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": 0.005321496791306334,
+      "grad_norm": 6.302913836794566,
       "learning_rate": 1.7421602787456448e-07,
-      "logits/chosen": -11.625,
-      "logits/rejected": -11.625,
-      "logps/chosen": -480.0,
-      "logps/rejected": -480.0,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -8.625,
+      "logps/chosen": -418.0,
+      "logps/rejected": -2096.0,
       "loss": 0.6914,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
@@ -25,4750 +25,4750 @@
     },
     {
       "epoch": 0.010471204188481676,
-      "grad_norm": 0.00546444725683883,
+      "grad_norm": 5.5975640316574955,
       "learning_rate": 1.7421602787456445e-06,
-      "logits/chosen": -11.5,
-      "logits/rejected": -11.5,
-      "logps/chosen": -456.0,
-      "logps/rejected": -456.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": 0.000614166259765625,
-      "rewards/margins": 0.0,
-      "rewards/rejected": 0.000614166259765625,
+      "logits/chosen": -10.9375,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -394.0,
+      "logps/rejected": -2288.0,
+      "loss": 0.689,
+      "rewards/accuracies": 0.2152777761220932,
+      "rewards/chosen": 0.0004119873046875,
+      "rewards/margins": 0.005889892578125,
+      "rewards/rejected": -0.005462646484375,
       "step": 10
     },
     {
       "epoch": 0.020942408376963352,
-      "grad_norm": 0.005083128367567884,
+      "grad_norm": 5.476938643595916,
       "learning_rate": 3.484320557491289e-06,
-      "logits/chosen": -11.125,
-      "logits/rejected": -11.125,
-      "logps/chosen": -412.0,
-      "logps/rejected": -412.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": 0.000362396240234375,
-      "rewards/margins": 0.0,
-      "rewards/rejected": 0.000362396240234375,
+      "logits/chosen": -10.625,
+      "logits/rejected": -7.90625,
+      "logps/chosen": -374.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.6182,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.0012054443359375,
+      "rewards/margins": 0.1513671875,
+      "rewards/rejected": -0.150390625,
       "step": 20
     },
     {
       "epoch": 0.031413612565445025,
-      "grad_norm": 0.005147109815316337,
+      "grad_norm": 1.4178786962721925,
       "learning_rate": 5.226480836236934e-06,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -442.0,
-      "logps/rejected": -442.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.00201416015625,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.00201416015625,
+      "logits/chosen": -10.875,
+      "logits/rejected": -8.375,
+      "logps/chosen": -392.0,
+      "logps/rejected": -2384.0,
+      "loss": 0.2863,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -0.0859375,
+      "rewards/margins": 1.3671875,
+      "rewards/rejected": -1.453125,
       "step": 30
     },
     {
       "epoch": 0.041884816753926704,
-      "grad_norm": 0.0056419213348981945,
+      "grad_norm": 0.6458101524912945,
       "learning_rate": 6.968641114982578e-06,
-      "logits/chosen": -11.375,
-      "logits/rejected": -11.375,
-      "logps/chosen": -438.0,
-      "logps/rejected": -438.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.00775146484375,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.00775146484375,
+      "logits/chosen": -11.3125,
+      "logits/rejected": -8.875,
+      "logps/chosen": -454.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0823,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -0.65234375,
+      "rewards/margins": 6.25,
+      "rewards/rejected": -6.90625,
       "step": 40
     },
     {
       "epoch": 0.05235602094240838,
-      "grad_norm": 0.006279514902204525,
+      "grad_norm": 1.016957357387982,
       "learning_rate": 8.710801393728225e-06,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -442.0,
-      "logps/rejected": -442.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.0177001953125,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.0177001953125,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -500.0,
+      "logps/rejected": -4032.0,
+      "loss": 0.0473,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -1.1953125,
+      "rewards/margins": 14.3125,
+      "rewards/rejected": -15.5,
       "step": 50
     },
     {
       "epoch": 0.06282722513089005,
-      "grad_norm": 0.00559343395701337,
+      "grad_norm": 0.5417065152709729,
       "learning_rate": 1.0452961672473868e-05,
-      "logits/chosen": -11.125,
-      "logits/rejected": -11.125,
-      "logps/chosen": -362.0,
-      "logps/rejected": -362.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.037109375,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.037109375,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -12.125,
+      "logps/chosen": -428.0,
+      "logps/rejected": -3472.0,
+      "loss": 0.0718,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.0859375,
+      "rewards/margins": 13.1875,
+      "rewards/rejected": -14.3125,
       "step": 60
     },
     {
       "epoch": 0.07329842931937172,
-      "grad_norm": 0.0060647133585819525,
+      "grad_norm": 0.5938876898152976,
       "learning_rate": 1.2195121951219513e-05,
-      "logits/chosen": -11.375,
-      "logits/rejected": -11.375,
-      "logps/chosen": -452.0,
-      "logps/rejected": -452.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.06005859375,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.06005859375,
+      "logits/chosen": -14.75,
+      "logits/rejected": -11.25,
+      "logps/chosen": -488.0,
+      "logps/rejected": -3424.0,
+      "loss": 0.073,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.94140625,
+      "rewards/margins": 11.0625,
+      "rewards/rejected": -12.0,
       "step": 70
     },
     {
       "epoch": 0.08376963350785341,
-      "grad_norm": 0.005554759925523654,
+      "grad_norm": 0.5361178356973734,
       "learning_rate": 1.3937282229965156e-05,
-      "logits/chosen": -11.0625,
-      "logits/rejected": -11.0625,
-      "logps/chosen": -420.0,
-      "logps/rejected": -420.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.11181640625,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.11181640625,
+      "logits/chosen": -13.875,
+      "logits/rejected": -11.5,
+      "logps/chosen": -482.0,
+      "logps/rejected": -3456.0,
+      "loss": 0.0518,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.1640625,
+      "rewards/margins": 12.0625,
+      "rewards/rejected": -13.25,
       "step": 80
     },
     {
       "epoch": 0.09424083769633508,
-      "grad_norm": 0.006098408514533971,
+      "grad_norm": 1.113255150623837,
       "learning_rate": 1.56794425087108e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.875,
-      "logps/chosen": -396.0,
-      "logps/rejected": -396.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.2021484375,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.2021484375,
+      "logits/chosen": -12.125,
+      "logits/rejected": -10.125,
+      "logps/chosen": -466.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0424,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -1.40625,
+      "rewards/margins": 14.8125,
+      "rewards/rejected": -16.25,
       "step": 90
     },
     {
       "epoch": 0.10471204188481675,
-      "grad_norm": 0.0061472278571238605,
+      "grad_norm": 0.22872919496659547,
       "learning_rate": 1.742160278745645e-05,
       "logits/chosen": -11.125,
-      "logits/rejected": -11.125,
-      "logps/chosen": -532.0,
-      "logps/rejected": -532.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.322265625,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.322265625,
+      "logits/rejected": -9.5,
+      "logps/chosen": -612.0,
+      "logps/rejected": -4000.0,
+      "loss": 0.0747,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -1.8203125,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -17.875,
       "step": 100
     },
     {
       "epoch": 0.10471204188481675,
       "eval_logits/chosen": -10.1875,
-      "eval_logits/rejected": -9.8125,
-      "eval_logps/chosen": -356.0,
-      "eval_logps/rejected": -328.0,
-      "eval_loss": 0.6925625205039978,
-      "eval_rewards/accuracies": 0.44921875,
-      "eval_rewards/chosen": -0.3828125,
-      "eval_rewards/margins": 0.00567626953125,
-      "eval_rewards/rejected": -0.388671875,
-      "eval_runtime": 48.3668,
-      "eval_samples_per_second": 41.351,
-      "eval_steps_per_second": 0.662,
+      "eval_logits/rejected": -9.875,
+      "eval_logps/chosen": -442.0,
+      "eval_logps/rejected": -428.0,
+      "eval_loss": 0.7318750023841858,
+      "eval_rewards/accuracies": 0.51953125,
+      "eval_rewards/chosen": -1.234375,
+      "eval_rewards/margins": 0.1494140625,
+      "eval_rewards/rejected": -1.390625,
+      "eval_runtime": 49.2553,
+      "eval_samples_per_second": 40.605,
+      "eval_steps_per_second": 0.65,
       "step": 100
     },
     {
       "epoch": 0.11518324607329843,
-      "grad_norm": 0.004907606140631277,
+      "grad_norm": 0.4962729739775567,
       "learning_rate": 1.9163763066202093e-05,
-      "logits/chosen": -10.4375,
-      "logits/rejected": -10.4375,
-      "logps/chosen": -480.0,
-      "logps/rejected": -480.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.412109375,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.412109375,
+      "logits/chosen": -11.375,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -498.0,
+      "logps/rejected": -3984.0,
+      "loss": 0.0415,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -1.2421875,
+      "rewards/margins": 15.0,
+      "rewards/rejected": -16.25,
       "step": 110
     },
     {
       "epoch": 0.1256544502617801,
-      "grad_norm": 0.005630524048281556,
+      "grad_norm": 1.108026595797128,
       "learning_rate": 2.0905923344947736e-05,
-      "logits/chosen": -10.25,
-      "logits/rejected": -10.25,
-      "logps/chosen": -448.0,
-      "logps/rejected": -448.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.56640625,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.56640625,
+      "logits/chosen": -13.0,
+      "logits/rejected": -11.1875,
+      "logps/chosen": -486.0,
+      "logps/rejected": -3968.0,
+      "loss": 0.0419,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.515625,
+      "rewards/margins": 16.25,
+      "rewards/rejected": -17.75,
       "step": 120
     },
     {
       "epoch": 0.13612565445026178,
-      "grad_norm": 0.006523402026746691,
+      "grad_norm": 0.47919150084105466,
       "learning_rate": 2.264808362369338e-05,
-      "logits/chosen": -9.25,
-      "logits/rejected": -9.25,
-      "logps/chosen": -516.0,
-      "logps/rejected": -516.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -0.78125,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -0.78125,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -11.375,
+      "logps/chosen": -592.0,
+      "logps/rejected": -4192.0,
+      "loss": 0.0416,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -2.15625,
+      "rewards/margins": 17.375,
+      "rewards/rejected": -19.5,
       "step": 130
     },
     {
       "epoch": 0.14659685863874344,
-      "grad_norm": 0.006154865969018647,
+      "grad_norm": 3.0814685778687516,
       "learning_rate": 2.4390243902439026e-05,
-      "logits/chosen": -8.625,
-      "logits/rejected": -8.625,
-      "logps/chosen": -596.0,
-      "logps/rejected": -596.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -1.234375,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -1.234375,
+      "logits/chosen": -14.75,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -728.0,
+      "logps/rejected": -4512.0,
+      "loss": 0.0462,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.25,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -23.5,
       "step": 140
     },
     {
       "epoch": 0.15706806282722513,
-      "grad_norm": 0.0090966480620495,
+      "grad_norm": 0.5025819544319339,
       "learning_rate": 2.6132404181184672e-05,
-      "logits/chosen": -7.1875,
-      "logits/rejected": -7.1875,
-      "logps/chosen": -636.0,
-      "logps/rejected": -636.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -1.78125,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -1.78125,
+      "logits/chosen": -13.375,
+      "logits/rejected": -11.625,
+      "logps/chosen": -588.0,
+      "logps/rejected": -3952.0,
+      "loss": 0.0675,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -2.0,
+      "rewards/margins": 15.25,
+      "rewards/rejected": -17.25,
       "step": 150
     },
     {
       "epoch": 0.16753926701570682,
-      "grad_norm": 0.02178305249243194,
+      "grad_norm": 1.883837818839591,
       "learning_rate": 2.7874564459930312e-05,
-      "logits/chosen": -6.78125,
-      "logits/rejected": -6.78125,
-      "logps/chosen": -676.0,
-      "logps/rejected": -676.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -2.640625,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -2.640625,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -544.0,
+      "logps/rejected": -3968.0,
+      "loss": 0.0498,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -1.8515625,
+      "rewards/margins": 15.3125,
+      "rewards/rejected": -17.125,
       "step": 160
     },
     {
       "epoch": 0.17801047120418848,
-      "grad_norm": 0.029324077006416743,
+      "grad_norm": 4.20808280085136,
       "learning_rate": 2.9616724738675962e-05,
-      "logits/chosen": -6.625,
-      "logits/rejected": -6.625,
-      "logps/chosen": -984.0,
-      "logps/rejected": -984.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -5.28125,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -5.28125,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -11.25,
+      "logps/chosen": -692.0,
+      "logps/rejected": -4000.0,
+      "loss": 0.0738,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.046875,
+      "rewards/margins": 15.625,
+      "rewards/rejected": -18.625,
       "step": 170
     },
     {
       "epoch": 0.18848167539267016,
-      "grad_norm": 0.1294752389922843,
+      "grad_norm": 0.23931093036320397,
       "learning_rate": 3.13588850174216e-05,
-      "logits/chosen": -5.6875,
-      "logits/rejected": -5.6875,
-      "logps/chosen": -1632.0,
-      "logps/rejected": -1632.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -11.75,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -11.75,
+      "logits/chosen": -13.8125,
+      "logits/rejected": -12.1875,
+      "logps/chosen": -608.0,
+      "logps/rejected": -3440.0,
+      "loss": 0.0602,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.140625,
+      "rewards/margins": 11.1875,
+      "rewards/rejected": -13.3125,
       "step": 180
     },
     {
       "epoch": 0.19895287958115182,
-      "grad_norm": 0.6591495992849863,
+      "grad_norm": 0.8092450589102698,
       "learning_rate": 3.310104529616725e-05,
-      "logits/chosen": -4.71875,
-      "logits/rejected": -4.71875,
-      "logps/chosen": -7552.0,
-      "logps/rejected": -7552.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -71.5,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -71.5,
+      "logits/chosen": -14.375,
+      "logits/rejected": -12.25,
+      "logps/chosen": -532.0,
+      "logps/rejected": -3984.0,
+      "loss": 0.0367,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -1.703125,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -17.75,
       "step": 190
     },
     {
       "epoch": 0.2094240837696335,
-      "grad_norm": 0.42030548233435494,
+      "grad_norm": 0.5412938932017806,
       "learning_rate": 3.48432055749129e-05,
-      "logits/chosen": -0.828125,
-      "logits/rejected": -0.828125,
-      "logps/chosen": -24064.0,
-      "logps/rejected": -24064.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -236.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -236.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -11.0625,
+      "logps/chosen": -588.0,
+      "logps/rejected": -3840.0,
+      "loss": 0.0623,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.7734375,
+      "rewards/margins": 14.1875,
+      "rewards/rejected": -15.9375,
       "step": 200
     },
     {
       "epoch": 0.2094240837696335,
-      "eval_logits/chosen": -0.828125,
-      "eval_logits/rejected": -0.92578125,
-      "eval_logps/chosen": -24704.0,
-      "eval_logps/rejected": -21376.0,
-      "eval_loss": 70.48087310791016,
-      "eval_rewards/accuracies": 0.44921875,
-      "eval_rewards/chosen": -244.0,
-      "eval_rewards/margins": -32.75,
-      "eval_rewards/rejected": -211.0,
-      "eval_runtime": 46.6513,
-      "eval_samples_per_second": 42.871,
+      "eval_logits/chosen": -13.625,
+      "eval_logits/rejected": -13.3125,
+      "eval_logps/chosen": -434.0,
+      "eval_logps/rejected": -420.0,
+      "eval_loss": 0.7325624823570251,
+      "eval_rewards/accuracies": 0.513671875,
+      "eval_rewards/chosen": -1.1640625,
+      "eval_rewards/margins": 0.1494140625,
+      "eval_rewards/rejected": -1.3125,
+      "eval_runtime": 46.6509,
+      "eval_samples_per_second": 42.872,
       "eval_steps_per_second": 0.686,
       "step": 200
     },
     {
       "epoch": 0.2198952879581152,
-      "grad_norm": 0.10286941786834401,
+      "grad_norm": 0.4707440844489683,
       "learning_rate": 3.6585365853658535e-05,
-      "logits/chosen": -0.263671875,
-      "logits/rejected": -0.263671875,
-      "logps/chosen": -29696.0,
-      "logps/rejected": -29696.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -294.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -294.0,
+      "logits/chosen": -13.4375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -490.0,
+      "logps/rejected": -3536.0,
+      "loss": 0.0387,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -1.5,
+      "rewards/margins": 11.0625,
+      "rewards/rejected": -12.5625,
       "step": 210
     },
     {
       "epoch": 0.23036649214659685,
-      "grad_norm": 0.05986187050034185,
+      "grad_norm": 0.11185140011851975,
       "learning_rate": 3.8327526132404185e-05,
-      "logits/chosen": 0.2890625,
-      "logits/rejected": 0.2890625,
-      "logps/chosen": -35584.0,
-      "logps/rejected": -35584.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -352.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -352.0,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -608.0,
+      "logps/rejected": -4048.0,
+      "loss": 0.0319,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.203125,
+      "rewards/margins": 15.6875,
+      "rewards/rejected": -17.875,
       "step": 220
     },
     {
       "epoch": 0.24083769633507854,
-      "grad_norm": 0.05303359745536982,
+      "grad_norm": 0.7262193841420953,
       "learning_rate": 4.006968641114983e-05,
-      "logits/chosen": 0.2578125,
-      "logits/rejected": 0.2578125,
-      "logps/chosen": -36864.0,
-      "logps/rejected": -36864.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -364.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -364.0,
+      "logits/chosen": -13.125,
+      "logits/rejected": -10.875,
+      "logps/chosen": -612.0,
+      "logps/rejected": -4192.0,
+      "loss": 0.0476,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -2.609375,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -20.625,
       "step": 230
     },
     {
       "epoch": 0.2513089005235602,
-      "grad_norm": 0.03623718953905268,
+      "grad_norm": 0.6905966587452627,
       "learning_rate": 4.181184668989547e-05,
-      "logits/chosen": -0.103515625,
-      "logits/rejected": -0.103515625,
-      "logps/chosen": -38656.0,
-      "logps/rejected": -38656.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -382.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -382.0,
+      "logits/chosen": -12.8125,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -544.0,
+      "logps/rejected": -3760.0,
+      "loss": 0.0407,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -1.984375,
+      "rewards/margins": 14.75,
+      "rewards/rejected": -16.75,
       "step": 240
     },
     {
       "epoch": 0.2617801047120419,
-      "grad_norm": 0.029017243695675224,
+      "grad_norm": 0.5805451338598272,
       "learning_rate": 4.3554006968641115e-05,
-      "logits/chosen": -0.20703125,
-      "logits/rejected": -0.20703125,
-      "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": -13.4375,
+      "logits/rejected": -11.125,
+      "logps/chosen": -664.0,
+      "logps/rejected": -4352.0,
+      "loss": 0.0391,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.578125,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -22.375,
       "step": 250
     },
     {
       "epoch": 0.27225130890052357,
-      "grad_norm": 0.021433217189061358,
+      "grad_norm": 1.061649161369204,
       "learning_rate": 4.529616724738676e-05,
-      "logits/chosen": -0.34765625,
-      "logits/rejected": -0.34765625,
-      "logps/chosen": -34816.0,
-      "logps/rejected": -34816.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -346.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -346.0,
+      "logits/chosen": -12.8125,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -796.0,
+      "logps/rejected": -5920.0,
+      "loss": 0.0766,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 31.25,
+      "rewards/rejected": -35.75,
       "step": 260
     },
     {
       "epoch": 0.28272251308900526,
-      "grad_norm": 0.023982427318443814,
+      "grad_norm": 0.9202559012098821,
       "learning_rate": 4.703832752613241e-05,
-      "logits/chosen": -0.5234375,
-      "logits/rejected": -0.5234375,
-      "logps/chosen": -37120.0,
-      "logps/rejected": -37120.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -368.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -368.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -10.9375,
+      "logps/chosen": -540.0,
+      "logps/rejected": -3792.0,
+      "loss": 0.0878,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -1.921875,
+      "rewards/margins": 13.9375,
+      "rewards/rejected": -15.875,
       "step": 270
     },
     {
       "epoch": 0.2931937172774869,
-      "grad_norm": 0.017314809161788524,
+      "grad_norm": 0.499121790752496,
       "learning_rate": 4.878048780487805e-05,
-      "logits/chosen": -0.64453125,
-      "logits/rejected": -0.64453125,
-      "logps/chosen": -36608.0,
-      "logps/rejected": -36608.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -362.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -362.0,
+      "logits/chosen": -15.25,
+      "logits/rejected": -11.375,
+      "logps/chosen": -506.0,
+      "logps/rejected": -3024.0,
+      "loss": 0.0581,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -1.7421875,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -9.9375,
       "step": 280
     },
     {
       "epoch": 0.3036649214659686,
-      "grad_norm": 0.01555782856683386,
+      "grad_norm": 0.3505358032885464,
       "learning_rate": 4.999983293481418e-05,
-      "logits/chosen": -0.73828125,
-      "logits/rejected": -0.73828125,
-      "logps/chosen": -38912.0,
-      "logps/rejected": -38912.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -384.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -384.0,
+      "logits/chosen": -15.125,
+      "logits/rejected": -10.9375,
+      "logps/chosen": -672.0,
+      "logps/rejected": -4224.0,
+      "loss": 0.0272,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.03125,
+      "rewards/margins": 16.75,
+      "rewards/rejected": -19.75,
       "step": 290
     },
     {
       "epoch": 0.31413612565445026,
-      "grad_norm": 0.01618671480836389,
+      "grad_norm": 1.4788106337385913,
       "learning_rate": 4.9996862949181595e-05,
-      "logits/chosen": -0.796875,
-      "logits/rejected": -0.796875,
-      "logps/chosen": -37632.0,
-      "logps/rejected": -37632.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -372.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -372.0,
+      "logits/chosen": -12.5,
+      "logits/rejected": -9.5,
+      "logps/chosen": -800.0,
+      "logps/rejected": -4864.0,
+      "loss": 0.0804,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -4.28125,
+      "rewards/margins": 23.25,
+      "rewards/rejected": -27.5,
       "step": 300
     },
     {
       "epoch": 0.31413612565445026,
-      "eval_logits/chosen": -0.78515625,
-      "eval_logits/rejected": -0.7578125,
-      "eval_logps/chosen": -33536.0,
-      "eval_logps/rejected": -29056.0,
-      "eval_loss": 98.02375030517578,
-      "eval_rewards/accuracies": 0.435546875,
-      "eval_rewards/chosen": -332.0,
-      "eval_rewards/margins": -45.25,
-      "eval_rewards/rejected": -288.0,
-      "eval_runtime": 46.6337,
-      "eval_samples_per_second": 42.887,
+      "eval_logits/chosen": -11.3125,
+      "eval_logits/rejected": -10.9375,
+      "eval_logps/chosen": -756.0,
+      "eval_logps/rejected": -748.0,
+      "eval_loss": 1.1384999752044678,
+      "eval_rewards/accuracies": 0.486328125,
+      "eval_rewards/chosen": -4.375,
+      "eval_rewards/margins": 0.2275390625,
+      "eval_rewards/rejected": -4.59375,
+      "eval_runtime": 46.6465,
+      "eval_samples_per_second": 42.876,
       "eval_steps_per_second": 0.686,
       "step": 300
     },
     {
       "epoch": 0.32460732984293195,
-      "grad_norm": 0.019237968587820158,
+      "grad_norm": 0.8540019871652407,
       "learning_rate": 4.9990180911526896e-05,
-      "logits/chosen": -0.8984375,
-      "logits/rejected": -0.8984375,
-      "logps/chosen": -37632.0,
-      "logps/rejected": -37632.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -372.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -372.0,
+      "logits/chosen": -10.9375,
+      "logits/rejected": -9.125,
+      "logps/chosen": -820.0,
+      "logps/rejected": -5440.0,
+      "loss": 0.0322,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 26.375,
+      "rewards/rejected": -31.125,
       "step": 310
     },
     {
       "epoch": 0.33507853403141363,
-      "grad_norm": 0.014383880242345944,
+      "grad_norm": 0.698629303905687,
       "learning_rate": 4.997978781413745e-05,
-      "logits/chosen": -0.8515625,
-      "logits/rejected": -0.8515625,
-      "logps/chosen": -40960.0,
-      "logps/rejected": -40960.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -406.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -406.0,
+      "logits/chosen": -9.9375,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -664.0,
+      "logps/rejected": -4512.0,
