diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -10,7 +10,7 @@
   "log_history": [
     {
       "epoch": 0.0010471204188481676,
-      "grad_norm": 1.582245517402759,
+      "grad_norm": 1.5822327138520569,
       "learning_rate": 1.7421602787456448e-07,
       "logits/chosen": -11.375,
       "logits/rejected": -11.6875,
@@ -25,119 +25,119 @@
     },
     {
       "epoch": 0.010471204188481676,
-      "grad_norm": 1.5263738363102872,
+      "grad_norm": 1.531601426535834,
       "learning_rate": 1.7421602787456445e-06,
       "logits/chosen": -11.3125,
-      "logits/rejected": -10.9375,
+      "logits/rejected": -10.875,
       "logps/chosen": -438.0,
       "logps/rejected": -548.0,
-      "loss": 0.6921,
+      "loss": 0.6928,
       "rewards/accuracies": 0.1458333283662796,
-      "rewards/chosen": 0.00107574462890625,
-      "rewards/margins": -0.000644683837890625,
-      "rewards/rejected": 0.0017242431640625,
+      "rewards/chosen": 0.000301361083984375,
+      "rewards/margins": -0.00067901611328125,
+      "rewards/rejected": 0.0009765625,
       "step": 10
     },
     {
       "epoch": 0.020942408376963352,
-      "grad_norm": 1.5021681553716235,
+      "grad_norm": 1.5026120891800383,
       "learning_rate": 3.484320557491289e-06,
       "logits/chosen": -11.125,
       "logits/rejected": -11.0,
       "logps/chosen": -404.0,
       "logps/rejected": -456.0,
-      "loss": 0.6904,
-      "rewards/accuracies": 0.29374998807907104,
-      "rewards/chosen": -0.00091552734375,
-      "rewards/margins": 0.005828857421875,
+      "loss": 0.6908,
+      "rewards/accuracies": 0.33125001192092896,
+      "rewards/chosen": -0.001708984375,
+      "rewards/margins": 0.005035400390625,
       "rewards/rejected": -0.006744384765625,
       "step": 20
     },
     {
       "epoch": 0.031413612565445025,
-      "grad_norm": 1.5002050170675545,
+      "grad_norm": 1.5008623344885976,
       "learning_rate": 5.226480836236934e-06,
       "logits/chosen": -11.5,
       "logits/rejected": -11.625,
       "logps/chosen": -422.0,
       "logps/rejected": -466.0,
-      "loss": 0.6876,
-      "rewards/accuracies": 0.4749999940395355,
-      "rewards/chosen": -0.0206298828125,
-      "rewards/margins": 0.00933837890625,
-      "rewards/rejected": -0.030029296875,
+      "loss": 0.6872,
+      "rewards/accuracies": 0.4625000059604645,
+      "rewards/chosen": -0.020263671875,
+      "rewards/margins": 0.00958251953125,
+      "rewards/rejected": -0.02978515625,
       "step": 30
     },
     {
       "epoch": 0.041884816753926704,
-      "grad_norm": 1.8590919500425669,
+      "grad_norm": 1.8572467389108622,
       "learning_rate": 6.968641114982578e-06,
       "logits/chosen": -12.1875,
       "logits/rejected": -11.9375,
       "logps/chosen": -442.0,
       "logps/rejected": -536.0,
-      "loss": 0.6716,
-      "rewards/accuracies": 0.606249988079071,
+      "loss": 0.6718,
+      "rewards/accuracies": 0.59375,
       "rewards/chosen": -0.1240234375,
-      "rewards/margins": 0.05615234375,
+      "rewards/margins": 0.05517578125,
       "rewards/rejected": -0.1796875,
       "step": 40
     },
     {
       "epoch": 0.05235602094240838,
-      "grad_norm": 2.0828530277262693,
+      "grad_norm": 2.089244375860063,
       "learning_rate": 8.710801393728225e-06,
       "logits/chosen": -12.25,
       "logits/rejected": -12.1875,
       "logps/chosen": -450.0,
       "logps/rejected": -592.0,
-      "loss": 0.6404,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -0.3203125,
-      "rewards/margins": 0.16796875,
-      "rewards/rejected": -0.48828125,
+      "loss": 0.6405,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -0.322265625,
+      "rewards/margins": 0.1669921875,
+      "rewards/rejected": -0.490234375,
       "step": 50
     },
     {
       "epoch": 0.06282722513089005,
-      "grad_norm": 2.150163508377863,
+      "grad_norm": 2.152513960510543,
       "learning_rate": 1.0452961672473868e-05,
       "logits/chosen": -11.125,
       "logits/rejected": -10.9375,
       "logps/chosen": -396.0,
       "logps/rejected": -644.0,
-      "loss": 0.603,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -0.443359375,
+      "loss": 0.6028,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.447265625,
       "rewards/margins": 0.369140625,
-      "rewards/rejected": -0.8125,
+      "rewards/rejected": -0.81640625,
       "step": 60
     },
     {
       "epoch": 0.07329842931937172,
-      "grad_norm": 2.259195204037021,
+      "grad_norm": 2.259291225046477,
       "learning_rate": 1.2195121951219513e-05,
       "logits/chosen": -11.5,
       "logits/rejected": -11.3125,
       "logps/chosen": -516.0,
       "logps/rejected": -628.0,
-      "loss": 0.6005,
-      "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -0.81640625,
-      "rewards/margins": 0.279296875,
+      "loss": 0.6012,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.8203125,
+      "rewards/margins": 0.27734375,
       "rewards/rejected": -1.09375,
       "step": 70
     },
     {
       "epoch": 0.08376963350785341,
-      "grad_norm": 2.0904376715861623,
+      "grad_norm": 2.0772617383198098,
       "learning_rate": 1.3937282229965156e-05,
-      "logits/chosen": -11.875,
+      "logits/chosen": -11.9375,
       "logits/rejected": -11.875,
       "logps/chosen": -490.0,
       "logps/rejected": -716.0,
-      "loss": 0.5793,
-      "rewards/accuracies": 0.75,
+      "loss": 0.5787,
+      "rewards/accuracies": 0.737500011920929,
       "rewards/chosen": -0.95703125,
       "rewards/margins": 0.75390625,
       "rewards/rejected": -1.7109375,
@@ -145,4630 +145,4630 @@
     },
     {
       "epoch": 0.09424083769633508,
-      "grad_norm": 2.2647507456401836,
+      "grad_norm": 2.2251238951560084,
       "learning_rate": 1.56794425087108e-05,
-      "logits/chosen": -11.9375,
-      "logits/rejected": -11.9375,
+      "logits/chosen": -12.0,
+      "logits/rejected": -12.0,
       "logps/chosen": -496.0,
       "logps/rejected": -636.0,
-      "loss": 0.565,
-      "rewards/accuracies": 0.643750011920929,
+      "loss": 0.5655,
+      "rewards/accuracies": 0.637499988079071,
       "rewards/chosen": -1.3359375,
-      "rewards/margins": 0.53125,
-      "rewards/rejected": -1.8671875,
+      "rewards/margins": 0.5234375,
+      "rewards/rejected": -1.859375,
       "step": 90
     },
     {
       "epoch": 0.10471204188481675,
-      "grad_norm": 3.2515110482990646,
+      "grad_norm": 2.615215763906515,
       "learning_rate": 1.742160278745645e-05,
-      "logits/chosen": -12.375,
-      "logits/rejected": -12.25,
-      "logps/chosen": -640.0,
-      "logps/rejected": -788.0,
-      "loss": 0.5942,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -1.65625,
+      "logits/chosen": -12.6875,
+      "logits/rejected": -12.5625,
+      "logps/chosen": -636.0,
+      "logps/rejected": -784.0,
+      "loss": 0.5941,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -1.625,
       "rewards/margins": 0.63671875,
-      "rewards/rejected": -2.296875,
+      "rewards/rejected": -2.265625,
       "step": 100
     },
     {
       "epoch": 0.10471204188481675,
-      "eval_logits/chosen": -11.4375,
-      "eval_logits/rejected": -11.125,
-      "eval_logps/chosen": -454.0,
-      "eval_logps/rejected": -446.0,
-      "eval_loss": 0.677984356880188,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -1.359375,
-      "eval_rewards/margins": 0.205078125,
-      "eval_rewards/rejected": -1.5625,
-      "eval_runtime": 54.7843,
-      "eval_samples_per_second": 36.507,
-      "eval_steps_per_second": 0.584,
+      "eval_logits/chosen": -12.3125,
+      "eval_logits/rejected": -12.0625,
+      "eval_logps/chosen": -444.0,
+      "eval_logps/rejected": -434.0,
+      "eval_loss": 0.678640604019165,
+      "eval_rewards/accuracies": 0.572265625,
+      "eval_rewards/chosen": -1.2578125,
+      "eval_rewards/margins": 0.1845703125,
+      "eval_rewards/rejected": -1.4453125,
+      "eval_runtime": 49.1196,
+      "eval_samples_per_second": 40.717,
+      "eval_steps_per_second": 0.651,
       "step": 100
     },
     {
       "epoch": 0.11518324607329843,
-      "grad_norm": 3.044731501422064,
+      "grad_norm": 3.153063933559375,
       "learning_rate": 1.9163763066202093e-05,
-      "logits/chosen": -11.0,
-      "logits/rejected": -10.9375,
+      "logits/chosen": -11.5625,
+      "logits/rejected": -11.4375,
       "logps/chosen": -552.0,
       "logps/rejected": -736.0,
-      "loss": 0.566,
-      "rewards/accuracies": 0.737500011920929,
+      "loss": 0.5682,
+      "rewards/accuracies": 0.7437499761581421,
       "rewards/chosen": -1.3125,
-      "rewards/margins": 0.5859375,
-      "rewards/rejected": -1.8984375,
+      "rewards/margins": 0.59375,
+      "rewards/rejected": -1.90625,
       "step": 110
     },
     {
       "epoch": 0.1256544502617801,
-      "grad_norm": 3.3312722337425034,
+      "grad_norm": 2.787131513028861,
       "learning_rate": 2.0905923344947736e-05,
-      "logits/chosen": -11.5,
-      "logits/rejected": -11.375,
-      "logps/chosen": -528.0,
-      "logps/rejected": -680.0,
-      "loss": 0.5536,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.578125,
-      "rewards/margins": 0.6328125,
-      "rewards/rejected": -2.21875,
+      "logits/chosen": -10.8125,
+      "logits/rejected": -10.75,
+      "logps/chosen": -532.0,
+      "logps/rejected": -676.0,
+      "loss": 0.5648,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -1.6015625,
+      "rewards/margins": 0.5546875,
+      "rewards/rejected": -2.15625,
       "step": 120
     },
     {
       "epoch": 0.13612565445026178,
