917454998,
       "learning_rate": 2.3052190072081493e-05,
-      "logits/chosen": 3.9375,
-      "logits/rejected": 3.9375,
-      "logps/chosen": -74240.0,
-      "logps/rejected": -74240.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -736.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -736.0,
+      "logits/chosen": -13.25,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -772.0,
+      "logps/rejected": -3264.0,
+      "loss": 0.0401,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 7.9375,
+      "rewards/rejected": -10.875,
       "step": 1640
     },
     {
       "epoch": 1.7277486910994764,
-      "grad_norm": 0.012215213712601652,
+      "grad_norm": 1.3273003659694136,
       "learning_rate": 2.2748614240937864e-05,
-      "logits/chosen": 3.765625,
-      "logits/rejected": 3.765625,
-      "logps/chosen": -80896.0,
-      "logps/rejected": -80896.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -804.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -804.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -748.0,
+      "logps/rejected": -3072.0,
+      "loss": 0.0446,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.6875,
+      "rewards/margins": 7.53125,
+      "rewards/rejected": -10.1875,
       "step": 1650
     },
     {
       "epoch": 1.738219895287958,
-      "grad_norm": 0.013503379462193472,
+      "grad_norm": 0.23019286697417038,
       "learning_rate": 2.2445372742189334e-05,
-      "logits/chosen": 3.890625,
-      "logits/rejected": 3.890625,
-      "logps/chosen": -77312.0,
-      "logps/rejected": -77312.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -768.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -768.0,
+      "logits/chosen": -13.0,
+      "logits/rejected": -9.5,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0139,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.609375,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -10.75,
       "step": 1660
     },
     {
       "epoch": 1.7486910994764397,
-      "grad_norm": 0.01306072185484255,
+      "grad_norm": 0.43006415236324225,
       "learning_rate": 2.214251060741528e-05,
-      "logits/chosen": 4.125,
-      "logits/rejected": 4.125,
-      "logps/chosen": -79360.0,
-      "logps/rejected": -79360.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -784.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -784.0,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3296.0,
+      "loss": 0.0427,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.75,
+      "rewards/margins": 7.90625,
+      "rewards/rejected": -10.625,
       "step": 1670
     },
     {
       "epoch": 1.7591623036649215,
-      "grad_norm": 0.011592390154166306,
+      "grad_norm": 0.2619475218293731,
       "learning_rate": 2.1840072811859302e-05,
-      "logits/chosen": 3.5,
-      "logits/rejected": 3.5,
-      "logps/chosen": -86016.0,
-      "logps/rejected": -86016.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -856.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -856.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -9.625,
+      "logps/chosen": -748.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.0285,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.5,
+      "rewards/margins": 6.9375,
+      "rewards/rejected": -9.4375,
       "step": 1680
     },
     {
       "epoch": 1.7696335078534031,
-      "grad_norm": 0.011153049812459572,
+      "grad_norm": 0.11053365547409333,
       "learning_rate": 2.1538104267750287e-05,
-      "logits/chosen": 4.21875,
-      "logits/rejected": 4.21875,
-      "logps/chosen": -78336.0,
-      "logps/rejected": -78336.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -780.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -780.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -9.5,
+      "logps/chosen": -632.0,
+      "logps/rejected": -3312.0,
+      "loss": 0.0115,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.265625,
+      "rewards/margins": 7.875,
+      "rewards/rejected": -10.125,
       "step": 1690
     },
     {
       "epoch": 1.7801047120418847,
-      "grad_norm": 0.010350298495160943,
+      "grad_norm": 0.25584214657604637,
       "learning_rate": 2.1236649817632952e-05,
-      "logits/chosen": 3.890625,
-      "logits/rejected": 3.890625,
-      "logps/chosen": -76800.0,
-      "logps/rejected": -76800.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -764.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -764.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.125,
+      "logps/chosen": -680.0,
+      "logps/rejected": -3264.0,
+      "loss": 0.0304,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.515625,
+      "rewards/margins": 7.96875,
+      "rewards/rejected": -10.5,
       "step": 1700
     },
     {
       "epoch": 1.7801047120418847,
-      "eval_logits/chosen": 5.3125,
-      "eval_logits/rejected": 5.78125,
-      "eval_logps/chosen": -62464.0,
-      "eval_logps/rejected": -54016.0,
-      "eval_loss": 183.2412567138672,
-      "eval_rewards/accuracies": 0.4296875,
-      "eval_rewards/chosen": -620.0,
-      "eval_rewards/margins": -84.5,
-      "eval_rewards/rejected": -536.0,
-      "eval_runtime": 47.1822,
-      "eval_samples_per_second": 42.389,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -13.75,
+      "eval_logits/rejected": -13.5,
+      "eval_logps/chosen": -556.0,
+      "eval_logps/rejected": -536.0,
+      "eval_loss": 0.9087187647819519,
+      "eval_rewards/accuracies": 0.490234375,
+      "eval_rewards/chosen": -2.375,
+      "eval_rewards/margins": 0.08203125,
+      "eval_rewards/rejected": -2.453125,
+      "eval_runtime": 46.5775,
+      "eval_samples_per_second": 42.939,
+      "eval_steps_per_second": 0.687,
       "step": 1700
     },
     {
       "epoch": 1.7905759162303665,
-      "grad_norm": 0.011168945630210973,
+      "grad_norm": 0.9648087990007245,
       "learning_rate": 2.093575422770872e-05,
-      "logits/chosen": 4.4375,
-      "logits/rejected": 4.4375,
-      "logps/chosen": -77312.0,
-      "logps/rejected": -77312.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -768.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -768.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3168.0,
+      "loss": 0.0212,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.53125,
+      "rewards/margins": 7.96875,
+      "rewards/rejected": -10.5,
       "step": 1710
     },
     {
       "epoch": 1.8010471204188483,
-      "grad_norm": 0.01149126837864788,
+      "grad_norm": 1.0081394718938281,
       "learning_rate": 2.0635462181187827e-05,
-      "logits/chosen": 4.0625,
-      "logits/rejected": 4.0625,
-      "logps/chosen": -75264.0,
-      "logps/rejected": -75264.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -748.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -748.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3360.0,
+      "loss": 0.0172,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.59375,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -11.25,
       "step": 1720
     },
     {
       "epoch": 1.8115183246073299,
-      "grad_norm": 0.010566954173902991,
+      "grad_norm": 0.18797543737430802,
       "learning_rate": 2.0335818271653917e-05,
-      "logits/chosen": 4.03125,
-      "logits/rejected": 4.03125,
-      "logps/chosen": -72704.0,
-      "logps/rejected": -72704.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -720.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -720.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -648.0,
+      "logps/rejected": -3328.0,
+      "loss": 0.0221,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.484375,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -11.5,
       "step": 1730
     },
     {
       "epoch": 1.8219895287958114,
-      "grad_norm": 0.012736091705353916,
+      "grad_norm": 0.16860467639320212,
       "learning_rate": 2.0036866996441815e-05,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "logps/chosen": -73728.0,
-      "logps/rejected": -73728.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -732.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -732.0,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.025,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.65625,
+      "rewards/margins": 8.375,
+      "rewards/rejected": -11.0625,
       "step": 1740
     },
     {
       "epoch": 1.8324607329842932,
-      "grad_norm": 0.01123246873131985,
+      "grad_norm": 0.9329828424413206,
       "learning_rate": 1.9738652750029677e-05,
-      "logits/chosen": 3.953125,
-      "logits/rejected": 3.953125,
-      "logps/chosen": -79872.0,
-      "logps/rejected": -79872.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -796.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -796.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -764.0,
+      "logps/rejected": -3136.0,
+      "loss": 0.0179,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.828125,
+      "rewards/margins": 7.59375,
+      "rewards/rejected": -10.4375,
       "step": 1750
     },
     {
       "epoch": 1.8429319371727748,
-      "grad_norm": 0.012370469201444162,
+      "grad_norm": 0.09630025226284178,
       "learning_rate": 1.944121981744631e-05,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "logps/chosen": -70656.0,
-      "logps/rejected": -70656.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -704.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -704.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3216.0,
+      "loss": 0.0186,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.5,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -11.125,
       "step": 1760
     },
     {
       "epoch": 1.8534031413612566,
-      "grad_norm": 0.010795735367592214,
+      "grad_norm": 0.29770614579765814,
       "learning_rate": 1.9144612367694887e-05,
-      "logits/chosen": 3.96875,
-      "logits/rejected": 3.96875,
-      "logps/chosen": -70144.0,
-      "logps/rejected": -70144.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -696.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -696.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -692.0,
+      "logps/rejected": -3104.0,
+      "loss": 0.0231,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.5,
+      "rewards/margins": 8.0,
+      "rewards/rejected": -10.5,
       "step": 1770
     },
     {
       "epoch": 1.8638743455497382,
-      "grad_norm": 0.011397248452403063,
+      "grad_norm": 0.5425820005935259,
       "learning_rate": 1.8848874447193803e-05,
-      "logits/chosen": 4.09375,
-      "logits/rejected": 4.09375,
-      "logps/chosen": -80384.0,
-      "logps/rejected": -80384.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -800.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -800.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3232.0,
+      "loss": 0.0115,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.59375,
+      "rewards/margins": 8.5625,
+      "rewards/rejected": -11.125,
       "step": 1780
     },
     {
       "epoch": 1.8743455497382198,
-      "grad_norm": 0.013497064247927699,
+      "grad_norm": 0.647072199849212,
       "learning_rate": 1.855404997323571e-05,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "logps/chosen": -81408.0,
-      "logps/rejected": -81408.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -808.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -808.0,
+      "logits/chosen": -13.75,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0163,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.78125,
+      "rewards/margins": 8.9375,
+      "rewards/rejected": -11.75,
       "step": 1790
     },
     {