+      "loss": 0.0524,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.515625,
+      "rewards/margins": 20.875,
+      "rewards/rejected": -23.5,
       "step": 320
     },
     {
       "epoch": 0.34554973821989526,
-      "grad_norm": 0.012678003777085308,
+      "grad_norm": 6.93838070420919,
       "learning_rate": 4.996568520039565e-05,
-      "logits/chosen": -0.91015625,
-      "logits/rejected": -0.91015625,
-      "logps/chosen": -42752.0,
-      "logps/rejected": -42752.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -424.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -424.0,
+      "logits/chosen": -9.5,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -692.0,
+      "logps/rejected": -4288.0,
+      "loss": 0.0622,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 17.5,
+      "rewards/rejected": -20.625,
       "step": 330
     },
     {
       "epoch": 0.35602094240837695,
-      "grad_norm": 0.015688470782026575,
+      "grad_norm": 0.6348597490782729,
       "learning_rate": 4.994787516454973e-05,
-      "logits/chosen": -0.83984375,
-      "logits/rejected": -0.83984375,
-      "logps/chosen": -41472.0,
-      "logps/rejected": -41472.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -410.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -410.0,
+      "logits/chosen": -8.625,
+      "logits/rejected": -6.65625,
+      "logps/chosen": -780.0,
+      "logps/rejected": -4992.0,
+      "loss": 0.0399,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -4.1875,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -27.5,
       "step": 340
     },
     {
       "epoch": 0.36649214659685864,
-      "grad_norm": 0.015961134904159947,
+      "grad_norm": 0.8946971684670513,
       "learning_rate": 4.992636035140273e-05,
-      "logits/chosen": -0.7109375,
-      "logits/rejected": -0.7109375,
-      "logps/chosen": -41216.0,
-      "logps/rejected": -41216.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -408.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -408.0,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -6.3125,
+      "logps/chosen": -716.0,
+      "logps/rejected": -4416.0,
+      "loss": 0.0344,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.78125,
+      "rewards/margins": 17.875,
+      "rewards/rejected": -21.625,
       "step": 350
     },
     {
       "epoch": 0.3769633507853403,
-      "grad_norm": 0.014661617836548764,
+      "grad_norm": 0.12075685121446786,
       "learning_rate": 4.99011439559198e-05,
-      "logits/chosen": -0.65234375,
-      "logits/rejected": -0.65234375,
-      "logps/chosen": -44032.0,
-      "logps/rejected": -44032.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -436.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -436.0,
+      "logits/chosen": -7.875,
+      "logits/rejected": -5.71875,
+      "logps/chosen": -800.0,
+      "logps/rejected": -4960.0,
+      "loss": 0.0271,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -4.28125,
+      "rewards/margins": 23.25,
+      "rewards/rejected": -27.5,
       "step": 360
     },
     {
       "epoch": 0.387434554973822,
-      "grad_norm": 0.01599841693976505,
+      "grad_norm": 0.5851342767470975,
       "learning_rate": 4.987222972275369e-05,
-      "logits/chosen": -0.7109375,
-      "logits/rejected": -0.7109375,
-      "logps/chosen": -43520.0,
-      "logps/rejected": -43520.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -432.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -432.0,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -6.15625,
+      "logps/chosen": -804.0,
+      "logps/rejected": -4896.0,
+      "loss": 0.0293,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -4.40625,
+      "rewards/margins": 23.125,
+      "rewards/rejected": -27.625,
       "step": 370
     },
     {
       "epoch": 0.39790575916230364,
-      "grad_norm": 0.0255118604098875,
+      "grad_norm": 0.34141007194531986,
       "learning_rate": 4.983962194568872e-05,
-      "logits/chosen": -0.5703125,
-      "logits/rejected": -0.5703125,
-      "logps/chosen": -44032.0,
-      "logps/rejected": -44032.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -438.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -438.0,
+      "logits/chosen": -11.0,
+      "logits/rejected": -8.25,
+      "logps/chosen": -784.0,
+      "logps/rejected": -4992.0,
+      "loss": 0.0676,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -4.03125,
+      "rewards/margins": 24.0,
+      "rewards/rejected": -28.0,
       "step": 380
     },
     {
       "epoch": 0.4083769633507853,
-      "grad_norm": 0.018642454073440458,
+      "grad_norm": 0.1393886891579739,
       "learning_rate": 4.9803325467003084e-05,
-      "logits/chosen": -0.5859375,
-      "logits/rejected": -0.5859375,
-      "logps/chosen": -46336.0,
-      "logps/rejected": -46336.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -460.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -460.0,
+      "logits/chosen": -12.9375,
+      "logits/rejected": -10.0,
+      "logps/chosen": -716.0,
+      "logps/rejected": -4736.0,
+      "loss": 0.0286,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.359375,
+      "rewards/margins": 21.875,
+      "rewards/rejected": -25.25,
       "step": 390
     },
     {
       "epoch": 0.418848167539267,
-      "grad_norm": 0.01521769846964594,
+      "grad_norm": 0.3005530849527355,
       "learning_rate": 4.9763345676749816e-05,
-      "logits/chosen": -0.62109375,
-      "logits/rejected": -0.62109375,
-      "logps/chosen": -42752.0,
-      "logps/rejected": -42752.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -424.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -424.0,
+      "logits/chosen": -12.875,
+      "logits/rejected": -9.375,
+      "logps/chosen": -688.0,
+      "logps/rejected": -4288.0,
+      "loss": 0.1502,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -20.875,
       "step": 400
     },
     {
       "epoch": 0.418848167539267,
-      "eval_logits/chosen": -0.341796875,
-      "eval_logits/rejected": -0.2431640625,
-      "eval_logps/chosen": -35840.0,
-      "eval_logps/rejected": -30848.0,
-      "eval_loss": 104.31487274169922,
-      "eval_rewards/accuracies": 0.435546875,
-      "eval_rewards/chosen": -354.0,
-      "eval_rewards/margins": -48.25,
-      "eval_rewards/rejected": -306.0,
-      "eval_runtime": 46.6455,
-      "eval_samples_per_second": 42.877,
+      "eval_logits/chosen": -11.9375,
+      "eval_logits/rejected": -11.375,
+      "eval_logps/chosen": -640.0,
+      "eval_logps/rejected": -620.0,
+      "eval_loss": 0.980093777179718,
+      "eval_rewards/accuracies": 0.484375,
+      "eval_rewards/chosen": -3.203125,
+      "eval_rewards/margins": 0.09912109375,
+      "eval_rewards/rejected": -3.3125,
+      "eval_runtime": 46.6531,
+      "eval_samples_per_second": 42.87,
       "eval_steps_per_second": 0.686,
       "step": 400
     },
     {
       "epoch": 0.4293193717277487,
-      "grad_norm": 0.019813228956494042,
+      "grad_norm": 1.2078529312118853,
       "learning_rate": 4.9719688511956374e-05,
-      "logits/chosen": -0.408203125,
-      "logits/rejected": -0.408203125,
-      "logps/chosen": -43264.0,
-      "logps/rejected": -43264.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -428.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -428.0,
+      "logits/chosen": -12.0,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -732.0,
+      "logps/rejected": -4672.0,
+      "loss": 0.0513,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 19.625,
+      "rewards/rejected": -23.0,
       "step": 410
     },
     {
       "epoch": 0.4397905759162304,
-      "grad_norm": 0.01459213750043445,
+      "grad_norm": 0.7776483169771649,
       "learning_rate": 4.967236045574294e-05,
-      "logits/chosen": -0.267578125,
-      "logits/rejected": -0.267578125,
-      "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": -11.375,
+      "logits/rejected": -8.1875,
+      "logps/chosen": -688.0,
+      "logps/rejected": -4080.0,
+      "loss": 0.0587,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 14.75,
+      "rewards/rejected": -17.875,
       "step": 420
     },
     {
       "epoch": 0.450261780104712,
-      "grad_norm": 0.02096728046906915,
+      "grad_norm": 1.8238426035865865,
       "learning_rate": 4.96213685363597e-05,
-      "logits/chosen": -0.15234375,
-      "logits/rejected": -0.15234375,
-      "logps/chosen": -46592.0,
-      "logps/rejected": -46592.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -462.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -462.0,
+      "logits/chosen": -11.9375,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -688.0,
+      "logps/rejected": -4320.0,
+      "loss": 0.0568,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 17.75,
+      "rewards/rejected": -21.0,
       "step": 430
     },
     {
       "epoch": 0.4607329842931937,
-      "grad_norm": 0.016175485334847098,
+      "grad_norm": 0.8517204127038045,
       "learning_rate": 4.956672032614315e-05,
-      "logits/chosen": -0.203125,
-      "logits/rejected": -0.203125,
-      "logps/chosen": -44544.0,
-      "logps/rejected": -44544.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -442.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -442.0,
+      "logits/chosen": -12.5,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -772.0,
+      "logps/rejected": -4320.0,
+      "loss": 0.0376,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.8125,
+      "rewards/margins": 17.5,
+      "rewards/rejected": -21.25,
       "step": 440
     },
     {
       "epoch": 0.4712041884816754,
-      "grad_norm": 0.020467403034003687,
+      "grad_norm": 0.6397948866920252,
       "learning_rate": 4.950842394039157e-05,
-      "logits/chosen": -0.03662109375,
-      "logits/rejected": -0.03662109375,
-      "logps/chosen": -44288.0,
-      "logps/rejected": -44288.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -440.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -440.0,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -10.375,
+      "logps/chosen": -724.0,
+      "logps/rejected": -4320.0,
+      "loss": 0.0401,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.75,
+      "rewards/margins": 16.875,
+      "rewards/rejected": -20.625,
       "step": 450
     },
     {
       "epoch": 0.4816753926701571,
-      "grad_norm": 0.0160458094749787,
+      "grad_norm": 0.34860288685526086,
       "learning_rate": 4.944648803615994e-05,
-      "logits/chosen": 0.2431640625,
-      "logits/rejected": 0.2431640625,
-      "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": -14.875,
+      "logits/rejected": -11.75,
+      "logps/chosen": -716.0,
+      "logps/rejected": -4320.0,
+      "loss": 0.0472,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 16.75,
+      "rewards/rejected": -20.0,
       "step": 460
     },
     {
       "epoch": 0.49214659685863876,
-      "grad_norm": 0.015122791248975012,
+      "grad_norm": 0.3538097512815358,
       "learning_rate": 4.938092181097433e-05,
-      "logits/chosen": 0.2373046875,
-      "logits/rejected": 0.2373046875,
-      "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": -14.75,
+      "logits/rejected": -11.0625,
+      "logps/chosen": -716.0,
+      "logps/rejected": -4608.0,
+      "loss": 0.0576,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 20.75,
+      "rewards/rejected": -24.25,
       "step": 470
     },
     {
       "epoch": 0.5026178010471204,
-      "grad_norm": 0.01532192029609866,
+      "grad_norm": 0.3816177982312744,
       "learning_rate": 4.931173500146607e-05,
-      "logits/chosen": 0.4921875,
-      "logits/rejected": 0.4921875,
-      "logps/chosen": -43008.0,
-      "logps/rejected": -43008.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -428.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -428.0,
+      "logits/chosen": -13.0,
+      "logits/rejected": -10.0,
+      "logps/chosen": -692.0,
+      "logps/rejected": -5024.0,
+      "loss": 0.07,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.46875,
+      "rewards/margins": 24.125,
+      "rewards/rejected": -27.5,
       "step": 480
     },
     {
       "epoch": 0.5130890052356021,
-      "grad_norm": 0.018546204985322143,
+      "grad_norm": 0.37129419260040497,
       "learning_rate": 4.923893788192582e-05,
-      "logits/chosen": 0.349609375,
-      "logits/rejected": 0.349609375,
-      "logps/chosen": -46848.0,
-      "logps/rejected": -46848.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -464.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -464.0,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -9.375,
+      "logps/chosen": -688.0,
+      "logps/rejected": -4416.0,
+      "loss": 0.0397,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 19.0,
+      "rewards/rejected": -22.125,
       "step": 490
     },
     {
       "epoch": 0.5235602094240838,
-      "grad_norm": 0.013381355641213297,
+      "grad_norm": 0.5221131080893542,
       "learning_rate": 4.9162541262777906e-05,
-      "logits/chosen": 0.470703125,
-      "logits/rejected": 0.470703125,
-      "logps/chosen": -50944.0,
-      "logps/rejected": -50944.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -504.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -504.0,
+      "logits/chosen": -13.875,
+      "logits/rejected": -10.0,
+      "logps/chosen": -668.0,
+      "logps/rejected": -4352.0,
+      "loss": 0.0464,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -3.0,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -21.875,
       "step": 500
     },
     {
       "epoch": 0.5235602094240838,
-      "eval_logits/chosen": 1.046875,
-      "eval_logits/rejected": 1.2421875,
-      "eval_logps/chosen": -37888.0,
-      "eval_logps/rejected": -32768.0,
-      "eval_loss": 111.05012512207031,
-      "eval_rewards/accuracies": 0.435546875,
-      "eval_rewards/chosen": -376.0,
-      "eval_rewards/margins": -51.25,
-      "eval_rewards/rejected": -326.0,
-      "eval_runtime": 46.6358,
-      "eval_samples_per_second": 42.886,
+      "eval_logits/chosen": -13.5625,
+      "eval_logits/rejected": -13.125,
+      "eval_logps/chosen": -588.0,
+      "eval_logps/rejected": -568.0,
+      "eval_loss": 0.9622499942779541,
+      "eval_rewards/accuracies": 0.48046875,
+      "eval_rewards/chosen": -2.6875,
+      "eval_rewards/margins": 0.107421875,
+      "eval_rewards/rejected": -2.796875,
+      "eval_runtime": 46.6656,
+      "eval_samples_per_second": 42.858,
       "eval_steps_per_second": 0.686,
       "step": 500
     },
     {
       "epoch": 0.5340314136125655,
-      "grad_norm": 0.0153900888694924,
+      "grad_norm": 0.16193646910210274,
       "learning_rate": 4.90825564889749e-05,
-      "logits/chosen": 0.52734375,
-      "logits/rejected": 0.52734375,
-      "logps/chosen": -46080.0,
-      "logps/rejected": -46080.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -456.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -456.0,
+      "logits/chosen": -13.8125,
+      "logits/rejected": -8.75,
+      "logps/chosen": -720.0,
+      "logps/rejected": -4576.0,
+      "loss": 0.0547,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.140625,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -23.75,
       "step": 510
     },
     {
       "epoch": 0.5445026178010471,
-      "grad_norm": 0.018673355122588966,
+      "grad_norm": 0.7376985259809453,
       "learning_rate": 4.899899543831288e-05,
-      "logits/chosen": 0.490234375,
-      "logits/rejected": 0.490234375,
-      "logps/chosen": -47104.0,
-      "logps/rejected": -47104.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -468.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -468.0,
+      "logits/chosen": -14.875,
+      "logits/rejected": -9.75,
+      "logps/chosen": -720.0,
+      "logps/rejected": -4704.0,
+      "loss": 0.0434,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.109375,
+      "rewards/margins": 20.875,
+      "rewards/rejected": -24.0,
       "step": 520
     },
     {
       "epoch": 0.5549738219895288,
-      "grad_norm": 0.024502942657144262,
+      "grad_norm": 0.5911347851619779,
       "learning_rate": 4.891187051966765e-05,
-      "logits/chosen": 0.33203125,
-      "logits/rejected": 0.33203125,
-      "logps/chosen": -43776.0,
-      "logps/rejected": -43776.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -432.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -432.0,
+      "logits/chosen": -15.5,
+      "logits/rejected": -10.875,
+      "logps/chosen": -744.0,
+      "logps/rejected": -4640.0,
+      "loss": 0.031,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.5625,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -24.125,
       "step": 530
     },
     {
       "epoch": 0.5654450261780105,
-      "grad_norm": 0.015859203742240858,
+      "grad_norm": 0.7854787560908408,
       "learning_rate": 4.882119467115191e-05,
-      "logits/chosen": 0.4921875,
-      "logits/rejected": 0.4921875,
-      "logps/chosen": -47360.0,
-      "logps/rejected": -47360.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -468.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -468.0,
+      "logits/chosen": -15.75,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -712.0,
+      "logps/rejected": -4928.0,
+      "loss": 0.0327,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -26.875,
       "step": 540
     },
     {
       "epoch": 0.5759162303664922,
-      "grad_norm": 0.018010458300495524,
+      "grad_norm": 0.3632275826792097,
       "learning_rate": 4.872698135819402e-05,
-      "logits/chosen": 0.60546875,
-      "logits/rejected": 0.60546875,
-      "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": -15.0,
+      "logits/rejected": -9.875,
+      "logps/chosen": -728.0,
+      "logps/rejected": -5088.0,
+      "loss": 0.0323,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -3.640625,
+      "rewards/margins": 25.375,
+      "rewards/rejected": -29.0,
       "step": 550
     },
     {
       "epoch": 0.5863874345549738,
-      "grad_norm": 0.017896842999353638,
+      "grad_norm": 0.49922961033893526,
       "learning_rate": 4.862924457153831e-05,
-      "logits/chosen": 0.98828125,
-      "logits/rejected": 0.98828125,
-      "logps/chosen": -41984.0,
-      "logps/rejected": -41984.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -416.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -416.0,
+      "logits/chosen": -14.8125,
+      "logits/rejected": -10.75,
+      "logps/chosen": -652.0,
+      "logps/rejected": -5024.0,
+      "loss": 0.03,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.34375,
+      "rewards/margins": 24.375,
+      "rewards/rejected": -27.625,
       "step": 560
     },
     {
       "epoch": 0.5968586387434555,
-      "grad_norm": 0.01810445830755415,
+      "grad_norm": 0.6552967210552291,
       "learning_rate": 4.8527998825167484e-05,
-      "logits/chosen": 0.87890625,
-      "logits/rejected": 0.87890625,
-      "logps/chosen": -41984.0,
-      "logps/rejected": -41984.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -416.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -416.0,
+      "logits/chosen": -12.9375,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -756.0,
+      "logps/rejected": -5056.0,
+      "loss": 0.0384,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -4.0,
+      "rewards/margins": 25.0,
+      "rewards/rejected": -29.0,
       "step": 570
     },
     {
       "epoch": 0.6073298429319371,
-      "grad_norm": 0.015321972260950964,
+      "grad_norm": 0.5103692702035204,
       "learning_rate": 4.842325915414728e-05,
-      "logits/chosen": 1.0,
-      "logits/rejected": 1.0,
-      "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": -12.625,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -812.0,
+      "logps/rejected": -5440.0,
+      "loss": 0.038,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 26.75,
+      "rewards/rejected": -31.5,
       "step": 580
     },
     {
       "epoch": 0.6178010471204188,
-      "grad_norm": 0.01893490414033851,
+      "grad_norm": 0.6181215765830905,
       "learning_rate": 4.831504111239373e-05,
-      "logits/chosen": 1.0390625,
-      "logits/rejected": 1.0390625,
-      "logps/chosen": -50688.0,
-      "logps/rejected": -50688.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -504.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -504.0,
+      "logits/chosen": -12.9375,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -740.0,
+      "logps/rejected": -4288.0,
+      "loss": 0.048,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.359375,
+      "rewards/margins": 16.5,
+      "rewards/rejected": -19.875,
       "step": 590
     },
     {
       "epoch": 0.6282722513089005,
-      "grad_norm": 0.014787816608472221,
+      "grad_norm": 0.3894028709694742,
       "learning_rate": 4.8203360770363425e-05,
-      "logits/chosen": 1.1484375,
-      "logits/rejected": 1.1484375,
-      "logps/chosen": -49920.0,
-      "logps/rejected": -49920.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -496.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -496.0,
+      "logits/chosen": -13.875,
+      "logits/rejected": -10.875,
+      "logps/chosen": -664.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0636,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.8125,
+      "rewards/margins": 15.1875,
+      "rewards/rejected": -18.0,
       "step": 600
     },
     {
       "epoch": 0.6282722513089005,
-      "eval_logits/chosen": 1.6875,
-      "eval_logits/rejected": 1.9375,
-      "eval_logps/chosen": -40192.0,
-      "eval_logps/rejected": -34816.0,
-      "eval_loss": 117.88849639892578,
-      "eval_rewards/accuracies": 0.4296875,
-      "eval_rewards/chosen": -400.0,
-      "eval_rewards/margins": -54.5,
-      "eval_rewards/rejected": -346.0,
-      "eval_runtime": 46.6423,
-      "eval_samples_per_second": 42.88,
+      "eval_logits/chosen": -14.0,
+      "eval_logits/rejected": -13.6875,
+      "eval_logps/chosen": -560.0,
+      "eval_logps/rejected": -532.0,
+      "eval_loss": 1.0378280878067017,
+      "eval_rewards/accuracies": 0.47265625,
+      "eval_rewards/chosen": -2.40625,
+      "eval_rewards/margins": 0.0263671875,
+      "eval_rewards/rejected": -2.4375,
+      "eval_runtime": 46.6572,
+      "eval_samples_per_second": 42.866,
       "eval_steps_per_second": 0.686,
       "step": 600
     },
     {
       "epoch": 0.6387434554973822,
-      "grad_norm": 0.019828257370018434,
+      "grad_norm": 0.6671924458556426,
       "learning_rate": 4.808823471266702e-05,
-      "logits/chosen": 1.5,
-      "logits/rejected": 1.5,
-      "logps/chosen": -49408.0,
-      "logps/rejected": -49408.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -490.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -490.0,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -11.125,
+      "logps/chosen": -616.0,
+      "logps/rejected": -4032.0,
+      "loss": 0.0411,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.53125,
+      "rewards/margins": 16.125,
+      "rewards/rejected": -18.625,
       "step": 610
     },
     {
       "epoch": 0.6492146596858639,
-      "grad_norm": 0.01710745293875366,
+      "grad_norm": 0.07062985082472016,
       "learning_rate": 4.796968003560639e-05,
-      "logits/chosen": 1.3828125,
-      "logits/rejected": 1.3828125,
-      "logps/chosen": -49408.0,
-      "logps/rejected": -49408.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -488.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -488.0,
+      "logits/chosen": -15.0,
+      "logits/rejected": -11.5625,
+      "logps/chosen": -696.0,
+      "logps/rejected": -4608.0,
+      "loss": 0.0708,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -23.875,
       "step": 620
     },
     {
       "epoch": 0.6596858638743456,
-      "grad_norm": 0.01770760866062657,
+      "grad_norm": 0.49486884660592556,
       "learning_rate": 4.7847714344635864e-05,
-      "logits/chosen": 1.9140625,
-      "logits/rejected": 1.9140625,
-      "logps/chosen": -48128.0,
-      "logps/rejected": -48128.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -476.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -476.0,
+      "logits/chosen": -15.25,
+      "logits/rejected": -11.4375,
+      "logps/chosen": -652.0,
+      "logps/rejected": -3872.0,
+      "loss": 0.0522,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.9375,
+      "rewards/margins": 14.8125,
+      "rewards/rejected": -17.75,
       "step": 630
     },
     {
       "epoch": 0.6701570680628273,
-      "grad_norm": 0.01861556894328915,
+      "grad_norm": 0.33129324916533526,
       "learning_rate": 4.772235575174776e-05,
-      "logits/chosen": 1.4375,
-      "logits/rejected": 1.4375,
-      "logps/chosen": -55296.0,
-      "logps/rejected": -55296.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -548.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -548.0,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -728.0,
+      "logps/rejected": -3936.0,
+      "loss": 0.0391,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 14.25,
+      "rewards/rejected": -17.625,
       "step": 640
     },
     {
       "epoch": 0.680628272251309,
-      "grad_norm": 0.02094390647262886,
+      "grad_norm": 2.4085256834262756,
       "learning_rate": 4.7593622872782754e-05,
-      "logits/chosen": 1.96875,
-      "logits/rejected": 1.96875,
-      "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": -14.6875,
+      "logits/rejected": -10.75,
+      "logps/chosen": -700.0,
+      "logps/rejected": -4224.0,
+      "loss": 0.0482,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.546875,
+      "rewards/margins": 17.375,
+      "rewards/rejected": -20.875,
       "step": 650
     },
     {
       "epoch": 0.6910994764397905,
-      "grad_norm": 0.02006081016715149,
+      "grad_norm": 0.6569028747685713,
       "learning_rate": 4.746153482466542e-05,
-      "logits/chosen": 2.015625,
-      "logits/rejected": 2.015625,
-      "logps/chosen": -49920.0,
-      "logps/rejected": -49920.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -496.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -496.0,
+      "logits/chosen": -14.75,
+      "logits/rejected": -10.25,
+      "logps/chosen": -624.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0492,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -2.734375,
+      "rewards/margins": 14.6875,
+      "rewards/rejected": -17.375,
       "step": 660
     },
     {