-      "grad_norm": 2.321069691360544,
+      "grad_norm": 2.1974844483373115,
       "learning_rate": 2.264808362369338e-05,
-      "logits/chosen": -11.875,
-      "logits/rejected": -11.75,
-      "logps/chosen": -600.0,
-      "logps/rejected": -736.0,
-      "loss": 0.6059,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -1.8046875,
-      "rewards/margins": 0.474609375,
-      "rewards/rejected": -2.28125,
+      "logits/chosen": -10.9375,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -564.0,
+      "logps/rejected": -700.0,
+      "loss": 0.6114,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -1.46875,
+      "rewards/margins": 0.4609375,
+      "rewards/rejected": -1.9296875,
       "step": 130
     },
     {
       "epoch": 0.14659685863874344,
-      "grad_norm": 3.07675020541296,
+      "grad_norm": 3.284890279661188,
       "learning_rate": 2.4390243902439026e-05,
-      "logits/chosen": -13.5,
-      "logits/rejected": -13.0,
-      "logps/chosen": -576.0,
-      "logps/rejected": -800.0,
-      "loss": 0.5517,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -1.296875,
-      "rewards/margins": 0.58203125,
-      "rewards/rejected": -1.875,
+      "logits/chosen": -12.875,
+      "logits/rejected": -12.5,
+      "logps/chosen": -588.0,
+      "logps/rejected": -816.0,
+      "loss": 0.553,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -1.4296875,
+      "rewards/margins": 0.60546875,
+      "rewards/rejected": -2.03125,
       "step": 140
     },
     {
       "epoch": 0.15706806282722513,
-      "grad_norm": 2.6654975683704993,
+      "grad_norm": 2.543892875769603,
       "learning_rate": 2.6132404181184672e-05,
-      "logits/chosen": -12.375,
-      "logits/rejected": -12.25,
-      "logps/chosen": -592.0,
-      "logps/rejected": -772.0,
-      "loss": 0.5593,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -1.625,
+      "logits/chosen": -12.75,
+      "logits/rejected": -12.625,
+      "logps/chosen": -576.0,
+      "logps/rejected": -756.0,
+      "loss": 0.5625,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -1.4453125,
       "rewards/margins": 0.71484375,
-      "rewards/rejected": -2.34375,
+      "rewards/rejected": -2.15625,
       "step": 150
     },
     {
       "epoch": 0.16753926701570682,
-      "grad_norm": 2.40614357482195,
+      "grad_norm": 2.276739267731798,
       "learning_rate": 2.7874564459930312e-05,
-      "logits/chosen": -11.4375,
-      "logits/rejected": -11.1875,
-      "logps/chosen": -552.0,
-      "logps/rejected": -728.0,
-      "loss": 0.5387,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -1.5546875,
-      "rewards/margins": 0.66015625,
-      "rewards/rejected": -2.21875,
+      "logits/chosen": -12.25,
+      "logits/rejected": -12.0,
+      "logps/chosen": -560.0,
+      "logps/rejected": -736.0,
+      "loss": 0.5369,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -1.6015625,
+      "rewards/margins": 0.7109375,
+      "rewards/rejected": -2.3125,
       "step": 160
     },
     {
       "epoch": 0.17801047120418848,
-      "grad_norm": 2.4356542812615305,
+      "grad_norm": 2.068097035502852,
       "learning_rate": 2.9616724738675962e-05,
-      "logits/chosen": -8.375,
-      "logits/rejected": -8.4375,
-      "logps/chosen": -680.0,
-      "logps/rejected": -876.0,
-      "loss": 0.5229,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -2.59375,
-      "rewards/margins": 0.83203125,
-      "rewards/rejected": -3.421875,
+      "logits/chosen": -10.8125,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -600.0,
+      "logps/rejected": -780.0,
+      "loss": 0.5147,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -1.765625,
+      "rewards/margins": 0.73046875,
+      "rewards/rejected": -2.5,
       "step": 170
     },
     {
       "epoch": 0.18848167539267016,
-      "grad_norm": 1.9778410054097122,
+      "grad_norm": 1.8624479424447675,
       "learning_rate": 3.13588850174216e-05,
-      "logits/chosen": -8.5,
-      "logits/rejected": -8.3125,
-      "logps/chosen": -760.0,
-      "logps/rejected": -900.0,
-      "loss": 0.5352,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -3.15625,
-      "rewards/margins": 0.8359375,
-      "rewards/rejected": -4.0,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -11.0,
+      "logps/chosen": -632.0,
+      "logps/rejected": -776.0,
+      "loss": 0.5447,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -1.8984375,
+      "rewards/margins": 0.83203125,
+      "rewards/rejected": -2.734375,
       "step": 180
     },
     {
       "epoch": 0.19895287958115182,
-      "grad_norm": 2.2609862761572064,
+      "grad_norm": 2.174695676945664,
       "learning_rate": 3.310104529616725e-05,
-      "logits/chosen": -10.75,
-      "logits/rejected": -10.75,
-      "logps/chosen": -660.0,
-      "logps/rejected": -864.0,
-      "loss": 0.5454,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -2.671875,
-      "rewards/margins": 0.72265625,
-      "rewards/rejected": -3.390625,
+      "logits/chosen": -10.875,
+      "logits/rejected": -10.9375,
+      "logps/chosen": -596.0,
+      "logps/rejected": -800.0,
+      "loss": 0.5448,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -2.015625,
+      "rewards/margins": 0.71484375,
+      "rewards/rejected": -2.71875,
       "step": 190
     },
     {
       "epoch": 0.2094240837696335,
-      "grad_norm": 2.3177692878659264,
+      "grad_norm": 2.2971537846383305,
       "learning_rate": 3.48432055749129e-05,
-      "logits/chosen": -12.25,
-      "logits/rejected": -12.125,
-      "logps/chosen": -660.0,
-      "logps/rejected": -820.0,
-      "loss": 0.561,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -1.953125,
-      "rewards/margins": 0.68359375,
-      "rewards/rejected": -2.640625,
+      "logits/chosen": -12.375,
+      "logits/rejected": -12.25,
+      "logps/chosen": -624.0,
+      "logps/rejected": -792.0,
+      "loss": 0.5458,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -1.609375,
+      "rewards/margins": 0.73828125,
+      "rewards/rejected": -2.34375,
       "step": 200
     },
     {
       "epoch": 0.2094240837696335,
-      "eval_logits/chosen": -10.25,
-      "eval_logits/rejected": -9.75,
-      "eval_logps/chosen": -540.0,
-      "eval_logps/rejected": -540.0,
-      "eval_loss": 0.6944218873977661,
-      "eval_rewards/accuracies": 0.60546875,
-      "eval_rewards/chosen": -2.21875,
-      "eval_rewards/margins": 0.294921875,
-      "eval_rewards/rejected": -2.515625,
-      "eval_runtime": 46.6306,
-      "eval_samples_per_second": 42.89,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -12.125,
+      "eval_logits/rejected": -11.625,
+      "eval_logps/chosen": -478.0,
+      "eval_logps/rejected": -470.0,
+      "eval_loss": 0.7028281092643738,
+      "eval_rewards/accuracies": 0.59765625,
+      "eval_rewards/chosen": -1.6015625,
+      "eval_rewards/margins": 0.212890625,
+      "eval_rewards/rejected": -1.8203125,
+      "eval_runtime": 47.0772,
+      "eval_samples_per_second": 42.483,
+      "eval_steps_per_second": 0.68,
       "step": 200
     },
     {
       "epoch": 0.2198952879581152,
-      "grad_norm": 2.9921467716825108,
+      "grad_norm": 2.772674624080697,
       "learning_rate": 3.6585365853658535e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.8125,
-      "logps/chosen": -624.0,
-      "logps/rejected": -748.0,
-      "loss": 0.5526,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -2.5625,
-      "rewards/margins": 0.56640625,
-      "rewards/rejected": -3.140625,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -11.125,
+      "logps/chosen": -556.0,
+      "logps/rejected": -684.0,
+      "loss": 0.5538,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -1.875,
+      "rewards/margins": 0.609375,
+      "rewards/rejected": -2.484375,
       "step": 210
     },
     {
       "epoch": 0.23036649214659685,
-      "grad_norm": 2.534227548628948,
+      "grad_norm": 2.5204961669920465,
       "learning_rate": 3.8327526132404185e-05,
-      "logits/chosen": -13.1875,
-      "logits/rejected": -12.75,
-      "logps/chosen": -636.0,
-      "logps/rejected": -776.0,
-      "loss": 0.5526,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -2.15625,
-      "rewards/margins": 0.76171875,
-      "rewards/rejected": -2.921875,
+      "logits/chosen": -11.1875,
+      "logits/rejected": -10.8125,
+      "logps/chosen": -612.0,
+      "logps/rejected": -748.0,
+      "loss": 0.5447,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -1.9140625,
+      "rewards/margins": 0.73828125,
+      "rewards/rejected": -2.65625,
       "step": 220
     },
     {
       "epoch": 0.24083769633507854,
-      "grad_norm": 2.538407636798547,
+      "grad_norm": 2.4297067855291195,
       "learning_rate": 4.006968641114983e-05,
-      "logits/chosen": -12.3125,
-      "logits/rejected": -12.125,
-      "logps/chosen": -696.0,
-      "logps/rejected": -856.0,
-      "loss": 0.5366,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -3.078125,
-      "rewards/margins": 0.6875,
-      "rewards/rejected": -3.765625,
+      "logits/chosen": -12.0625,
+      "logits/rejected": -11.8125,
+      "logps/chosen": -576.0,
+      "logps/rejected": -736.0,
+      "loss": 0.529,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -1.9140625,
+      "rewards/margins": 0.66796875,
+      "rewards/rejected": -2.578125,
       "step": 230
     },
     {
       "epoch": 0.2513089005235602,
-      "grad_norm": 2.7177700431099385,
+      "grad_norm": 2.672803519197462,
       "learning_rate": 4.181184668989547e-05,
-      "logits/chosen": -11.9375,