       "epoch": 1.8848167539267016,
-      "grad_norm": 0.01385941341057707,
+      "grad_norm": 0.5988047084665278,
       "learning_rate": 1.82601827274658e-05,
-      "logits/chosen": 3.78125,
-      "logits/rejected": 3.78125,
-      "logps/chosen": -81408.0,
-      "logps/rejected": -81408.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -808.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -808.0,
+      "logits/chosen": -13.0,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3440.0,
+      "loss": 0.0189,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 8.4375,
+      "rewards/rejected": -11.25,
       "step": 1800
     },
     {
       "epoch": 1.8848167539267016,
-      "eval_logits/chosen": 5.15625,
-      "eval_logits/rejected": 5.59375,
-      "eval_logps/chosen": -64000.0,
-      "eval_logps/rejected": -55040.0,
-      "eval_loss": 186.8874969482422,
-      "eval_rewards/accuracies": 0.427734375,
-      "eval_rewards/chosen": -636.0,
-      "eval_rewards/margins": -86.0,
-      "eval_rewards/rejected": -548.0,
-      "eval_runtime": 47.1778,
-      "eval_samples_per_second": 42.393,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -13.375,
+      "eval_logits/rejected": -13.0,
+      "eval_logps/chosen": -584.0,
+      "eval_logps/rejected": -568.0,
+      "eval_loss": 0.8958749771118164,
+      "eval_rewards/accuracies": 0.50390625,
+      "eval_rewards/chosen": -2.671875,
+      "eval_rewards/margins": 0.13671875,
+      "eval_rewards/rejected": -2.8125,
+      "eval_runtime": 46.5907,
+      "eval_samples_per_second": 42.927,
+      "eval_steps_per_second": 0.687,
       "step": 1800
     },
     {
       "epoch": 1.8952879581151834,
-      "grad_norm": 0.013632010028096435,
+      "grad_norm": 0.17098314735531286,
       "learning_rate": 1.7967316349380213e-05,
-      "logits/chosen": 3.953125,
-      "logits/rejected": 3.953125,
-      "logps/chosen": -84992.0,
-      "logps/rejected": -84992.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -844.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -844.0,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3120.0,
+      "loss": 0.0213,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.796875,
+      "rewards/margins": 7.625,
+      "rewards/rejected": -10.4375,
       "step": 1810
     },
     {
       "epoch": 1.905759162303665,
-      "grad_norm": 0.01223874157528922,
+      "grad_norm": 0.06935447293402892,
       "learning_rate": 1.7675494329845516e-05,
-      "logits/chosen": 3.90625,
-      "logits/rejected": 3.90625,
-      "logps/chosen": -78336.0,
-      "logps/rejected": -78336.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -776.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -776.0,
+      "logits/chosen": -12.875,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3360.0,
+      "loss": 0.029,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.921875,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -11.5,
       "step": 1820
     },
     {
       "epoch": 1.9162303664921465,
-      "grad_norm": 0.010980815497028295,
+      "grad_norm": 0.11739092594291739,
       "learning_rate": 1.738476000464026e-05,
-      "logits/chosen": 4.0,
-      "logits/rejected": 4.0,
-      "logps/chosen": -75264.0,
-      "logps/rejected": -75264.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -748.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -748.0,
+      "logits/chosen": -12.8125,
+      "logits/rejected": -9.375,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3376.0,
+      "loss": 0.0351,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 8.75,
+      "rewards/rejected": -11.625,
       "step": 1830
     },
     {
       "epoch": 1.9267015706806283,
-      "grad_norm": 0.011549320010919203,
+      "grad_norm": 0.1838408720062665,
       "learning_rate": 1.7095156548019646e-05,
-      "logits/chosen": 4.0625,
-      "logits/rejected": 4.0625,
-      "logps/chosen": -83456.0,
-      "logps/rejected": -83456.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -828.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -828.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -760.0,
+      "logps/rejected": -3328.0,
+      "loss": 0.0229,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -11.5,
       "step": 1840
     },
     {
       "epoch": 1.93717277486911,
-      "grad_norm": 0.011799740032443851,
+      "grad_norm": 0.28910958447459284,
       "learning_rate": 1.6806726966304064e-05,
-      "logits/chosen": 4.09375,
-      "logits/rejected": 4.09375,
-      "logps/chosen": -77312.0,
-      "logps/rejected": -77312.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -768.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -768.0,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3248.0,
+      "loss": 0.0405,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -10.875,
       "step": 1850
     },
     {
       "epoch": 1.9476439790575917,
-      "grad_norm": 0.011719887052809977,
+      "grad_norm": 1.2674093527891848,
       "learning_rate": 1.6519514091492625e-05,
-      "logits/chosen": 3.890625,
-      "logits/rejected": 3.890625,
-      "logps/chosen": -76800.0,
-      "logps/rejected": -76800.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -764.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -764.0,
+      "logits/chosen": -12.875,
+      "logits/rejected": -9.75,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3344.0,
+      "loss": 0.0258,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.65625,
+      "rewards/margins": 7.84375,
+      "rewards/rejected": -10.5,
       "step": 1860
     },
     {
       "epoch": 1.9581151832460733,
-      "grad_norm": 0.015257754897417871,
+      "grad_norm": 0.9413467962900552,
       "learning_rate": 1.6233560574902627e-05,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "logps/chosen": -86016.0,
-      "logps/rejected": -86016.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -852.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -852.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -768.0,
+      "logps/rejected": -3040.0,
+      "loss": 0.0235,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.78125,
+      "rewards/margins": 7.71875,
+      "rewards/rejected": -10.5,
       "step": 1870
     },
     {
       "epoch": 1.9685863874345548,
-      "grad_norm": 0.015070117851695488,
+      "grad_norm": 0.6240464515354742,
       "learning_rate": 1.5948908880835753e-05,
-      "logits/chosen": 4.46875,
-      "logits/rejected": 4.46875,
-      "logps/chosen": -77312.0,
-      "logps/rejected": -77312.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -768.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -768.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3424.0,
+      "loss": 0.0383,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -11.375,
       "step": 1880
     },
     {
       "epoch": 1.9790575916230366,
-      "grad_norm": 0.011534981923145459,
+      "grad_norm": 0.05542960533750751,
       "learning_rate": 1.5665601280272124e-05,
-      "logits/chosen": 4.40625,
-      "logits/rejected": 4.40625,
-      "logps/chosen": -82432.0,
-      "logps/rejected": -82432.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -820.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -820.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3552.0,
+      "loss": 0.0089,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.828125,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -12.0,
       "step": 1890
     },
     {
       "epoch": 1.9895287958115184,
-      "grad_norm": 0.01351828623410575,
+      "grad_norm": 0.06820617401303433,
       "learning_rate": 1.538367984459298e-05,
-      "logits/chosen": 4.25,
-      "logits/rejected": 4.25,
-      "logps/chosen": -78848.0,
-      "logps/rejected": -78848.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -780.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -780.0,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -9.75,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3104.0,
+      "loss": 0.0113,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.875,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -11.5625,
       "step": 1900
     },
     {
       "epoch": 1.9895287958115184,
-      "eval_logits/chosen": 5.375,
-      "eval_logits/rejected": 5.8125,
-      "eval_logps/chosen": -65024.0,
-      "eval_logps/rejected": -56064.0,
-      "eval_loss": 190.4392547607422,
-      "eval_rewards/accuracies": 0.431640625,
-      "eval_rewards/chosen": -648.0,
-      "eval_rewards/margins": -87.5,
-      "eval_rewards/rejected": -560.0,
-      "eval_runtime": 47.1773,
-      "eval_samples_per_second": 42.393,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -13.625,
+      "eval_logits/rejected": -13.375,
+      "eval_logps/chosen": -596.0,
+      "eval_logps/rejected": -584.0,
+      "eval_loss": 0.909375011920929,
+      "eval_rewards/accuracies": 0.51171875,
+      "eval_rewards/chosen": -2.78125,
+      "eval_rewards/margins": 0.173828125,
+      "eval_rewards/rejected": -2.953125,
+      "eval_runtime": 46.5809,
+      "eval_samples_per_second": 42.936,
+      "eval_steps_per_second": 0.687,
       "step": 1900
     },
     {
       "epoch": 2.0,
-      "grad_norm": 0.013303361461992536,
+      "grad_norm": 0.7923736138071442,
       "learning_rate": 1.5103186439333133e-05,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "logps/chosen": -79872.0,
-      "logps/rejected": -79872.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -792.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -792.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -9.625,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3440.0,
+      "loss": 0.0141,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.546875,
+      "rewards/margins": 10.0625,
+      "rewards/rejected": -12.625,
       "step": 1910
     },
     {
       "epoch": 2.0104712041884816,
-      "grad_norm": 0.012739875394382255,
+      "grad_norm": 0.03797413952811105,
       "learning_rate": 1.4824162717963829e-05,
-      "logits/chosen": 3.828125,
-      "logits/rejected": 3.828125,
-      "logps/chosen": -88576.0,
-      "logps/rejected": -88576.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -880.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -880.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -772.0,
+      "logps/rejected": -3280.0,
+      "loss": 0.0021,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.765625,
+      "rewards/margins": 9.625,
+      "rewards/rejected": -12.375,
       "step": 1920
     },
     {
       "epoch": 2.020942408376963,
-      "grad_norm": 0.012091785703087271,
+      "grad_norm": 0.08445522476020367,
       "learning_rate": 1.4546650115707192e-05,
-      "logits/chosen": 4.375,
-      "logits/rejected": 4.375,
-      "logps/chosen": -78336.0,
-      "logps/rejected": -78336.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -776.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -776.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.25,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0024,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.640625,
+      "rewards/margins": 10.375,
+      "rewards/rejected": -13.0625,
       "step": 1930
     },
     {
       "epoch": 2.031413612565445,
-      "grad_norm": 0.011322707737967544,
+      "grad_norm": 0.008763710192524561,
       "learning_rate": 1.4270689843383111e-05,
-      "logits/chosen": 3.890625,
-      "logits/rejected": 3.890625,
-      "logps/chosen": -80384.0,
-      "logps/rejected": -80384.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -800.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -800.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -764.0,