       "epoch": 0.7015706806282722,
-      "grad_norm": 0.019188455212096184,
+      "grad_norm": 1.9041712458199032,
       "learning_rate": 4.732611122256537e-05,
-      "logits/chosen": 1.921875,
-      "logits/rejected": 1.921875,
-      "logps/chosen": -54016.0,
-      "logps/rejected": -54016.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -536.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -536.0,
+      "logits/chosen": -14.8125,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3984.0,
+      "loss": 0.042,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.71875,
+      "rewards/margins": 15.625,
+      "rewards/rejected": -18.375,
       "step": 670
     },
     {
       "epoch": 0.7120418848167539,
-      "grad_norm": 0.017895312478557714,
+      "grad_norm": 0.4552961797267801,
       "learning_rate": 4.718737217698435e-05,
-      "logits/chosen": 1.875,
-      "logits/rejected": 1.875,
-      "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": -14.625,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -656.0,
+      "logps/rejected": -4288.0,
+      "loss": 0.0688,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -20.625,
       "step": 680
     },
     {
       "epoch": 0.7225130890052356,
-      "grad_norm": 0.01769914915725165,
+      "grad_norm": 0.30627525636905933,
       "learning_rate": 4.704533829076982e-05,
-      "logits/chosen": 1.8984375,
-      "logits/rejected": 1.8984375,
-      "logps/chosen": -52224.0,
-      "logps/rejected": -52224.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -516.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -516.0,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -10.375,
+      "logps/chosen": -636.0,
+      "logps/rejected": -3776.0,
+      "loss": 0.0751,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.671875,
+      "rewards/margins": 12.4375,
+      "rewards/rejected": -15.125,
       "step": 690
     },
     {
       "epoch": 0.7329842931937173,
-      "grad_norm": 0.017992198005302676,
+      "grad_norm": 0.9646469500045647,
       "learning_rate": 4.6900030656055455e-05,
-      "logits/chosen": 1.9921875,
-      "logits/rejected": 1.9921875,
-      "logps/chosen": -51456.0,
-      "logps/rejected": -51456.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -512.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -512.0,
+      "logits/chosen": -14.0,
+      "logits/rejected": -9.875,
+      "logps/chosen": -608.0,
+      "logps/rejected": -3424.0,
+      "loss": 0.0638,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -2.296875,
+      "rewards/margins": 10.1875,
+      "rewards/rejected": -12.4375,
       "step": 700
     },
     {
       "epoch": 0.7329842931937173,
-      "eval_logits/chosen": 2.859375,
-      "eval_logits/rejected": 3.203125,
-      "eval_logps/chosen": -42752.0,
-      "eval_logps/rejected": -36864.0,
-      "eval_loss": 124.87437438964844,
-      "eval_rewards/accuracies": 0.43359375,
-      "eval_rewards/chosen": -424.0,
-      "eval_rewards/margins": -57.75,
-      "eval_rewards/rejected": -366.0,
-      "eval_runtime": 46.6367,
-      "eval_samples_per_second": 42.885,
+      "eval_logits/chosen": -13.875,
+      "eval_logits/rejected": -13.6875,
+      "eval_logps/chosen": -532.0,
+      "eval_logps/rejected": -504.0,
+      "eval_loss": 0.8977500200271606,
+      "eval_rewards/accuracies": 0.50390625,
+      "eval_rewards/chosen": -2.15625,
+      "eval_rewards/margins": 0.003692626953125,
+      "eval_rewards/rejected": -2.15625,
+      "eval_runtime": 46.6482,
+      "eval_samples_per_second": 42.874,
       "eval_steps_per_second": 0.686,
       "step": 700
     },
     {
       "epoch": 0.743455497382199,
-      "grad_norm": 0.017045818276090623,
+      "grad_norm": 0.38775239867598665,
       "learning_rate": 4.6751470851128906e-05,
-      "logits/chosen": 1.921875,
-      "logits/rejected": 1.921875,
-      "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": -9.9375,
+      "logps/chosen": -636.0,
+      "logps/rejected": -3600.0,
+      "loss": 0.0492,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -2.359375,
+      "rewards/margins": 10.6875,
+      "rewards/rejected": -13.0625,
       "step": 710
     },
     {
       "epoch": 0.7539267015706806,
-      "grad_norm": 0.017823751951018818,
+      "grad_norm": 0.312291238843419,
       "learning_rate": 4.6599680937227454e-05,
-      "logits/chosen": 2.171875,
-      "logits/rejected": 2.171875,
-      "logps/chosen": -53248.0,
-      "logps/rejected": -53248.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -528.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -528.0,
+      "logits/chosen": -14.8125,
+      "logits/rejected": -10.875,
+      "logps/chosen": -636.0,
+      "logps/rejected": -3952.0,
+      "loss": 0.036,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.515625,
+      "rewards/margins": 14.4375,
+      "rewards/rejected": -17.0,
       "step": 720
     },
     {
       "epoch": 0.7643979057591623,
-      "grad_norm": 0.016985845818632998,
+      "grad_norm": 0.8880424590500766,
       "learning_rate": 4.644468345526183e-05,
-      "logits/chosen": 2.171875,
-      "logits/rejected": 2.171875,
-      "logps/chosen": -55296.0,
-      "logps/rejected": -55296.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -548.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -548.0,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -11.125,
+      "logps/chosen": -652.0,
+      "logps/rejected": -4128.0,
+      "loss": 0.0386,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -2.734375,
+      "rewards/margins": 16.0,
+      "rewards/rejected": -18.75,
       "step": 730
     },
     {
       "epoch": 0.774869109947644,
-      "grad_norm": 0.016871940650217698,
+      "grad_norm": 0.5688666808177001,
       "learning_rate": 4.628650142246894e-05,
-      "logits/chosen": 2.34375,
-      "logits/rejected": 2.34375,
-      "logps/chosen": -48640.0,
-      "logps/rejected": -48640.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -482.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -482.0,
+      "logits/chosen": -15.875,
+      "logits/rejected": -12.0,
+      "logps/chosen": -620.0,
+      "logps/rejected": -3968.0,
+      "loss": 0.0321,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.671875,
+      "rewards/margins": 15.6875,
+      "rewards/rejected": -18.375,
       "step": 740
     },
     {
       "epoch": 0.7853403141361257,
-      "grad_norm": 0.018904677740562464,
+      "grad_norm": 4.074697850871747,
       "learning_rate": 4.6125158328993765e-05,
-      "logits/chosen": 2.234375,
-      "logits/rejected": 2.234375,
-      "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": -17.0,
+      "logits/rejected": -12.5625,
+      "logps/chosen": -676.0,
+      "logps/rejected": -4384.0,
+      "loss": 0.0409,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -21.125,
       "step": 750
     },
     {
       "epoch": 0.7958115183246073,
-      "grad_norm": 0.015792160252367262,
+      "grad_norm": 1.468236053388505,
       "learning_rate": 4.5960678134401064e-05,
-      "logits/chosen": 2.609375,
-      "logits/rejected": 2.609375,
-      "logps/chosen": -57344.0,
-      "logps/rejected": -57344.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -568.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -568.0,
+      "logits/chosen": -16.375,
+      "logits/rejected": -10.375,
+      "logps/chosen": -724.0,
+      "logps/rejected": -4640.0,
+      "loss": 0.0485,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 21.375,
+      "rewards/rejected": -24.875,
       "step": 760
     },
     {
       "epoch": 0.806282722513089,
-      "grad_norm": 0.01791951592603191,
+      "grad_norm": 0.15847846136806318,
       "learning_rate": 4.579308526411734e-05,
-      "logits/chosen": 3.21875,
-      "logits/rejected": 3.21875,
-      "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": -15.5,
+      "logits/rejected": -9.875,
+      "logps/chosen": -684.0,
+      "logps/rejected": -4864.0,
+      "loss": 0.0426,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 22.75,
+      "rewards/rejected": -26.0,
       "step": 770
     },
     {
       "epoch": 0.8167539267015707,
-      "grad_norm": 0.018561258510443056,
+      "grad_norm": 2.0119322591053357,
       "learning_rate": 4.562240460580367e-05,
-      "logits/chosen": 2.734375,
-      "logits/rejected": 2.734375,
-      "logps/chosen": -55296.0,
-      "logps/rejected": -55296.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -548.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -548.0,
+      "logits/chosen": -15.625,
+      "logits/rejected": -10.375,
+      "logps/chosen": -680.0,
+      "logps/rejected": -4544.0,
+      "loss": 0.0461,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.09375,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -22.5,
       "step": 780
     },
     {
       "epoch": 0.8272251308900523,
-      "grad_norm": 0.019975426108116028,
+      "grad_norm": 1.3776285180015304,
       "learning_rate": 4.544866150565985e-05,
-      "logits/chosen": 3.1875,
-      "logits/rejected": 3.1875,
-      "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": -15.0625,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3904.0,
+      "loss": 0.0439,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.15625,
+      "rewards/margins": 14.0,
+      "rewards/rejected": -17.125,
       "step": 790
     },
     {
       "epoch": 0.837696335078534,
-      "grad_norm": 0.0178963507722851,
+      "grad_norm": 0.778290635061097,
       "learning_rate": 4.527188176466051e-05,
-      "logits/chosen": 2.8125,
-      "logits/rejected": 2.8125,
-      "logps/chosen": -55808.0,
-      "logps/rejected": -55808.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -552.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -552.0,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -732.0,
+      "logps/rejected": -3680.0,
+      "loss": 0.0552,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.40625,
+      "rewards/margins": 11.875,
+      "rewards/rejected": -15.25,
       "step": 800
     },
     {
       "epoch": 0.837696335078534,
-      "eval_logits/chosen": 3.828125,
-      "eval_logits/rejected": 4.21875,
-      "eval_logps/chosen": -45056.0,
-      "eval_logps/rejected": -38912.0,
-      "eval_loss": 132.08238220214844,
-      "eval_rewards/accuracies": 0.4296875,
-      "eval_rewards/chosen": -448.0,
-      "eval_rewards/margins": -61.0,
-      "eval_rewards/rejected": -388.0,
-      "eval_runtime": 46.6488,
-      "eval_samples_per_second": 42.874,
+      "eval_logits/chosen": -13.5625,
+      "eval_logits/rejected": -13.0625,
+      "eval_logps/chosen": -664.0,
+      "eval_logps/rejected": -636.0,
+      "eval_loss": 0.9712187647819519,
+      "eval_rewards/accuracies": 0.498046875,
+      "eval_rewards/chosen": -3.4375,
+      "eval_rewards/margins": 0.033203125,
+      "eval_rewards/rejected": -3.46875,
+      "eval_runtime": 46.644,
+      "eval_samples_per_second": 42.878,
       "eval_steps_per_second": 0.686,
       "step": 800
     },
     {
       "epoch": 0.8481675392670157,
-      "grad_norm": 0.016494461567137887,
+      "grad_norm": 0.3128438962468461,
       "learning_rate": 4.509209163472357e-05,
-      "logits/chosen": 2.890625,
-      "logits/rejected": 2.890625,
-      "logps/chosen": -58880.0,
-      "logps/rejected": -58880.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -584.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -584.0,
+      "logits/chosen": -13.75,
+      "logits/rejected": -7.84375,
+      "logps/chosen": -736.0,
+      "logps/rejected": -3728.0,
+      "loss": 0.0309,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 11.4375,
+      "rewards/rejected": -15.0,
       "step": 810
     },
     {
       "epoch": 0.8586387434554974,
-      "grad_norm": 0.015672697577052794,
+      "grad_norm": 1.7684713233269638,
       "learning_rate": 4.4909317814811896e-05,
-      "logits/chosen": 3.09375,
-      "logits/rejected": 3.09375,
-      "logps/chosen": -55808.0,
-      "logps/rejected": -55808.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -552.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -552.0,
+      "logits/chosen": -14.5,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -704.0,
+      "logps/rejected": -3968.0,
+      "loss": 0.0312,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.28125,
+      "rewards/margins": 13.125,
+      "rewards/rejected": -16.375,
       "step": 820
     },
     {
       "epoch": 0.8691099476439791,
-      "grad_norm": 0.01665230251194738,
+      "grad_norm": 0.9671547600463704,
       "learning_rate": 4.472358744696845e-05,
-      "logits/chosen": 3.0625,
-      "logits/rejected": 3.0625,
-      "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": -15.5,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -728.0,
+      "logps/rejected": -4032.0,
+      "loss": 0.0304,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.578125,
+      "rewards/margins": 15.9375,
+      "rewards/rejected": -19.5,
       "step": 830
     },
     {
       "epoch": 0.8795811518324608,
-      "grad_norm": 0.018363596504792717,
+      "grad_norm": 0.8997822991547062,
       "learning_rate": 4.4534928112285645e-05,
-      "logits/chosen": 3.125,
-      "logits/rejected": 3.125,
-      "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": -15.4375,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -736.0,
+      "logps/rejected": -4512.0,
+      "loss": 0.0244,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.640625,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -23.375,
       "step": 840
     },
     {
       "epoch": 0.8900523560209425,
-      "grad_norm": 0.018846844290474726,
+      "grad_norm": 0.3500933562017802,
       "learning_rate": 4.43433678268096e-05,
-      "logits/chosen": 2.8125,
-      "logits/rejected": 2.8125,
-      "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": -15.625,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -752.0,
+      "logps/rejected": -4800.0,
+      "loss": 0.0319,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.9375,
+      "rewards/margins": 20.75,
+      "rewards/rejected": -24.75,
       "step": 850
     },
     {
       "epoch": 0.900523560209424,
-      "grad_norm": 0.018283137124858666,
+      "grad_norm": 1.0316863432554277,
       "learning_rate": 4.414893503737968e-05,
-      "logits/chosen": 3.109375,
-      "logits/rejected": 3.109375,
-      "logps/chosen": -53248.0,
-      "logps/rejected": -53248.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -528.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -528.0,
+      "logits/chosen": -16.125,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -760.0,
+      "logps/rejected": -4928.0,
+      "loss": 0.0395,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -4.1875,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -27.625,
       "step": 860
     },
     {
       "epoch": 0.9109947643979057,
-      "grad_norm": 0.020465139796148966,
+      "grad_norm": 0.8328382747668167,
       "learning_rate": 4.395165861740414e-05,
-      "logits/chosen": 2.984375,
-      "logits/rejected": 2.984375,
-      "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": -15.9375,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -724.0,
+      "logps/rejected": -4992.0,
+      "loss": 0.0571,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -3.90625,
+      "rewards/margins": 24.25,
+      "rewards/rejected": -28.25,
       "step": 870
     },
     {
       "epoch": 0.9214659685863874,
-      "grad_norm": 0.01880113374820466,
+      "grad_norm": 0.741672208779536,
       "learning_rate": 4.375156786257241e-05,
-      "logits/chosen": 2.765625,
-      "logits/rejected": 2.765625,
-      "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": -14.125,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -864.0,
+      "logps/rejected": -4672.0,
+      "loss": 0.0279,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -24.75,
       "step": 880
     },
     {
       "epoch": 0.9319371727748691,
-      "grad_norm": 0.018217686240705935,
+      "grad_norm": 0.4782644365241539,
       "learning_rate": 4.354869248650466e-05,
-      "logits/chosen": 3.140625,
-      "logits/rejected": 3.140625,
-      "logps/chosen": -58880.0,
-      "logps/rejected": -58880.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -584.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -584.0,
+      "logits/chosen": -12.375,
+      "logits/rejected": -8.125,
+      "logps/chosen": -904.0,
+      "logps/rejected": -4640.0,
+      "loss": 0.0521,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -5.15625,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -24.0,
       "step": 890
     },
     {
       "epoch": 0.9424083769633508,
-      "grad_norm": 0.018045951064972983,
+      "grad_norm": 1.7268532151599367,
       "learning_rate": 4.334306261633931e-05,
-      "logits/chosen": 3.046875,
-      "logits/rejected": 3.046875,
-      "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": -13.5,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -816.0,
+      "logps/rejected": -4352.0,
+      "loss": 0.0459,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -4.125,
+      "rewards/margins": 16.375,
+      "rewards/rejected": -20.5,
       "step": 900
     },
     {
       "epoch": 0.9424083769633508,
-      "eval_logits/chosen": 4.3125,
-      "eval_logits/rejected": 4.6875,
-      "eval_logps/chosen": -47616.0,
-      "eval_logps/rejected": -40960.0,
-      "eval_loss": 139.28286743164062,
-      "eval_rewards/accuracies": 0.4296875,
-      "eval_rewards/chosen": -472.0,
-      "eval_rewards/margins": -64.5,
-      "eval_rewards/rejected": -408.0,
-      "eval_runtime": 46.6517,
-      "eval_samples_per_second": 42.871,
+      "eval_logits/chosen": -13.0,
+      "eval_logits/rejected": -12.625,
+      "eval_logps/chosen": -788.0,
+      "eval_logps/rejected": -784.0,
+      "eval_loss": 1.0447499752044678,
+      "eval_rewards/accuracies": 0.51171875,
+      "eval_rewards/chosen": -4.71875,
+      "eval_rewards/margins": 0.2490234375,
+      "eval_rewards/rejected": -4.96875,
+      "eval_runtime": 46.646,
+      "eval_samples_per_second": 42.876,
       "eval_steps_per_second": 0.686,
       "step": 900
     },
     {
       "epoch": 0.9528795811518325,
-      "grad_norm": 0.01652647304768983,
+      "grad_norm": 2.881059069126684,
       "learning_rate": 4.313470878825911e-05,
-      "logits/chosen": 3.1875,
-      "logits/rejected": 3.1875,
-      "logps/chosen": -57344.0,
-      "logps/rejected": -57344.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -568.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -568.0,
+      "logits/chosen": -13.0,
+      "logits/rejected": -9.125,
+      "logps/chosen": -840.0,
+      "logps/rejected": -4736.0,
+      "loss": 0.0395,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -4.78125,
+      "rewards/margins": 20.375,
+      "rewards/rejected": -25.125,
       "step": 910
     },
     {
       "epoch": 0.9633507853403142,
-      "grad_norm": 0.024583461320126055,
+      "grad_norm": 0.5897972554591244,
       "learning_rate": 4.292366194295652e-05,
-      "logits/chosen": 3.625,
-      "logits/rejected": 3.625,
-      "logps/chosen": -57344.0,
-      "logps/rejected": -57344.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -568.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -568.0,
+      "logits/chosen": -13.6875,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -824.0,
+      "logps/rejected": -4512.0,
+      "loss": 0.0795,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -4.53125,
+      "rewards/margins": 19.0,
+      "rewards/rejected": -23.625,
       "step": 920
     },
     {
       "epoch": 0.9738219895287958,
-      "grad_norm": 0.01938759368846602,
+      "grad_norm": 0.5736792508898516,
       "learning_rate": 4.2709953421038964e-05,
-      "logits/chosen": 3.625,
-      "logits/rejected": 3.625,
-      "logps/chosen": -57344.0,
-      "logps/rejected": -57344.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -568.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -568.0,
+      "logits/chosen": -11.75,
+      "logits/rejected": -8.25,
+      "logps/chosen": -676.0,
+      "logps/rejected": -3920.0,
+      "loss": 0.0478,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -3.171875,
+      "rewards/margins": 13.5,
+      "rewards/rejected": -16.625,
       "step": 930
     },
     {
       "epoch": 0.9842931937172775,
-      "grad_norm": 0.01701357609442511,
+      "grad_norm": 1.0930940664794397,
       "learning_rate": 4.249361495837478e-05,
-      "logits/chosen": 3.15625,
-      "logits/rejected": 3.15625,
-      "logps/chosen": -55808.0,
-      "logps/rejected": -55808.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -552.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -552.0,
+      "logits/chosen": -10.8125,
+      "logits/rejected": -7.53125,
+      "logps/chosen": -672.0,
+      "logps/rejected": -3712.0,
+      "loss": 0.0416,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 11.6875,
+      "rewards/rejected": -14.8125,
       "step": 940
     },
     {
       "epoch": 0.9947643979057592,
-      "grad_norm": 0.016047671474244896,
+      "grad_norm": 0.7635076512096929,
       "learning_rate": 4.227467868138035e-05,
-      "logits/chosen": 3.421875,
-      "logits/rejected": 3.421875,
-      "logps/chosen": -59392.0,
-      "logps/rejected": -59392.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -588.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -588.0,
+      "logits/chosen": -9.5625,
+      "logits/rejected": -6.84375,
+      "logps/chosen": -844.0,
+      "logps/rejected": -4192.0,
+      "loss": 0.0344,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -4.65625,
+      "rewards/margins": 14.3125,
+      "rewards/rejected": -19.0,
       "step": 950
     },
     {
       "epoch": 1.0052356020942408,
-      "grad_norm": 0.01714425565416694,
+      "grad_norm": 0.6567951235566086,
       "learning_rate": 4.205317710224938e-05,
-      "logits/chosen": 3.203125,
-      "logits/rejected": 3.203125,
-      "logps/chosen": -59136.0,
-      "logps/rejected": -59136.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -588.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -588.0,
+      "logits/chosen": -11.25,
+      "logits/rejected": -7.0,
+      "logps/chosen": -976.0,
+      "logps/rejected": -4896.0,
+      "loss": 0.0193,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -5.90625,
+      "rewards/margins": 20.875,
+      "rewards/rejected": -26.875,
       "step": 960
     },
     {
       "epoch": 1.0157068062827226,
-      "grad_norm": 0.024613821163355816,
+      "grad_norm": 4.131220810079968,
       "learning_rate": 4.1829143114124733e-05,
-      "logits/chosen": 3.59375,
-      "logits/rejected": 3.59375,
-      "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": -13.875,
+      "logits/rejected": -8.875,
+      "logps/chosen": -896.0,
+      "logps/rejected": -5088.0,
+      "loss": 0.0086,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.3125,
+      "rewards/margins": 24.25,
+      "rewards/rejected": -29.625,
       "step": 970
     },
     {
       "epoch": 1.0261780104712042,
-      "grad_norm": 0.1865341935085455,
+      "grad_norm": 0.005007248627096267,
       "learning_rate": 4.160260998621387e-05,
-      "logits/chosen": 3.1875,
-      "logits/rejected": 3.1875,
-      "logps/chosen": -68096.0,
-      "logps/rejected": -68096.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -680.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -680.0,
+      "logits/chosen": -15.5,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -884.0,
+      "logps/rejected": -5312.0,
+      "loss": 0.0028,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 26.375,
+      "rewards/rejected": -31.375,
       "step": 980
     },
     {
       "epoch": 1.036649214659686,
-      "grad_norm": 0.2146283883475591,
+      "grad_norm": 0.11149014204285719,
       "learning_rate": 4.137361135884824e-05,
-      "logits/chosen": 1.9609375,
-      "logits/rejected": 1.9609375,
-      "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": -14.8125,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -872.0,
+      "logps/rejected": -5632.0,
+      "loss": 0.0044,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 28.0,
+      "rewards/rejected": -33.0,
       "step": 990
     },
     {
       "epoch": 1.0471204188481675,
-      "grad_norm": 0.15496308010464696,
+      "grad_norm": 0.03244520122361913,
       "learning_rate": 4.114218123848777e-05,
-      "logits/chosen": 1.546875,
-      "logits/rejected": 1.546875,
-      "logps/chosen": -53760.0,
-      "logps/rejected": -53760.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -536.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -536.0,
+      "logits/chosen": -15.0625,
+      "logits/rejected": -9.0,
+      "logps/chosen": -828.0,
+      "logps/rejected": -5408.0,
+      "loss": 0.0041,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 26.5,
+      "rewards/rejected": -31.125,
       "step": 1000
     },
     {
       "epoch": 1.0471204188481675,
-      "eval_logits/chosen": 1.8671875,
-      "eval_logits/rejected": 2.21875,
-      "eval_logps/chosen": -46848.0,
-      "eval_logps/rejected": -40448.0,
-      "eval_loss": 134.4352569580078,
-      "eval_rewards/accuracies": 0.44140625,
-      "eval_rewards/chosen": -464.0,
-      "eval_rewards/margins": -63.5,
-      "eval_rewards/rejected": -400.0,
-      "eval_runtime": 46.6382,
-      "eval_samples_per_second": 42.883,
+      "eval_logits/chosen": -14.9375,
+      "eval_logits/rejected": -14.8125,
+      "eval_logps/chosen": -816.0,
+      "eval_logps/rejected": -808.0,
+      "eval_loss": 1.3026875257492065,
+      "eval_rewards/accuracies": 0.478515625,
+      "eval_rewards/chosen": -4.96875,
+      "eval_rewards/margins": 0.23828125,
+      "eval_rewards/rejected": -5.1875,
+      "eval_runtime": 46.6377,
+      "eval_samples_per_second": 42.884,
       "eval_steps_per_second": 0.686,
       "step": 1000
     },
     {
       "epoch": 1.057591623036649,
-      "grad_norm": 0.1123488946758681,
+      "grad_norm": 0.06515079695357202,
       "learning_rate": 4.0908353992670824e-05,
-      "logits/chosen": 1.8828125,
-      "logits/rejected": 1.8828125,
-      "logps/chosen": -55296.0,