-      "logits/rejected": -11.6875,
-      "logps/chosen": -720.0,
-      "logps/rejected": -872.0,
-      "loss": 0.5854,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -3.453125,
-      "rewards/margins": 0.7890625,
-      "rewards/rejected": -4.25,
+      "logits/chosen": -12.25,
+      "logits/rejected": -11.875,
+      "logps/chosen": -672.0,
+      "logps/rejected": -832.0,
+      "loss": 0.5643,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -2.984375,
+      "rewards/margins": 0.83203125,
+      "rewards/rejected": -3.8125,
       "step": 240
     },
     {
       "epoch": 0.2617801047120419,
-      "grad_norm": 3.3750564834129504,
+      "grad_norm": 4.469305847162535,
       "learning_rate": 4.3554006968641115e-05,
-      "logits/chosen": -12.375,
-      "logits/rejected": -12.0625,
-      "logps/chosen": -704.0,
-      "logps/rejected": -868.0,
-      "loss": 0.5305,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -2.484375,
-      "rewards/margins": 0.85546875,
-      "rewards/rejected": -3.34375,
+      "logits/chosen": -11.75,
+      "logits/rejected": -11.5,
+      "logps/chosen": -864.0,
+      "logps/rejected": -1040.0,
+      "loss": 0.524,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -4.0625,
+      "rewards/margins": 1.0,
+      "rewards/rejected": -5.0625,
       "step": 250
     },
     {
       "epoch": 0.27225130890052357,
-      "grad_norm": 2.3187384311598565,
+      "grad_norm": 2.039361428872278,
       "learning_rate": 4.529616724738676e-05,
-      "logits/chosen": -10.4375,
-      "logits/rejected": -10.125,
-      "logps/chosen": -768.0,
-      "logps/rejected": -1020.0,
-      "loss": 0.5257,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -3.90625,
-      "rewards/margins": 1.1953125,
-      "rewards/rejected": -5.09375,
+      "logits/chosen": -14.0,
+      "logits/rejected": -13.4375,
+      "logps/chosen": -632.0,
+      "logps/rejected": -864.0,
+      "loss": 0.5188,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -2.5625,
+      "rewards/margins": 0.9609375,
+      "rewards/rejected": -3.515625,
       "step": 260
     },
     {
       "epoch": 0.28272251308900526,
-      "grad_norm": 1.959143521875051,
+      "grad_norm": 3.0188224001209534,
       "learning_rate": 4.703832752613241e-05,
-      "logits/chosen": -11.75,
-      "logits/rejected": -11.5625,
-      "logps/chosen": -732.0,
-      "logps/rejected": -888.0,
-      "loss": 0.548,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -3.515625,
-      "rewards/margins": 0.83203125,
-      "rewards/rejected": -4.34375,
+      "logits/chosen": -11.375,
+      "logits/rejected": -11.375,
+      "logps/chosen": -664.0,
+      "logps/rejected": -836.0,
+      "loss": 0.5458,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -2.828125,
+      "rewards/margins": 1.0234375,
+      "rewards/rejected": -3.84375,
       "step": 270
     },
     {
       "epoch": 0.2931937172774869,
-      "grad_norm": 2.548392657608926,
+      "grad_norm": 2.4087568400909234,
       "learning_rate": 4.878048780487805e-05,
-      "logits/chosen": -12.0,
-      "logits/rejected": -11.625,
-      "logps/chosen": -632.0,
-      "logps/rejected": -800.0,
-      "loss": 0.5448,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -2.640625,
-      "rewards/margins": 0.69921875,
-      "rewards/rejected": -3.328125,
+      "logits/chosen": -13.625,
+      "logits/rejected": -13.25,
+      "logps/chosen": -596.0,
+      "logps/rejected": -756.0,
+      "loss": 0.5712,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -2.296875,
+      "rewards/margins": 0.59375,
+      "rewards/rejected": -2.890625,
       "step": 280
     },
     {
       "epoch": 0.3036649214659686,
-      "grad_norm": 2.620820857416364,
+      "grad_norm": 3.4594251103083082,
       "learning_rate": 4.999983293481418e-05,
-      "logits/chosen": -14.375,
-      "logits/rejected": -14.25,
-      "logps/chosen": -708.0,
-      "logps/rejected": -920.0,
-      "loss": 0.5595,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -3.125,
-      "rewards/margins": 0.9140625,
-      "rewards/rejected": -4.03125,
+      "logits/chosen": -13.0625,
+      "logits/rejected": -12.8125,
+      "logps/chosen": -728.0,
+      "logps/rejected": -940.0,
+      "loss": 0.5368,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -3.328125,
+      "rewards/margins": 0.94140625,
+      "rewards/rejected": -4.28125,
       "step": 290
     },
     {
       "epoch": 0.31413612565445026,
-      "grad_norm": 2.4811851838998034,
+      "grad_norm": 2.778915469156192,
       "learning_rate": 4.9996862949181595e-05,
-      "logits/chosen": -9.875,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -840.0,
-      "logps/rejected": -992.0,
-      "loss": 0.5506,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -4.28125,
-      "rewards/margins": 0.6953125,
-      "rewards/rejected": -5.0,
+      "logits/chosen": -10.1875,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -780.0,
+      "logps/rejected": -968.0,
+      "loss": 0.5451,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 1.0546875,
+      "rewards/rejected": -4.75,
       "step": 300
     },
     {
       "epoch": 0.31413612565445026,
-      "eval_logits/chosen": -8.3125,
-      "eval_logits/rejected": -7.5,
-      "eval_logps/chosen": -764.0,
-      "eval_logps/rejected": -772.0,
-      "eval_loss": 0.7181406021118164,
-      "eval_rewards/accuracies": 0.607421875,
-      "eval_rewards/chosen": -4.4375,
-      "eval_rewards/margins": 0.400390625,
-      "eval_rewards/rejected": -4.84375,
-      "eval_runtime": 46.6454,
-      "eval_samples_per_second": 42.877,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -9.75,
+      "eval_logits/rejected": -9.125,
+      "eval_logps/chosen": -668.0,
+      "eval_logps/rejected": -672.0,
+      "eval_loss": 0.7219062447547913,
+      "eval_rewards/accuracies": 0.6015625,
+      "eval_rewards/chosen": -3.5,
+      "eval_rewards/margins": 0.3515625,
+      "eval_rewards/rejected": -3.84375,
+      "eval_runtime": 47.0913,
+      "eval_samples_per_second": 42.471,
+      "eval_steps_per_second": 0.68,
       "step": 300
     },
     {
       "epoch": 0.32460732984293195,
-      "grad_norm": 2.728167916044458,
+      "grad_norm": 2.5080958046825255,
       "learning_rate": 4.9990180911526896e-05,
-      "logits/chosen": -10.5,
-      "logits/rejected": -10.0625,
-      "logps/chosen": -680.0,
-      "logps/rejected": -844.0,
-      "loss": 0.5617,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -3.046875,
-      "rewards/margins": 0.53515625,
-      "rewards/rejected": -3.59375,
+      "logits/chosen": -10.625,
+      "logits/rejected": -10.3125,
+      "logps/chosen": -644.0,
+      "logps/rejected": -812.0,
+      "loss": 0.5709,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -2.703125,
+      "rewards/margins": 0.578125,
+      "rewards/rejected": -3.28125,
       "step": 310
     },
     {
       "epoch": 0.33507853403141363,
-      "grad_norm": 3.0150065333219325,
+      "grad_norm": 2.878935660965702,
       "learning_rate": 4.997978781413745e-05,
-      "logits/chosen": -9.3125,
-      "logits/rejected": -9.125,
-      "logps/chosen": -888.0,
-      "logps/rejected": -1064.0,
-      "loss": 0.5592,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -4.4375,
-      "rewards/margins": 1.09375,
-      "rewards/rejected": -5.53125,
+      "logits/chosen": -10.375,
+      "logits/rejected": -10.125,
+      "logps/chosen": -684.0,
+      "logps/rejected": -840.0,
+      "loss": 0.607,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -2.40625,
+      "rewards/margins": 0.8359375,
+      "rewards/rejected": -3.234375,
       "step": 320
     },
     {
       "epoch": 0.34554973821989526,
-      "grad_norm": 2.5804615362734467,
+      "grad_norm": 2.95715712725854,
       "learning_rate": 4.996568520039565e-05,
-      "logits/chosen": -10.1875,
+      "logits/chosen": -10.125,
       "logits/rejected": -10.125,
-      "logps/chosen": -740.0,
-      "logps/rejected": -948.0,
-      "loss": 0.5393,
-      "rewards/accuracies": 0.65625,
-      "rewards/chosen": -3.234375,
-      "rewards/margins": 0.94140625,
-      "rewards/rejected": -4.1875,
+      "logps/chosen": -696.0,
+      "logps/rejected": -904.0,
+      "loss": 0.5341,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -2.796875,
+      "rewards/margins": 0.93359375,
+      "rewards/rejected": -3.734375,
       "step": 330
     },
     {
       "epoch": 0.35602094240837695,
-      "grad_norm": 2.9949010065088353,
+      "grad_norm": 2.879146775985956,
       "learning_rate": 4.994787516454973e-05,
-      "logits/chosen": -11.0,
-      "logits/rejected": -10.5,
-      "logps/chosen": -716.0,
-      "logps/rejected": -872.0,
-      "loss": 0.525,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -3.125,
-      "rewards/margins": 0.87890625,
-      "rewards/rejected": -4.0,
+      "logits/chosen": -9.375,
+      "logits/rejected": -9.125,
+      "logps/chosen": -776.0,
+      "logps/rejected": -944.0,
+      "loss": 0.5361,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -3.71875,
+      "rewards/margins": 1.0234375,
+      "rewards/rejected": -4.75,
       "step": 340
     },
     {
       "epoch": 0.36649214659685864,
-      "grad_norm": 3.6024122625196804,
+      "grad_norm": 4.055740125273308,
       "learning_rate": 4.992636035140273e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -700.0,
-      "logps/rejected": -928.0,
-      "loss": 0.534,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -3.28125,
-      "rewards/margins": 1.0,
-      "rewards/rejected": -4.28125,
+      "logits/chosen": -10.125,