+      "logps/rejected": -3568.0,
+      "loss": 0.0015,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.734375,
+      "rewards/margins": 10.8125,
+      "rewards/rejected": -13.5,
       "step": 1940
     },
     {
       "epoch": 2.0418848167539267,
-      "grad_norm": 0.011673988201261594,
+      "grad_norm": 0.010871862750517299,
       "learning_rate": 1.3996322881289347e-05,
-      "logits/chosen": 3.953125,
-      "logits/rejected": 3.953125,
-      "logps/chosen": -84992.0,
-      "logps/rejected": -84992.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -844.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -844.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.25,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0012,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 10.625,
+      "rewards/rejected": -13.5,
       "step": 1950
     },
     {
       "epoch": 2.0523560209424083,
-      "grad_norm": 0.010916176953963597,
+      "grad_norm": 0.0355862679650303,
       "learning_rate": 1.3723589973115963e-05,
-      "logits/chosen": 4.375,
-      "logits/rejected": 4.375,
-      "logps/chosen": -73728.0,
-      "logps/rejected": -73728.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -736.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -736.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.125,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3600.0,
+      "loss": 0.001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -14.3125,
       "step": 1960
     },
     {
       "epoch": 2.06282722513089,
-      "grad_norm": 0.012312436647789624,
+      "grad_norm": 0.012331372934647332,
       "learning_rate": 1.3452531619894837e-05,
-      "logits/chosen": 3.796875,
-      "logits/rejected": 3.796875,
-      "logps/chosen": -80384.0,
-      "logps/rejected": -80384.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -800.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -800.0,
+      "logits/chosen": -13.25,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3536.0,
+      "loss": 0.0012,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 11.25,
+      "rewards/rejected": -14.25,
       "step": 1970
     },
     {
       "epoch": 2.073298429319372,
-      "grad_norm": 0.01054549451073691,
+      "grad_norm": 0.03788379705316475,
       "learning_rate": 1.3183188073985245e-05,
-      "logits/chosen": 4.0,
-      "logits/rejected": 4.0,
-      "logps/chosen": -80896.0,
-      "logps/rejected": -80896.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -800.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -800.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -10.125,
+      "logps/chosen": -816.0,
+      "logps/rejected": -3568.0,
+      "loss": 0.0017,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 11.4375,
+      "rewards/rejected": -14.3125,
       "step": 1980
     },
     {
       "epoch": 2.0837696335078535,
-      "grad_norm": 0.012397519696288283,
+      "grad_norm": 0.020425162583961123,
       "learning_rate": 1.2915599333096351e-05,
-      "logits/chosen": 3.8125,
-      "logits/rejected": 3.8125,
-      "logps/chosen": -86528.0,
-      "logps/rejected": -86528.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -856.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -856.0,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3440.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 10.875,
+      "rewards/rejected": -13.875,
       "step": 1990
     },
     {
       "epoch": 2.094240837696335,
-      "grad_norm": 0.011245007062421108,
+      "grad_norm": 0.043482431208357666,
       "learning_rate": 1.264980513434752e-05,
-      "logits/chosen": 4.03125,
-      "logits/rejected": 4.03125,
-      "logps/chosen": -77824.0,
-      "logps/rejected": -77824.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -776.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -776.0,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.125,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3600.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 11.4375,
+      "rewards/rejected": -14.5,
       "step": 2000
     },
     {
       "epoch": 2.094240837696335,
-      "eval_logits/chosen": 5.125,
-      "eval_logits/rejected": 5.53125,
-      "eval_logps/chosen": -66048.0,
-      "eval_logps/rejected": -57088.0,
-      "eval_loss": 193.28050231933594,
-      "eval_rewards/accuracies": 0.4296875,
-      "eval_rewards/chosen": -656.0,
-      "eval_rewards/margins": -89.0,
-      "eval_rewards/rejected": -568.0,
-      "eval_runtime": 47.0655,
-      "eval_samples_per_second": 42.494,
-      "eval_steps_per_second": 0.68,
+      "eval_logits/chosen": -13.75,
+      "eval_logits/rejected": -13.5,
+      "eval_logps/chosen": -616.0,
+      "eval_logps/rejected": -608.0,
+      "eval_loss": 0.9535468816757202,
+      "eval_rewards/accuracies": 0.513671875,
+      "eval_rewards/chosen": -2.96875,
+      "eval_rewards/margins": 0.220703125,
+      "eval_rewards/rejected": -3.1875,
+      "eval_runtime": 46.5716,
+      "eval_samples_per_second": 42.945,
+      "eval_steps_per_second": 0.687,
       "step": 2000
     },
     {
       "epoch": 2.1047120418848166,
-      "grad_norm": 0.011764721444151442,
+      "grad_norm": 0.021990040282395953,
       "learning_rate": 1.2385844948367324e-05,
-      "logits/chosen": 3.875,
-      "logits/rejected": 3.875,
-      "logps/chosen": -90624.0,
-      "logps/rejected": -90624.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -900.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -900.0,
+      "logits/chosen": -13.75,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0011,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.953125,
+      "rewards/margins": 11.75,
+      "rewards/rejected": -14.75,
       "step": 2010
     },
     {
       "epoch": 2.115183246073298,
-      "grad_norm": 0.01065344051577734,
+      "grad_norm": 0.02681732808537171,
       "learning_rate": 1.2123757973432114e-05,
-      "logits/chosen": 3.984375,
-      "logits/rejected": 3.984375,
-      "logps/chosen": -84480.0,
-      "logps/rejected": -84480.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -840.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -840.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.375,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3488.0,
+      "loss": 0.0049,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 11.4375,
+      "rewards/rejected": -14.25,
       "step": 2020
     },
     {
       "epoch": 2.1256544502617802,
-      "grad_norm": 0.0122317107772418,
+      "grad_norm": 0.02804504925376927,
       "learning_rate": 1.1863583129645106e-05,
-      "logits/chosen": 3.921875,
-      "logits/rejected": 3.921875,
-      "logps/chosen": -81408.0,
-      "logps/rejected": -81408.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -808.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -808.0,
+      "logits/chosen": -12.9375,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -800.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0015,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.078125,
+      "rewards/margins": 11.75,
+      "rewards/rejected": -14.875,
       "step": 2030
     },
     {
       "epoch": 2.136125654450262,
-      "grad_norm": 0.011882871899091839,
+      "grad_norm": 0.8075496941248711,
       "learning_rate": 1.1605359053156607e-05,
-      "logits/chosen": 3.9375,
-      "logits/rejected": 3.9375,
-      "logps/chosen": -87552.0,
-      "logps/rejected": -87552.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -868.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -868.0,
+      "logits/chosen": -13.75,
+      "logits/rejected": -10.25,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3424.0,
+      "loss": 0.0017,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.796875,
+      "rewards/margins": 11.375,
+      "rewards/rejected": -14.1875,
       "step": 2040
     },
     {
       "epoch": 2.1465968586387434,
-      "grad_norm": 0.012143921144895754,
+      "grad_norm": 0.01902900455234704,
       "learning_rate": 1.1349124090426644e-05,
-      "logits/chosen": 3.46875,
-      "logits/rejected": 3.46875,
-      "logps/chosen": -88064.0,
-      "logps/rejected": -88064.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -872.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -872.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -9.875,
+      "logps/chosen": -816.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0016,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -15.1875,
       "step": 2050
     },
     {
       "epoch": 2.157068062827225,
-      "grad_norm": 0.011789699985689014,
+      "grad_norm": 0.0283077420771619,
       "learning_rate": 1.1094916292530406e-05,
-      "logits/chosen": 3.765625,
-      "logits/rejected": 3.765625,
-      "logps/chosen": -81408.0,
-      "logps/rejected": -81408.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -812.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -812.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0018,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.796875,
+      "rewards/margins": 12.1875,
+      "rewards/rejected": -15.0,
       "step": 2060
     },
     {
       "epoch": 2.167539267015707,
-      "grad_norm": 0.011899300277907681,
+      "grad_norm": 0.012130048267476778,
       "learning_rate": 1.0842773409507623e-05,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "logps/chosen": -75776.0,
-      "logps/rejected": -75776.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -752.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -752.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.5,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3600.0,
+      "loss": 0.001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.6875,
+      "rewards/margins": 12.75,
+      "rewards/rejected": -15.4375,
       "step": 2070
     },
     {
       "epoch": 2.1780104712041886,
-      "grad_norm": 0.011218008599867242,
+      "grad_norm": 0.11186142873174292,
       "learning_rate": 1.0592732884756754e-05,
-      "logits/chosen": 4.03125,
-      "logits/rejected": 4.03125,
-      "logps/chosen": -81920.0,
-      "logps/rejected": -81920.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -812.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -812.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.0,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0014,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.96875,
+      "rewards/margins": 11.9375,
+      "rewards/rejected": -14.9375,
       "step": 2080
     },
     {
       "epoch": 2.18848167539267,
-      "grad_norm": 0.010640763845735215,
+      "grad_norm": 0.022020673007143294,
       "learning_rate": 1.0344831849474507e-05,
-      "logits/chosen": 4.28125,
-      "logits/rejected": 4.28125,
-      "logps/chosen": -75264.0,
-      "logps/rejected": -75264.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -748.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -748.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.0,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0009,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -15.25,
       "step": 2090
     },
     {
       "epoch": 2.1989528795811517,
-      "grad_norm": 0.011048073998396699,
+      "grad_norm": 0.031190943948052354,
       "learning_rate": 1.0099107117141879e-05,
-      "logits/chosen": 3.9375,
-      "logits/rejected": 3.9375,
-      "logps/chosen": -86016.0,