-      "logps/rejected": -55296.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -548.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -548.0,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -9.25,
+      "logps/chosen": -800.0,
+      "logps/rejected": -5504.0,
+      "loss": 0.0047,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.4375,
+      "rewards/margins": 28.25,
+      "rewards/rejected": -32.75,
       "step": 1010
     },
     {
       "epoch": 1.068062827225131,
-      "grad_norm": 0.13810511139289133,
+      "grad_norm": 0.033317918659648206,
       "learning_rate": 4.067216434491059e-05,
-      "logits/chosen": 1.875,
-      "logits/rejected": 1.875,
-      "logps/chosen": -59136.0,
-      "logps/rejected": -59136.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -588.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -588.0,
+      "logits/chosen": -14.5,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -884.0,
+      "logps/rejected": -5504.0,
+      "loss": 0.0036,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.28125,
+      "rewards/margins": 27.5,
+      "rewards/rejected": -32.75,
       "step": 1020
     },
     {
       "epoch": 1.0785340314136125,
-      "grad_norm": 0.3148148203562111,
+      "grad_norm": 0.20094781152778735,
       "learning_rate": 4.043364736953866e-05,
-      "logits/chosen": 2.28125,
-      "logits/rejected": 2.28125,
-      "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": -14.5,
+      "logits/rejected": -9.125,
+      "logps/chosen": -1000.0,
+      "logps/rejected": -5344.0,
+      "loss": 0.0175,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -5.8125,
+      "rewards/margins": 25.875,
+      "rewards/rejected": -31.75,
       "step": 1030
     },
     {
       "epoch": 1.0890052356020943,
-      "grad_norm": 0.04453155125291067,
+      "grad_norm": 0.18948033638496353,
       "learning_rate": 4.019283848649643e-05,
-      "logits/chosen": 2.59375,
-      "logits/rejected": 2.59375,
-      "logps/chosen": -61440.0,
-      "logps/rejected": -61440.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -612.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -612.0,
+      "logits/chosen": -14.75,
+      "logits/rejected": -9.875,
+      "logps/chosen": -872.0,
+      "logps/rejected": -5408.0,
+      "loss": 0.0083,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 26.25,
+      "rewards/rejected": -31.375,
       "step": 1040
     },
     {
       "epoch": 1.0994764397905759,
-      "grad_norm": 0.04203978428602959,
+      "grad_norm": 0.006055788003645607,
       "learning_rate": 3.9949773456075215e-05,
-      "logits/chosen": 3.09375,
-      "logits/rejected": 3.09375,
-      "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": -14.8125,
+      "logits/rejected": -9.125,
+      "logps/chosen": -792.0,
+      "logps/rejected": -5280.0,
+      "loss": 0.0078,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.375,
+      "rewards/margins": 25.25,
+      "rewards/rejected": -29.625,
       "step": 1050
     },
     {
       "epoch": 1.1099476439790577,
-      "grad_norm": 0.042549103047232105,
+      "grad_norm": 0.5828151557653697,
       "learning_rate": 3.970448837360585e-05,
-      "logits/chosen": 3.15625,
-      "logits/rejected": 3.15625,
-      "logps/chosen": -65280.0,
-      "logps/rejected": -65280.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -648.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -648.0,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -10.25,
+      "logps/chosen": -796.0,
+      "logps/rejected": -4416.0,
+      "loss": 0.0063,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.890625,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -23.0,
       "step": 1060
     },
     {
       "epoch": 1.1204188481675392,
-      "grad_norm": 0.0759673225034136,
+      "grad_norm": 0.03668440792122428,
       "learning_rate": 3.945701966409846e-05,
-      "logits/chosen": 3.40625,
-      "logits/rejected": 3.40625,
-      "logps/chosen": -64768.0,
-      "logps/rejected": -64768.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -644.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -644.0,
+      "logits/chosen": -15.75,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -748.0,
+      "logps/rejected": -4864.0,
+      "loss": 0.0108,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.796875,
+      "rewards/margins": 21.375,
+      "rewards/rejected": -25.125,
       "step": 1070
     },
     {
       "epoch": 1.130890052356021,
-      "grad_norm": 0.02235749542795559,
+      "grad_norm": 0.06099644777926882,
       "learning_rate": 3.9207404076833374e-05,
-      "logits/chosen": 3.203125,
-      "logits/rejected": 3.203125,
-      "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": -15.1875,
+      "logits/rejected": -10.5,
+      "logps/chosen": -820.0,
+      "logps/rejected": -4640.0,
+      "loss": 0.0053,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.15625,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -23.875,
       "step": 1080
     },
     {
       "epoch": 1.1413612565445026,
-      "grad_norm": 0.098094825752945,
+      "grad_norm": 0.013867607604378999,
       "learning_rate": 3.895567867990379e-05,
-      "logits/chosen": 3.734375,
-      "logits/rejected": 3.734375,
-      "logps/chosen": -71168.0,
-      "logps/rejected": -71168.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -708.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -708.0,
+      "logits/chosen": -15.0,
+      "logits/rejected": -10.125,
+      "logps/chosen": -844.0,
+      "logps/rejected": -4736.0,
+      "loss": 0.0044,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.34375,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -24.875,
       "step": 1090
     },
     {
       "epoch": 1.1518324607329844,
-      "grad_norm": 0.024299217362579516,
+      "grad_norm": 0.05627965199346601,
       "learning_rate": 3.870188085471114e-05,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "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": -14.625,
+      "logits/rejected": -10.25,
+      "logps/chosen": -820.0,
+      "logps/rejected": -4896.0,
+      "loss": 0.0032,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.40625,
+      "rewards/margins": 23.25,
+      "rewards/rejected": -27.625,
       "step": 1100
     },
     {
       "epoch": 1.1518324607329844,
-      "eval_logits/chosen": 4.6875,
-      "eval_logits/rejected": 5.0625,
-      "eval_logps/chosen": -51712.0,
-      "eval_logps/rejected": -44544.0,
-      "eval_loss": 151.29299926757812,
-      "eval_rewards/accuracies": 0.43359375,
-      "eval_rewards/chosen": -512.0,
-      "eval_rewards/margins": -70.0,
-      "eval_rewards/rejected": -444.0,
-      "eval_runtime": 46.6476,
-      "eval_samples_per_second": 42.875,
+      "eval_logits/chosen": -15.3125,
+      "eval_logits/rejected": -15.0,
+      "eval_logps/chosen": -768.0,
+      "eval_logps/rejected": -752.0,
+      "eval_loss": 1.152093768119812,
+      "eval_rewards/accuracies": 0.509765625,
+      "eval_rewards/chosen": -4.5,
+      "eval_rewards/margins": 0.1455078125,
+      "eval_rewards/rejected": -4.625,
+      "eval_runtime": 46.6711,
+      "eval_samples_per_second": 42.853,
       "eval_steps_per_second": 0.686,
       "step": 1100
     },
     {
       "epoch": 1.162303664921466,
-      "grad_norm": 0.02471888133613041,
+      "grad_norm": 0.15797635521632786,
       "learning_rate": 3.844604829041395e-05,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "logps/chosen": -59136.0,
-      "logps/rejected": -59136.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -588.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -588.0,
+      "logits/chosen": -15.8125,
+      "logits/rejected": -10.9375,
+      "logps/chosen": -756.0,
+      "logps/rejected": -4992.0,
+      "loss": 0.0078,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -4.15625,
+      "rewards/margins": 23.875,
+      "rewards/rejected": -28.0,
       "step": 1110
     },
     {
       "epoch": 1.1727748691099475,
-      "grad_norm": 0.022107780872822187,
+      "grad_norm": 0.08863875146430751,
       "learning_rate": 3.818821897833095e-05,
-      "logits/chosen": 3.78125,
-      "logits/rejected": 3.78125,
-      "logps/chosen": -64256.0,
-      "logps/rejected": -64256.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -640.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -640.0,
+      "logits/chosen": -15.625,
+      "logits/rejected": -11.0,
+      "logps/chosen": -792.0,
+      "logps/rejected": -5056.0,
+      "loss": 0.0039,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.125,
+      "rewards/margins": 24.375,
+      "rewards/rejected": -28.5,
       "step": 1120
     },
     {
       "epoch": 1.1832460732984293,
-      "grad_norm": 0.021288048856570225,
+      "grad_norm": 0.1808617457250385,
       "learning_rate": 3.792843120629935e-05,
-      "logits/chosen": 3.78125,
-      "logits/rejected": 3.78125,
-      "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": -15.6875,
+      "logits/rejected": -11.0625,
+      "logps/chosen": -764.0,
+      "logps/rejected": -4832.0,
+      "loss": 0.0058,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -4.03125,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -27.5,
       "step": 1130
     },
     {
       "epoch": 1.193717277486911,
-      "grad_norm": 0.02207295102924724,
+      "grad_norm": 0.11471672370312695,
       "learning_rate": 3.766672355298907e-05,
-      "logits/chosen": 4.09375,
-      "logits/rejected": 4.09375,
-      "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": -15.9375,
+      "logits/rejected": -11.4375,
+      "logps/chosen": -692.0,
+      "logps/rejected": -4864.0,
+      "loss": 0.0166,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 23.0,
+      "rewards/rejected": -26.5,
       "step": 1140
     },
     {
       "epoch": 1.2041884816753927,
-      "grad_norm": 0.0247712371897416,
+      "grad_norm": 0.15270269836453526,
       "learning_rate": 3.740313488217373e-05,
-      "logits/chosen": 4.09375,
-      "logits/rejected": 4.09375,
-      "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": -15.375,
+      "logits/rejected": -10.5,
+      "logps/chosen": -704.0,
+      "logps/rejected": -4864.0,
+      "loss": 0.0075,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.609375,
+      "rewards/margins": 22.875,
+      "rewards/rejected": -26.5,
       "step": 1150
     },
     {
       "epoch": 1.2146596858638743,
-      "grad_norm": 0.02140936658502045,
+      "grad_norm": 0.03742264302196276,
       "learning_rate": 3.7137704336959464e-05,
-      "logits/chosen": 4.25,
-      "logits/rejected": 4.25,
-      "logps/chosen": -60672.0,
-      "logps/rejected": -60672.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -604.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -604.0,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -732.0,
+      "logps/rejected": -5024.0,
+      "loss": 0.0163,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.90625,
+      "rewards/margins": 23.625,
+      "rewards/rejected": -27.5,
       "step": 1160
     },
     {
       "epoch": 1.225130890052356,
-      "grad_norm": 0.02288910812836358,
+      "grad_norm": 1.545319063184461,
       "learning_rate": 3.687047133397201e-05,
-      "logits/chosen": 3.8125,
-      "logits/rejected": 3.8125,
-      "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": -15.375,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -708.0,
+      "logps/rejected": -4832.0,
+      "loss": 0.0142,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.484375,
+      "rewards/margins": 21.375,
+      "rewards/rejected": -24.875,
       "step": 1170
     },
     {
       "epoch": 1.2356020942408377,
-      "grad_norm": 0.026138315049158294,
+      "grad_norm": 0.026635808132564935,
       "learning_rate": 3.6601475557503415e-05,
-      "logits/chosen": 3.96875,
-      "logits/rejected": 3.96875,
-      "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": -15.3125,
+      "logits/rejected": -10.875,
+      "logps/chosen": -664.0,
+      "logps/rejected": -4480.0,
+      "loss": 0.0186,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -22.75,
       "step": 1180
     },
     {
       "epoch": 1.2460732984293195,
-      "grad_norm": 0.02505461725644418,
+      "grad_norm": 0.9840860721087831,
       "learning_rate": 3.633075695361882e-05,
-      "logits/chosen": 4.34375,
-      "logits/rejected": 4.34375,
-      "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": -15.125,
+      "logits/rejected": -11.25,
+      "logps/chosen": -640.0,
+      "logps/rejected": -4672.0,
+      "loss": 0.014,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -2.90625,
+      "rewards/margins": 20.125,
+      "rewards/rejected": -23.0,
       "step": 1190
     },
     {
       "epoch": 1.256544502617801,
-      "grad_norm": 0.020387496360151323,
+      "grad_norm": 1.662098128607907,
       "learning_rate": 3.6058355724224477e-05,
-      "logits/chosen": 4.03125,
-      "logits/rejected": 4.03125,
-      "logps/chosen": -63744.0,
-      "logps/rejected": -63744.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -632.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -632.0,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -9.25,
+      "logps/chosen": -748.0,
+      "logps/rejected": -4576.0,
+      "loss": 0.0068,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.828125,
+      "rewards/margins": 19.25,
+      "rewards/rejected": -23.125,
       "step": 1200
     },
     {
       "epoch": 1.256544502617801,
-      "eval_logits/chosen": 4.71875,
-      "eval_logits/rejected": 5.09375,
-      "eval_logps/chosen": -52992.0,
-      "eval_logps/rejected": -45824.0,
-      "eval_loss": 154.90586853027344,
-      "eval_rewards/accuracies": 0.431640625,
-      "eval_rewards/chosen": -528.0,
-      "eval_rewards/margins": -71.5,
-      "eval_rewards/rejected": -454.0,
-      "eval_runtime": 46.6647,
-      "eval_samples_per_second": 42.859,
+      "eval_logits/chosen": -9.625,
+      "eval_logits/rejected": -8.5,
+      "eval_logps/chosen": -768.0,
+      "eval_logps/rejected": -764.0,
+      "eval_loss": 0.961187481880188,
+      "eval_rewards/accuracies": 0.53125,
+      "eval_rewards/chosen": -4.5,
+      "eval_rewards/margins": 0.26171875,
+      "eval_rewards/rejected": -4.75,
+      "eval_runtime": 46.6613,
+      "eval_samples_per_second": 42.862,
       "eval_steps_per_second": 0.686,
       "step": 1200
     },
     {
       "epoch": 1.2670157068062826,
-      "grad_norm": 0.021703191455058127,
+      "grad_norm": 0.02933612518039438,
       "learning_rate": 3.5784312321097744e-05,
-      "logits/chosen": 3.84375,
-      "logits/rejected": 3.84375,
-      "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": -10.125,
+      "logits/rejected": -7.21875,
+      "logps/chosen": -816.0,
+      "logps/rejected": -4704.0,
+      "loss": 0.0043,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.46875,
+      "rewards/margins": 21.375,
+      "rewards/rejected": -25.75,
       "step": 1210
     },
     {
       "epoch": 1.2774869109947644,
-      "grad_norm": 0.021696685154825035,
+      "grad_norm": 0.753909893193408,
       "learning_rate": 3.550866743987994e-05,
-      "logits/chosen": 4.25,
-      "logits/rejected": 4.25,
-      "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": -12.125,
+      "logits/rejected": -7.5,
+      "logps/chosen": -912.0,
+      "logps/rejected": -4992.0,
+      "loss": 0.0196,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -5.25,
+      "rewards/margins": 23.875,
+      "rewards/rejected": -29.125,
       "step": 1220
     },
     {
       "epoch": 1.2879581151832462,
-      "grad_norm": 0.020427303930140858,
+      "grad_norm": 0.11664642243794159,
       "learning_rate": 3.5231462014033024e-05,
-      "logits/chosen": 4.3125,
-      "logits/rejected": 4.3125,
-      "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": -14.8125,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -836.0,
+      "logps/rejected": -5728.0,
+      "loss": 0.0038,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 30.375,
+      "rewards/rejected": -35.25,
       "step": 1230
     },
     {
       "epoch": 1.2984293193717278,
-      "grad_norm": 0.020180074655970878,
+      "grad_norm": 3.416568620514007,
       "learning_rate": 3.495273720876095e-05,
-      "logits/chosen": 4.625,
-      "logits/rejected": 4.625,
-      "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": -16.0,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -912.0,
+      "logps/rejected": -6016.0,
+      "loss": 0.0293,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.28125,
+      "rewards/margins": 31.875,
+      "rewards/rejected": -37.25,
       "step": 1240
     },
     {
       "epoch": 1.3089005235602094,
-      "grad_norm": 0.020512519046242877,
+      "grad_norm": 0.15409732170738263,
       "learning_rate": 3.4672534414896616e-05,
-      "logits/chosen": 4.71875,
-      "logits/rejected": 4.71875,
-      "logps/chosen": -72704.0,
-      "logps/rejected": -72704.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -724.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -724.0,
+      "logits/chosen": -16.875,
+      "logits/rejected": -12.375,
+      "logps/chosen": -832.0,
+      "logps/rejected": -5408.0,
+      "loss": 0.0035,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.46875,
+      "rewards/margins": 27.75,
+      "rewards/rejected": -32.25,
       "step": 1250
     },
     {
       "epoch": 1.3193717277486912,
-      "grad_norm": 0.019725779629412972,
+      "grad_norm": 0.07633860131365038,
       "learning_rate": 3.439089524275526e-05,
-      "logits/chosen": 4.28125,
-      "logits/rejected": 4.28125,
-      "logps/chosen": -65536.0,
-      "logps/rejected": -65536.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -652.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -652.0,
+      "logits/chosen": -16.375,
+      "logits/rejected": -12.125,
+      "logps/chosen": -728.0,
+      "logps/rejected": -5152.0,
+      "loss": 0.0032,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.78125,
+      "rewards/margins": 25.125,
+      "rewards/rejected": -29.0,
       "step": 1260
     },
     {
       "epoch": 1.3298429319371727,
-      "grad_norm": 0.014419046181179304,
+      "grad_norm": 0.2803778286404917,
       "learning_rate": 3.410786151595533e-05,
-      "logits/chosen": 4.25,
-      "logits/rejected": 4.25,
-      "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": -17.0,
+      "logits/rejected": -12.9375,
+      "logps/chosen": -692.0,
+      "logps/rejected": -4960.0,
+      "loss": 0.0037,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 24.375,
+      "rewards/rejected": -27.75,
       "step": 1270
     },
     {
       "epoch": 1.3403141361256545,
-      "grad_norm": 0.01753945034268306,
+      "grad_norm": 0.08488901813878603,
       "learning_rate": 3.382347526520765e-05,
-      "logits/chosen": 4.40625,
-      "logits/rejected": 4.40625,
-      "logps/chosen": -68608.0,
-      "logps/rejected": -68608.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -680.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -680.0,
+      "logits/chosen": -17.25,
+      "logits/rejected": -12.625,
+      "logps/chosen": -704.0,
+      "logps/rejected": -4896.0,
+      "loss": 0.0053,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -26.75,
       "step": 1280
     },
     {
       "epoch": 1.350785340314136,
-      "grad_norm": 0.017937631817950002,
+      "grad_norm": 0.2294996129893291,
       "learning_rate": 3.353777872207381e-05,
-      "logits/chosen": 4.8125,
-      "logits/rejected": 4.8125,
-      "logps/chosen": -67584.0,
-      "logps/rejected": -67584.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -672.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -672.0,
+      "logits/chosen": -16.875,
+      "logits/rejected": -12.4375,
+      "logps/chosen": -708.0,
+      "logps/rejected": -4832.0,
+      "loss": 0.0039,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.453125,
+      "rewards/margins": 23.5,
+      "rewards/rejected": -27.0,
       "step": 1290
     },
     {
       "epoch": 1.3612565445026177,
-      "grad_norm": 0.016461818752574697,
-      "learning_rate": 3.325081431269475e-05,
-      "logits/chosen": 4.78125,
-      "logits/rejected": 4.78125,
-      "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,
+      "grad_norm": 0.06398524618685396,
+      "learning_rate": 3.325081431269475e-05,
+      "logits/chosen": -16.125,
+      "logits/rejected": -11.9375,
+      "logps/chosen": -688.0,
+      "logps/rejected": -5152.0,
+      "loss": 0.0038,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 25.25,
+      "rewards/rejected": -28.75,
       "step": 1300
     },
     {
       "epoch": 1.3612565445026177,
-      "eval_logits/chosen": 5.5,
-      "eval_logits/rejected": 5.90625,
-      "eval_logps/chosen": -54272.0,
-      "eval_logps/rejected": -46848.0,
-      "eval_loss": 159.00799560546875,
-      "eval_rewards/accuracies": 0.43359375,
-      "eval_rewards/chosen": -540.0,
-      "eval_rewards/margins": -73.5,
-      "eval_rewards/rejected": -466.0,
-      "eval_runtime": 46.641,
-      "eval_samples_per_second": 42.881,
+      "eval_logits/chosen": -16.75,
+      "eval_logits/rejected": -16.75,
+      "eval_logps/chosen": -680.0,
+      "eval_logps/rejected": -672.0,
+      "eval_loss": 1.0891094207763672,
+      "eval_rewards/accuracies": 0.533203125,
+      "eval_rewards/chosen": -3.609375,
+      "eval_rewards/margins": 0.2470703125,
+      "eval_rewards/rejected": -3.84375,
+      "eval_runtime": 46.6527,
+      "eval_samples_per_second": 42.87,
       "eval_steps_per_second": 0.686,
       "step": 1300
     },
     {
       "epoch": 1.3717277486910995,
-      "grad_norm": 0.017705054630315378,
+      "grad_norm": 0.10610381164510099,
       "learning_rate": 3.296262465149046e-05,
-      "logits/chosen": 4.125,
-      "logits/rejected": 4.125,
-      "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": -16.25,
+      "logits/rejected": -12.1875,
+      "logps/chosen": -756.0,
+      "logps/rejected": -5376.0,
+      "loss": 0.0034,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.765625,
+      "rewards/margins": 26.875,
+      "rewards/rejected": -30.75,
       "step": 1310
     },
     {
       "epoch": 1.3821989528795813,
-      "grad_norm": 0.017290450018014176,
+      "grad_norm": 0.36812406562635513,
       "learning_rate": 3.2673252534831687e-05,
-      "logits/chosen": 4.4375,
-      "logits/rejected": 4.4375,
-      "logps/chosen": -67584.0,
-      "logps/rejected": -67584.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -672.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -672.0,
+      "logits/chosen": -16.75,
+      "logits/rejected": -13.125,
+      "logps/chosen": -744.0,
+      "logps/rejected": -5248.0,
+      "loss": 0.0042,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.734375,
+      "rewards/margins": 26.75,
+      "rewards/rejected": -30.5,
       "step": 1320
     },
     {
       "epoch": 1.3926701570680629,
-      "grad_norm": 0.018982525601240054,
+      "grad_norm": 0.0006368892137268903,
       "learning_rate": 3.2382740934684696e-05,
-      "logits/chosen": 4.8125,
-      "logits/rejected": 4.8125,
-      "logps/chosen": -65536.0,
-      "logps/rejected": -65536.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -652.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -652.0,
+      "logits/chosen": -16.5,
+      "logits/rejected": -12.3125,
+      "logps/chosen": -724.0,
+      "logps/rejected": -5248.0,
+      "loss": 0.0666,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.609375,
+      "rewards/margins": 26.125,
+      "rewards/rejected": -29.75,
       "step": 1330
     },
     {
       "epoch": 1.4031413612565444,
-      "grad_norm": 0.017675593897000807,
+      "grad_norm": 0.027988648274773178,
       "learning_rate": 3.209113299222982e-05,
-      "logits/chosen": 4.84375,
-      "logits/rejected": 4.84375,
-      "logps/chosen": -59136.0,
-      "logps/rejected": -59136.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -588.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -588.0,
+      "logits/chosen": -16.125,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -676.0,
+      "logps/rejected": -5184.0,
+      "loss": 0.0097,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.5625,
+      "rewards/margins": 25.0,
+      "rewards/rejected": -28.5,
       "step": 1340
     },
     {
       "epoch": 1.4136125654450262,
-      "grad_norm": 0.01895734258603,
+      "grad_norm": 0.010222100330675865,
       "learning_rate": 3.179847201145505e-05,
-      "logits/chosen": 4.875,
-      "logits/rejected": 4.875,
-      "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": -16.125,
+      "logits/rejected": -11.25,
+      "logps/chosen": -788.0,
+      "logps/rejected": -5216.0,
+      "loss": 0.0047,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.03125,
+      "rewards/margins": 25.375,
+      "rewards/rejected": -29.375,
       "step": 1350
     },
     {
       "epoch": 1.4240837696335078,
-      "grad_norm": 0.025537213349144463,
+      "grad_norm": 0.07865871200931096,
       "learning_rate": 3.150480145272528e-05,
-      "logits/chosen": 4.8125,
-      "logits/rejected": 4.8125,
-      "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": -15.8125,
+      "logits/rejected": -11.625,
+      "logps/chosen": -800.0,
+      "logps/rejected": -5088.0,
+      "loss": 0.0139,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -4.34375,
+      "rewards/margins": 23.625,
+      "rewards/rejected": -28.0,
       "step": 1360
     },
     {
       "epoch": 1.4345549738219896,
-      "grad_norm": 0.022332688579143143,
+      "grad_norm": 0.06476340711008267,
       "learning_rate": 3.121016492632848e-05,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "logps/chosen": -67584.0,
-      "logps/rejected": -67584.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -672.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -672.0,
+      "logits/chosen": -15.5625,
+      "logits/rejected": -11.4375,
+      "logps/chosen": -812.0,
+      "logps/rejected": -5056.0,
+      "loss": 0.0069,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.375,
+      "rewards/margins": 24.875,
+      "rewards/rejected": -29.25,
       "step": 1370
     },
     {
       "epoch": 1.4450261780104712,
-      "grad_norm": 0.020049113528768526,
+      "grad_norm": 0.031948390717315846,