+      "logits/rejected": -10.0,
+      "logps/chosen": -648.0,
+      "logps/rejected": -872.0,
+      "loss": 0.531,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": -2.765625,
+      "rewards/margins": 0.953125,
+      "rewards/rejected": -3.71875,
       "step": 350
     },
     {
       "epoch": 0.3769633507853403,
-      "grad_norm": 2.650227753877416,
+      "grad_norm": 2.759892635957954,
       "learning_rate": 4.99011439559198e-05,
-      "logits/chosen": -12.1875,
-      "logits/rejected": -11.625,
-      "logps/chosen": -780.0,
-      "logps/rejected": -1032.0,
-      "loss": 0.5353,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -3.71875,
-      "rewards/margins": 1.1640625,
-      "rewards/rejected": -4.875,
+      "logits/chosen": -10.25,
+      "logits/rejected": -9.9375,
+      "logps/chosen": -692.0,
+      "logps/rejected": -936.0,
+      "loss": 0.5453,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -2.84375,
+      "rewards/margins": 1.0859375,
+      "rewards/rejected": -3.9375,
       "step": 360
     },
     {
       "epoch": 0.387434554973822,
-      "grad_norm": 2.4010085612350345,
+      "grad_norm": 2.607360224332401,
       "learning_rate": 4.987222972275369e-05,
-      "logits/chosen": -8.5625,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -908.0,
-      "logps/rejected": -1096.0,
-      "loss": 0.5455,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -4.96875,
-      "rewards/margins": 1.03125,
-      "rewards/rejected": -6.0,
+      "logits/chosen": -10.875,
+      "logits/rejected": -10.4375,
+      "logps/chosen": -648.0,
+      "logps/rejected": -824.0,
+      "loss": 0.5336,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -2.375,
+      "rewards/margins": 0.96484375,
+      "rewards/rejected": -3.34375,
       "step": 370
     },
     {
       "epoch": 0.39790575916230364,
-      "grad_norm": 2.034008157585721,
+      "grad_norm": 2.2638929515665374,
       "learning_rate": 4.983962194568872e-05,
-      "logits/chosen": -9.9375,
-      "logits/rejected": -9.5,
-      "logps/chosen": -864.0,
-      "logps/rejected": -1128.0,
-      "loss": 0.5344,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -4.4375,
-      "rewards/margins": 1.078125,
-      "rewards/rejected": -5.53125,
+      "logits/chosen": -10.625,
+      "logits/rejected": -10.375,
+      "logps/chosen": -732.0,
+      "logps/rejected": -1004.0,
+      "loss": 0.5456,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -3.125,
+      "rewards/margins": 1.125,
+      "rewards/rejected": -4.25,
       "step": 380
     },
     {
       "epoch": 0.4083769633507853,
-      "grad_norm": 2.4174156914986704,
+      "grad_norm": 2.4936922601973746,
       "learning_rate": 4.9803325467003084e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -832.0,
-      "logps/rejected": -1088.0,
-      "loss": 0.5098,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -4.03125,
-      "rewards/margins": 1.171875,
-      "rewards/rejected": -5.21875,
+      "logits/chosen": -10.375,
+      "logits/rejected": -10.0625,
+      "logps/chosen": -692.0,
+      "logps/rejected": -952.0,
+      "loss": 0.5036,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -2.625,
+      "rewards/margins": 1.2421875,
+      "rewards/rejected": -3.875,
       "step": 390
     },
     {
       "epoch": 0.418848167539267,
-      "grad_norm": 3.3476915694916918,
+      "grad_norm": 3.3684547528014033,
       "learning_rate": 4.9763345676749816e-05,
-      "logits/chosen": -10.25,
-      "logits/rejected": -9.875,
-      "logps/chosen": -928.0,
-      "logps/rejected": -1208.0,
-      "loss": 0.5233,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -5.25,
-      "rewards/margins": 1.4453125,
-      "rewards/rejected": -6.6875,
+      "logits/chosen": -8.3125,
+      "logits/rejected": -8.0625,
+      "logps/chosen": -696.0,
+      "logps/rejected": -980.0,
+      "loss": 0.5221,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -2.9375,
+      "rewards/margins": 1.453125,
+      "rewards/rejected": -4.375,
       "step": 400
     },
     {
       "epoch": 0.418848167539267,
-      "eval_logits/chosen": -13.625,
-      "eval_logits/rejected": -12.875,
-      "eval_logps/chosen": -720.0,
-      "eval_logps/rejected": -736.0,
-      "eval_loss": 0.8159219026565552,
-      "eval_rewards/accuracies": 0.583984375,
-      "eval_rewards/chosen": -4.03125,
-      "eval_rewards/margins": 0.466796875,
-      "eval_rewards/rejected": -4.5,
-      "eval_runtime": 46.6586,
-      "eval_samples_per_second": 42.865,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -6.375,
+      "eval_logits/rejected": -5.28125,
+      "eval_logps/chosen": -640.0,
+      "eval_logps/rejected": -664.0,
+      "eval_loss": 0.8154218792915344,
+      "eval_rewards/accuracies": 0.5859375,
+      "eval_rewards/chosen": -3.21875,
+      "eval_rewards/margins": 0.54296875,
+      "eval_rewards/rejected": -3.765625,
+      "eval_runtime": 47.0669,
+      "eval_samples_per_second": 42.493,
+      "eval_steps_per_second": 0.68,
       "step": 400
     },
     {
       "epoch": 0.4293193717277487,
-      "grad_norm": 2.586399973896129,
+      "grad_norm": 3.518407195448941,
       "learning_rate": 4.9719688511956374e-05,
-      "logits/chosen": -13.9375,
-      "logits/rejected": -13.375,
-      "logps/chosen": -816.0,
-      "logps/rejected": -1088.0,
-      "loss": 0.5349,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -3.921875,
-      "rewards/margins": 1.3203125,
-      "rewards/rejected": -5.25,
-      "step": 410
-    },
+      "logits/chosen": -6.625,
+      "logits/rejected": -6.3125,
+      "logps/chosen": -808.0,
+      "logps/rejected": -1112.0,
+      "loss": 0.5043,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -3.859375,
+      "rewards/margins": 1.578125,
+      "rewards/rejected": -5.4375,
+      "step": 410
+    },
     {
       "epoch": 0.4397905759162304,
-      "grad_norm": 2.9277764977659055,
+      "grad_norm": 2.5281940574743085,
       "learning_rate": 4.967236045574294e-05,
-      "logits/chosen": -10.6875,
-      "logits/rejected": -10.0,
-      "logps/chosen": -888.0,
-      "logps/rejected": -1104.0,
-      "loss": 0.5166,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -4.75,
-      "rewards/margins": 1.1875,
-      "rewards/rejected": -5.9375,
+      "logits/chosen": -7.5625,
+      "logits/rejected": -7.125,
+      "logps/chosen": -696.0,
+      "logps/rejected": -908.0,
+      "loss": 0.5167,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -2.828125,
+      "rewards/margins": 1.171875,
+      "rewards/rejected": -4.0,
       "step": 420
     },
     {
       "epoch": 0.450261780104712,
-      "grad_norm": 2.1729239398577644,
+      "grad_norm": 2.3552559450669794,
       "learning_rate": 4.96213685363597e-05,
-      "logits/chosen": -9.8125,
+      "logits/chosen": -9.5,
       "logits/rejected": -9.4375,
-      "logps/chosen": -856.0,
-      "logps/rejected": -1016.0,
-      "loss": 0.5763,
+      "logps/chosen": -644.0,
+      "logps/rejected": -808.0,
+      "loss": 0.5317,
       "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -4.53125,
-      "rewards/margins": 0.86328125,
-      "rewards/rejected": -5.40625,
+      "rewards/chosen": -2.40625,
+      "rewards/margins": 0.9140625,
+      "rewards/rejected": -3.3125,
       "step": 430
     },
     {
       "epoch": 0.4607329842931937,
-      "grad_norm": 2.7623880256483684,
+      "grad_norm": 3.7503712248895305,
       "learning_rate": 4.956672032614315e-05,
-      "logits/chosen": -12.0,
-      "logits/rejected": -11.625,
-      "logps/chosen": -760.0,
-      "logps/rejected": -936.0,
-      "loss": 0.5593,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -3.4375,
-      "rewards/margins": 0.7578125,
-      "rewards/rejected": -4.1875,
+      "logits/chosen": -8.25,
+      "logits/rejected": -8.375,
+      "logps/chosen": -720.0,
+      "logps/rejected": -916.0,
+      "loss": 0.5139,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -3.0625,
+      "rewards/margins": 0.94140625,
+      "rewards/rejected": -4.0,
       "step": 440
     },
     {
       "epoch": 0.4712041884816754,
-      "grad_norm": 2.8537544022057633,
+      "grad_norm": 2.2733254933065936,
       "learning_rate": 4.950842394039157e-05,
-      "logits/chosen": -11.8125,
-      "logits/rejected": -11.3125,
-      "logps/chosen": -904.0,
-      "logps/rejected": -1120.0,
-      "loss": 0.5358,
+      "logits/chosen": -7.84375,
+      "logits/rejected": -7.96875,
+      "logps/chosen": -752.0,
+      "logps/rejected": -952.0,
+      "loss": 0.5338,
       "rewards/accuracies": 0.6875,
-      "rewards/chosen": -5.1875,
-      "rewards/margins": 1.171875,
-      "rewards/rejected": -6.34375,
+      "rewards/chosen": -3.640625,
+      "rewards/margins": 1.0546875,
+      "rewards/rejected": -4.6875,
       "step": 450
     },
     {
       "epoch": 0.4816753926701571,
-      "grad_norm": 2.491837555416849,
+      "grad_norm": 2.1365975408243094,
       "learning_rate": 4.944648803615994e-05,
-      "logits/chosen": -13.8125,
-      "logits/rejected": -13.1875,
-      "logps/chosen": -820.0,
-      "logps/rejected": -1016.0,
-      "loss": 0.5427,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -4.03125,
-      "rewards/margins": 1.1796875,
-      "rewards/rejected": -5.21875,
+      "logits/chosen": -8.125,
+      "logits/rejected": -7.5,
+      "logps/chosen": -736.0,
+      "logps/rejected": -924.0,
+      "loss": 0.5372,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -3.1875,
+      "rewards/margins": 1.1015625,
+      "rewards/rejected": -4.28125,