-      "logps/rejected": -86016.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -852.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -852.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -796.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0014,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -15.625,
       "step": 2100
     },
     {
       "epoch": 2.1989528795811517,
-      "eval_logits/chosen": 5.0625,
-      "eval_logits/rejected": 5.46875,
-      "eval_logps/chosen": -66560.0,
-      "eval_logps/rejected": -57600.0,
-      "eval_loss": 195.64700317382812,
-      "eval_rewards/accuracies": 0.427734375,
-      "eval_rewards/chosen": -664.0,
-      "eval_rewards/margins": -90.5,
-      "eval_rewards/rejected": -576.0,
-      "eval_runtime": 46.985,
-      "eval_samples_per_second": 42.567,
-      "eval_steps_per_second": 0.681,
+      "eval_logits/chosen": -13.5625,
+      "eval_logits/rejected": -13.3125,
+      "eval_logps/chosen": -632.0,
+      "eval_logps/rejected": -628.0,
+      "eval_loss": 0.979812502861023,
+      "eval_rewards/accuracies": 0.517578125,
+      "eval_rewards/chosen": -3.15625,
+      "eval_rewards/margins": 0.228515625,
+      "eval_rewards/rejected": -3.375,
+      "eval_runtime": 46.5669,
+      "eval_samples_per_second": 42.949,
+      "eval_steps_per_second": 0.687,
       "step": 2100
     },
     {
       "epoch": 2.2094240837696333,
-      "grad_norm": 0.013211617100480777,
+      "grad_norm": 0.02086347682769286,
       "learning_rate": 9.855595178057334e-06,
-      "logits/chosen": 3.890625,
-      "logits/rejected": 3.890625,
-      "logps/chosen": -88576.0,
-      "logps/rejected": -88576.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -880.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -880.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -848.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.0015,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -15.4375,
       "step": 2110
     },
     {
       "epoch": 2.2198952879581153,
-      "grad_norm": 0.011000601190165538,
+      "grad_norm": 0.0714355062071931,
       "learning_rate": 9.614332193917949e-06,
-      "logits/chosen": 3.703125,
-      "logits/rejected": 3.703125,
-      "logps/chosen": -82432.0,
-      "logps/rejected": -82432.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -820.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -820.0,
+      "logits/chosen": -13.25,
+      "logits/rejected": -10.125,
+      "logps/chosen": -836.0,
+      "logps/rejected": -3648.0,
+      "loss": 0.0011,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 12.0625,
+      "rewards/rejected": -15.375,
       "step": 2120
     },
     {
       "epoch": 2.230366492146597,
-      "grad_norm": 0.011090527108191948,
+      "grad_norm": 0.010214424707746925,
       "learning_rate": 9.375353992449385e-06,
-      "logits/chosen": 3.953125,
-      "logits/rejected": 3.953125,
-      "logps/chosen": -77824.0,
-      "logps/rejected": -77824.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -776.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -776.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -9.75,
+      "logps/chosen": -712.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.0007,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -15.3125,
       "step": 2130
     },
     {
       "epoch": 2.2408376963350785,
-      "grad_norm": 0.01179929388143516,
+      "grad_norm": 0.009449610139048667,
       "learning_rate": 9.138696062085442e-06,
-      "logits/chosen": 3.8125,
-      "logits/rejected": 3.8125,
-      "logps/chosen": -81408.0,
-      "logps/rejected": -81408.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -808.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -808.0,
+      "logits/chosen": -13.25,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 13.0,
+      "rewards/rejected": -16.125,
       "step": 2140
     },
     {
       "epoch": 2.25130890052356,
-      "grad_norm": 0.009902880127591012,
+      "grad_norm": 0.030870851684990906,
       "learning_rate": 8.904393546698007e-06,
-      "logits/chosen": 3.734375,
-      "logits/rejected": 3.734375,
-      "logps/chosen": -83456.0,
-      "logps/rejected": -83456.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -828.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -828.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -820.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -15.9375,
       "step": 2150
     },
     {
       "epoch": 2.261780104712042,
-      "grad_norm": 0.014441028301653449,
+      "grad_norm": 0.017928103149379482,
       "learning_rate": 8.672481240378142e-06,
-      "logits/chosen": 4.0,
-      "logits/rejected": 4.0,
-      "logps/chosen": -83968.0,
-      "logps/rejected": -83968.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -832.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -832.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -856.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.0016,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 12.375,
+      "rewards/rejected": -15.6875,
       "step": 2160
     },
     {
       "epoch": 2.2722513089005236,
-      "grad_norm": 0.01143666886692562,
+      "grad_norm": 0.020979844872565743,
       "learning_rate": 8.44299358226919e-06,
-      "logits/chosen": 3.875,
-      "logits/rejected": 3.875,
-      "logps/chosen": -89088.0,
-      "logps/rejected": -89088.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -888.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -888.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -836.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0019,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.375,
+      "rewards/margins": 13.3125,
+      "rewards/rejected": -16.625,
       "step": 2170
     },
     {
       "epoch": 2.282722513089005,
-      "grad_norm": 0.012903796937529372,
+      "grad_norm": 0.01132743031264989,
       "learning_rate": 8.215964651452457e-06,
-      "logits/chosen": 3.890625,
-      "logits/rejected": 3.890625,
-      "logps/chosen": -86528.0,
-      "logps/rejected": -86528.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -860.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -860.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -824.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.328125,
+      "rewards/margins": 13.375,
+      "rewards/rejected": -16.75,
       "step": 2180
     },
     {
       "epoch": 2.2931937172774868,
-      "grad_norm": 0.011498246186801158,
+      "grad_norm": 0.008195016345104533,
       "learning_rate": 7.991428161886502e-06,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "logps/chosen": -82944.0,
-      "logps/rejected": -82944.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -824.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -824.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.25,
+      "logps/chosen": -820.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0009,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 13.0625,
+      "rewards/rejected": -16.375,
       "step": 2190
     },
     {
       "epoch": 2.303664921465969,
-      "grad_norm": 0.011369959736655442,
+      "grad_norm": 0.007235192617224554,
       "learning_rate": 7.76941745740061e-06,
-      "logits/chosen": 4.125,
-      "logits/rejected": 4.125,
-      "logps/chosen": -77312.0,
-      "logps/rejected": -77312.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -772.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -772.0,
+      "logits/chosen": -13.125,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 13.125,
+      "rewards/rejected": -16.375,
       "step": 2200
     },
     {
       "epoch": 2.303664921465969,
-      "eval_logits/chosen": 5.0625,
-      "eval_logits/rejected": 5.46875,
-      "eval_logps/chosen": -67584.0,
-      "eval_logps/rejected": -58368.0,
-      "eval_loss": 197.70675659179688,
-      "eval_rewards/accuracies": 0.423828125,
-      "eval_rewards/chosen": -672.0,
-      "eval_rewards/margins": -91.0,
-      "eval_rewards/rejected": -580.0,
-      "eval_runtime": 47.1527,
-      "eval_samples_per_second": 42.415,
-      "eval_steps_per_second": 0.679,
+      "eval_logits/chosen": -13.5625,
+      "eval_logits/rejected": -13.3125,
+      "eval_logps/chosen": -652.0,
+      "eval_logps/rejected": -648.0,
+      "eval_loss": 0.9977812767028809,
+      "eval_rewards/accuracies": 0.513671875,
+      "eval_rewards/chosen": -3.328125,
+      "eval_rewards/margins": 0.240234375,
+      "eval_rewards/rejected": -3.578125,
+      "eval_runtime": 46.5722,
+      "eval_samples_per_second": 42.944,
+      "eval_steps_per_second": 0.687,
       "step": 2200
     },
     {
       "epoch": 2.3141361256544504,
-      "grad_norm": 0.012300556094064644,
+      "grad_norm": 0.019723181322208762,
       "learning_rate": 7.549965506743123e-06,
-      "logits/chosen": 4.46875,
-      "logits/rejected": 4.46875,
-      "logps/chosen": -70144.0,
-      "logps/rejected": -70144.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -696.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -696.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -9.75,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0012,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 13.0,
+      "rewards/rejected": -16.25,
       "step": 2210
     },
     {
       "epoch": 2.324607329842932,
-      "grad_norm": 0.009562440829912389,
+      "grad_norm": 0.014077934151234522,
       "learning_rate": 7.333104898685631e-06,
-      "logits/chosen": 4.65625,
-      "logits/rejected": 4.65625,
-      "logps/chosen": -75776.0,
-      "logps/rejected": -75776.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -752.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -752.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 13.0625,
+      "rewards/rejected": -16.125,
       "step": 2220
     },
     {
       "epoch": 2.3350785340314135,
-      "grad_norm": 0.01224682041863487,
+      "grad_norm": 0.02530387524099349,
       "learning_rate": 7.118867837183518e-06,
-      "logits/chosen": 3.921875,
-      "logits/rejected": 3.921875,
-      "logps/chosen": -82944.0,
-      "logps/rejected": -82944.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -824.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -824.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -9.625,
+      "logps/chosen": -696.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0012,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 13.5625,
+      "rewards/rejected": -16.625,
       "step": 2230
     },
     {
       "epoch": 2.345549738219895,
-      "grad_norm": 0.012426913583186911,
+      "grad_norm": 0.010256460710128305,
       "learning_rate": 6.9072861365936045e-06,
-      "logits/chosen": 4.21875,
-      "logits/rejected": 4.21875,
-      "logps/chosen": -79872.0,
-      "logps/rejected": -79872.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -792.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -792.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.0,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0014,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 13.0,
+      "rewards/rejected": -16.25,
       "step": 2240
     },
     {
       "epoch": 2.356020942408377,
-      "grad_norm": 0.01097683277718859,
+      "grad_norm": 0.06519330938521245,
       "learning_rate": 6.6983912169497005e-06,
-      "logits/chosen": 3.890625,
-      "logits/rejected": 3.890625,
-      "logps/chosen": -83968.0,
-      "logps/rejected": -83968.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -832.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -832.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -796.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.5,
+      "rewards/margins": 12.5625,