       "learning_rate": 3.091460618599951e-05,
-      "logits/chosen": 4.40625,
-      "logits/rejected": 4.40625,
-      "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": -15.8125,
+      "logits/rejected": -11.25,
+      "logps/chosen": -740.0,
+      "logps/rejected": -4864.0,
+      "loss": 0.035,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.828125,
+      "rewards/margins": 23.0,
+      "rewards/rejected": -26.875,
       "step": 1380
     },
     {
       "epoch": 1.4554973821989527,
-      "grad_norm": 0.020666755032095602,
+      "grad_norm": 0.2081880057931379,
       "learning_rate": 3.061816912242265e-05,
-      "logits/chosen": 4.4375,
-      "logits/rejected": 4.4375,
-      "logps/chosen": -65280.0,
-      "logps/rejected": -65280.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -648.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -648.0,
+      "logits/chosen": -15.5625,
+      "logits/rejected": -11.25,
+      "logps/chosen": -716.0,
+      "logps/rejected": -4416.0,
+      "loss": 0.0041,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.390625,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -22.375,
       "step": 1390
     },
     {
       "epoch": 1.4659685863874345,
-      "grad_norm": 0.017883172214469462,
+      "grad_norm": 0.001867097424650572,
       "learning_rate": 3.0320897756713785e-05,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "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": -15.3125,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -736.0,
+      "logps/rejected": -4544.0,
+      "loss": 0.0036,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 19.0,
+      "rewards/rejected": -22.5,
       "step": 1400
     },
     {
       "epoch": 1.4659685863874345,
-      "eval_logits/chosen": 5.34375,
-      "eval_logits/rejected": 5.75,
-      "eval_logps/chosen": -55552.0,
-      "eval_logps/rejected": -47872.0,
-      "eval_loss": 162.57138061523438,
-      "eval_rewards/accuracies": 0.4296875,
-      "eval_rewards/chosen": -552.0,
-      "eval_rewards/margins": -75.0,
-      "eval_rewards/rejected": -476.0,
-      "eval_runtime": 46.6482,
-      "eval_samples_per_second": 42.874,
+      "eval_logits/chosen": -15.75,
+      "eval_logits/rejected": -15.6875,
+      "eval_logps/chosen": -688.0,
+      "eval_logps/rejected": -676.0,
+      "eval_loss": 1.0725312232971191,
+      "eval_rewards/accuracies": 0.525390625,
+      "eval_rewards/chosen": -3.6875,
+      "eval_rewards/margins": 0.1884765625,
+      "eval_rewards/rejected": -3.875,
+      "eval_runtime": 46.6456,
+      "eval_samples_per_second": 42.876,
       "eval_steps_per_second": 0.686,
       "step": 1400
     },
     {
       "epoch": 1.4764397905759163,
-      "grad_norm": 0.01847421547521472,
+      "grad_norm": 0.014930538503708454,
       "learning_rate": 3.002283623388332e-05,
-      "logits/chosen": 4.625,
-      "logits/rejected": 4.625,
-      "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": -15.375,
+      "logits/rejected": -11.0,
+      "logps/chosen": -720.0,
+      "logps/rejected": -4480.0,
+      "loss": 0.032,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.734375,
+      "rewards/margins": 18.25,
+      "rewards/rejected": -22.0,
       "step": 1410
     },
     {
       "epoch": 1.486910994764398,
-      "grad_norm": 0.017018450522741084,
+      "grad_norm": 0.43545595192598147,
       "learning_rate": 2.9724028816280507e-05,
-      "logits/chosen": 4.875,
-      "logits/rejected": 4.875,
-      "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": -15.3125,
+      "logits/rejected": -10.875,
+      "logps/chosen": -708.0,
+      "logps/rejected": -3888.0,
+      "loss": 0.0033,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 14.6875,
+      "rewards/rejected": -18.25,
       "step": 1420
     },
     {
       "epoch": 1.4973821989528795,
-      "grad_norm": 0.01768516604645206,
+      "grad_norm": 0.015077784385837047,
       "learning_rate": 2.9424519877020524e-05,
-      "logits/chosen": 4.3125,
-      "logits/rejected": 4.3125,
-      "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": -15.125,
+      "logits/rejected": -10.125,
+      "logps/chosen": -752.0,
+      "logps/rejected": -4128.0,
+      "loss": 0.0054,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.546875,
+      "rewards/margins": 14.8125,
+      "rewards/rejected": -18.375,
       "step": 1430
     },
     {
       "epoch": 1.5078534031413613,
-      "grad_norm": 0.017163960504199546,
+      "grad_norm": 0.07256596971927733,
       "learning_rate": 2.912435389339504e-05,
-      "logits/chosen": 4.84375,
-      "logits/rejected": 4.84375,
-      "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": -15.0,
+      "logits/rejected": -10.375,
+      "logps/chosen": -704.0,
+      "logps/rejected": -4048.0,
+      "loss": 0.0043,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 15.25,
+      "rewards/rejected": -18.625,
       "step": 1440
     },
     {
       "epoch": 1.518324607329843,
-      "grad_norm": 0.014000942120854268,
+      "grad_norm": 0.19898639223750217,
       "learning_rate": 2.8823575440267226e-05,
-      "logits/chosen": 4.5625,
-      "logits/rejected": 4.5625,
-      "logps/chosen": -68608.0,
-      "logps/rejected": -68608.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -680.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -680.0,
+      "logits/chosen": -14.875,
+      "logits/rejected": -10.375,
+      "logps/chosen": -700.0,
+      "logps/rejected": -4448.0,
+      "loss": 0.0024,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.46875,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -21.0,
       "step": 1450
     },
     {
       "epoch": 1.5287958115183247,
-      "grad_norm": 0.016960866155657613,
+      "grad_norm": 0.08285464260306774,
       "learning_rate": 2.8522229183452465e-05,
-      "logits/chosen": 4.6875,
-      "logits/rejected": 4.6875,
-      "logps/chosen": -68608.0,
-      "logps/rejected": -68608.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -680.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -680.0,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -724.0,
+      "logps/rejected": -4192.0,
+      "loss": 0.0324,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.5,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -20.5,
       "step": 1460
     },
     {
       "epoch": 1.5392670157068062,
-      "grad_norm": 0.015252685035156078,
+      "grad_norm": 0.019550912367633424,
       "learning_rate": 2.8220359873085373e-05,
-      "logits/chosen": 4.59375,
-      "logits/rejected": 4.59375,
-      "logps/chosen": -69632.0,
-      "logps/rejected": -69632.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -696.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -696.0,
+      "logits/chosen": -14.875,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -792.0,
+      "logps/rejected": -4448.0,
+      "loss": 0.0063,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -4.15625,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -21.125,
       "step": 1470
     },
     {
       "epoch": 1.5497382198952878,
-      "grad_norm": 0.017885864100773648,
+      "grad_norm": 4.811141522580551,
       "learning_rate": 2.7918012336974388e-05,
-      "logits/chosen": 4.65625,
-      "logits/rejected": 4.65625,
-      "logps/chosen": -63744.0,
-      "logps/rejected": -63744.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -636.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -636.0,
+      "logits/chosen": -14.875,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -728.0,
+      "logps/rejected": -4096.0,
+      "loss": 0.0403,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.625,
+      "rewards/margins": 16.0,
+      "rewards/rejected": -19.625,
       "step": 1480
     },
     {
       "epoch": 1.5602094240837696,
-      "grad_norm": 0.01765387281810315,
+      "grad_norm": 0.042194855821360386,
       "learning_rate": 2.7615231473944815e-05,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "logps/chosen": -66560.0,
-      "logps/rejected": -66560.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -664.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -664.0,
+      "logits/chosen": -14.6875,
+      "logits/rejected": -10.1875,
+      "logps/chosen": -704.0,
+      "logps/rejected": -4128.0,
+      "loss": 0.0172,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.53125,
+      "rewards/margins": 14.625,
+      "rewards/rejected": -18.125,
       "step": 1490
     },
     {
       "epoch": 1.5706806282722514,
-      "grad_norm": 0.017959135244778778,
+      "grad_norm": 0.8571314494772534,
       "learning_rate": 2.7312062247171327e-05,
-      "logits/chosen": 4.875,
-      "logits/rejected": 4.875,
-      "logps/chosen": -68608.0,
-      "logps/rejected": -68608.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -680.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -680.0,
+      "logits/chosen": -14.875,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -712.0,
+      "logps/rejected": -4000.0,
+      "loss": 0.0067,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.71875,
+      "rewards/margins": 14.0,
+      "rewards/rejected": -17.75,
       "step": 1500
     },
     {
       "epoch": 1.5706806282722514,
-      "eval_logits/chosen": 5.46875,
-      "eval_logits/rejected": 5.90625,
-      "eval_logps/chosen": -56576.0,
-      "eval_logps/rejected": -48896.0,
-      "eval_loss": 166.05850219726562,
-      "eval_rewards/accuracies": 0.435546875,
-      "eval_rewards/chosen": -564.0,
-      "eval_rewards/margins": -76.5,
-      "eval_rewards/rejected": -486.0,
-      "eval_runtime": 46.6541,
-      "eval_samples_per_second": 42.869,
+      "eval_logits/chosen": -14.875,
+      "eval_logits/rejected": -14.625,
+      "eval_logps/chosen": -712.0,
+      "eval_logps/rejected": -704.0,
+      "eval_loss": 1.0606563091278076,
+      "eval_rewards/accuracies": 0.51171875,
+      "eval_rewards/chosen": -3.953125,
+      "eval_rewards/margins": 0.2158203125,
+      "eval_rewards/rejected": -4.15625,
+      "eval_runtime": 46.6445,
+      "eval_samples_per_second": 42.877,
       "eval_steps_per_second": 0.686,
       "step": 1500
     },
     {
       "epoch": 1.581151832460733,
-      "grad_norm": 0.013963938476825954,
+      "grad_norm": 0.07737887416249943,
       "learning_rate": 2.7008549677500876e-05,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "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": -15.25,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -780.0,
+      "logps/rejected": -4032.0,
+      "loss": 0.0058,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -3.984375,
+      "rewards/margins": 14.8125,
+      "rewards/rejected": -18.75,
       "step": 1510
     },
     {
       "epoch": 1.5916230366492146,
-      "grad_norm": 0.01746622601501269,
+      "grad_norm": 0.024990811611027272,
       "learning_rate": 2.670473883676709e-05,
-      "logits/chosen": 4.875,
-      "logits/rejected": 4.875,
-      "logps/chosen": -65536.0,
-      "logps/rejected": -65536.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -652.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -652.0,
+      "logits/chosen": -14.875,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -784.0,
+      "logps/rejected": -4256.0,
+      "loss": 0.0037,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.03125,
+      "rewards/margins": 15.875,
+      "rewards/rejected": -19.875,
       "step": 1520
     },
     {
       "epoch": 1.6020942408376964,
-      "grad_norm": 0.01704258488505808,
+      "grad_norm": 0.09412729255212647,
       "learning_rate": 2.6400674841097074e-05,
-      "logits/chosen": 4.28125,
-      "logits/rejected": 4.28125,
-      "logps/chosen": -72704.0,
-      "logps/rejected": -72704.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -724.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -724.0,
+      "logits/chosen": -14.25,
+      "logits/rejected": -9.625,
+      "logps/chosen": -840.0,
+      "logps/rejected": -4672.0,
+      "loss": 0.0015,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 18.125,
+      "rewards/rejected": -22.75,
       "step": 1530
     },
     {
       "epoch": 1.6125654450261782,
-      "grad_norm": 0.015671734998850804,
+      "grad_norm": 0.13052629487915446,
       "learning_rate": 2.609640284421158e-05,
-      "logits/chosen": 4.0625,
-      "logits/rejected": 4.0625,
-      "logps/chosen": -72192.0,
-      "logps/rejected": -72192.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -720.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -720.0,
+      "logits/chosen": -14.8125,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -796.0,
+      "logps/rejected": -4608.0,
+      "loss": 0.0073,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.375,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -23.5,
       "step": 1540
     },
     {
       "epoch": 1.6230366492146597,
-      "grad_norm": 0.01790819740648182,
+      "grad_norm": 0.010941685864745627,
       "learning_rate": 2.5791968030719726e-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": -736.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -736.0,
+      "logits/chosen": -14.9375,
+      "logits/rejected": -9.875,
+      "logps/chosen": -872.0,
+      "logps/rejected": -4704.0,
+      "loss": 0.0032,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 20.75,
+      "rewards/rejected": -25.375,
       "step": 1550
     },
     {
       "epoch": 1.6335078534031413,
-      "grad_norm": 0.020027154192073553,
+      "grad_norm": 0.019551015897882336,
       "learning_rate": 2.548741560940902e-05,
-      "logits/chosen": 3.953125,
-      "logits/rejected": 3.953125,
-      "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": -15.0625,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -828.0,
+      "logps/rejected": -4896.0,
+      "loss": 0.0018,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 20.625,
+      "rewards/rejected": -25.375,
       "step": 1560
     },
     {
       "epoch": 1.6439790575916229,
-      "grad_norm": 0.019606468220474998,
+      "grad_norm": 0.3775305703670633,
       "learning_rate": 2.5182790806531787e-05,
-      "logits/chosen": 4.5625,
-      "logits/rejected": 4.5625,
-      "logps/chosen": -71680.0,
-      "logps/rejected": -71680.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -716.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -716.0,
+      "logits/chosen": -15.4375,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -856.0,
+      "logps/rejected": -4672.0,
+      "loss": 0.0056,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.65625,
+      "rewards/margins": 21.25,
+      "rewards/rejected": -25.875,
       "step": 1570
     },
     {
       "epoch": 1.6544502617801047,
-      "grad_norm": 0.02812516620822696,
+      "grad_norm": 0.0865947369426246,
       "learning_rate": 2.4878138859089074e-05,
-      "logits/chosen": 4.4375,
-      "logits/rejected": 4.4375,
-      "logps/chosen": -68608.0,
-      "logps/rejected": -68608.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -684.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -684.0,
+      "logits/chosen": -15.375,
+      "logits/rejected": -10.0,
+      "logps/chosen": -828.0,
+      "logps/rejected": -4960.0,
+      "loss": 0.0012,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 22.375,
+      "rewards/rejected": -26.875,
       "step": 1580
     },
     {
       "epoch": 1.6649214659685865,
-      "grad_norm": 0.025448382620601093,
+      "grad_norm": 0.00571163456416579,
       "learning_rate": 2.4573505008112924e-05,
-      "logits/chosen": 4.625,
-      "logits/rejected": 4.625,
-      "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": -15.375,
+      "logits/rejected": -10.5,
+      "logps/chosen": -844.0,
+      "logps/rejected": -4672.0,
+      "loss": 0.0055,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -4.75,
+      "rewards/margins": 20.875,
+      "rewards/rejected": -25.625,
       "step": 1590
     },
     {
       "epoch": 1.675392670157068,
-      "grad_norm": 0.025445491261678284,
+      "grad_norm": 2.547767221673039,
       "learning_rate": 2.4268934491948032e-05,
-      "logits/chosen": 4.40625,
-      "logits/rejected": 4.40625,
-      "logps/chosen": -68608.0,
-      "logps/rejected": -68608.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -680.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -680.0,
+      "logits/chosen": -15.125,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -872.0,
+      "logps/rejected": -4608.0,
+      "loss": 0.0068,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -25.0,
       "step": 1600
     },
     {
       "epoch": 1.675392670157068,
-      "eval_logits/chosen": 5.125,
-      "eval_logits/rejected": 5.5,
-      "eval_logps/chosen": -57856.0,
-      "eval_logps/rejected": -49920.0,
-      "eval_loss": 169.17837524414062,
-      "eval_rewards/accuracies": 0.427734375,
-      "eval_rewards/chosen": -576.0,
-      "eval_rewards/margins": -78.0,
-      "eval_rewards/rejected": -496.0,
-      "eval_runtime": 46.653,
-      "eval_samples_per_second": 42.87,
+      "eval_logits/chosen": -15.5,
+      "eval_logits/rejected": -15.1875,
+      "eval_logps/chosen": -776.0,
+      "eval_logps/rejected": -780.0,
+      "eval_loss": 1.1895625591278076,
+      "eval_rewards/accuracies": 0.513671875,
+      "eval_rewards/chosen": -4.59375,
+      "eval_rewards/margins": 0.31640625,
+      "eval_rewards/rejected": -4.90625,
+      "eval_runtime": 46.6587,
+      "eval_samples_per_second": 42.864,
       "eval_steps_per_second": 0.686,
       "step": 1600
     },
     {
       "epoch": 1.6858638743455496,
-      "grad_norm": 0.023142039651226087,
+      "grad_norm": 0.14066633077195337,
       "learning_rate": 2.396447253953385e-05,
-      "logits/chosen": 3.71875,
-      "logits/rejected": 3.71875,
-      "logps/chosen": -76288.0,
-      "logps/rejected": -76288.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -760.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -760.0,
+      "logits/chosen": -15.125,
+      "logits/rejected": -9.75,
+      "logps/chosen": -860.0,
+      "logps/rejected": -4800.0,
+      "loss": 0.0011,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.625,
+      "rewards/margins": 19.25,
+      "rewards/rejected": -23.875,
       "step": 1610
     },
     {
       "epoch": 1.6963350785340314,
-      "grad_norm": 0.020337258745866008,
+      "grad_norm": 0.5084057295064235,
       "learning_rate": 2.3660164363687998e-05,
-      "logits/chosen": 3.875,
-      "logits/rejected": 3.875,
-      "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": -15.1875,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -864.0,
+      "logps/rejected": -4608.0,
+      "loss": 0.0193,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -4.53125,
+      "rewards/margins": 18.5,
+      "rewards/rejected": -23.0,
       "step": 1620
     },
     {
       "epoch": 1.7068062827225132,
-      "grad_norm": 0.020419139184703373,
+      "grad_norm": 0.3836893350640311,
       "learning_rate": 2.33560551543922e-05,
-      "logits/chosen": 3.890625,
-      "logits/rejected": 3.890625,
-      "logps/chosen": -72704.0,
-      "logps/rejected": -72704.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -724.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -724.0,
+      "logits/chosen": -15.25,
+      "logits/rejected": -10.5,
+      "logps/chosen": -812.0,
+      "logps/rejected": -4288.0,
+      "loss": 0.0067,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -4.09375,
+      "rewards/margins": 17.0,
+      "rewards/rejected": -21.125,
       "step": 1630
     },
     {
       "epoch": 1.7172774869109948,
-      "grad_norm": 0.0177910670171398,
+      "grad_norm": 0.469631956469339,
       "learning_rate": 2.3052190072081493e-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": -14.8125,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -812.0,
+      "logps/rejected": -4192.0,
+      "loss": 0.0031,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.25,
+      "rewards/margins": 16.25,
+      "rewards/rejected": -20.5,
       "step": 1640
     },
     {
       "epoch": 1.7277486910994764,
-      "grad_norm": 0.01878977570693321,
+      "grad_norm": 0.25231071075868305,
       "learning_rate": 2.2748614240937864e-05,
-      "logits/chosen": 3.9375,
-      "logits/rejected": 3.9375,
-      "logps/chosen": -68096.0,
-      "logps/rejected": -68096.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -680.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -680.0,
+      "logits/chosen": -14.75,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -740.0,
+      "logps/rejected": -4448.0,
+      "loss": 0.0043,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.890625,
+      "rewards/margins": 18.0,
+      "rewards/rejected": -22.0,
       "step": 1650
     },
     {
       "epoch": 1.738219895287958,
-      "grad_norm": 0.019149515399832492,
+      "grad_norm": 0.019518477334146713,
       "learning_rate": 2.2445372742189334e-05,
-      "logits/chosen": 4.0,
-      "logits/rejected": 4.0,
-      "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": -15.1875,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -776.0,
+      "logps/rejected": -4256.0,
+      "loss": 0.0021,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.890625,
+      "rewards/margins": 17.25,
+      "rewards/rejected": -21.125,
       "step": 1660
     },
     {
       "epoch": 1.7486910994764397,
-      "grad_norm": 0.019597785163003277,
+      "grad_norm": 0.0076843833318803265,
       "learning_rate": 2.214251060741528e-05,
-      "logits/chosen": 4.34375,
-      "logits/rejected": 4.34375,
-      "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": -15.6875,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -824.0,
+      "logps/rejected": -4448.0,
+      "loss": 0.0042,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.25,
+      "rewards/margins": 18.625,
+      "rewards/rejected": -22.75,
       "step": 1670
     },
     {
       "epoch": 1.7591623036649215,
-      "grad_norm": 0.01825381403252034,
+      "grad_norm": 0.5198127340052583,
       "learning_rate": 2.1840072811859302e-05,
-      "logits/chosen": 4.15625,
-      "logits/rejected": 4.15625,
-      "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": -15.8125,
+      "logits/rejected": -10.75,
+      "logps/chosen": -820.0,
+      "logps/rejected": -4608.0,
+      "loss": 0.0076,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.21875,
+      "rewards/margins": 19.25,
+      "rewards/rejected": -23.5,
       "step": 1680
     },
     {
       "epoch": 1.7696335078534031,
-      "grad_norm": 0.02164150012310038,
+      "grad_norm": 0.4739061581911072,
       "learning_rate": 2.1538104267750287e-05,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "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": -16.25,
+      "logits/rejected": -10.6875,
+      "logps/chosen": -784.0,
+      "logps/rejected": -4864.0,
+      "loss": 0.0023,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.21875,
+      "rewards/margins": 21.25,
+      "rewards/rejected": -25.5,
       "step": 1690
     },
     {
       "epoch": 1.7801047120418847,
-      "grad_norm": 0.017943673329487417,
+      "grad_norm": 0.13626249277651167,
       "learning_rate": 2.1236649817632952e-05,
-      "logits/chosen": 4.34375,
-      "logits/rejected": 4.34375,
-      "logps/chosen": -65536.0,
-      "logps/rejected": -65536.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -652.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -652.0,
+      "logits/chosen": -15.5,
+      "logits/rejected": -10.5,
+      "logps/chosen": -768.0,
+      "logps/rejected": -4832.0,
+      "loss": 0.0042,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -4.25,
+      "rewards/margins": 21.875,
+      "rewards/rejected": -26.125,
       "step": 1700
     },
     {
       "epoch": 1.7801047120418847,
-      "eval_logits/chosen": 5.09375,
-      "eval_logits/rejected": 5.46875,
-      "eval_logps/chosen": -58880.0,
-      "eval_logps/rejected": -50688.0,
-      "eval_loss": 172.13162231445312,
-      "eval_rewards/accuracies": 0.435546875,
-      "eval_rewards/chosen": -584.0,
-      "eval_rewards/margins": -79.5,
-      "eval_rewards/rejected": -504.0,
-      "eval_runtime": 46.6523,
-      "eval_samples_per_second": 42.87,
+      "eval_logits/chosen": -15.875,
+      "eval_logits/rejected": -15.375,
+      "eval_logps/chosen": -760.0,
+      "eval_logps/rejected": -756.0,
+      "eval_loss": 1.128812551498413,
+      "eval_rewards/accuracies": 0.52734375,
+      "eval_rewards/chosen": -4.40625,
+      "eval_rewards/margins": 0.267578125,
+      "eval_rewards/rejected": -4.65625,
+      "eval_runtime": 46.6489,
+      "eval_samples_per_second": 42.874,
       "eval_steps_per_second": 0.686,
       "step": 1700
     },
     {
       "epoch": 1.7905759162303665,
-      "grad_norm": 0.020404965271585335,
+      "grad_norm": 0.1469600910894821,
       "learning_rate": 2.093575422770872e-05,
-      "logits/chosen": 4.3125,
-      "logits/rejected": 4.3125,
-      "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": -15.3125,
+      "logits/rejected": -10.25,
+      "logps/chosen": -748.0,
+      "logps/rejected": -4672.0,
+      "loss": 0.0068,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -4.03125,
+      "rewards/margins": 20.75,
+      "rewards/rejected": -24.75,
       "step": 1710
     },
     {
       "epoch": 1.8010471204188483,
-      "grad_norm": 0.019079986028164173,
+      "grad_norm": 0.03999484452633173,
       "learning_rate": 2.0635462181187827e-05,
-      "logits/chosen": 3.75,
-      "logits/rejected": 3.75,
-      "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": -16.0,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -784.0,
+      "logps/rejected": -4736.0,
+      "loss": 0.0032,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.125,
+      "rewards/margins": 19.75,
+      "rewards/rejected": -23.875,
       "step": 1720
     },
     {
       "epoch": 1.8115183246073299,
-      "grad_norm": 0.02153072060197361,
+      "grad_norm": 0.4644228913573545,
       "learning_rate": 2.0335818271653917e-05,
-      "logits/chosen": 4.375,
-      "logits/rejected": 4.375,
-      "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": -15.9375,
+      "logits/rejected": -11.0,
+      "logps/chosen": -760.0,
+      "logps/rejected": -4512.0,
+      "loss": 0.0125,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -3.8125,