       "step": 460
     },
     {
       "epoch": 0.49214659685863876,
-      "grad_norm": 3.1316085810060597,
+      "grad_norm": 3.0159579075617797,
       "learning_rate": 4.938092181097433e-05,
-      "logits/chosen": -14.25,
-      "logits/rejected": -13.9375,
-      "logps/chosen": -800.0,
-      "logps/rejected": -976.0,
-      "loss": 0.4985,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -3.984375,
-      "rewards/margins": 1.0,
-      "rewards/rejected": -5.0,
+      "logits/chosen": -7.84375,
+      "logits/rejected": -7.5,
+      "logps/chosen": -756.0,
+      "logps/rejected": -944.0,
+      "loss": 0.4818,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -3.5625,
+      "rewards/margins": 1.1171875,
+      "rewards/rejected": -4.6875,
       "step": 470
     },
     {
       "epoch": 0.5026178010471204,
-      "grad_norm": 4.236106193310983,
+      "grad_norm": 4.635367261254427,
       "learning_rate": 4.931173500146607e-05,
-      "logits/chosen": -13.25,
-      "logits/rejected": -12.75,
-      "logps/chosen": -868.0,
-      "logps/rejected": -1144.0,
-      "loss": 0.5287,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -4.8125,
-      "rewards/margins": 1.4765625,
-      "rewards/rejected": -6.28125,
+      "logits/chosen": -5.96875,
+      "logits/rejected": -6.03125,
+      "logps/chosen": -916.0,
+      "logps/rejected": -1208.0,
+      "loss": 0.5162,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -5.28125,
+      "rewards/margins": 1.6796875,
+      "rewards/rejected": -6.96875,
       "step": 480
     },
     {
       "epoch": 0.5130890052356021,
-      "grad_norm": 2.644398046825569,
+      "grad_norm": 2.6383752601276527,
       "learning_rate": 4.923893788192582e-05,
-      "logits/chosen": -13.75,
-      "logits/rejected": -13.25,
-      "logps/chosen": -760.0,
-      "logps/rejected": -928.0,
-      "loss": 0.5147,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -3.375,
-      "rewards/margins": 1.015625,
-      "rewards/rejected": -4.40625,
+      "logits/chosen": -8.875,
+      "logits/rejected": -8.5,
+      "logps/chosen": -752.0,
+      "logps/rejected": -936.0,
+      "loss": 0.503,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -3.3125,
+      "rewards/margins": 1.15625,
+      "rewards/rejected": -4.46875,
       "step": 490
     },
     {
       "epoch": 0.5235602094240838,
-      "grad_norm": 4.059056197201719,
+      "grad_norm": 2.877733310489956,
       "learning_rate": 4.9162541262777906e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -780.0,
-      "logps/rejected": -984.0,
-      "loss": 0.5164,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -3.578125,
-      "rewards/margins": 0.9609375,
-      "rewards/rejected": -4.53125,
+      "logits/chosen": -9.4375,
+      "logits/rejected": -9.25,
+      "logps/chosen": -712.0,
+      "logps/rejected": -920.0,
+      "loss": 0.5128,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -2.890625,
+      "rewards/margins": 1.0,
+      "rewards/rejected": -3.890625,
       "step": 500
     },
     {
       "epoch": 0.5235602094240838,
-      "eval_logits/chosen": -8.3125,
-      "eval_logits/rejected": -6.65625,
-      "eval_logps/chosen": -636.0,
-      "eval_logps/rejected": -672.0,
-      "eval_loss": 0.7724218964576721,
-      "eval_rewards/accuracies": 0.626953125,
-      "eval_rewards/chosen": -3.15625,
-      "eval_rewards/margins": 0.66015625,
-      "eval_rewards/rejected": -3.828125,
-      "eval_runtime": 46.6427,
-      "eval_samples_per_second": 42.879,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -6.59375,
+      "eval_logits/rejected": -5.53125,
+      "eval_logps/chosen": -652.0,
+      "eval_logps/rejected": -680.0,
+      "eval_loss": 0.7688124775886536,
+      "eval_rewards/accuracies": 0.609375,
+      "eval_rewards/chosen": -3.328125,
+      "eval_rewards/margins": 0.5859375,
+      "eval_rewards/rejected": -3.921875,
+      "eval_runtime": 47.0697,
+      "eval_samples_per_second": 42.49,
+      "eval_steps_per_second": 0.68,
       "step": 500
     },
     {
       "epoch": 0.5340314136125655,
-      "grad_norm": 2.662031273800223,
+      "grad_norm": 2.580765266923729,
       "learning_rate": 4.90825564889749e-05,
-      "logits/chosen": -11.3125,
-      "logits/rejected": -10.6875,
-      "logps/chosen": -712.0,
-      "logps/rejected": -888.0,
-      "loss": 0.5222,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -2.75,
-      "rewards/margins": 0.9765625,
-      "rewards/rejected": -3.71875,
+      "logits/chosen": -8.75,
+      "logits/rejected": -8.8125,
+      "logps/chosen": -740.0,
+      "logps/rejected": -928.0,
+      "loss": 0.5027,
+      "rewards/accuracies": 0.8187500238418579,
+      "rewards/chosen": -3.015625,
+      "rewards/margins": 1.078125,
+      "rewards/rejected": -4.09375,
       "step": 510
     },
     {
       "epoch": 0.5445026178010471,
-      "grad_norm": 3.226497544027977,
+      "grad_norm": 3.365819226866314,
       "learning_rate": 4.899899543831288e-05,
-      "logits/chosen": -9.875,
-      "logits/rejected": -8.875,
-      "logps/chosen": -804.0,
-      "logps/rejected": -1048.0,
-      "loss": 0.4952,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -3.609375,
-      "rewards/margins": 1.46875,
-      "rewards/rejected": -5.0625,
+      "logits/chosen": -8.8125,
+      "logits/rejected": -8.25,
+      "logps/chosen": -852.0,
+      "logps/rejected": -1080.0,
+      "loss": 0.4866,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -4.0625,
+      "rewards/margins": 1.3203125,
+      "rewards/rejected": -5.375,
       "step": 520
     },
     {
       "epoch": 0.5549738219895288,
-      "grad_norm": 3.154187459868018,
+      "grad_norm": 2.984255334202947,
       "learning_rate": 4.891187051966765e-05,
-      "logits/chosen": -9.125,
-      "logits/rejected": -8.5625,
-      "logps/chosen": -812.0,
-      "logps/rejected": -940.0,
-      "loss": 0.5335,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -3.828125,
-      "rewards/margins": 0.8125,
-      "rewards/rejected": -4.625,
+      "logits/chosen": -9.375,
+      "logits/rejected": -9.0625,
+      "logps/chosen": -780.0,
+      "logps/rejected": -920.0,
+      "loss": 0.5005,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -3.484375,
+      "rewards/margins": 0.94140625,
+      "rewards/rejected": -4.4375,
       "step": 530
     },
     {
       "epoch": 0.5654450261780105,
-      "grad_norm": 3.170157111167069,
+      "grad_norm": 3.1258462704618744,
       "learning_rate": 4.882119467115191e-05,
-      "logits/chosen": -10.0,
-      "logits/rejected": -9.625,
-      "logps/chosen": -748.0,
-      "logps/rejected": -972.0,
-      "loss": 0.5404,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -3.453125,
+      "logits/chosen": -9.3125,
+      "logits/rejected": -9.1875,
+      "logps/chosen": -752.0,
+      "logps/rejected": -976.0,
+      "loss": 0.5339,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -3.53125,
       "rewards/margins": 1.0078125,
-      "rewards/rejected": -4.46875,
+      "rewards/rejected": -4.53125,
       "step": 540
     },
     {
       "epoch": 0.5759162303664922,
-      "grad_norm": 2.634840823127897,
+      "grad_norm": 2.8902618505091806,
       "learning_rate": 4.872698135819402e-05,
-      "logits/chosen": -10.125,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -812.0,
-      "logps/rejected": -1048.0,
-      "loss": 0.5301,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -3.96875,
-      "rewards/margins": 1.03125,
-      "rewards/rejected": -5.0,
+      "logits/chosen": -9.25,
+      "logits/rejected": -8.875,
+      "logps/chosen": -772.0,
+      "logps/rejected": -1012.0,
+      "loss": 0.525,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -3.578125,
+      "rewards/margins": 1.0703125,
+      "rewards/rejected": -4.65625,
       "step": 550
     },
     {
       "epoch": 0.5863874345549738,
-      "grad_norm": 2.8169975227892716,
+      "grad_norm": 2.741386327878407,
       "learning_rate": 4.862924457153831e-05,
-      "logits/chosen": -9.0625,
+      "logits/chosen": -8.9375,
       "logits/rejected": -8.625,
-      "logps/chosen": -784.0,
-      "logps/rejected": -1032.0,
-      "loss": 0.4821,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -4.40625,
-      "rewards/margins": 1.1796875,
-      "rewards/rejected": -5.59375,
+      "logps/chosen": -616.0,
+      "logps/rejected": -860.0,
+      "loss": 0.4982,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -2.734375,
+      "rewards/margins": 1.1171875,
+      "rewards/rejected": -3.84375,
       "step": 560
     },
     {
       "epoch": 0.5968586387434555,
-      "grad_norm": 2.404754553912183,
+      "grad_norm": 2.386529847235471,
       "learning_rate": 4.8527998825167484e-05,
-      "logits/chosen": -7.8125,
-      "logits/rejected": -7.6875,
-      "logps/chosen": -924.0,
-      "logps/rejected": -1144.0,
-      "loss": 0.4674,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -5.375,
-      "rewards/margins": 1.421875,
-      "rewards/rejected": -6.8125,
+      "logits/chosen": -7.125,
+      "logits/rejected": -7.28125,
+      "logps/chosen": -748.0,
+      "logps/rejected": -984.0,
+      "loss": 0.4914,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -3.640625,
+      "rewards/margins": 1.5703125,
+      "rewards/rejected": -5.21875,
       "step": 570
     },
     {
       "epoch": 0.6073298429319371,
-      "grad_norm": 2.496733491642304,
+      "grad_norm": 2.4132965487300666,
       "learning_rate": 4.842325915414728e-05,
-      "logits/chosen": -10.0,
-      "logits/rejected": -9.375,
-      "logps/chosen": -832.0,