+      "rewards/rejected": -16.0,
       "step": 2250
     },
     {
       "epoch": 2.3664921465968587,
-      "grad_norm": 0.011713192862590266,
+      "grad_norm": 0.02535199174674999,
       "learning_rate": 6.49221409929677e-06,
-      "logits/chosen": 3.53125,
-      "logits/rejected": 3.53125,
-      "logps/chosen": -89600.0,
-      "logps/rejected": -89600.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -892.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -892.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -816.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0009,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.359375,
+      "rewards/margins": 12.9375,
+      "rewards/rejected": -16.25,
       "step": 2260
     },
     {
       "epoch": 2.3769633507853403,
-      "grad_norm": 0.012266448497528984,
+      "grad_norm": 0.008646543365720447,
       "learning_rate": 6.2887854010842075e-06,
-      "logits/chosen": 4.34375,
-      "logits/rejected": 4.34375,
-      "logps/chosen": -83968.0,
-      "logps/rejected": -83968.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -832.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -832.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0018,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 13.25,
+      "rewards/rejected": -16.625,
       "step": 2270
     },
     {
       "epoch": 2.387434554973822,
-      "grad_norm": 0.012425535198393681,
+      "grad_norm": 0.048063864804168376,
       "learning_rate": 6.0881353316191385e-06,
-      "logits/chosen": 3.90625,
-      "logits/rejected": 3.90625,
-      "logps/chosen": -82944.0,
-      "logps/rejected": -82944.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -824.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -824.0,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -804.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0007,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.609375,
+      "rewards/margins": 12.6875,
+      "rewards/rejected": -16.375,
       "step": 2280
     },
     {
       "epoch": 2.3979057591623034,
-      "grad_norm": 0.010960822674507145,
+      "grad_norm": 0.02029365870305144,
       "learning_rate": 5.890293687580381e-06,
-      "logits/chosen": 4.0625,
-      "logits/rejected": 4.0625,
-      "logps/chosen": -84480.0,
-      "logps/rejected": -84480.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -840.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -840.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.125,
+      "logps/chosen": -744.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.484375,
+      "rewards/margins": 13.125,
+      "rewards/rejected": -16.625,
       "step": 2290
     },
     {
       "epoch": 2.4083769633507854,
-      "grad_norm": 0.010789499053331356,
+      "grad_norm": 0.021633310444226825,
       "learning_rate": 5.695289848593532e-06,
-      "logits/chosen": 3.765625,
-      "logits/rejected": 3.765625,
-      "logps/chosen": -86016.0,
-      "logps/rejected": -86016.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -852.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -852.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -824.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.0011,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.515625,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -15.75,
       "step": 2300
     },
     {
       "epoch": 2.4083769633507854,
-      "eval_logits/chosen": 5.125,
-      "eval_logits/rejected": 5.5,
-      "eval_logps/chosen": -68096.0,
-      "eval_logps/rejected": -58880.0,
-      "eval_loss": 199.1925048828125,
-      "eval_rewards/accuracies": 0.423828125,
-      "eval_rewards/chosen": -676.0,
-      "eval_rewards/margins": -92.0,
-      "eval_rewards/rejected": -584.0,
-      "eval_runtime": 47.1976,
-      "eval_samples_per_second": 42.375,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -13.5625,
+      "eval_logits/rejected": -13.3125,
+      "eval_logps/chosen": -664.0,
+      "eval_logps/rejected": -660.0,
+      "eval_loss": 1.008062481880188,
+      "eval_rewards/accuracies": 0.513671875,
+      "eval_rewards/chosen": -3.453125,
+      "eval_rewards/margins": 0.2470703125,
+      "eval_rewards/rejected": -3.6875,
+      "eval_runtime": 46.5711,
+      "eval_samples_per_second": 42.945,
+      "eval_steps_per_second": 0.687,
       "step": 2300
     },
     {
       "epoch": 2.418848167539267,
-      "grad_norm": 0.010172048287037959,
+      "grad_norm": 0.038924291142115805,
       "learning_rate": 5.50315277286812e-06,
-      "logits/chosen": 3.90625,
-      "logits/rejected": 3.90625,
-      "logps/chosen": -77824.0,
-      "logps/rejected": -77824.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -776.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -776.0,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -9.875,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3872.0,
+      "loss": 0.0019,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 13.25,
+      "rewards/rejected": -16.625,
       "step": 2310
     },
     {
       "epoch": 2.4293193717277486,
-      "grad_norm": 0.011526767658226127,
+      "grad_norm": 0.027840950067718296,
       "learning_rate": 5.313910992897281e-06,
-      "logits/chosen": 4.03125,
-      "logits/rejected": 4.03125,
-      "logps/chosen": -86528.0,
-      "logps/rejected": -86528.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -860.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -860.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -9.875,
+      "logps/chosen": -756.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0019,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 12.5,
+      "rewards/rejected": -15.75,
       "step": 2320
     },
     {
       "epoch": 2.4397905759162306,
-      "grad_norm": 0.01398760897533521,
+      "grad_norm": 0.013668323394273506,
       "learning_rate": 5.127592611220658e-06,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "logps/chosen": -79360.0,
-      "logps/rejected": -79360.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -788.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -788.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -784.0,
+      "logps/rejected": -3920.0,
+      "loss": 0.0011,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.484375,
+      "rewards/margins": 12.5625,
+      "rewards/rejected": -16.0,
       "step": 2330
     },
     {
       "epoch": 2.450261780104712,
-      "grad_norm": 0.014922273429823912,
+      "grad_norm": 0.015624084426943893,
       "learning_rate": 4.944225296251159e-06,
-      "logits/chosen": 4.125,
-      "logits/rejected": 4.125,
-      "logps/chosen": -88576.0,
-      "logps/rejected": -88576.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -880.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -880.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -9.5,
+      "logps/chosen": -800.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0009,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.375,
+      "rewards/margins": 12.5,
+      "rewards/rejected": -15.9375,
       "step": 2340
     },
     {
       "epoch": 2.4607329842931938,
-      "grad_norm": 0.010899378377946092,
+      "grad_norm": 0.0330934045068677,
       "learning_rate": 4.763836278166172e-06,
-      "logits/chosen": 3.765625,
-      "logits/rejected": 3.765625,
-      "logps/chosen": -84992.0,
-      "logps/rejected": -84992.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -844.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -844.0,
+      "logits/chosen": -13.25,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -792.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0012,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.421875,
+      "rewards/margins": 12.6875,
+      "rewards/rejected": -16.125,
       "step": 2350
     },
     {
       "epoch": 2.4712041884816753,
-      "grad_norm": 0.012294265354840546,
+      "grad_norm": 0.21612560380681003,
       "learning_rate": 4.586452344863876e-06,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "logps/chosen": -88576.0,
-      "logps/rejected": -88576.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -880.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -880.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -788.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 13.125,
+      "rewards/rejected": -16.375,
       "step": 2360
     },
     {
       "epoch": 2.481675392670157,
-      "grad_norm": 0.011614597196631039,
+      "grad_norm": 0.038060395714203235,
       "learning_rate": 4.412099837985192e-06,
-      "logits/chosen": 4.125,
-      "logits/rejected": 4.125,
-      "logps/chosen": -78336.0,
-      "logps/rejected": -78336.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -780.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -780.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -9.5,
+      "logps/chosen": -724.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0015,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.1875,
+      "rewards/margins": 12.9375,
+      "rewards/rejected": -16.125,
       "step": 2370
     },
     {
       "epoch": 2.492146596858639,
-      "grad_norm": 0.012548048125286846,
+      "grad_norm": 0.01808326965074876,
       "learning_rate": 4.240804649002089e-06,
-      "logits/chosen": 4.03125,
-      "logits/rejected": 4.03125,
-      "logps/chosen": -78848.0,
-      "logps/rejected": -78848.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -784.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -784.0,
+      "logits/chosen": -13.25,
+      "logits/rejected": -9.625,
+      "logps/chosen": -748.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0011,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 13.0,
+      "rewards/rejected": -16.375,
       "step": 2380
     },
     {
       "epoch": 2.5026178010471205,
-      "grad_norm": 0.012395987750038067,
+      "grad_norm": 0.02466175839503429,
       "learning_rate": 4.072592215372598e-06,
-      "logits/chosen": 3.828125,
-      "logits/rejected": 3.828125,
-      "logps/chosen": -88576.0,
-      "logps/rejected": -88576.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -884.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -884.0,
+      "logits/chosen": -13.8125,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -812.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.0012,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 12.4375,
+      "rewards/rejected": -15.75,
       "step": 2390
     },
     {
       "epoch": 2.513089005235602,
-      "grad_norm": 0.011715420318706999,
+      "grad_norm": 0.0070712892991022045,
       "learning_rate": 3.907487516763389e-06,
-      "logits/chosen": 4.5,
-      "logits/rejected": 4.5,
-      "logps/chosen": -81408.0,
-      "logps/rejected": -81408.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -808.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -808.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -796.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0017,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.328125,
+      "rewards/margins": 13.0625,
+      "rewards/rejected": -16.375,
       "step": 2400
     },
     {
       "epoch": 2.513089005235602,
-      "eval_logits/chosen": 5.125,
-      "eval_logits/rejected": 5.53125,
-      "eval_logps/chosen": -68096.0,
-      "eval_logps/rejected": -59136.0,
-      "eval_loss": 200.09774780273438,
-      "eval_rewards/accuracies": 0.42578125,
-      "eval_rewards/chosen": -680.0,
-      "eval_rewards/margins": -92.5,
-      "eval_rewards/rejected": -588.0,
-      "eval_runtime": 47.1955,