+      "rewards/margins": 19.125,
+      "rewards/rejected": -22.875,
       "step": 1730
     },
     {
       "epoch": 1.8219895287958114,
-      "grad_norm": 0.0152402200751631,
+      "grad_norm": 0.01132245620408721,
       "learning_rate": 2.0036866996441815e-05,
-      "logits/chosen": 4.25,
-      "logits/rejected": 4.25,
-      "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": -16.125,
+      "logits/rejected": -10.5625,
+      "logps/chosen": -824.0,
+      "logps/rejected": -5024.0,
+      "loss": 0.0065,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.46875,
+      "rewards/margins": 22.125,
+      "rewards/rejected": -26.625,
       "step": 1740
     },
     {
       "epoch": 1.8324607329842932,
-      "grad_norm": 0.017047959925777262,
+      "grad_norm": 0.017561276439441113,
       "learning_rate": 1.9738652750029677e-05,
-      "logits/chosen": 4.34375,
-      "logits/rejected": 4.34375,
-      "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": -15.375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -848.0,
+      "logps/rejected": -4864.0,
+      "loss": 0.0017,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.53125,
+      "rewards/margins": 22.25,
+      "rewards/rejected": -26.875,
       "step": 1750
     },
     {
       "epoch": 1.8429319371727748,
-      "grad_norm": 0.016000988313882455,
+      "grad_norm": 0.08982405562270285,
       "learning_rate": 1.944121981744631e-05,
-      "logits/chosen": 4.21875,
-      "logits/rejected": 4.21875,
-      "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": -15.4375,
+      "logits/rejected": -9.75,
+      "logps/chosen": -884.0,
+      "logps/rejected": -4928.0,
+      "loss": 0.0031,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 23.5,
+      "rewards/rejected": -28.125,
       "step": 1760
     },
     {
       "epoch": 1.8534031413612566,
-      "grad_norm": 0.018050655227925514,
+      "grad_norm": 0.021576930494333128,
       "learning_rate": 1.9144612367694887e-05,
-      "logits/chosen": 4.59375,
-      "logits/rejected": 4.59375,
-      "logps/chosen": -72704.0,
-      "logps/rejected": -72704.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -724.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -724.0,
+      "logits/chosen": -15.9375,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -812.0,
+      "logps/rejected": -5152.0,
+      "loss": 0.0014,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.46875,
+      "rewards/margins": 25.125,
+      "rewards/rejected": -29.625,
       "step": 1770
     },
     {
       "epoch": 1.8638743455497382,
-      "grad_norm": 0.01403281023263878,
+      "grad_norm": 0.026091805062134623,
       "learning_rate": 1.8848874447193803e-05,
-      "logits/chosen": 4.1875,
-      "logits/rejected": 4.1875,
-      "logps/chosen": -73216.0,
-      "logps/rejected": -73216.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -728.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -728.0,
+      "logits/chosen": -15.1875,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -812.0,
+      "logps/rejected": -5248.0,
+      "loss": 0.0031,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -4.5625,
+      "rewards/margins": 25.5,
+      "rewards/rejected": -30.0,
       "step": 1780
     },
     {
       "epoch": 1.8743455497382198,
-      "grad_norm": 0.018021511140449756,
+      "grad_norm": 0.05755355234367501,
       "learning_rate": 1.855404997323571e-05,
-      "logits/chosen": 4.4375,
-      "logits/rejected": 4.4375,
-      "logps/chosen": -73216.0,
-      "logps/rejected": -73216.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -728.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -728.0,
+      "logits/chosen": -15.375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -900.0,
+      "logps/rejected": -5248.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.21875,
+      "rewards/margins": 24.875,
+      "rewards/rejected": -30.125,
       "step": 1790
     },
     {
       "epoch": 1.8848167539267016,
-      "grad_norm": 0.022705488900012692,
+      "grad_norm": 0.005212819059006313,
       "learning_rate": 1.82601827274658e-05,
-      "logits/chosen": 4.8125,
-      "logits/rejected": 4.8125,
-      "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": -15.75,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -876.0,
+      "logps/rejected": -5408.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.21875,
+      "rewards/margins": 27.375,
+      "rewards/rejected": -32.5,
       "step": 1800
     },
     {
       "epoch": 1.8848167539267016,
-      "eval_logits/chosen": 5.4375,
-      "eval_logits/rejected": 5.8125,
-      "eval_logps/chosen": -59648.0,
-      "eval_logps/rejected": -51456.0,
-      "eval_loss": 174.91287231445312,
-      "eval_rewards/accuracies": 0.431640625,
-      "eval_rewards/chosen": -592.0,
-      "eval_rewards/margins": -80.5,
-      "eval_rewards/rejected": -512.0,
-      "eval_runtime": 46.6472,
-      "eval_samples_per_second": 42.875,
+      "eval_logits/chosen": -15.1875,
+      "eval_logits/rejected": -14.6875,
+      "eval_logps/chosen": -848.0,
+      "eval_logps/rejected": -852.0,
+      "eval_loss": 1.300874948501587,
+      "eval_rewards/accuracies": 0.505859375,
+      "eval_rewards/chosen": -5.3125,
+      "eval_rewards/margins": 0.3203125,
+      "eval_rewards/rejected": -5.625,
+      "eval_runtime": 46.6588,
+      "eval_samples_per_second": 42.864,
       "eval_steps_per_second": 0.686,
       "step": 1800
     },
     {
       "epoch": 1.8952879581151834,
-      "grad_norm": 0.020794941548187895,
+      "grad_norm": 0.0167056601158033,
       "learning_rate": 1.7967316349380213e-05,
-      "logits/chosen": 4.46875,
-      "logits/rejected": 4.46875,
-      "logps/chosen": -69632.0,
-      "logps/rejected": -69632.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -696.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -696.0,
+      "logits/chosen": -15.75,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -840.0,
+      "logps/rejected": -5312.0,
+      "loss": 0.0045,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.90625,
+      "rewards/margins": 26.375,
+      "rewards/rejected": -31.375,
       "step": 1810
     },
     {
       "epoch": 1.905759162303665,
-      "grad_norm": 0.021150166184970225,
+      "grad_norm": 0.11763488705535029,
       "learning_rate": 1.7675494329845516e-05,
-      "logits/chosen": 4.375,
-      "logits/rejected": 4.375,
-      "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": -15.3125,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -884.0,
+      "logps/rejected": -5472.0,
+      "loss": 0.0142,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -5.15625,
+      "rewards/margins": 26.375,
+      "rewards/rejected": -31.625,
       "step": 1820
     },
     {
       "epoch": 1.9162303664921465,
-      "grad_norm": 0.015857780257693032,
+      "grad_norm": 0.02896352342275872,
       "learning_rate": 1.738476000464026e-05,
-      "logits/chosen": 4.46875,
-      "logits/rejected": 4.46875,
-      "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": -15.25,
+      "logits/rejected": -9.5,
+      "logps/chosen": -836.0,
+      "logps/rejected": -5248.0,
+      "loss": 0.0043,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 25.125,
+      "rewards/rejected": -29.75,
       "step": 1830
     },
     {
       "epoch": 1.9267015706806283,
-      "grad_norm": 0.01815342699173491,
+      "grad_norm": 0.04602475561984972,
       "learning_rate": 1.7095156548019646e-05,
-      "logits/chosen": 4.3125,
-      "logits/rejected": 4.3125,
-      "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": -15.4375,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -868.0,
+      "logps/rejected": -5472.0,
+      "loss": 0.0049,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 26.125,
+      "rewards/rejected": -31.125,
       "step": 1840
     },
     {
       "epoch": 1.93717277486911,
-      "grad_norm": 0.019316624316359752,
+      "grad_norm": 0.014734120690145092,
       "learning_rate": 1.6806726966304064e-05,
-      "logits/chosen": 4.4375,
-      "logits/rejected": 4.4375,
-      "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": -15.3125,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -896.0,
+      "logps/rejected": -5312.0,
+      "loss": 0.0031,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 25.875,
+      "rewards/rejected": -30.875,
       "step": 1850
     },
     {
       "epoch": 1.9476439790575917,
-      "grad_norm": 0.020177389257171104,
+      "grad_norm": 0.018826093704605848,
       "learning_rate": 1.6519514091492625e-05,
-      "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": -700.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -700.0,
+      "logits/chosen": -14.5625,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -832.0,
+      "logps/rejected": -5280.0,
+      "loss": 0.0015,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 26.25,
+      "rewards/rejected": -31.125,
       "step": 1860
     },
     {
       "epoch": 1.9581151832460733,
-      "grad_norm": 0.01734100105989557,
+      "grad_norm": 0.06339552508541217,
       "learning_rate": 1.6233560574902627e-05,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "logps/chosen": -68608.0,
-      "logps/rejected": -68608.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -684.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -684.0,
+      "logits/chosen": -14.75,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -800.0,
+      "logps/rejected": -5376.0,
+      "loss": 0.0017,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 26.5,
+      "rewards/rejected": -31.25,
       "step": 1870
     },
     {
       "epoch": 1.9685863874345548,
-      "grad_norm": 0.01638017639691173,
+      "grad_norm": 3.465458900254028,
       "learning_rate": 1.5948908880835753e-05,
-      "logits/chosen": 4.71875,
-      "logits/rejected": 4.71875,
-      "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": -14.75,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -864.0,
+      "logps/rejected": -5376.0,
+      "loss": 0.0111,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.0,
+      "rewards/margins": 26.375,
+      "rewards/rejected": -31.375,
       "step": 1880
     },
     {
       "epoch": 1.9790575916230366,
-      "grad_norm": 0.016683761879610036,
+      "grad_norm": 0.0067156629005057445,
       "learning_rate": 1.5665601280272124e-05,
-      "logits/chosen": 4.65625,
-      "logits/rejected": 4.65625,
-      "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": -14.4375,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -832.0,
+      "logps/rejected": -5248.0,
+      "loss": 0.004,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.8125,
+      "rewards/margins": 25.625,
+      "rewards/rejected": -30.5,
       "step": 1890
     },
     {
       "epoch": 1.9895287958115184,
-      "grad_norm": 0.01666879339993686,
+      "grad_norm": 0.9329843538776248,
       "learning_rate": 1.538367984459298e-05,
-      "logits/chosen": 4.5,
-      "logits/rejected": 4.5,
-      "logps/chosen": -67584.0,
-      "logps/rejected": -67584.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -672.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -672.0,
+      "logits/chosen": -13.8125,
+      "logits/rejected": -8.625,
+      "logps/chosen": -844.0,
+      "logps/rejected": -4960.0,
+      "loss": 0.002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 23.25,
+      "rewards/rejected": -28.25,
       "step": 1900
     },
     {
       "epoch": 1.9895287958115184,
-      "eval_logits/chosen": 5.4375,
-      "eval_logits/rejected": 5.84375,
-      "eval_logps/chosen": -60416.0,
-      "eval_logps/rejected": -52224.0,
-      "eval_loss": 177.3406219482422,
-      "eval_rewards/accuracies": 0.427734375,
-      "eval_rewards/chosen": -600.0,
-      "eval_rewards/margins": -81.5,
-      "eval_rewards/rejected": -520.0,
-      "eval_runtime": 46.6497,
-      "eval_samples_per_second": 42.873,
+      "eval_logits/chosen": -14.3125,
+      "eval_logits/rejected": -13.625,
+      "eval_logps/chosen": -804.0,
+      "eval_logps/rejected": -800.0,
+      "eval_loss": 1.2142187356948853,
+      "eval_rewards/accuracies": 0.515625,
+      "eval_rewards/chosen": -4.84375,
+      "eval_rewards/margins": 0.287109375,
+      "eval_rewards/rejected": -5.125,
+      "eval_runtime": 46.6556,
+      "eval_samples_per_second": 42.867,
       "eval_steps_per_second": 0.686,
       "step": 1900
     },
     {
       "epoch": 2.0,
-      "grad_norm": 0.0164542821536098,
+      "grad_norm": 0.016570498817847217,
       "learning_rate": 1.5103186439333133e-05,
-      "logits/chosen": 4.8125,
-      "logits/rejected": 4.8125,
-      "logps/chosen": -66560.0,
-      "logps/rejected": -66560.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -664.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -664.0,
+      "logits/chosen": -14.5,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -804.0,
+      "logps/rejected": -5024.0,
+      "loss": 0.0412,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.4375,
+      "rewards/margins": 24.875,
+      "rewards/rejected": -29.375,
       "step": 1910
     },
     {
       "epoch": 2.0104712041884816,
-      "grad_norm": 0.019184016634089557,
+      "grad_norm": 0.003521268451389636,
       "learning_rate": 1.4824162717963829e-05,
-      "logits/chosen": 4.71875,
-      "logits/rejected": 4.71875,
-      "logps/chosen": -73216.0,
-      "logps/rejected": -73216.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -728.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -728.0,
+      "logits/chosen": -14.25,
+      "logits/rejected": -9.0,
+      "logps/chosen": -804.0,
+      "logps/rejected": -5088.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 23.625,
+      "rewards/rejected": -28.125,
       "step": 1920
     },
     {
       "epoch": 2.020942408376963,
-      "grad_norm": 0.019264426296760884,
+      "grad_norm": 0.003970136283634851,
       "learning_rate": 1.4546650115707192e-05,
-      "logits/chosen": 4.28125,
-      "logits/rejected": 4.28125,
-      "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": -14.4375,
+      "logits/rejected": -8.125,
+      "logps/chosen": -776.0,
+      "logps/rejected": -5056.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.3125,
+      "rewards/margins": 24.875,
+      "rewards/rejected": -29.125,
       "step": 1930
     },
     {
       "epoch": 2.031413612565445,
-      "grad_norm": 0.01733221602407002,
+      "grad_norm": 0.5385083625686191,
       "learning_rate": 1.4270689843383111e-05,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "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": -13.5,
+      "logits/rejected": -8.75,
+      "logps/chosen": -816.0,
+      "logps/rejected": -4896.0,
+      "loss": 0.0011,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.4375,
+      "rewards/margins": 23.0,
+      "rewards/rejected": -27.5,
       "step": 1940
     },
     {
       "epoch": 2.0418848167539267,
-      "grad_norm": 0.016229724800380724,
+      "grad_norm": 0.010250717042781814,
       "learning_rate": 1.3996322881289347e-05,
-      "logits/chosen": 4.65625,
-      "logits/rejected": 4.65625,
-      "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": -14.375,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -772.0,
+      "logps/rejected": -4864.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.375,
+      "rewards/margins": 22.875,
+      "rewards/rejected": -27.25,
       "step": 1950
     },
     {
       "epoch": 2.0523560209424083,
-      "grad_norm": 0.019362786355152803,
+      "grad_norm": 0.00135818706962715,
       "learning_rate": 1.3723589973115963e-05,
-      "logits/chosen": 4.78125,
-      "logits/rejected": 4.78125,
-      "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": -14.4375,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -780.0,
+      "logps/rejected": -5056.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.34375,
+      "rewards/margins": 23.875,
+      "rewards/rejected": -28.125,
       "step": 1960
     },
     {
       "epoch": 2.06282722513089,
-      "grad_norm": 0.01927928552402117,
+      "grad_norm": 0.00976079580946157,
       "learning_rate": 1.3452531619894837e-05,
-      "logits/chosen": 4.40625,
-      "logits/rejected": 4.40625,
-      "logps/chosen": -73216.0,
-      "logps/rejected": -73216.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -732.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -732.0,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -9.125,
+      "logps/chosen": -800.0,
+      "logps/rejected": -4960.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.375,
+      "rewards/margins": 22.75,
+      "rewards/rejected": -27.125,
       "step": 1970
     },
     {
       "epoch": 2.073298429319372,
-      "grad_norm": 0.018824065709863894,
+      "grad_norm": 0.0015483156322630724,
       "learning_rate": 1.3183188073985245e-05,
-      "logits/chosen": 4.28125,
-      "logits/rejected": 4.28125,
-      "logps/chosen": -72704.0,
-      "logps/rejected": -72704.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -724.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -724.0,
+      "logits/chosen": -14.625,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -784.0,
+      "logps/rejected": -5216.0,
+      "loss": 0.0023,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.40625,
+      "rewards/margins": 23.75,
+      "rewards/rejected": -28.125,
       "step": 1980
     },
     {
       "epoch": 2.0837696335078535,
-      "grad_norm": 0.018823402442436857,
+      "grad_norm": 0.0023137314170755966,
       "learning_rate": 1.2915599333096351e-05,
-      "logits/chosen": 4.21875,
-      "logits/rejected": 4.21875,
-      "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": -14.8125,
+      "logits/rejected": -9.625,
+      "logps/chosen": -836.0,
+      "logps/rejected": -5152.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.5,
+      "rewards/margins": 24.5,
+      "rewards/rejected": -29.0,
       "step": 1990
     },
     {
       "epoch": 2.094240837696335,
-      "grad_norm": 0.019338946527642193,
+      "grad_norm": 0.0077125965657208,
       "learning_rate": 1.264980513434752e-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": -14.5,
+      "logits/rejected": -8.875,
+      "logps/chosen": -868.0,
+      "logps/rejected": -5312.0,
+      "loss": 0.0004,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.8125,
+      "rewards/margins": 24.5,
+      "rewards/rejected": -29.25,
       "step": 2000
     },
     {
       "epoch": 2.094240837696335,
-      "eval_logits/chosen": 5.46875,
-      "eval_logits/rejected": 5.875,
-      "eval_logps/chosen": -61184.0,
-      "eval_logps/rejected": -52992.0,
-      "eval_loss": 179.19924926757812,
-      "eval_rewards/accuracies": 0.42578125,
-      "eval_rewards/chosen": -608.0,
-      "eval_rewards/margins": -82.5,
-      "eval_rewards/rejected": -524.0,
-      "eval_runtime": 46.6546,
-      "eval_samples_per_second": 42.868,
+      "eval_logits/chosen": -14.125,
+      "eval_logits/rejected": -13.4375,
+      "eval_logps/chosen": -804.0,
+      "eval_logps/rejected": -804.0,
+      "eval_loss": 1.229968786239624,
+      "eval_rewards/accuracies": 0.513671875,
+      "eval_rewards/chosen": -4.84375,
+      "eval_rewards/margins": 0.296875,
+      "eval_rewards/rejected": -5.15625,
+      "eval_runtime": 46.6361,
+      "eval_samples_per_second": 42.885,
       "eval_steps_per_second": 0.686,
       "step": 2000
     },
     {
       "epoch": 2.1047120418848166,
-      "grad_norm": 0.019213423331078644,
+      "grad_norm": 0.018545541057029637,
       "learning_rate": 1.2385844948367324e-05,
-      "logits/chosen": 4.6875,
-      "logits/rejected": 4.6875,
-      "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": -14.0625,
+      "logits/rejected": -9.5,
+      "logps/chosen": -788.0,
+      "logps/rejected": -5184.0,
+      "loss": 0.0005,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.59375,
+      "rewards/margins": 25.0,
+      "rewards/rejected": -29.625,
       "step": 2010
     },
     {
       "epoch": 2.115183246073298,
-      "grad_norm": 0.018859244719165228,
+      "grad_norm": 0.005398588736489164,
       "learning_rate": 1.2123757973432114e-05,
-      "logits/chosen": 4.125,
-      "logits/rejected": 4.125,
-      "logps/chosen": -77824.0,
-      "logps/rejected": -77824.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -772.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -772.0,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -900.0,
+      "logps/rejected": -5280.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 24.625,
+      "rewards/rejected": -29.75,
       "step": 2020
     },
     {
       "epoch": 2.1256544502617802,
-      "grad_norm": 0.015956302759664767,
+      "grad_norm": 0.008798234077560531,
       "learning_rate": 1.1863583129645106e-05,
-      "logits/chosen": 4.59375,
-      "logits/rejected": 4.59375,
-      "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": -14.1875,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -836.0,
+      "logps/rejected": -5280.0,
+      "loss": 0.0006,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 25.0,
+      "rewards/rejected": -29.75,
       "step": 2030
     },
     {
       "epoch": 2.136125654450262,
-      "grad_norm": 0.018171068993714772,
+      "grad_norm": 0.002124735070796494,
       "learning_rate": 1.1605359053156607e-05,
-      "logits/chosen": 4.5625,
-      "logits/rejected": 4.5625,
-      "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": -14.6875,
+      "logits/rejected": -9.25,
+      "logps/chosen": -828.0,
+      "logps/rejected": -5216.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.5625,
+      "rewards/margins": 24.875,
+      "rewards/rejected": -29.375,
       "step": 2040
     },
     {
       "epoch": 2.1465968586387434,
-      "grad_norm": 0.01763829553452198,
+      "grad_norm": 0.052506752814364024,
       "learning_rate": 1.1349124090426644e-05,
-      "logits/chosen": 4.40625,
-      "logits/rejected": 4.40625,
-      "logps/chosen": -76800.0,
-      "logps/rejected": -76800.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -760.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -760.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -8.25,
+      "logps/chosen": -848.0,
+      "logps/rejected": -5184.0,
+      "loss": 0.0005,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.78125,
+      "rewards/margins": 24.75,
+      "rewards/rejected": -29.5,
       "step": 2050
     },
     {
       "epoch": 2.157068062827225,
-      "grad_norm": 0.013338312592012798,
+      "grad_norm": 0.0005001003063130318,
       "learning_rate": 1.1094916292530406e-05,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "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": -14.3125,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -872.0,
+      "logps/rejected": -5152.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.90625,
+      "rewards/margins": 24.0,
+      "rewards/rejected": -29.0,
       "step": 2060
     },
     {
       "epoch": 2.167539267015707,
-      "grad_norm": 0.01299547561143703,
-      "learning_rate": 1.0842773409507623e-05,
-      "logits/chosen": 4.71875,
-      "logits/rejected": 4.71875,
-      "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,
+      "grad_norm": 0.024208547579075666,
+      "learning_rate": 1.0842773409507623e-05,
+      "logits/chosen": -13.8125,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -800.0,
+      "logps/rejected": -5184.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.59375,
+      "rewards/margins": 25.5,
+      "rewards/rejected": -30.125,
       "step": 2070
     },
     {
       "epoch": 2.1780104712041886,
-      "grad_norm": 0.014832211428225928,
+      "grad_norm": 0.02578309892298803,
       "learning_rate": 1.0592732884756754e-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": -14.625,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -820.0,
+      "logps/rejected": -5216.0,
+      "loss": 0.0004,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.46875,
+      "rewards/margins": 24.875,
+      "rewards/rejected": -29.375,
       "step": 2080
     },
     {
       "epoch": 2.18848167539267,
-      "grad_norm": 0.015752398291318612,
+      "grad_norm": 0.001606040097015124,
       "learning_rate": 1.0344831849474507e-05,
-      "logits/chosen": 4.71875,
-      "logits/rejected": 4.71875,
-      "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": -13.8125,
+      "logits/rejected": -9.125,
+      "logps/chosen": -888.0,
+      "logps/rejected": -5184.0,
+      "loss": 0.0004,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 24.75,
+      "rewards/rejected": -29.75,
       "step": 2090
     },
     {
       "epoch": 2.1989528795811517,
-      "grad_norm": 0.01741805251599387,
+      "grad_norm": 0.008909151410766935,
       "learning_rate": 1.0099107117141879e-05,
-      "logits/chosen": 5.09375,
-      "logits/rejected": 5.09375,
-      "logps/chosen": -71168.0,
-      "logps/rejected": -71168.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -708.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -708.0,
+      "logits/chosen": -14.125,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -816.0,
+      "logps/rejected": -5248.0,
+      "loss": 0.0148,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 25.75,
+      "rewards/rejected": -30.375,
       "step": 2100
     },
     {
       "epoch": 2.1989528795811517,
-      "eval_logits/chosen": 5.5,
-      "eval_logits/rejected": 5.875,
-      "eval_logps/chosen": -61696.0,
-      "eval_logps/rejected": -53504.0,
-      "eval_loss": 180.8938751220703,
-      "eval_rewards/accuracies": 0.4296875,
-      "eval_rewards/chosen": -616.0,
-      "eval_rewards/margins": -83.5,
-      "eval_rewards/rejected": -532.0,
-      "eval_runtime": 46.6591,
-      "eval_samples_per_second": 42.864,
+      "eval_logits/chosen": -13.9375,
+      "eval_logits/rejected": -13.25,
+      "eval_logps/chosen": -824.0,
+      "eval_logps/rejected": -828.0,
+      "eval_loss": 1.2569375038146973,
+      "eval_rewards/accuracies": 0.513671875,
+      "eval_rewards/chosen": -5.0625,
+      "eval_rewards/margins": 0.322265625,
+      "eval_rewards/rejected": -5.375,
+      "eval_runtime": 46.6532,
+      "eval_samples_per_second": 42.87,
       "eval_steps_per_second": 0.686,