-      "logps/rejected": -1112.0,
-      "loss": 0.5142,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -4.5625,
-      "rewards/margins": 1.6484375,
-      "rewards/rejected": -6.21875,
+      "logits/chosen": -7.21875,
+      "logits/rejected": -7.15625,
+      "logps/chosen": -740.0,
+      "logps/rejected": -1016.0,
+      "loss": 0.4865,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -3.640625,
+      "rewards/margins": 1.578125,
+      "rewards/rejected": -5.21875,
       "step": 580
     },
     {
       "epoch": 0.6178010471204188,
-      "grad_norm": 3.282918671224393,
+      "grad_norm": 3.1705623322741507,
       "learning_rate": 4.831504111239373e-05,
-      "logits/chosen": -10.625,
-      "logits/rejected": -9.75,
-      "logps/chosen": -908.0,
-      "logps/rejected": -1208.0,
-      "loss": 0.4821,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -4.875,
-      "rewards/margins": 1.75,
-      "rewards/rejected": -6.625,
+      "logits/chosen": -7.65625,
+      "logits/rejected": -7.5,
+      "logps/chosen": -856.0,
+      "logps/rejected": -1120.0,
+      "loss": 0.4857,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -4.375,
+      "rewards/margins": 1.375,
+      "rewards/rejected": -5.75,
       "step": 590
     },
     {
       "epoch": 0.6282722513089005,
-      "grad_norm": 2.4131390450477173,
+      "grad_norm": 3.0899697624274753,
       "learning_rate": 4.8203360770363425e-05,
-      "logits/chosen": -7.65625,
+      "logits/chosen": -7.5625,
       "logits/rejected": -7.25,
-      "logps/chosen": -1056.0,
-      "logps/rejected": -1320.0,
-      "loss": 0.5142,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -6.34375,
-      "rewards/margins": 1.7734375,
-      "rewards/rejected": -8.125,
+      "logps/chosen": -808.0,
+      "logps/rejected": -1040.0,
+      "loss": 0.5178,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -3.859375,
+      "rewards/margins": 1.421875,
+      "rewards/rejected": -5.28125,
       "step": 600
     },
     {
       "epoch": 0.6282722513089005,
-      "eval_logits/chosen": -7.40625,
-      "eval_logits/rejected": -6.4375,
-      "eval_logps/chosen": -852.0,
-      "eval_logps/rejected": -860.0,
-      "eval_loss": 0.7633437514305115,
-      "eval_rewards/accuracies": 0.564453125,
-      "eval_rewards/chosen": -5.34375,
-      "eval_rewards/margins": 0.376953125,
-      "eval_rewards/rejected": -5.71875,
-      "eval_runtime": 46.6524,
-      "eval_samples_per_second": 42.87,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -6.6875,
+      "eval_logits/rejected": -5.59375,
+      "eval_logps/chosen": -704.0,
+      "eval_logps/rejected": -724.0,
+      "eval_loss": 0.7540000081062317,
+      "eval_rewards/accuracies": 0.58203125,
+      "eval_rewards/chosen": -3.859375,
+      "eval_rewards/margins": 0.490234375,
+      "eval_rewards/rejected": -4.34375,
+      "eval_runtime": 47.088,
+      "eval_samples_per_second": 42.474,
+      "eval_steps_per_second": 0.68,
       "step": 600
     },
     {
       "epoch": 0.6387434554973822,
-      "grad_norm": 2.8929559033137595,
+      "grad_norm": 4.688425563712276,
       "learning_rate": 4.808823471266702e-05,
-      "logits/chosen": -9.4375,
-      "logits/rejected": -9.25,
-      "logps/chosen": -864.0,
-      "logps/rejected": -1080.0,
-      "loss": 0.5153,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -4.59375,
-      "rewards/margins": 1.171875,
-      "rewards/rejected": -5.78125,
+      "logits/chosen": -7.9375,
+      "logits/rejected": -7.65625,
+      "logps/chosen": -772.0,
+      "logps/rejected": -1004.0,
+      "loss": 0.5154,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -3.6875,
+      "rewards/margins": 1.328125,
+      "rewards/rejected": -5.03125,
       "step": 610
     },
     {
       "epoch": 0.6492146596858639,
-      "grad_norm": 2.4910936225080227,
+      "grad_norm": 2.2885995431879866,
       "learning_rate": 4.796968003560639e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.375,
-      "logps/chosen": -864.0,
-      "logps/rejected": -1184.0,
-      "loss": 0.5141,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -4.5625,
-      "rewards/margins": 1.71875,
-      "rewards/rejected": -6.28125,
+      "logits/chosen": -8.5,
+      "logits/rejected": -8.3125,
+      "logps/chosen": -780.0,
+      "logps/rejected": -1064.0,
+      "loss": 0.5079,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -3.75,
+      "rewards/margins": 1.328125,
+      "rewards/rejected": -5.09375,
       "step": 620
     },
     {
       "epoch": 0.6596858638743456,
-      "grad_norm": 2.2899945014197467,
+      "grad_norm": 2.0263178929465706,
       "learning_rate": 4.7847714344635864e-05,
-      "logits/chosen": -11.5625,
-      "logits/rejected": -11.0,
-      "logps/chosen": -900.0,
-      "logps/rejected": -1088.0,
-      "loss": 0.5145,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -5.09375,
-      "rewards/margins": 1.203125,
-      "rewards/rejected": -6.28125,
+      "logits/chosen": -8.5,
+      "logits/rejected": -8.0,
+      "logps/chosen": -720.0,
+      "logps/rejected": -896.0,
+      "loss": 0.4845,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -3.296875,
+      "rewards/margins": 1.1015625,
+      "rewards/rejected": -4.40625,
       "step": 630
     },
     {
       "epoch": 0.6701570680628273,
-      "grad_norm": 2.1500746204108623,
+      "grad_norm": 2.9074898085739984,
       "learning_rate": 4.772235575174776e-05,
-      "logits/chosen": -11.125,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -1024.0,
-      "logps/rejected": -1304.0,
-      "loss": 0.4773,
+      "logits/chosen": -7.125,
+      "logits/rejected": -6.53125,
+      "logps/chosen": -876.0,
+      "logps/rejected": -1176.0,
+      "loss": 0.4613,
       "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -5.84375,
-      "rewards/margins": 1.4140625,
-      "rewards/rejected": -7.25,
+      "rewards/chosen": -4.40625,
+      "rewards/margins": 1.609375,
+      "rewards/rejected": -6.0,
       "step": 640
     },
     {
       "epoch": 0.680628272251309,
-      "grad_norm": 2.633705064835583,
+      "grad_norm": 2.1363694966752247,
       "learning_rate": 4.7593622872782754e-05,
-      "logits/chosen": -10.25,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -908.0,
-      "logps/rejected": -1112.0,
-      "loss": 0.4948,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -5.28125,
-      "rewards/margins": 0.94140625,
-      "rewards/rejected": -6.21875,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -5.40625,
+      "logps/chosen": -816.0,
+      "logps/rejected": -1024.0,
+      "loss": 0.4983,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -4.3125,
+      "rewards/margins": 1.0078125,
+      "rewards/rejected": -5.34375,
       "step": 650
     },
     {
       "epoch": 0.6910994764397905,
-      "grad_norm": 2.2951995186399277,
+      "grad_norm": 1.9222077515363032,
       "learning_rate": 4.746153482466542e-05,
-      "logits/chosen": -8.875,
-      "logits/rejected": -8.1875,
-      "logps/chosen": -980.0,
-      "logps/rejected": -1152.0,
-      "loss": 0.5016,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -5.84375,
+      "logits/chosen": -5.46875,
+      "logits/rejected": -5.03125,
+      "logps/chosen": -800.0,
+      "logps/rejected": -968.0,
+      "loss": 0.502,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -4.03125,
       "rewards/margins": 1.0859375,
-      "rewards/rejected": -6.9375,
+      "rewards/rejected": -5.125,
       "step": 660
     },
     {
       "epoch": 0.7015706806282722,
-      "grad_norm": 2.1689085624054703,
+      "grad_norm": 2.4134427675685775,
       "learning_rate": 4.732611122256537e-05,
-      "logits/chosen": -10.3125,
-      "logits/rejected": -9.875,
-      "logps/chosen": -960.0,
-      "logps/rejected": -1160.0,
-      "loss": 0.4773,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -5.28125,
-      "rewards/margins": 1.2734375,
-      "rewards/rejected": -6.53125,
+      "logits/chosen": -7.8125,
+      "logits/rejected": -7.46875,
+      "logps/chosen": -856.0,
+      "logps/rejected": -1048.0,
+      "loss": 0.4779,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -4.21875,
+      "rewards/margins": 1.2421875,
+      "rewards/rejected": -5.46875,
       "step": 670
     },
     {
       "epoch": 0.7120418848167539,
-      "grad_norm": 2.237995573105497,
+      "grad_norm": 2.989113982491697,
       "learning_rate": 4.718737217698435e-05,
-      "logits/chosen": -10.75,
-      "logits/rejected": -10.5,
-      "logps/chosen": -824.0,
-      "logps/rejected": -1016.0,
-      "loss": 0.5256,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -4.46875,
-      "rewards/margins": 1.0078125,
-      "rewards/rejected": -5.46875,
+      "logits/chosen": -6.34375,
+      "logits/rejected": -6.3125,
+      "logps/chosen": -808.0,
+      "logps/rejected": -1004.0,
+      "loss": 0.5391,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -4.28125,
+      "rewards/margins": 1.046875,
+      "rewards/rejected": -5.34375,
       "step": 680
     },
     {
       "epoch": 0.7225130890052356,
-      "grad_norm": 2.8369807600016763,
+      "grad_norm": 3.815188284030519,
       "learning_rate": 4.704533829076982e-05,
-      "logits/chosen": -11.75,
-      "logits/rejected": -10.8125,
-      "logps/chosen": -816.0,
-      "logps/rejected": -1160.0,
-      "loss": 0.4752,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -4.15625,
-      "rewards/margins": 1.6796875,
-      "rewards/rejected": -5.84375,
+      "logits/chosen": -7.03125,
+      "logits/rejected": -6.5,