-      "eval_samples_per_second": 42.377,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -13.5625,
+      "eval_logits/rejected": -13.3125,
+      "eval_logps/chosen": -660.0,
+      "eval_logps/rejected": -656.0,
+      "eval_loss": 1.0045312643051147,
+      "eval_rewards/accuracies": 0.51171875,
+      "eval_rewards/chosen": -3.421875,
+      "eval_rewards/margins": 0.24609375,
+      "eval_rewards/rejected": -3.671875,
+      "eval_runtime": 46.5818,
+      "eval_samples_per_second": 42.935,
+      "eval_steps_per_second": 0.687,
       "step": 2400
     },
     {
       "epoch": 2.5235602094240837,
-      "grad_norm": 0.012170819547956865,
+      "grad_norm": 0.051153708522702936,
       "learning_rate": 3.745515071340286e-06,
-      "logits/chosen": 3.921875,
-      "logits/rejected": 3.921875,
-      "logps/chosen": -84480.0,
-      "logps/rejected": -84480.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -840.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -840.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.5,
+      "logps/chosen": -864.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0014,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 12.9375,
+      "rewards/rejected": -16.375,
       "step": 2410
     },
     {
       "epoch": 2.5340314136125652,
-      "grad_norm": 0.01013767511428227,
+      "grad_norm": 0.006987118197520675,
       "learning_rate": 3.586698932127236e-06,
-      "logits/chosen": 3.984375,
-      "logits/rejected": 3.984375,
-      "logps/chosen": -84992.0,
-      "logps/rejected": -84992.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -844.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -844.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -800.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 13.0,
+      "rewards/rejected": -16.375,
       "step": 2420
     },
     {
       "epoch": 2.5445026178010473,
-      "grad_norm": 0.010346684971359031,
+      "grad_norm": 0.02454198188363595,
       "learning_rate": 3.4310626834344744e-06,
-      "logits/chosen": 3.953125,
-      "logits/rejected": 3.953125,
-      "logps/chosen": -82432.0,
-      "logps/rejected": -82432.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -820.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -820.0,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -864.0,
+      "logps/rejected": -3616.0,
+      "loss": 0.008,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -15.5625,
       "step": 2430
     },
     {
       "epoch": 2.554973821989529,
-      "grad_norm": 0.011711607679201853,
+      "grad_norm": 0.03616404967795384,
       "learning_rate": 3.2786294373562343e-06,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "logps/chosen": -83968.0,
-      "logps/rejected": -83968.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -832.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -832.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -748.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 12.75,
+      "rewards/rejected": -16.0,
       "step": 2440
     },
     {
       "epoch": 2.5654450261780104,
-      "grad_norm": 0.009611865050915056,
+      "grad_norm": 0.018868867279177436,
       "learning_rate": 3.1294218303385526e-06,
-      "logits/chosen": 4.3125,
-      "logits/rejected": 4.3125,
-      "logps/chosen": -86016.0,
-      "logps/rejected": -86016.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -856.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -856.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -784.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0009,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -15.6875,
       "step": 2450
     },
     {
       "epoch": 2.5759162303664924,
-      "grad_norm": 0.010379359056333607,
+      "grad_norm": 0.09786363961612284,
       "learning_rate": 2.9834620198177483e-06,
-      "logits/chosen": 3.84375,
-      "logits/rejected": 3.84375,
-      "logps/chosen": -80896.0,
-      "logps/rejected": -80896.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -804.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -804.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -776.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.203125,
+      "rewards/margins": 12.5,
+      "rewards/rejected": -15.75,
       "step": 2460
     },
     {
       "epoch": 2.5863874345549736,
-      "grad_norm": 0.012711981552008889,
+      "grad_norm": 0.0425125160908712,
       "learning_rate": 2.840771680930068e-06,
-      "logits/chosen": 3.90625,
-      "logits/rejected": 3.90625,
-      "logps/chosen": -79360.0,
-      "logps/rejected": -79360.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -788.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -788.0,
+      "logits/chosen": -13.125,
+      "logits/rejected": -9.75,
+      "logps/chosen": -796.0,
+      "logps/rejected": -3552.0,
+      "loss": 0.001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -15.625,
       "step": 2470
     },
     {
       "epoch": 2.5968586387434556,
-      "grad_norm": 0.011011091157280974,
+      "grad_norm": 0.016020161645640844,
       "learning_rate": 2.701372003292868e-06,
-      "logits/chosen": 4.0,
-      "logits/rejected": 4.0,
-      "logps/chosen": -90624.0,
-      "logps/rejected": -90624.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -904.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -904.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -9.75,
+      "logps/chosen": -800.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0006,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 12.375,
+      "rewards/rejected": -15.8125,
       "step": 2480
     },
     {
       "epoch": 2.607329842931937,
-      "grad_norm": 0.011739967570472394,
+      "grad_norm": 0.05485646786669231,
       "learning_rate": 2.56528368785795e-06,
-      "logits/chosen": 3.875,
-      "logits/rejected": 3.875,
-      "logps/chosen": -83456.0,
-      "logps/rejected": -83456.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -828.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -828.0,
+      "logits/chosen": -13.125,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -808.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0015,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.375,
+      "rewards/margins": 12.4375,
+      "rewards/rejected": -15.8125,
       "step": 2490
     },
     {
       "epoch": 2.6178010471204187,
-      "grad_norm": 0.011299962837751844,
+      "grad_norm": 0.029447658857205895,
       "learning_rate": 2.432526943837471e-06,
-      "logits/chosen": 4.375,
-      "logits/rejected": 4.375,
-      "logps/chosen": -80384.0,
-      "logps/rejected": -80384.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -800.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -800.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -9.875,
+      "logps/chosen": -720.0,
+      "logps/rejected": -3968.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 13.5625,
+      "rewards/rejected": -16.875,
       "step": 2500
     },
     {
       "epoch": 2.6178010471204187,
-      "eval_logits/chosen": 5.15625,
-      "eval_logits/rejected": 5.5625,
-      "eval_logps/chosen": -68608.0,
-      "eval_logps/rejected": -59392.0,
-      "eval_loss": 200.89999389648438,
-      "eval_rewards/accuracies": 0.427734375,
-      "eval_rewards/chosen": -684.0,
-      "eval_rewards/margins": -92.5,
-      "eval_rewards/rejected": -588.0,
-      "eval_runtime": 47.196,
-      "eval_samples_per_second": 42.376,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -13.5625,
+      "eval_logits/rejected": -13.3125,
+      "eval_logps/chosen": -660.0,
+      "eval_logps/rejected": -656.0,
+      "eval_loss": 1.0018750429153442,
+      "eval_rewards/accuracies": 0.51171875,
+      "eval_rewards/chosen": -3.40625,
+      "eval_rewards/margins": 0.2431640625,
+      "eval_rewards/rejected": -3.65625,
+      "eval_runtime": 46.5655,
+      "eval_samples_per_second": 42.95,
+      "eval_steps_per_second": 0.687,
       "step": 2500
     },
     {
       "epoch": 2.6282722513089007,
-      "grad_norm": 0.011446445862642912,
+      "grad_norm": 0.005800482150465527,
       "learning_rate": 2.3031214857028446e-06,
-      "logits/chosen": 4.0625,
-      "logits/rejected": 4.0625,
-      "logps/chosen": -83456.0,
-      "logps/rejected": -83456.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -828.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -828.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -9.875,
+      "logps/chosen": -772.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.359375,
+      "rewards/margins": 12.375,
+      "rewards/rejected": -15.75,
       "step": 2510
     },
     {
       "epoch": 2.6387434554973823,
-      "grad_norm": 0.0109794087178545,
+      "grad_norm": 0.013866432950815213,
       "learning_rate": 2.177086530257139e-06,
-      "logits/chosen": 4.28125,
-      "logits/rejected": 4.28125,
-      "logps/chosen": -82432.0,
-      "logps/rejected": -82432.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -820.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -820.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0007,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 12.4375,
+      "rewards/rejected": -15.75,
       "step": 2520
     },
     {
       "epoch": 2.649214659685864,
-      "grad_norm": 0.010914063370695058,
+      "grad_norm": 0.05500628156454972,
       "learning_rate": 2.054440793781354e-06,
-      "logits/chosen": 4.09375,
-      "logits/rejected": 4.09375,
-      "logps/chosen": -85504.0,
-      "logps/rejected": -85504.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -852.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -852.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -820.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0014,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.578125,
+      "rewards/margins": 12.4375,
+      "rewards/rejected": -16.0,
       "step": 2530
     },
     {
       "epoch": 2.6596858638743455,
-      "grad_norm": 0.011488897566256455,
+      "grad_norm": 0.004091710513700706,
       "learning_rate": 1.9352024892550587e-06,
-      "logits/chosen": 4.125,
-      "logits/rejected": 4.125,
-      "logps/chosen": -85504.0,
-      "logps/rejected": -85504.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -848.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -848.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -812.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0019,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.484375,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -16.375,
       "step": 2540
     },
     {
       "epoch": 2.670157068062827,
-      "grad_norm": 0.012114029674365121,
+      "grad_norm": 0.6674184407955883,
       "learning_rate": 1.8193893236517379e-06,
-      "logits/chosen": 4.5625,
-      "logits/rejected": 4.5625,
-      "logps/chosen": -76800.0,
-      "logps/rejected": -76800.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -764.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -764.0,
+      "logits/chosen": -13.1875,
+      "logits/rejected": -9.875,
+      "logps/chosen": -772.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0037,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.359375,
+      "rewards/margins": 12.5,
+      "rewards/rejected": -15.875,
       "step": 2550
     },
     {
       "epoch": 2.680628272251309,
-      "grad_norm": 0.01126989760869917,
+      "grad_norm": 0.01364424265323226,
       "learning_rate": 1.7070184953092956e-06,
-      "logits/chosen": 4.34375,
-      "logits/rejected": 4.34375,
-      "logps/chosen": -84480.0,