       "step": 2100
     },
     {
       "epoch": 2.2094240837696333,
-      "grad_norm": 0.017012582118169233,
+      "grad_norm": 0.004864805793481847,
       "learning_rate": 9.855595178057334e-06,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "logps/chosen": -76288.0,
-      "logps/rejected": -76288.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -760.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -760.0,
+      "logits/chosen": -13.8125,
+      "logits/rejected": -8.5,
+      "logps/chosen": -884.0,
+      "logps/rejected": -5376.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 25.5,
+      "rewards/rejected": -30.625,
       "step": 2110
     },
     {
       "epoch": 2.2198952879581153,
-      "grad_norm": 0.013526614247754679,
+      "grad_norm": 0.0005751217184025536,
       "learning_rate": 9.614332193917949e-06,
-      "logits/chosen": 4.90625,
-      "logits/rejected": 4.90625,
-      "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": -14.4375,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -924.0,
+      "logps/rejected": -5248.0,
+      "loss": 0.0001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.25,
+      "rewards/margins": 26.0,
+      "rewards/rejected": -31.125,
       "step": 2120
     },
     {
       "epoch": 2.230366492146597,
-      "grad_norm": 0.014474001543836488,
+      "grad_norm": 0.002870277724714611,
       "learning_rate": 9.375353992449385e-06,
-      "logits/chosen": 4.625,
-      "logits/rejected": 4.625,
-      "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": -14.5625,
+      "logits/rejected": -9.375,
+      "logps/chosen": -840.0,
+      "logps/rejected": -5120.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 25.375,
+      "rewards/rejected": -30.0,
       "step": 2130
     },
     {
       "epoch": 2.2408376963350785,
-      "grad_norm": 0.012669194624872806,
+      "grad_norm": 0.007916065177424592,
       "learning_rate": 9.138696062085442e-06,
-      "logits/chosen": 4.40625,
-      "logits/rejected": 4.40625,
-      "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": -14.0625,
+      "logits/rejected": -9.0,
+      "logps/chosen": -844.0,
+      "logps/rejected": -5472.0,
+      "loss": 0.0001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 26.375,
+      "rewards/rejected": -31.25,
       "step": 2140
     },
     {
       "epoch": 2.25130890052356,
-      "grad_norm": 0.015372948612642951,
+      "grad_norm": 0.012133065926144127,
       "learning_rate": 8.904393546698007e-06,
-      "logits/chosen": 4.84375,
-      "logits/rejected": 4.84375,
-      "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": -14.6875,
+      "logits/rejected": -9.125,
+      "logps/chosen": -864.0,
+      "logps/rejected": -5440.0,
+      "loss": 0.0022,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.78125,
+      "rewards/margins": 26.75,
+      "rewards/rejected": -31.5,
       "step": 2150
     },
     {
       "epoch": 2.261780104712042,
-      "grad_norm": 0.01614638409738981,
+      "grad_norm": 0.019872927702526286,
       "learning_rate": 8.672481240378142e-06,
-      "logits/chosen": 4.78125,
-      "logits/rejected": 4.78125,
-      "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": -13.75,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -800.0,
+      "logps/rejected": -5248.0,
+      "loss": 0.0007,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.75,
+      "rewards/margins": 26.375,
+      "rewards/rejected": -31.125,
       "step": 2160
     },
     {
       "epoch": 2.2722513089005236,
-      "grad_norm": 0.017071955214960435,
+      "grad_norm": 0.02019655286149391,
       "learning_rate": 8.44299358226919e-06,
-      "logits/chosen": 4.75,
-      "logits/rejected": 4.75,
-      "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": -13.875,
+      "logits/rejected": -9.125,
+      "logps/chosen": -848.0,
+      "logps/rejected": -5440.0,
+      "loss": 0.0004,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 26.25,
+      "rewards/rejected": -31.125,
       "step": 2170
     },
     {
       "epoch": 2.282722513089005,
-      "grad_norm": 0.017444160257691944,
+      "grad_norm": 0.004271729548724879,
       "learning_rate": 8.215964651452457e-06,
-      "logits/chosen": 4.8125,
-      "logits/rejected": 4.8125,
-      "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": -13.6875,
+      "logits/rejected": -8.5625,
+      "logps/chosen": -856.0,
+      "logps/rejected": -5376.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 26.25,
+      "rewards/rejected": -31.25,
       "step": 2180
     },
     {
       "epoch": 2.2931937172774868,
-      "grad_norm": 0.015696858883980966,
+      "grad_norm": 0.006157230056925586,
       "learning_rate": 7.991428161886502e-06,
-      "logits/chosen": 4.84375,
-      "logits/rejected": 4.84375,
-      "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": -13.625,
+      "logits/rejected": -9.125,
+      "logps/chosen": -896.0,
+      "logps/rejected": -5248.0,
+      "loss": 0.0001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.21875,
+      "rewards/margins": 26.0,
+      "rewards/rejected": -31.125,
       "step": 2190
     },
     {
       "epoch": 2.303664921465969,
-      "grad_norm": 0.02169863826527065,
+      "grad_norm": 0.0019931931649125353,
       "learning_rate": 7.76941745740061e-06,
-      "logits/chosen": 4.625,
-      "logits/rejected": 4.625,
-      "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": -9.125,
+      "logps/chosen": -940.0,
+      "logps/rejected": -5216.0,
+      "loss": 0.0009,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.5625,
+      "rewards/margins": 25.125,
+      "rewards/rejected": -30.75,
       "step": 2200
     },
     {
       "epoch": 2.303664921465969,
-      "eval_logits/chosen": 5.6875,
-      "eval_logits/rejected": 6.09375,
-      "eval_logps/chosen": -62208.0,
-      "eval_logps/rejected": -53760.0,
-      "eval_loss": 182.2845001220703,
-      "eval_rewards/accuracies": 0.431640625,
-      "eval_rewards/chosen": -620.0,
-      "eval_rewards/margins": -84.0,
-      "eval_rewards/rejected": -536.0,
-      "eval_runtime": 46.6626,
-      "eval_samples_per_second": 42.861,
+      "eval_logits/chosen": -13.5625,
+      "eval_logits/rejected": -12.8125,
+      "eval_logps/chosen": -848.0,
+      "eval_logps/rejected": -852.0,
+      "eval_loss": 1.2545000314712524,
+      "eval_rewards/accuracies": 0.505859375,
+      "eval_rewards/chosen": -5.3125,
+      "eval_rewards/margins": 0.318359375,
+      "eval_rewards/rejected": -5.625,
+      "eval_runtime": 46.634,
+      "eval_samples_per_second": 42.887,
       "eval_steps_per_second": 0.686,
       "step": 2200
     },
     {
       "epoch": 2.3141361256544504,
-      "grad_norm": 0.01712367394158948,
+      "grad_norm": 0.002455853735228954,
       "learning_rate": 7.549965506743123e-06,
-      "logits/chosen": 5.0,
-      "logits/rejected": 5.0,
-      "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": -14.4375,
+      "logits/rejected": -8.625,
+      "logps/chosen": -848.0,
+      "logps/rejected": -5536.0,
+      "loss": 0.0007,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 27.375,
+      "rewards/rejected": -32.25,
       "step": 2210
     },
     {
       "epoch": 2.324607329842932,
-      "grad_norm": 0.016157193781100625,
+      "grad_norm": 0.012715716334360722,
       "learning_rate": 7.333104898685631e-06,
-      "logits/chosen": 4.40625,
-      "logits/rejected": 4.40625,
-      "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": -14.125,
+      "logits/rejected": -9.125,
+      "logps/chosen": -840.0,
+      "logps/rejected": -5376.0,
+      "loss": 0.0015,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.8125,
+      "rewards/margins": 26.75,
+      "rewards/rejected": -31.5,
       "step": 2220
     },
     {
       "epoch": 2.3350785340314135,
-      "grad_norm": 0.014741980606352857,
+      "grad_norm": 0.014136099730776284,
       "learning_rate": 7.118867837183518e-06,
-      "logits/chosen": 4.375,
-      "logits/rejected": 4.375,
-      "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": -14.375,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -948.0,
+      "logps/rejected": -5312.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.25,
+      "rewards/margins": 25.25,
+      "rewards/rejected": -30.375,
       "step": 2230
     },
     {
       "epoch": 2.345549738219895,
-      "grad_norm": 0.016920119597022932,
+      "grad_norm": 0.005154299832928575,
       "learning_rate": 6.9072861365936045e-06,
-      "logits/chosen": 4.46875,
-      "logits/rejected": 4.46875,
-      "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": -15.125,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -920.0,
+      "logps/rejected": -5152.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 25.125,
+      "rewards/rejected": -30.25,
       "step": 2240
     },
     {
       "epoch": 2.356020942408377,
-      "grad_norm": 0.016990354211749423,
+      "grad_norm": 0.014836319793634006,
       "learning_rate": 6.6983912169497005e-06,
-      "logits/chosen": 4.75,
-      "logits/rejected": 4.75,
-      "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": -14.3125,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -900.0,
+      "logps/rejected": -5248.0,
+      "loss": 0.004,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 25.125,
+      "rewards/rejected": -30.125,
       "step": 2250
     },
     {
       "epoch": 2.3664921465968587,
-      "grad_norm": 0.014048844434311326,
+      "grad_norm": 0.0031057821169706936,
       "learning_rate": 6.49221409929677e-06,
-      "logits/chosen": 5.0,
-      "logits/rejected": 5.0,
-      "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": -13.8125,
+      "logits/rejected": -9.25,
+      "logps/chosen": -848.0,
+      "logps/rejected": -5536.0,
+      "loss": 0.0013,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.125,
+      "rewards/margins": 27.375,
+      "rewards/rejected": -32.5,
       "step": 2260
     },
     {
       "epoch": 2.3769633507853403,
-      "grad_norm": 0.014435654426911035,
+      "grad_norm": 0.02504209812096854,
       "learning_rate": 6.2887854010842075e-06,
-      "logits/chosen": 4.9375,
-      "logits/rejected": 4.9375,
-      "logps/chosen": -72704.0,
-      "logps/rejected": -72704.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -724.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -724.0,
+      "logits/chosen": -13.625,
+      "logits/rejected": -8.75,
+      "logps/chosen": -852.0,
+      "logps/rejected": -5344.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.0,
+      "rewards/margins": 26.25,
+      "rewards/rejected": -31.375,
       "step": 2270
     },
     {
       "epoch": 2.387434554973822,
-      "grad_norm": 0.016145003179229397,
+      "grad_norm": 0.007593010245532049,
       "learning_rate": 6.0881353316191385e-06,
-      "logits/chosen": 4.46875,
-      "logits/rejected": 4.46875,
-      "logps/chosen": -74752.0,
-      "logps/rejected": -74752.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -744.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -744.0,
+      "logits/chosen": -13.875,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -908.0,
+      "logps/rejected": -5216.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.15625,
+      "rewards/margins": 25.75,
+      "rewards/rejected": -30.875,
       "step": 2280
     },
     {
       "epoch": 2.3979057591623034,
-      "grad_norm": 0.014636905508994533,
+      "grad_norm": 0.02757464947403492,
       "learning_rate": 5.890293687580381e-06,
-      "logits/chosen": 4.6875,
-      "logits/rejected": 4.6875,
-      "logps/chosen": -76288.0,
-      "logps/rejected": -76288.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -760.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -760.0,
+      "logits/chosen": -13.875,
+      "logits/rejected": -9.625,
+      "logps/chosen": -916.0,
+      "logps/rejected": -5312.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.375,
+      "rewards/margins": 25.875,
+      "rewards/rejected": -31.125,
       "step": 2290
     },
     {
       "epoch": 2.4083769633507854,
-      "grad_norm": 0.01714586784655505,
+      "grad_norm": 0.03428740018810136,
       "learning_rate": 5.695289848593532e-06,
-      "logits/chosen": 4.375,
-      "logits/rejected": 4.375,
-      "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": -13.9375,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -816.0,
+      "logps/rejected": -5216.0,
+      "loss": 0.0006,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.625,
+      "rewards/margins": 26.5,
+      "rewards/rejected": -31.125,
       "step": 2300
     },
     {
       "epoch": 2.4083769633507854,
-      "eval_logits/chosen": 5.5625,
-      "eval_logits/rejected": 5.96875,
-      "eval_logps/chosen": -62464.0,
-      "eval_logps/rejected": -54016.0,
-      "eval_loss": 183.29150390625,
-      "eval_rewards/accuracies": 0.4375,
-      "eval_rewards/chosen": -624.0,
-      "eval_rewards/margins": -84.5,
-      "eval_rewards/rejected": -536.0,
-      "eval_runtime": 46.6586,
-      "eval_samples_per_second": 42.865,
+      "eval_logits/chosen": -13.6875,
+      "eval_logits/rejected": -12.9375,
+      "eval_logps/chosen": -840.0,
+      "eval_logps/rejected": -840.0,
+      "eval_loss": 1.2549687623977661,
+      "eval_rewards/accuracies": 0.509765625,
+      "eval_rewards/chosen": -5.25,
+      "eval_rewards/margins": 0.30078125,
+      "eval_rewards/rejected": -5.53125,
+      "eval_runtime": 46.6654,
+      "eval_samples_per_second": 42.858,
       "eval_steps_per_second": 0.686,
       "step": 2300
     },
     {
       "epoch": 2.418848167539267,
-      "grad_norm": 0.01625483829759587,
+      "grad_norm": 0.0013932961920628188,
       "learning_rate": 5.50315277286812e-06,
-      "logits/chosen": 4.9375,
-      "logits/rejected": 4.9375,
-      "logps/chosen": -75264.0,
-      "logps/rejected": -75264.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -752.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -752.0,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -844.0,
+      "logps/rejected": -5312.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 25.875,
+      "rewards/rejected": -31.0,
       "step": 2310
     },
     {
       "epoch": 2.4293193717277486,
-      "grad_norm": 0.015040163043270313,
+      "grad_norm": 0.11096596541540364,
       "learning_rate": 5.313910992897281e-06,
-      "logits/chosen": 4.5,
-      "logits/rejected": 4.5,
-      "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": -13.4375,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -824.0,
+      "logps/rejected": -5248.0,
+      "loss": 0.0007,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 26.375,
+      "rewards/rejected": -31.0,
       "step": 2320
     },
     {
       "epoch": 2.4397905759162306,
-      "grad_norm": 0.016651122757673102,
+      "grad_norm": 0.005803935668758239,
       "learning_rate": 5.127592611220658e-06,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "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": -13.9375,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -892.0,
+      "logps/rejected": -5312.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 26.625,
+      "rewards/rejected": -31.75,
       "step": 2330
     },
     {
       "epoch": 2.450261780104712,
-      "grad_norm": 0.014075888133178645,
+      "grad_norm": 0.009377808643088494,
       "learning_rate": 4.944225296251159e-06,
-      "logits/chosen": 4.75,
-      "logits/rejected": 4.75,
-      "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": -13.875,
+      "logits/rejected": -8.6875,
+      "logps/chosen": -836.0,
+      "logps/rejected": -5568.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 27.625,
+      "rewards/rejected": -32.5,
       "step": 2340
     },
     {
       "epoch": 2.4607329842931938,
-      "grad_norm": 0.015778100245365224,
+      "grad_norm": 0.0010445959214331287,
       "learning_rate": 4.763836278166172e-06,
-      "logits/chosen": 4.84375,
-      "logits/rejected": 4.84375,
-      "logps/chosen": -74752.0,
-      "logps/rejected": -74752.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -744.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -744.0,
+      "logits/chosen": -13.8125,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -836.0,
+      "logps/rejected": -5216.0,
+      "loss": 0.0004,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.84375,
+      "rewards/margins": 25.875,
+      "rewards/rejected": -30.625,
       "step": 2350
     },
     {
       "epoch": 2.4712041884816753,
-      "grad_norm": 0.016399744233765227,
+      "grad_norm": 0.003406081019883494,
       "learning_rate": 4.586452344863876e-06,
-      "logits/chosen": 4.65625,
-      "logits/rejected": 4.65625,
-      "logps/chosen": -76288.0,
-      "logps/rejected": -76288.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -760.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -760.0,
+      "logits/chosen": -14.4375,
+      "logits/rejected": -9.375,
+      "logps/chosen": -856.0,
+      "logps/rejected": -5312.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 26.5,
+      "rewards/rejected": -31.375,
       "step": 2360
     },
     {
       "epoch": 2.481675392670157,
-      "grad_norm": 0.015048950095585073,
+      "grad_norm": 0.03530616873682385,
       "learning_rate": 4.412099837985192e-06,
-      "logits/chosen": 4.75,
-      "logits/rejected": 4.75,
-      "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": -14.3125,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -860.0,
+      "logps/rejected": -5376.0,
+      "loss": 0.0006,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 26.375,
+      "rewards/rejected": -31.25,
       "step": 2370
     },
     {
       "epoch": 2.492146596858639,
-      "grad_norm": 0.016573180997983297,
+      "grad_norm": 0.006563381413062817,
       "learning_rate": 4.240804649002089e-06,
-      "logits/chosen": 4.6875,
-      "logits/rejected": 4.6875,
-      "logps/chosen": -71168.0,
-      "logps/rejected": -71168.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -708.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -708.0,
+      "logits/chosen": -14.0,
+      "logits/rejected": -8.5625,
+      "logps/chosen": -844.0,
+      "logps/rejected": -5504.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.125,
+      "rewards/margins": 27.5,
+      "rewards/rejected": -32.5,
       "step": 2380
     },
     {
       "epoch": 2.5026178010471205,
-      "grad_norm": 0.015863635435345965,
+      "grad_norm": 0.0016972205283148803,
       "learning_rate": 4.072592215372598e-06,
-      "logits/chosen": 4.46875,
-      "logits/rejected": 4.46875,
-      "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": -14.5,
+      "logits/rejected": -8.75,
+      "logps/chosen": -944.0,
+      "logps/rejected": -5312.0,
+      "loss": 0.0006,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.4375,
+      "rewards/margins": 25.75,
+      "rewards/rejected": -31.125,
       "step": 2390
     },
     {
       "epoch": 2.513089005235602,
-      "grad_norm": 0.014303133833648162,
+      "grad_norm": 0.005067233377919347,
       "learning_rate": 3.907487516763389e-06,
-      "logits/chosen": 4.96875,
-      "logits/rejected": 4.96875,
-      "logps/chosen": -74752.0,
-      "logps/rejected": -74752.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -744.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -744.0,
+      "logits/chosen": -14.5,
+      "logits/rejected": -9.75,
+      "logps/chosen": -892.0,
+      "logps/rejected": -5568.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.34375,
+      "rewards/margins": 27.5,
+      "rewards/rejected": -32.75,
       "step": 2400
     },
     {
       "epoch": 2.513089005235602,
-      "eval_logits/chosen": 5.59375,
-      "eval_logits/rejected": 6.0,
-      "eval_logps/chosen": -62976.0,
-      "eval_logps/rejected": -54272.0,
-      "eval_loss": 184.10775756835938,
-      "eval_rewards/accuracies": 0.43359375,
-      "eval_rewards/chosen": -624.0,
-      "eval_rewards/margins": -85.0,
-      "eval_rewards/rejected": -540.0,
-      "eval_runtime": 46.6723,
-      "eval_samples_per_second": 42.852,
+      "eval_logits/chosen": -13.875,
+      "eval_logits/rejected": -13.1875,
+      "eval_logps/chosen": -848.0,
+      "eval_logps/rejected": -848.0,
+      "eval_loss": 1.275781273841858,
+      "eval_rewards/accuracies": 0.509765625,
+      "eval_rewards/chosen": -5.28125,
+      "eval_rewards/margins": 0.322265625,
+      "eval_rewards/rejected": -5.625,
+      "eval_runtime": 46.6592,
+      "eval_samples_per_second": 42.864,
       "eval_steps_per_second": 0.686,
       "step": 2400
     },
     {
       "epoch": 2.5235602094240837,
-      "grad_norm": 0.01733060475606437,
+      "grad_norm": 0.004430289949009675,
       "learning_rate": 3.745515071340286e-06,
-      "logits/chosen": 4.9375,
-      "logits/rejected": 4.9375,
-      "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": -14.0625,
+      "logits/rejected": -8.75,
+      "logps/chosen": -964.0,
+      "logps/rejected": -5376.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.71875,
+      "rewards/margins": 27.0,
+      "rewards/rejected": -32.75,
       "step": 2410
     },
     {
       "epoch": 2.5340314136125652,
-      "grad_norm": 0.017634427199633762,
+      "grad_norm": 0.016912272942012255,
       "learning_rate": 3.586698932127236e-06,
-      "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": -14.0625,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -864.0,
+      "logps/rejected": -5536.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 27.125,
+      "rewards/rejected": -32.25,
       "step": 2420
     },
     {
       "epoch": 2.5445026178010473,
-      "grad_norm": 0.01608546277903839,
+      "grad_norm": 0.0185364869264746,
       "learning_rate": 3.4310626834344744e-06,
-      "logits/chosen": 4.59375,
-      "logits/rejected": 4.59375,
-      "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": -14.375,
+      "logits/rejected": -9.5625,
+      "logps/chosen": -900.0,
+      "logps/rejected": -5120.0,
+      "loss": 0.0007,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.1875,
+      "rewards/margins": 25.375,
+      "rewards/rejected": -30.5,
       "step": 2430
     },
     {
       "epoch": 2.554973821989529,
-      "grad_norm": 0.014451566980983009,
+      "grad_norm": 0.016171112964388112,
       "learning_rate": 3.2786294373562343e-06,
-      "logits/chosen": 4.625,
-      "logits/rejected": 4.625,
-      "logps/chosen": -76288.0,
-      "logps/rejected": -76288.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -760.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -760.0,
+      "logits/chosen": -13.75,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -896.0,
+      "logps/rejected": -5664.0,
+      "loss": 0.0004,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.21875,
+      "rewards/margins": 28.375,
+      "rewards/rejected": -33.5,
       "step": 2440
     },
     {
       "epoch": 2.5654450261780104,
-      "grad_norm": 0.012501769445046825,
+      "grad_norm": 0.012165879407141509,
       "learning_rate": 3.1294218303385526e-06,
-      "logits/chosen": 4.21875,
-      "logits/rejected": 4.21875,
-      "logps/chosen": -80384.0,
-      "logps/rejected": -80384.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -796.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -796.0,
+      "logits/chosen": -13.75,
+      "logits/rejected": -8.875,
+      "logps/chosen": -884.0,
+      "logps/rejected": -5504.0,
+      "loss": 0.0001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 26.75,
+      "rewards/rejected": -31.75,
       "step": 2450
     },
     {
       "epoch": 2.5759162303664924,
-      "grad_norm": 0.016228551304688792,
+      "grad_norm": 0.004416944054089835,
       "learning_rate": 2.9834620198177483e-06,
-      "logits/chosen": 4.03125,
-      "logits/rejected": 4.03125,
-      "logps/chosen": -76288.0,
-      "logps/rejected": -76288.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -756.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -756.0,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -8.875,
+      "logps/chosen": -892.0,
+      "logps/rejected": -5312.0,
+      "loss": 0.0004,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.1875,
+      "rewards/margins": 26.875,
+      "rewards/rejected": -32.0,
       "step": 2460
     },
     {
       "epoch": 2.5863874345549736,
-      "grad_norm": 0.016966901097151713,
+      "grad_norm": 0.00396393259218364,
       "learning_rate": 2.840771680930068e-06,
-      "logits/chosen": 4.625,
-      "logits/rejected": 4.625,
-      "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": -14.6875,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -948.0,
+      "logps/rejected": -5728.0,
+      "loss": 0.0001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.53125,
+      "rewards/margins": 29.125,
+      "rewards/rejected": -34.75,
       "step": 2470
     },
     {
       "epoch": 2.5968586387434556,
-      "grad_norm": 0.013604974412031411,
+      "grad_norm": 0.0024904932792960965,
       "learning_rate": 2.701372003292868e-06,
-      "logits/chosen": 4.875,
-      "logits/rejected": 4.875,
-      "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": -13.5,
+      "logits/rejected": -8.875,
+      "logps/chosen": -872.0,
+      "logps/rejected": -5472.0,
+      "loss": 0.0004,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.15625,
+      "rewards/margins": 26.875,
+      "rewards/rejected": -32.0,
       "step": 2480
     },
     {
       "epoch": 2.607329842931937,
-      "grad_norm": 0.019817921335395502,
+      "grad_norm": 0.0016280188423222486,
       "learning_rate": 2.56528368785795e-06,
-      "logits/chosen": 4.6875,
-      "logits/rejected": 4.6875,
-      "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": -14.5,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -884.0,
+      "logps/rejected": -5632.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.0,
+      "rewards/margins": 28.125,
+      "rewards/rejected": -33.0,
       "step": 2490
     },
     {
       "epoch": 2.6178010471204187,
-      "grad_norm": 0.01580063322795,
+      "grad_norm": 0.027726411034436783,
       "learning_rate": 2.432526943837471e-06,
-      "logits/chosen": 4.71875,
-      "logits/rejected": 4.71875,
-      "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": -14.0625,