+      "logps/chosen": -840.0,
+      "logps/rejected": -1184.0,
+      "loss": 0.4698,
+      "rewards/accuracies": 0.8187500238418579,
+      "rewards/chosen": -4.375,
+      "rewards/margins": 1.703125,
+      "rewards/rejected": -6.0625,
       "step": 690
     },
     {
       "epoch": 0.7329842931937173,
-      "grad_norm": 2.832657180557495,
+      "grad_norm": 2.9874056878103024,
       "learning_rate": 4.6900030656055455e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -904.0,
-      "logps/rejected": -1112.0,
-      "loss": 0.4846,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -4.875,
-      "rewards/margins": 1.1640625,
-      "rewards/rejected": -6.03125,
+      "logits/chosen": -6.375,
+      "logits/rejected": -6.25,
+      "logps/chosen": -884.0,
+      "logps/rejected": -1104.0,
+      "loss": 0.4831,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -4.6875,
+      "rewards/margins": 1.296875,
+      "rewards/rejected": -6.0,
       "step": 700
     },
     {
       "epoch": 0.7329842931937173,
-      "eval_logits/chosen": -8.625,
-      "eval_logits/rejected": -7.25,
-      "eval_logps/chosen": -888.0,
-      "eval_logps/rejected": -924.0,
-      "eval_loss": 0.813796877861023,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -5.6875,
-      "eval_rewards/margins": 0.6640625,
-      "eval_rewards/rejected": -6.34375,
-      "eval_runtime": 46.6535,
-      "eval_samples_per_second": 42.869,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -5.3125,
+      "eval_logits/rejected": -4.0625,
+      "eval_logps/chosen": -784.0,
+      "eval_logps/rejected": -824.0,
+      "eval_loss": 0.7994687557220459,
+      "eval_rewards/accuracies": 0.60546875,
+      "eval_rewards/chosen": -4.65625,
+      "eval_rewards/margins": 0.68359375,
+      "eval_rewards/rejected": -5.34375,
+      "eval_runtime": 47.0825,
+      "eval_samples_per_second": 42.479,
+      "eval_steps_per_second": 0.68,
       "step": 700
     },
     {
       "epoch": 0.743455497382199,
-      "grad_norm": 3.38180386213725,
+      "grad_norm": 3.326008081024393,
       "learning_rate": 4.6751470851128906e-05,
-      "logits/chosen": -9.75,
-      "logits/rejected": -8.75,
-      "logps/chosen": -952.0,
-      "logps/rejected": -1192.0,
-      "loss": 0.4763,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -5.15625,
-      "rewards/margins": 1.5,
-      "rewards/rejected": -6.65625,
+      "logits/chosen": -7.46875,
+      "logits/rejected": -6.75,
+      "logps/chosen": -800.0,
+      "logps/rejected": -1032.0,
+      "loss": 0.4767,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -3.65625,
+      "rewards/margins": 1.359375,
+      "rewards/rejected": -5.0,
       "step": 710
     },
     {
       "epoch": 0.7539267015706806,
-      "grad_norm": 2.135094059484422,
+      "grad_norm": 2.216478691110921,
       "learning_rate": 4.6599680937227454e-05,
-      "logits/chosen": -11.1875,
-      "logits/rejected": -10.125,
-      "logps/chosen": -880.0,
-      "logps/rejected": -1112.0,
-      "loss": 0.4684,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -4.5625,
-      "rewards/margins": 1.53125,
-      "rewards/rejected": -6.09375,
+      "logits/chosen": -7.625,
+      "logits/rejected": -6.625,
+      "logps/chosen": -792.0,
+      "logps/rejected": -1012.0,
+      "loss": 0.476,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -3.640625,
+      "rewards/margins": 1.421875,
+      "rewards/rejected": -5.0625,
       "step": 720
     },
     {
       "epoch": 0.7643979057591623,
-      "grad_norm": 2.7871919268924965,
+      "grad_norm": 2.578414792459822,
       "learning_rate": 4.644468345526183e-05,
-      "logits/chosen": -11.375,
-      "logits/rejected": -10.625,
-      "logps/chosen": -888.0,
-      "logps/rejected": -1208.0,
-      "loss": 0.5149,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -4.75,
-      "rewards/margins": 1.703125,
-      "rewards/rejected": -6.4375,
+      "logits/chosen": -6.78125,
+      "logits/rejected": -6.125,
+      "logps/chosen": -840.0,
+      "logps/rejected": -1184.0,
+      "loss": 0.4978,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -4.3125,
+      "rewards/margins": 1.8828125,
+      "rewards/rejected": -6.1875,
       "step": 730
     },
     {
       "epoch": 0.774869109947644,
-      "grad_norm": 2.824019091892997,
+      "grad_norm": 2.592714351626455,
       "learning_rate": 4.628650142246894e-05,
-      "logits/chosen": -9.9375,
-      "logits/rejected": -9.125,
-      "logps/chosen": -880.0,
-      "logps/rejected": -1112.0,
-      "loss": 0.4964,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -4.96875,
-      "rewards/margins": 1.3125,
-      "rewards/rejected": -6.28125,
+      "logits/chosen": -5.875,
+      "logits/rejected": -5.5,
+      "logps/chosen": -792.0,
+      "logps/rejected": -1024.0,
+      "loss": 0.5258,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -4.0625,
+      "rewards/margins": 1.3359375,
+      "rewards/rejected": -5.40625,
       "step": 740
     },
     {
       "epoch": 0.7853403141361257,
-      "grad_norm": 3.1209058777021714,
+      "grad_norm": 3.2258295718841365,
       "learning_rate": 4.6125158328993765e-05,
-      "logits/chosen": -10.0625,
-      "logits/rejected": -9.1875,
-      "logps/chosen": -848.0,
-      "logps/rejected": -1072.0,
-      "loss": 0.4976,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -4.59375,
-      "rewards/margins": 0.88671875,
-      "rewards/rejected": -5.5,
+      "logits/chosen": -8.375,
+      "logits/rejected": -7.6875,
+      "logps/chosen": -712.0,
+      "logps/rejected": -936.0,
+      "loss": 0.4968,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -3.234375,
+      "rewards/margins": 0.9140625,
+      "rewards/rejected": -4.15625,
       "step": 750
     },
     {
       "epoch": 0.7958115183246073,
-      "grad_norm": 3.4125851321038625,
+      "grad_norm": 2.7196352102155217,
       "learning_rate": 4.5960678134401064e-05,
-      "logits/chosen": -10.6875,
-      "logits/rejected": -9.6875,
-      "logps/chosen": -884.0,
-      "logps/rejected": -1144.0,
-      "loss": 0.4801,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -4.6875,
-      "rewards/margins": 1.3984375,
-      "rewards/rejected": -6.0625,
+      "logits/chosen": -8.1875,
+      "logits/rejected": -7.1875,
+      "logps/chosen": -836.0,
+      "logps/rejected": -1112.0,
+      "loss": 0.4775,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -4.21875,
+      "rewards/margins": 1.5546875,
+      "rewards/rejected": -5.75,
       "step": 760
     },
     {
       "epoch": 0.806282722513089,
-      "grad_norm": 3.762260741004996,
+      "grad_norm": 3.7046444022869234,
       "learning_rate": 4.579308526411734e-05,
-      "logits/chosen": -9.0,
-      "logits/rejected": -8.5,
-      "logps/chosen": -1004.0,
-      "logps/rejected": -1224.0,
-      "loss": 0.4636,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -6.0,
-      "rewards/margins": 1.484375,
-      "rewards/rejected": -7.5,
+      "logits/chosen": -5.21875,
+      "logits/rejected": -5.09375,
+      "logps/chosen": -932.0,
+      "logps/rejected": -1160.0,
+      "loss": 0.4482,
+      "rewards/accuracies": 0.8187500238418579,
+      "rewards/chosen": -5.28125,
+      "rewards/margins": 1.5546875,
+      "rewards/rejected": -6.84375,
       "step": 770
     },
     {
       "epoch": 0.8167539267015707,
-      "grad_norm": 3.9679923042468954,
+      "grad_norm": 3.8987870090067145,
       "learning_rate": 4.562240460580367e-05,
-      "logits/chosen": -10.5,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -892.0,
-      "logps/rejected": -1152.0,
-      "loss": 0.4316,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -4.9375,
-      "rewards/margins": 1.4375,
-      "rewards/rejected": -6.375,
+      "logits/chosen": -7.34375,
+      "logits/rejected": -7.1875,
+      "logps/chosen": -880.0,
+      "logps/rejected": -1144.0,
+      "loss": 0.4449,
+      "rewards/accuracies": 0.8062499761581421,
+      "rewards/chosen": -4.8125,
+      "rewards/margins": 1.5234375,
+      "rewards/rejected": -6.34375,
       "step": 780
     },
     {
       "epoch": 0.8272251308900523,
-      "grad_norm": 2.314774282426851,
+      "grad_norm": 2.79737929725398,
       "learning_rate": 4.544866150565985e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -9.5625,
-      "logps/chosen": -972.0,
-      "logps/rejected": -1272.0,
-      "loss": 0.5085,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -5.59375,
-      "rewards/margins": 1.6328125,
-      "rewards/rejected": -7.21875,
+      "logits/chosen": -8.4375,
+      "logits/rejected": -7.96875,
+      "logps/chosen": -940.0,
+      "logps/rejected": -1248.0,
+      "loss": 0.5153,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -5.25,
+      "rewards/margins": 1.671875,
+      "rewards/rejected": -6.9375,
       "step": 790
     },
     {
       "epoch": 0.837696335078534,
-      "grad_norm": 2.6804262409159385,
+      "grad_norm": 2.864905474588235,
       "learning_rate": 4.527188176466051e-05,
-      "logits/chosen": -11.0,
-      "logits/rejected": -10.75,
-      "logps/chosen": -908.0,
-      "logps/rejected": -1112.0,
-      "loss": 0.475,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -4.75,
-      "rewards/margins": 1.140625,
-      "rewards/rejected": -5.90625,
+      "logits/chosen": -9.5,
+      "logits/rejected": -9.3125,
+      "logps/chosen": -816.0,
+      "logps/rejected": -1020.0,
+      "loss": 0.5004,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -3.84375,
+      "rewards/margins": 1.1171875,
+      "rewards/rejected": -4.96875,