-      "logps/rejected": -84480.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -840.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -840.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.0,
+      "logps/chosen": -824.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -15.625,
       "step": 2560
     },
     {
       "epoch": 2.6910994764397906,
-      "grad_norm": 0.014949687221434502,
+      "grad_norm": 0.046943393549352876,
       "learning_rate": 1.5981066913760918e-06,
-      "logits/chosen": 3.984375,
-      "logits/rejected": 3.984375,
-      "logps/chosen": -78848.0,
-      "logps/rejected": -78848.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -784.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -784.0,
+      "logits/chosen": -13.125,
+      "logits/rejected": -9.5,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0007,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 11.875,
+      "rewards/rejected": -15.125,
       "step": 2570
     },
     {
       "epoch": 2.701570680628272,
-      "grad_norm": 0.010283455077057897,
+      "grad_norm": 0.014697967818783861,
       "learning_rate": 1.492670085332884e-06,
-      "logits/chosen": 4.375,
-      "logits/rejected": 4.375,
-      "logps/chosen": -85504.0,
-      "logps/rejected": -85504.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -848.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -848.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0012,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.375,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -15.6875,
       "step": 2580
     },
     {
       "epoch": 2.712041884816754,
-      "grad_norm": 0.011666353872101813,
+      "grad_norm": 0.2009031898381944,
       "learning_rate": 1.3907243345910787e-06,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "logps/chosen": -79872.0,
-      "logps/rejected": -79872.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -792.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -792.0,
+      "logits/chosen": -13.25,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -796.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0014,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 12.75,
+      "rewards/rejected": -16.0,
       "step": 2590
     },
     {
       "epoch": 2.7225130890052354,
-      "grad_norm": 0.013297037809018804,
+      "grad_norm": 0.0213273128699711,
       "learning_rate": 1.292284578167538e-06,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "logps/chosen": -76800.0,
-      "logps/rejected": -76800.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -764.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -764.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -716.0,
+      "logps/rejected": -3808.0,
+      "loss": 0.0038,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 12.8125,
+      "rewards/rejected": -16.0,
       "step": 2600
     },
     {
       "epoch": 2.7225130890052354,
-      "eval_logits/chosen": 5.1875,
-      "eval_logits/rejected": 5.59375,
-      "eval_logps/chosen": -68608.0,
-      "eval_logps/rejected": -59392.0,
-      "eval_loss": 201.17950439453125,
-      "eval_rewards/accuracies": 0.427734375,
-      "eval_rewards/chosen": -684.0,
-      "eval_rewards/margins": -92.5,
-      "eval_rewards/rejected": -592.0,
-      "eval_runtime": 47.1833,
-      "eval_samples_per_second": 42.388,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -13.625,
+      "eval_logits/rejected": -13.375,
+      "eval_logps/chosen": -660.0,
+      "eval_logps/rejected": -652.0,
+      "eval_loss": 1.0070937871932983,
+      "eval_rewards/accuracies": 0.505859375,
+      "eval_rewards/chosen": -3.40625,
+      "eval_rewards/margins": 0.2353515625,
+      "eval_rewards/rejected": -3.640625,
+      "eval_runtime": 46.5719,
+      "eval_samples_per_second": 42.944,
+      "eval_steps_per_second": 0.687,
       "step": 2600
     },
     {
       "epoch": 2.7329842931937174,
-      "grad_norm": 0.011068859783482855,
+      "grad_norm": 0.028491461200941728,
       "learning_rate": 1.1973654344364925e-06,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "logps/chosen": -83456.0,
-      "logps/rejected": -83456.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -828.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -828.0,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -788.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0012,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -15.4375,
       "step": 2610
     },
     {
       "epoch": 2.743455497382199,
-      "grad_norm": 0.009794226502324038,
+      "grad_norm": 0.02339833657575842,
       "learning_rate": 1.1059809989586583e-06,
-      "logits/chosen": 4.0625,
-      "logits/rejected": 4.0625,
-      "logps/chosen": -76800.0,
-      "logps/rejected": -76800.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -764.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -764.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.125,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.140625,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -16.0,
       "step": 2620
     },
     {
       "epoch": 2.7539267015706805,
-      "grad_norm": 0.011735701572725198,
+      "grad_norm": 0.03392001247428998,
       "learning_rate": 1.018144842388044e-06,
-      "logits/chosen": 4.09375,
-      "logits/rejected": 4.09375,
-      "logps/chosen": -79872.0,
-      "logps/rejected": -79872.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -796.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -796.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -9.875,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0007,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 12.5625,
+      "rewards/rejected": -15.875,
       "step": 2630
     },
     {
       "epoch": 2.7643979057591626,
-      "grad_norm": 0.011136647421523344,
+      "grad_norm": 0.04739797262632349,
       "learning_rate": 9.338700084567109e-07,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "logps/chosen": -78336.0,
-      "logps/rejected": -78336.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -780.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -780.0,
+      "logits/chosen": -13.3125,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -800.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -16.25,
       "step": 2640
     },
     {
       "epoch": 2.774869109947644,
-      "grad_norm": 0.010895112197306812,
+      "grad_norm": 0.06621260558475561,
       "learning_rate": 8.531690120377606e-07,
-      "logits/chosen": 4.125,
-      "logits/rejected": 4.125,
-      "logps/chosen": -86016.0,
-      "logps/rejected": -86016.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -856.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -856.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -808.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0017,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 13.0625,
+      "rewards/rejected": -16.125,
       "step": 2650
     },
     {
       "epoch": 2.7853403141361257,
-      "grad_norm": 0.01209828511124195,
+      "grad_norm": 0.019839730997041942,
       "learning_rate": 7.760538372868636e-07,
-      "logits/chosen": 4.4375,
-      "logits/rejected": 4.4375,
-      "logps/chosen": -77312.0,
-      "logps/rejected": -77312.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -768.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -768.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -10.125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0012,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 13.1875,
+      "rewards/rejected": -16.5,
       "step": 2660
     },
     {
       "epoch": 2.7958115183246073,
-      "grad_norm": 0.011626834886658423,
+      "grad_norm": 0.01559260064244627,
       "learning_rate": 7.025359358626166e-07,
-      "logits/chosen": 3.9375,
-      "logits/rejected": 3.9375,
-      "logps/chosen": -82432.0,
-      "logps/rejected": -82432.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -820.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -820.0,
+      "logits/chosen": -13.75,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -768.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.375,
+      "rewards/margins": 12.6875,
+      "rewards/rejected": -16.125,
       "step": 2670
     },
     {
       "epoch": 2.806282722513089,
-      "grad_norm": 0.013626689524492384,
+      "grad_norm": 0.01055973669572818,
       "learning_rate": 6.326262252259463e-07,
-      "logits/chosen": 4.25,
-      "logits/rejected": 4.25,
-      "logps/chosen": -90112.0,
-      "logps/rejected": -90112.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -900.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -900.0,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.25,
+      "logps/chosen": -824.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0012,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -16.125,
       "step": 2680
     },
     {
       "epoch": 2.816753926701571,
-      "grad_norm": 0.012773362335971641,
+      "grad_norm": 0.029388384625518307,
       "learning_rate": 5.66335087018871e-07,
-      "logits/chosen": 4.3125,
-      "logits/rejected": 4.3125,
-      "logps/chosen": -80896.0,
-      "logps/rejected": -80896.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -800.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -800.0,
+      "logits/chosen": -13.25,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -844.0,
+      "logps/rejected": -3824.0,
+      "loss": 0.0009,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.46875,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -16.375,
       "step": 2690
     },
     {
       "epoch": 2.8272251308900525,
-      "grad_norm": 0.0106162356642666,
+      "grad_norm": 0.011543059380749308,
       "learning_rate": 5.036723655228226e-07,
-      "logits/chosen": 3.96875,
-      "logits/rejected": 3.96875,
-      "logps/chosen": -85504.0,
-      "logps/rejected": -85504.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -848.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -848.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -9.5,
+      "logps/chosen": -792.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0009,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 11.75,
+      "rewards/rejected": -15.0625,
       "step": 2700
     },
     {
       "epoch": 2.8272251308900525,
-      "eval_logits/chosen": 5.1875,
-      "eval_logits/rejected": 5.59375,
-      "eval_logps/chosen": -68608.0,
-      "eval_logps/rejected": -59392.0,
-      "eval_loss": 201.3105010986328,
-      "eval_rewards/accuracies": 0.427734375,
-      "eval_rewards/chosen": -684.0,
-      "eval_rewards/margins": -93.0,
-      "eval_rewards/rejected": -592.0,
-      "eval_runtime": 47.1948,
-      "eval_samples_per_second": 42.378,
-      "eval_steps_per_second": 0.678,
+      "eval_logits/chosen": -13.625,
+      "eval_logits/rejected": -13.375,
+      "eval_logps/chosen": -660.0,
+      "eval_logps/rejected": -652.0,
+      "eval_loss": 1.008218765258789,
+      "eval_rewards/accuracies": 0.509765625,
+      "eval_rewards/chosen": -3.40625,
+      "eval_rewards/margins": 0.236328125,
+      "eval_rewards/rejected": -3.640625,
+      "eval_runtime": 46.5936,
+      "eval_samples_per_second": 42.924,
+      "eval_steps_per_second": 0.687,
       "step": 2700
     },
     {
       "epoch": 2.837696335078534,
-      "grad_norm": 0.01067105710332084,
+      "grad_norm": 0.027333554374588787,
       "learning_rate": 4.446473661967432e-07,
-      "logits/chosen": 3.859375,
-      "logits/rejected": 3.859375,
-      "logps/chosen": -88064.0,