+      "logits/rejected": -8.75,
+      "logps/chosen": -836.0,
+      "logps/rejected": -5472.0,
+      "loss": 0.0004,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.8125,
+      "rewards/margins": 27.625,
+      "rewards/rejected": -32.5,
       "step": 2500
     },
     {
       "epoch": 2.6178010471204187,
-      "eval_logits/chosen": 5.65625,
-      "eval_logits/rejected": 6.0625,
-      "eval_logps/chosen": -62976.0,
-      "eval_logps/rejected": -54528.0,
-      "eval_loss": 184.6541290283203,
-      "eval_rewards/accuracies": 0.423828125,
-      "eval_rewards/chosen": -628.0,
-      "eval_rewards/margins": -85.0,
-      "eval_rewards/rejected": -540.0,
-      "eval_runtime": 46.6628,
-      "eval_samples_per_second": 42.861,
+      "eval_logits/chosen": -13.875,
+      "eval_logits/rejected": -13.1875,
+      "eval_logps/chosen": -844.0,
+      "eval_logps/rejected": -848.0,
+      "eval_loss": 1.2773749828338623,
+      "eval_rewards/accuracies": 0.50390625,
+      "eval_rewards/chosen": -5.28125,
+      "eval_rewards/margins": 0.32421875,
+      "eval_rewards/rejected": -5.59375,
+      "eval_runtime": 46.6497,
+      "eval_samples_per_second": 42.873,
       "eval_steps_per_second": 0.686,
       "step": 2500
     },
     {
       "epoch": 2.6282722513089007,
-      "grad_norm": 0.015931506363039496,
+      "grad_norm": 0.02769769667932305,
       "learning_rate": 2.3031214857028446e-06,
-      "logits/chosen": 4.59375,
-      "logits/rejected": 4.59375,
-      "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": -14.0,
+      "logits/rejected": -8.875,
+      "logps/chosen": -892.0,
+      "logps/rejected": -5344.0,
+      "loss": 0.0038,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.0,
+      "rewards/margins": 26.875,
+      "rewards/rejected": -31.875,
       "step": 2510
     },
     {
       "epoch": 2.6387434554973823,
-      "grad_norm": 0.013881012695415284,
+      "grad_norm": 0.006169289216015707,
       "learning_rate": 2.177086530257139e-06,
-      "logits/chosen": 4.875,
-      "logits/rejected": 4.875,
-      "logps/chosen": -76288.0,
-      "logps/rejected": -76288.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -760.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -760.0,
+      "logits/chosen": -14.125,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -852.0,
+      "logps/rejected": -5600.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.0625,
+      "rewards/margins": 28.375,
+      "rewards/rejected": -33.5,
       "step": 2520
     },
     {
       "epoch": 2.649214659685864,
-      "grad_norm": 0.015961858033847396,
+      "grad_norm": 0.015381496790030457,
       "learning_rate": 2.054440793781354e-06,
-      "logits/chosen": 4.65625,
-      "logits/rejected": 4.65625,
-      "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": -14.0,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -856.0,
+      "logps/rejected": -5408.0,
+      "loss": 0.0004,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 27.125,
+      "rewards/rejected": -32.0,
       "step": 2530
     },
     {
       "epoch": 2.6596858638743455,
-      "grad_norm": 0.016259792363164875,
+      "grad_norm": 0.006699038864973256,
       "learning_rate": 1.9352024892550587e-06,
-      "logits/chosen": 4.75,
-      "logits/rejected": 4.75,
-      "logps/chosen": -74752.0,
-      "logps/rejected": -74752.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -744.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -744.0,
+      "logits/chosen": -13.375,
+      "logits/rejected": -8.875,
+      "logps/chosen": -908.0,
+      "logps/rejected": -5536.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.375,
+      "rewards/margins": 27.625,
+      "rewards/rejected": -33.0,
       "step": 2540
     },
     {
       "epoch": 2.670157068062827,
-      "grad_norm": 0.01379595720263763,
+      "grad_norm": 0.07067023282211729,
       "learning_rate": 1.8193893236517379e-06,
-      "logits/chosen": 4.28125,
-      "logits/rejected": 4.28125,
-      "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": -14.3125,
+      "logits/rejected": -9.375,
+      "logps/chosen": -888.0,
+      "logps/rejected": -5472.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 27.0,
+      "rewards/rejected": -32.0,
       "step": 2550
     },
     {
       "epoch": 2.680628272251309,
-      "grad_norm": 0.016414001023817397,
+      "grad_norm": 0.0035490897673744154,
       "learning_rate": 1.7070184953092956e-06,
-      "logits/chosen": 4.59375,
-      "logits/rejected": 4.59375,
-      "logps/chosen": -76288.0,
-      "logps/rejected": -76288.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -756.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -756.0,
+      "logits/chosen": -13.9375,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -856.0,
+      "logps/rejected": -5280.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 27.125,
+      "rewards/rejected": -32.0,
       "step": 2560
     },
     {
       "epoch": 2.6910994764397906,
-      "grad_norm": 0.013539743446248715,
+      "grad_norm": 0.0009483951569351297,
       "learning_rate": 1.5981066913760918e-06,
-      "logits/chosen": 4.90625,
-      "logits/rejected": 4.90625,
-      "logps/chosen": -65536.0,
-      "logps/rejected": -65536.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -652.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -652.0,
+      "logits/chosen": -13.5625,
+      "logits/rejected": -9.5,
+      "logps/chosen": -860.0,
+      "logps/rejected": -5504.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.125,
+      "rewards/margins": 27.25,
+      "rewards/rejected": -32.5,
       "step": 2570
     },
     {
       "epoch": 2.701570680628272,
-      "grad_norm": 0.016790388062982404,
+      "grad_norm": 0.08556280393733294,
       "learning_rate": 1.492670085332884e-06,
-      "logits/chosen": 4.5625,
-      "logits/rejected": 4.5625,
-      "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": -14.375,
+      "logits/rejected": -8.625,
+      "logps/chosen": -920.0,
+      "logps/rejected": -5760.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.34375,
+      "rewards/margins": 28.125,
+      "rewards/rejected": -33.5,
       "step": 2580
     },
     {
       "epoch": 2.712041884816754,
-      "grad_norm": 0.0160677185202212,
+      "grad_norm": 0.03733970451956642,
       "learning_rate": 1.3907243345910787e-06,
-      "logits/chosen": 4.78125,
-      "logits/rejected": 4.78125,
-      "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": -14.375,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -900.0,
+      "logps/rejected": -5472.0,
+      "loss": 0.0012,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.0,
+      "rewards/margins": 26.375,
+      "rewards/rejected": -31.375,
       "step": 2590
     },
     {
       "epoch": 2.7225130890052354,
-      "grad_norm": 0.015823164766396173,
+      "grad_norm": 0.0017046210827818176,
       "learning_rate": 1.292284578167538e-06,
-      "logits/chosen": 4.28125,
-      "logits/rejected": 4.28125,
-      "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": -13.9375,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -856.0,
+      "logps/rejected": -5376.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 27.25,
+      "rewards/rejected": -32.0,
       "step": 2600
     },
     {
       "epoch": 2.7225130890052354,
-      "eval_logits/chosen": 5.625,
-      "eval_logits/rejected": 6.03125,
-      "eval_logps/chosen": -63232.0,
-      "eval_logps/rejected": -54528.0,
-      "eval_loss": 185.0226287841797,
-      "eval_rewards/accuracies": 0.427734375,
-      "eval_rewards/chosen": -628.0,
-      "eval_rewards/margins": -85.5,
-      "eval_rewards/rejected": -544.0,
-      "eval_runtime": 46.6566,
-      "eval_samples_per_second": 42.866,
+      "eval_logits/chosen": -13.9375,
+      "eval_logits/rejected": -13.25,
+      "eval_logps/chosen": -844.0,
+      "eval_logps/rejected": -848.0,
+      "eval_loss": 1.2789686918258667,
+      "eval_rewards/accuracies": 0.50390625,
+      "eval_rewards/chosen": -5.28125,
+      "eval_rewards/margins": 0.328125,
+      "eval_rewards/rejected": -5.59375,
+      "eval_runtime": 46.6561,
+      "eval_samples_per_second": 42.867,
       "eval_steps_per_second": 0.686,
       "step": 2600
     },
     {
       "epoch": 2.7329842931937174,
-      "grad_norm": 0.017457001383893146,
+      "grad_norm": 0.01271477209319848,
       "learning_rate": 1.1973654344364925e-06,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "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": -14.4375,
+      "logits/rejected": -8.875,
+      "logps/chosen": -896.0,
+      "logps/rejected": -5376.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.125,
+      "rewards/margins": 26.875,
+      "rewards/rejected": -32.0,
       "step": 2610
     },
     {
       "epoch": 2.743455497382199,
-      "grad_norm": 0.016782987633735924,
+      "grad_norm": 0.021483968566332147,
       "learning_rate": 1.1059809989586583e-06,
-      "logits/chosen": 5.0,
-      "logits/rejected": 5.0,
-      "logps/chosen": -74752.0,
-      "logps/rejected": -74752.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -744.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -744.0,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -876.0,
+      "logps/rejected": -5440.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 26.625,
+      "rewards/rejected": -31.625,
       "step": 2620
     },
     {
       "epoch": 2.7539267015706805,
-      "grad_norm": 0.015404754596285269,
+      "grad_norm": 0.013043529395686317,
       "learning_rate": 1.018144842388044e-06,
-      "logits/chosen": 4.71875,
-      "logits/rejected": 4.71875,
-      "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": -13.625,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -908.0,
+      "logps/rejected": -5568.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 27.875,
+      "rewards/rejected": -33.0,
       "step": 2630
     },
     {
       "epoch": 2.7643979057591626,
-      "grad_norm": 0.01685160901855413,
+      "grad_norm": 0.0015820866140562899,
       "learning_rate": 9.338700084567109e-07,
-      "logits/chosen": 4.625,
-      "logits/rejected": 4.625,
-      "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": -14.0625,
+      "logits/rejected": -9.0,
+      "logps/chosen": -880.0,
+      "logps/rejected": -5536.0,
+      "loss": 0.0001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.1875,
+      "rewards/margins": 27.625,
+      "rewards/rejected": -32.75,
       "step": 2640
     },
     {
       "epoch": 2.774869109947644,
-      "grad_norm": 0.016932026378610687,
+      "grad_norm": 0.006693597297910744,
       "learning_rate": 8.531690120377606e-07,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "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": -14.0625,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -924.0,
+      "logps/rejected": -5408.0,
+      "loss": 0.0219,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.28125,
+      "rewards/margins": 26.75,
+      "rewards/rejected": -32.0,
       "step": 2650
     },
     {
       "epoch": 2.7853403141361257,
-      "grad_norm": 0.015931378554499352,
+      "grad_norm": 0.005032235482414184,
       "learning_rate": 7.760538372868636e-07,
-      "logits/chosen": 4.625,
-      "logits/rejected": 4.625,
-      "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": -14.9375,
+      "logits/rejected": -8.5625,
+      "logps/chosen": -848.0,
+      "logps/rejected": -5440.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.90625,
+      "rewards/margins": 27.125,
+      "rewards/rejected": -32.0,
       "step": 2660
     },
     {
       "epoch": 2.7958115183246073,
-      "grad_norm": 0.018071234054647396,
+      "grad_norm": 0.011930229964050118,
       "learning_rate": 7.025359358626166e-07,
-      "logits/chosen": 4.8125,
-      "logits/rejected": 4.8125,
-      "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": -13.9375,
+      "logits/rejected": -8.625,
+      "logps/chosen": -948.0,
+      "logps/rejected": -5504.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.53125,
+      "rewards/margins": 28.0,
+      "rewards/rejected": -33.5,
       "step": 2670
     },
     {
       "epoch": 2.806282722513089,
-      "grad_norm": 0.016633856627237726,
+      "grad_norm": 0.012605512499874861,
       "learning_rate": 6.326262252259463e-07,
-      "logits/chosen": 4.40625,
-      "logits/rejected": 4.40625,
-      "logps/chosen": -73216.0,
-      "logps/rejected": -73216.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -728.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -728.0,
+      "logits/chosen": -14.0625,
+      "logits/rejected": -9.0,
+      "logps/chosen": -908.0,
+      "logps/rejected": -5568.0,
+      "loss": 0.0009,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.125,
+      "rewards/margins": 27.75,
+      "rewards/rejected": -33.0,
       "step": 2680
     },
     {
       "epoch": 2.816753926701571,
-      "grad_norm": 0.017480316026428723,
+      "grad_norm": 0.013470367213375353,
       "learning_rate": 5.66335087018871e-07,
-      "logits/chosen": 4.40625,
-      "logits/rejected": 4.40625,
-      "logps/chosen": -76288.0,
-      "logps/rejected": -76288.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -756.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -756.0,
+      "logits/chosen": -14.375,
+      "logits/rejected": -9.375,
+      "logps/chosen": -900.0,
+      "logps/rejected": -5344.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.0625,
+      "rewards/margins": 26.5,
+      "rewards/rejected": -31.625,
       "step": 2690
     },
     {
       "epoch": 2.8272251308900525,
-      "grad_norm": 0.015485942195298297,
+      "grad_norm": 0.006947985440392243,
       "learning_rate": 5.036723655228226e-07,
-      "logits/chosen": 4.84375,
-      "logits/rejected": 4.84375,
-      "logps/chosen": -71168.0,
-      "logps/rejected": -71168.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -708.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -708.0,
+      "logits/chosen": -14.375,
+      "logits/rejected": -9.0,
+      "logps/chosen": -880.0,
+      "logps/rejected": -5408.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.25,
+      "rewards/margins": 27.0,
+      "rewards/rejected": -32.25,
       "step": 2700
     },
     {
       "epoch": 2.8272251308900525,
-      "eval_logits/chosen": 5.65625,
-      "eval_logits/rejected": 6.0625,
-      "eval_logps/chosen": -63232.0,
-      "eval_logps/rejected": -54528.0,
-      "eval_loss": 185.15574645996094,
-      "eval_rewards/accuracies": 0.4296875,
-      "eval_rewards/chosen": -628.0,
-      "eval_rewards/margins": -85.5,
-      "eval_rewards/rejected": -544.0,
-      "eval_runtime": 46.6532,
+      "eval_logits/chosen": -13.9375,
+      "eval_logits/rejected": -13.25,
+      "eval_logps/chosen": -844.0,
+      "eval_logps/rejected": -848.0,
+      "eval_loss": 1.2763124704360962,
+      "eval_rewards/accuracies": 0.501953125,
+      "eval_rewards/chosen": -5.28125,
+      "eval_rewards/margins": 0.33203125,
+      "eval_rewards/rejected": -5.59375,
+      "eval_runtime": 46.6541,
       "eval_samples_per_second": 42.869,
       "eval_steps_per_second": 0.686,
       "step": 2700
     },
     {
       "epoch": 2.837696335078534,
-      "grad_norm": 0.014581629089113578,
+      "grad_norm": 0.002001134770322125,
       "learning_rate": 4.446473661967432e-07,
-      "logits/chosen": 4.84375,
-      "logits/rejected": 4.84375,
-      "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": -14.75,
+      "logits/rejected": -9.4375,
+      "logps/chosen": -872.0,
+      "logps/rejected": -5312.0,
+      "loss": 0.0007,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.15625,
+      "rewards/margins": 25.75,
+      "rewards/rejected": -30.875,
       "step": 2710
     },
     {
       "epoch": 2.8481675392670156,
-      "grad_norm": 0.017152425658207706,
+      "grad_norm": 0.02409440907941081,
       "learning_rate": 3.892688542952411e-07,
-      "logits/chosen": 5.0,
-      "logits/rejected": 5.0,
-      "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": -14.375,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -816.0,
+      "logps/rejected": -5056.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.625,
+      "rewards/margins": 25.125,
+      "rewards/rejected": -29.75,
       "step": 2720
     },
     {
       "epoch": 2.858638743455497,
-      "grad_norm": 0.014878244485550376,
+      "grad_norm": 0.0022833452925910486,
       "learning_rate": 3.3754505356694e-07,
-      "logits/chosen": 4.46875,
-      "logits/rejected": 4.46875,
-      "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": -14.875,
+      "logits/rejected": -9.25,
+      "logps/chosen": -872.0,
+      "logps/rejected": -5600.0,
+      "loss": 0.0001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.03125,
+      "rewards/margins": 28.0,
+      "rewards/rejected": -33.0,
       "step": 2730
     },
     {
       "epoch": 2.869109947643979,
-      "grad_norm": 0.016380618369050567,
+      "grad_norm": 0.593967722802982,
       "learning_rate": 2.8948364503322276e-07,
-      "logits/chosen": 4.875,
-      "logits/rejected": 4.875,
-      "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": -14.3125,
+      "logits/rejected": -9.125,
+      "logps/chosen": -844.0,
+      "logps/rejected": -5536.0,
+      "loss": 0.0049,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 27.875,
+      "rewards/rejected": -33.0,
       "step": 2740
     },
     {
       "epoch": 2.8795811518324608,
-      "grad_norm": 0.014414973120714561,
+      "grad_norm": 0.00017724466857888153,
       "learning_rate": 2.4509176584762706e-07,
-      "logits/chosen": 4.65625,
-      "logits/rejected": 4.65625,
-      "logps/chosen": -74752.0,
-      "logps/rejected": -74752.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -744.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -744.0,
+      "logits/chosen": -14.375,
+      "logits/rejected": -8.4375,
+      "logps/chosen": -856.0,
+      "logps/rejected": -5504.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.0,
+      "rewards/margins": 27.25,
+      "rewards/rejected": -32.25,
       "step": 2750
     },
     {
       "epoch": 2.8900523560209423,
-      "grad_norm": 0.015440359256275139,
+      "grad_norm": 0.003997240440376325,
       "learning_rate": 2.0437600823595694e-07,
-      "logits/chosen": 4.625,
-      "logits/rejected": 4.625,
-      "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": -13.875,
+      "logits/rejected": -9.125,
+      "logps/chosen": -908.0,
+      "logps/rejected": -5440.0,
+      "loss": 0.0005,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 26.75,
+      "rewards/rejected": -31.875,
       "step": 2760
     },
     {
       "epoch": 2.900523560209424,
-      "grad_norm": 0.01830173070682299,
+      "grad_norm": 0.003981325015247753,
       "learning_rate": 1.6734241851733e-07,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "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": -14.0,
+      "logits/rejected": -8.625,
+      "logps/chosen": -944.0,
+      "logps/rejected": -5408.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.375,
+      "rewards/margins": 26.625,
+      "rewards/rejected": -32.0,
       "step": 2770
     },
     {
       "epoch": 2.9109947643979055,
-      "grad_norm": 0.015896618150146622,
+      "grad_norm": 0.007659446220854144,
       "learning_rate": 1.339964962062984e-07,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "logps/chosen": -75776.0,
-      "logps/rejected": -75776.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -756.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -756.0,
+      "logits/chosen": -14.1875,
+      "logits/rejected": -9.6875,
+      "logps/chosen": -832.0,
+      "logps/rejected": -5504.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.875,
+      "rewards/margins": 28.75,
+      "rewards/rejected": -33.75,
       "step": 2780
     },
     {
       "epoch": 2.9214659685863875,
-      "grad_norm": 0.015874461187826042,
+      "grad_norm": 0.002986219027401011,
       "learning_rate": 1.0434319319617136e-07,
-      "logits/chosen": 4.96875,
-      "logits/rejected": 4.96875,
-      "logps/chosen": -76288.0,
-      "logps/rejected": -76288.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -760.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -760.0,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -9.125,
+      "logps/chosen": -860.0,
+      "logps/rejected": -5504.0,
+      "loss": 0.0004,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.0625,
+      "rewards/margins": 27.5,
+      "rewards/rejected": -32.5,
       "step": 2790
     },
     {
       "epoch": 2.931937172774869,
-      "grad_norm": 0.013958448467088576,
+      "grad_norm": 0.00014825552055201214,
       "learning_rate": 7.838691302365363e-08,
-      "logits/chosen": 4.5,
-      "logits/rejected": 4.5,
-      "logps/chosen": -77824.0,
-      "logps/rejected": -77824.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -772.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -772.0,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -9.625,
+      "logps/chosen": -868.0,
+      "logps/rejected": -5280.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 26.375,
+      "rewards/rejected": -31.375,
       "step": 2800
     },
     {
       "epoch": 2.931937172774869,
-      "eval_logits/chosen": 5.65625,
-      "eval_logits/rejected": 6.0625,
-      "eval_logps/chosen": -63232.0,
-      "eval_logps/rejected": -54528.0,
-      "eval_loss": 185.19000244140625,
-      "eval_rewards/accuracies": 0.431640625,
-      "eval_rewards/chosen": -628.0,
-      "eval_rewards/margins": -85.5,
-      "eval_rewards/rejected": -544.0,
-      "eval_runtime": 46.6619,
-      "eval_samples_per_second": 42.862,
+      "eval_logits/chosen": -14.0,
+      "eval_logits/rejected": -13.25,
+      "eval_logps/chosen": -844.0,
+      "eval_logps/rejected": -848.0,
+      "eval_loss": 1.277500033378601,
+      "eval_rewards/accuracies": 0.50390625,
+      "eval_rewards/chosen": -5.28125,
+      "eval_rewards/margins": 0.330078125,
+      "eval_rewards/rejected": -5.59375,
+      "eval_runtime": 46.6742,
+      "eval_samples_per_second": 42.85,
       "eval_steps_per_second": 0.686,
       "step": 2800
     },
     {
       "epoch": 2.9424083769633507,
-      "grad_norm": 0.016429832368248316,
+      "grad_norm": 0.004909665218282632,
       "learning_rate": 5.613151021490459e-08,
-      "logits/chosen": 4.8125,
-      "logits/rejected": 4.8125,
-      "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": -14.25,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -788.0,
+      "logps/rejected": -5408.0,
+      "loss": 0.0001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.71875,
+      "rewards/margins": 27.5,
+      "rewards/rejected": -32.25,
       "step": 2810
     },
     {
       "epoch": 2.9528795811518327,
-      "grad_norm": 0.015014248861730257,
+      "grad_norm": 0.010188994266724465,
       "learning_rate": 3.7580289713159965e-08,
-      "logits/chosen": 4.875,
-      "logits/rejected": 4.875,
-      "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": -14.3125,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -892.0,
+      "logps/rejected": -5312.0,
+      "loss": 0.0003,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.0625,
+      "rewards/margins": 26.125,
+      "rewards/rejected": -31.25,
       "step": 2820
     },
     {
       "epoch": 2.9633507853403143,
-      "grad_norm": 0.017762532422474338,
+      "grad_norm": 0.008001944870352438,
       "learning_rate": 2.2736006387927277e-08,
-      "logits/chosen": 4.6875,
-      "logits/rejected": 4.6875,
-      "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": -14.25,
+      "logits/rejected": -9.0,
+      "logps/chosen": -888.0,
+      "logps/rejected": -5344.0,
+      "loss": 0.0002,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.0,
+      "rewards/margins": 26.625,
+      "rewards/rejected": -31.625,
       "step": 2830
     },
     {
       "epoch": 2.973821989528796,
-      "grad_norm": 0.014056152571939384,
+      "grad_norm": 0.01106107039922046,
       "learning_rate": 1.1600864625893604e-08,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "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": -14.5625,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -848.0,
+      "logps/rejected": -5568.0,
+      "loss": 0.0001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -4.84375,
+      "rewards/margins": 28.0,
+      "rewards/rejected": -32.75,
       "step": 2840
     },
     {
       "epoch": 2.9842931937172774,
-      "grad_norm": 0.015234205635484511,
+      "grad_norm": 0.006444542464492691,
       "learning_rate": 4.176518003573549e-09,
-      "logits/chosen": 4.53125,
-      "logits/rejected": 4.53125,
-      "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": -14.5625,
+      "logits/rejected": -8.9375,
+      "logps/chosen": -888.0,
+      "logps/rejected": -5440.0,
+      "loss": 0.0001,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.1875,
+      "rewards/margins": 26.75,
+      "rewards/rejected": -31.875,
       "step": 2850
     },
     {
       "epoch": 2.994764397905759,
-      "grad_norm": 0.01757459964805105,
+      "grad_norm": 0.22770292622257943,
       "learning_rate": 4.640690417528903e-10,
-      "logits/chosen": 4.78125,
-      "logits/rejected": 4.78125,
-      "logps/chosen": -71680.0,
-      "logps/rejected": -71680.0,
-      "loss": 0.6914,
-      "rewards/accuracies": 0.0,
-      "rewards/chosen": -716.0,
-      "rewards/margins": 0.0,
-      "rewards/rejected": -716.0,
+      "logits/chosen": -14.3125,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -856.0,
+      "logps/rejected": -5568.0,
+      "loss": 0.0008,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -5.09375,
+      "rewards/margins": 28.125,
+      "rewards/rejected": -33.25,
       "step": 2860
     },
     {
       "epoch": 3.0,
       "step": 2865,
       "total_flos": 0.0,
-      "train_loss": 0.69140625,
-      "train_runtime": 12163.3984,
-      "train_samples_per_second": 15.074,
-      "train_steps_per_second": 0.236
+      "train_loss": 0.024607474328222707,
+      "train_runtime": 12418.8835,
+      "train_samples_per_second": 14.764,
+      "train_steps_per_second": 0.231
     }
   ],
   "logging_steps": 10,