-      "logps/rejected": -88064.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -872.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -872.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -10.0,
+      "logps/chosen": -812.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 12.25,
+      "rewards/rejected": -15.75,
       "step": 2710
     },
     {
       "epoch": 2.8481675392670156,
-      "grad_norm": 0.01193783759118073,
+      "grad_norm": 0.015194606967890609,
       "learning_rate": 3.892688542952411e-07,
-      "logits/chosen": 3.96875,
-      "logits/rejected": 3.96875,
-      "logps/chosen": -92160.0,
-      "logps/rejected": -92160.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -912.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -912.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -844.0,
+      "logps/rejected": -3744.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 12.5625,
+      "rewards/rejected": -16.0,
       "step": 2720
     },
     {
       "epoch": 2.858638743455497,
-      "grad_norm": 0.013937913340001887,
+      "grad_norm": 0.01950443993574235,
       "learning_rate": 3.3754505356694e-07,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "logps/chosen": -80896.0,
-      "logps/rejected": -80896.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -800.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -800.0,
+      "logits/chosen": -13.25,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3664.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.1875,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -16.125,
       "step": 2730
     },
     {
       "epoch": 2.869109947643979,
-      "grad_norm": 0.013272674821208292,
+      "grad_norm": 0.022576475096029308,
       "learning_rate": 2.8948364503322276e-07,
-      "logits/chosen": 3.859375,
-      "logits/rejected": 3.859375,
-      "logps/chosen": -94208.0,
-      "logps/rejected": -94208.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -936.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -936.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -9.75,
+      "logps/chosen": -856.0,
+      "logps/rejected": -3856.0,
+      "loss": 0.0014,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.484375,
+      "rewards/margins": 12.3125,
+      "rewards/rejected": -15.8125,
       "step": 2740
     },
     {
       "epoch": 2.8795811518324608,
-      "grad_norm": 0.010285710570700459,
+      "grad_norm": 0.013850195199160394,
       "learning_rate": 2.4509176584762706e-07,
-      "logits/chosen": 4.0,
-      "logits/rejected": 4.0,
-      "logps/chosen": -79872.0,
-      "logps/rejected": -79872.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -792.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -792.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -9.875,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0009,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.21875,
+      "rewards/margins": 12.125,
+      "rewards/rejected": -15.375,
       "step": 2750
     },
     {
       "epoch": 2.8900523560209423,
-      "grad_norm": 0.012839652171316484,
+      "grad_norm": 0.04347102184590706,
       "learning_rate": 2.0437600823595694e-07,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "logps/chosen": -86528.0,
-      "logps/rejected": -86528.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -860.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -860.0,
+      "logits/chosen": -13.75,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -808.0,
+      "logps/rejected": -3968.0,
+      "loss": 0.0007,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 13.25,
+      "rewards/rejected": -16.625,
       "step": 2760
     },
     {
       "epoch": 2.900523560209424,
-      "grad_norm": 0.012019965155790551,
+      "grad_norm": 0.03540481134680033,
       "learning_rate": 1.6734241851733e-07,
-      "logits/chosen": 3.890625,
-      "logits/rejected": 3.890625,
-      "logps/chosen": -84480.0,
-      "logps/rejected": -84480.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -840.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -840.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -10.125,
+      "logps/chosen": -776.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -15.9375,
       "step": 2770
     },
     {
       "epoch": 2.9109947643979055,
-      "grad_norm": 0.013370353248449589,
+      "grad_norm": 0.01574807485170882,
       "learning_rate": 1.339964962062984e-07,
-      "logits/chosen": 4.25,
-      "logits/rejected": 4.25,
-      "logps/chosen": -88064.0,
-      "logps/rejected": -88064.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -876.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -876.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -820.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 12.875,
+      "rewards/rejected": -16.375,
       "step": 2780
     },
     {
       "epoch": 2.9214659685863875,
-      "grad_norm": 0.011564787650071283,
+      "grad_norm": 0.03463690015275206,
       "learning_rate": 1.0434319319617136e-07,
-      "logits/chosen": 4.09375,
-      "logits/rejected": 4.09375,
-      "logps/chosen": -87040.0,
-      "logps/rejected": -87040.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -864.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -864.0,
+      "logits/chosen": -13.875,
+      "logits/rejected": -10.0,
+      "logps/chosen": -768.0,
+      "logps/rejected": -3696.0,
+      "loss": 0.0006,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 12.5,
+      "rewards/rejected": -15.75,
       "step": 2790
     },
     {
       "epoch": 2.931937172774869,
-      "grad_norm": 0.011314175179467692,
+      "grad_norm": 0.0080680306236487,
       "learning_rate": 7.838691302365363e-08,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "logps/chosen": -86016.0,
-      "logps/rejected": -86016.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -852.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -852.0,
+      "logits/chosen": -13.5,
+      "logits/rejected": -9.875,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0011,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 12.625,
+      "rewards/rejected": -15.9375,
       "step": 2800
     },
     {
       "epoch": 2.931937172774869,
-      "eval_logits/chosen": 5.15625,
-      "eval_logits/rejected": 5.5625,
-      "eval_logps/chosen": -68608.0,
-      "eval_logps/rejected": -59392.0,
-      "eval_loss": 201.4324951171875,
-      "eval_rewards/accuracies": 0.427734375,
-      "eval_rewards/chosen": -684.0,
-      "eval_rewards/margins": -93.0,
-      "eval_rewards/rejected": -592.0,
-      "eval_runtime": 47.1564,
-      "eval_samples_per_second": 42.412,
-      "eval_steps_per_second": 0.679,
+      "eval_logits/chosen": -13.625,
+      "eval_logits/rejected": -13.375,
+      "eval_logps/chosen": -660.0,
+      "eval_logps/rejected": -652.0,
+      "eval_loss": 1.0087811946868896,
+      "eval_rewards/accuracies": 0.5078125,
+      "eval_rewards/chosen": -3.40625,
+      "eval_rewards/margins": 0.2353515625,
+      "eval_rewards/rejected": -3.640625,
+      "eval_runtime": 46.5682,
+      "eval_samples_per_second": 42.948,
+      "eval_steps_per_second": 0.687,
       "step": 2800
     },
     {
       "epoch": 2.9424083769633507,
-      "grad_norm": 0.011710129170735049,
+      "grad_norm": 0.022797110033489408,
       "learning_rate": 5.613151021490459e-08,
-      "logits/chosen": 3.984375,
-      "logits/rejected": 3.984375,
-      "logps/chosen": -89600.0,
-      "logps/rejected": -89600.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -892.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -892.0,
+      "logits/chosen": -13.75,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -780.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0014,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 12.0,
+      "rewards/rejected": -15.4375,
       "step": 2810
     },
     {
       "epoch": 2.9528795811518327,
-      "grad_norm": 0.010838216031673596,
+      "grad_norm": 0.012692747223129556,
       "learning_rate": 3.7580289713159965e-08,
-      "logits/chosen": 3.84375,
-      "logits/rejected": 3.84375,
-      "logps/chosen": -87552.0,
-      "logps/rejected": -87552.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -868.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -868.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0009,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 12.6875,
+      "rewards/rejected": -15.8125,
       "step": 2820
     },
     {
       "epoch": 2.9633507853403143,
-      "grad_norm": 0.01227548955929598,
+      "grad_norm": 1.7651805279527109,
       "learning_rate": 2.2736006387927277e-08,
-      "logits/chosen": 3.84375,
-      "logits/rejected": 3.84375,
-      "logps/chosen": -82944.0,
-      "logps/rejected": -82944.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -824.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -824.0,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -752.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0051,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 11.9375,
+      "rewards/rejected": -15.25,
       "step": 2830
     },
     {
       "epoch": 2.973821989528796,
-      "grad_norm": 0.010476071168246684,
+      "grad_norm": 0.04429116103492594,
       "learning_rate": 1.1600864625893604e-08,
-      "logits/chosen": 3.828125,
-      "logits/rejected": 3.828125,
-      "logps/chosen": -91136.0,
-      "logps/rejected": -91136.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -904.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -904.0,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -832.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0011,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.265625,
+      "rewards/margins": 12.0,
+      "rewards/rejected": -15.3125,
       "step": 2840
     },
     {
       "epoch": 2.9842931937172774,
-      "grad_norm": 0.010707399323313473,
+      "grad_norm": 0.011064137488642318,
       "learning_rate": 4.176518003573549e-09,
-      "logits/chosen": 4.0625,
-      "logits/rejected": 4.0625,
-      "logps/chosen": -88576.0,
-      "logps/rejected": -88576.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -880.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -880.0,
+      "logits/chosen": -13.75,
+      "logits/rejected": -10.0,
+      "logps/chosen": -784.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.421875,
+      "rewards/margins": 12.1875,
+      "rewards/rejected": -15.625,
       "step": 2850
     },
     {
       "epoch": 2.994764397905759,
-      "grad_norm": 0.010043854680677556,
+      "grad_norm": 0.03583142898507438,
       "learning_rate": 4.640690417528903e-10,
-      "logits/chosen": 4.09375,
-      "logits/rejected": 4.09375,
-      "logps/chosen": -86016.0,
-      "logps/rejected": -86016.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -852.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -852.0,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -10.5,
+      "logps/chosen": -740.0,
+      "logps/rejected": -3584.0,
+      "loss": 0.0018,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 12.1875,
+      "rewards/rejected": -15.3125,
       "step": 2860
     },
     {
       "epoch": 3.0,
       "step": 2865,
       "total_flos": 0.0,
-      "train_loss": 0.69140625,
-      "train_runtime": 12294.033,
-      "train_samples_per_second": 14.914,
-      "train_steps_per_second": 0.233
+      "train_loss": 0.06583595570358879,
+      "train_runtime": 12397.8277,
+      "train_samples_per_second": 14.789,
+      "train_steps_per_second": 0.231
     }
   ],
   "logging_steps": 10,