       "step": 800
     },
     {
       "epoch": 0.837696335078534,
-      "eval_logits/chosen": -10.0625,
-      "eval_logits/rejected": -8.9375,
-      "eval_logps/chosen": -868.0,
-      "eval_logps/rejected": -900.0,
-      "eval_loss": 0.7353593707084656,
-      "eval_rewards/accuracies": 0.61328125,
-      "eval_rewards/chosen": -5.5,
-      "eval_rewards/margins": 0.6015625,
-      "eval_rewards/rejected": -6.09375,
-      "eval_runtime": 46.6513,
-      "eval_samples_per_second": 42.871,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -7.4375,
+      "eval_logits/rejected": -6.125,
+      "eval_logps/chosen": -748.0,
+      "eval_logps/rejected": -776.0,
+      "eval_loss": 0.7361875176429749,
+      "eval_rewards/accuracies": 0.615234375,
+      "eval_rewards/chosen": -4.28125,
+      "eval_rewards/margins": 0.609375,
+      "eval_rewards/rejected": -4.90625,
+      "eval_runtime": 47.0854,
+      "eval_samples_per_second": 42.476,
+      "eval_steps_per_second": 0.68,
       "step": 800
     },
     {
       "epoch": 0.8481675392670157,
-      "grad_norm": 2.3642703551498863,
+      "grad_norm": 2.743114007303884,
       "learning_rate": 4.509209163472357e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -960.0,
-      "logps/rejected": -1296.0,
-      "loss": 0.4508,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -5.28125,
-      "rewards/margins": 1.703125,
-      "rewards/rejected": -7.0,
+      "logits/chosen": -9.8125,
+      "logits/rejected": -9.0,
+      "logps/chosen": -872.0,
+      "logps/rejected": -1208.0,
+      "loss": 0.4632,
+      "rewards/accuracies": 0.8125,
+      "rewards/chosen": -4.40625,
+      "rewards/margins": 1.671875,
+      "rewards/rejected": -6.09375,
       "step": 810
     },
     {
       "epoch": 0.8586387434554974,
-      "grad_norm": 1.9388731619599733,
+      "grad_norm": 2.3327882073276442,
       "learning_rate": 4.4909317814811896e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -9.5,
-      "logps/chosen": -980.0,
-      "logps/rejected": -1264.0,
-      "loss": 0.4925,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -5.6875,
+      "logits/chosen": -10.125,
+      "logits/rejected": -9.625,
+      "logps/chosen": -864.0,
+      "logps/rejected": -1144.0,
+      "loss": 0.4803,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": -4.53125,
       "rewards/margins": 1.5859375,
-      "rewards/rejected": -7.28125,
+      "rewards/rejected": -6.125,
       "step": 820
     },
     {
       "epoch": 0.8691099476439791,
-      "grad_norm": 2.795751920306338,
+      "grad_norm": 2.5159325968794652,
       "learning_rate": 4.472358744696845e-05,
-      "logits/chosen": -11.125,
-      "logits/rejected": -10.5625,
-      "logps/chosen": -940.0,
-      "logps/rejected": -1144.0,
-      "loss": 0.5011,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -5.375,
-      "rewards/margins": 1.328125,
-      "rewards/rejected": -6.71875,
+      "logits/chosen": -9.5,
+      "logits/rejected": -9.0,
+      "logps/chosen": -860.0,
+      "logps/rejected": -1064.0,
+      "loss": 0.5093,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -4.5625,
+      "rewards/margins": 1.3046875,
+      "rewards/rejected": -5.875,
       "step": 830
     },
     {
       "epoch": 0.8795811518324608,
-      "grad_norm": 2.4026863352022167,
+      "grad_norm": 2.776030172040102,
       "learning_rate": 4.4534928112285645e-05,
-      "logits/chosen": -10.3125,
-      "logits/rejected": -9.5,
-      "logps/chosen": -992.0,
-      "logps/rejected": -1224.0,
-      "loss": 0.4933,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -5.71875,
-      "rewards/margins": 1.28125,
-      "rewards/rejected": -7.0,
+      "logits/chosen": -7.53125,
+      "logits/rejected": -6.9375,
+      "logps/chosen": -932.0,
+      "logps/rejected": -1184.0,
+      "loss": 0.4766,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -5.125,
+      "rewards/margins": 1.4296875,
+      "rewards/rejected": -6.5625,
       "step": 840
     },
     {
       "epoch": 0.8900523560209425,
-      "grad_norm": 2.4775270195103496,
+      "grad_norm": 2.254098242142525,
       "learning_rate": 4.43433678268096e-05,
-      "logits/chosen": -9.125,
-      "logits/rejected": -8.4375,
-      "logps/chosen": -992.0,
-      "logps/rejected": -1216.0,
-      "loss": 0.4605,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -6.0,
-      "rewards/margins": 1.3359375,
-      "rewards/rejected": -7.3125,
+      "logits/chosen": -7.75,
+      "logits/rejected": -7.25,
+      "logps/chosen": -916.0,
+      "logps/rejected": -1152.0,
+      "loss": 0.4559,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -5.21875,
+      "rewards/margins": 1.4375,
+      "rewards/rejected": -6.65625,
       "step": 850
     },
     {
       "epoch": 0.900523560209424,
-      "grad_norm": 2.5385019657243424,
+      "grad_norm": 2.5126684239234613,
       "learning_rate": 4.414893503737968e-05,
-      "logits/chosen": -9.5,
-      "logits/rejected": -8.375,
-      "logps/chosen": -928.0,
-      "logps/rejected": -1128.0,
-      "loss": 0.4882,
+      "logits/chosen": -7.25,
+      "logits/rejected": -6.8125,
+      "logps/chosen": -868.0,
+      "logps/rejected": -1072.0,
+      "loss": 0.4808,
       "rewards/accuracies": 0.75,
-      "rewards/chosen": -5.5,
-      "rewards/margins": 1.359375,
-      "rewards/rejected": -6.84375,
+      "rewards/chosen": -4.9375,
+      "rewards/margins": 1.390625,
+      "rewards/rejected": -6.3125,
       "step": 860
     },
     {
       "epoch": 0.9109947643979057,
-      "grad_norm": 2.641691812834352,
+      "grad_norm": 3.147577045867666,
       "learning_rate": 4.395165861740414e-05,
-      "logits/chosen": -10.875,
-      "logits/rejected": -10.3125,
-      "logps/chosen": -812.0,
-      "logps/rejected": -1056.0,
-      "loss": 0.4784,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -4.34375,
-      "rewards/margins": 1.3984375,
-      "rewards/rejected": -5.75,
+      "logits/chosen": -7.78125,
+      "logits/rejected": -7.5625,
+      "logps/chosen": -880.0,
+      "logps/rejected": -1152.0,
+      "loss": 0.4583,
+      "rewards/accuracies": 0.793749988079071,
+      "rewards/chosen": -5.0,
+      "rewards/margins": 1.65625,
+      "rewards/rejected": -6.6875,
       "step": 870
     },
     {
       "epoch": 0.9214659685863874,
-      "grad_norm": 2.4341279971251977,
+      "grad_norm": 2.288785383433436,
       "learning_rate": 4.375156786257241e-05,
-      "logits/chosen": -11.0,
-      "logits/rejected": -10.5,
-      "logps/chosen": -928.0,
-      "logps/rejected": -1120.0,
-      "loss": 0.4985,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -4.8125,
-      "rewards/margins": 1.1484375,
-      "rewards/rejected": -5.96875,
+      "logits/chosen": -10.0625,
+      "logits/rejected": -9.875,
+      "logps/chosen": -832.0,
+      "logps/rejected": -1020.0,
+      "loss": 0.5186,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -3.875,
+      "rewards/margins": 1.046875,
+      "rewards/rejected": -4.9375,
       "step": 880
     },
     {
       "epoch": 0.9319371727748691,
-      "grad_norm": 2.122364195576178,
+      "grad_norm": 2.30506919773941,
       "learning_rate": 4.354869248650466e-05,
-      "logits/chosen": -10.75,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -864.0,
-      "logps/rejected": -1040.0,
-      "loss": 0.4762,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -4.53125,
-      "rewards/margins": 1.140625,
-      "rewards/rejected": -5.65625,
+      "logits/chosen": -10.3125,
+      "logits/rejected": -9.8125,
+      "logps/chosen": -756.0,
+      "logps/rejected": -912.0,
+      "loss": 0.4954,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -3.4375,
+      "rewards/margins": 0.9921875,
+      "rewards/rejected": -4.4375,
       "step": 890
     },
     {
       "epoch": 0.9424083769633508,
-      "grad_norm": 4.027936423662177,
+      "grad_norm": 3.968567644772862,
       "learning_rate": 4.334306261633931e-05,
-      "logits/chosen": -7.8125,
-      "logits/rejected": -6.875,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -1328.0,
-      "loss": 0.4734,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -6.71875,
-      "rewards/margins": 1.28125,
-      "rewards/rejected": -8.0,
+      "logits/chosen": -7.125,
+      "logits/rejected": -6.21875,
+      "logps/chosen": -1012.0,
+      "logps/rejected": -1240.0,
+      "loss": 0.4651,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -5.75,
+      "rewards/margins": 1.3515625,
+      "rewards/rejected": -7.09375,
       "step": 900
     },
     {
       "epoch": 0.9424083769633508,
-      "eval_logits/chosen": -7.34375,
-      "eval_logits/rejected": -5.75,
-      "eval_logps/chosen": -1072.0,
-      "eval_logps/rejected": -1120.0,
-      "eval_loss": 0.8620625138282776,
-      "eval_rewards/accuracies": 0.607421875,
-      "eval_rewards/chosen": -7.5625,
-      "eval_rewards/margins": 0.73046875,
-      "eval_rewards/rejected": -8.3125,
-      "eval_runtime": 46.6481,
-      "eval_samples_per_second": 42.874,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -5.0,
+      "eval_logits/rejected": -3.4375,
+      "eval_logps/chosen": -1000.0,
+      "eval_logps/rejected": -1048.0,
+      "eval_loss": 0.8776875138282776,
+      "eval_rewards/accuracies": 0.587890625,
+      "eval_rewards/chosen": -6.84375,
+      "eval_rewards/margins": 0.74609375,
+      "eval_rewards/rejected": -7.59375,
+      "eval_runtime": 47.0728,
+      "eval_samples_per_second": 42.487,
+      "eval_steps_per_second": 0.68,
       "step": 900
     },
     {
       "epoch": 0.9528795811518325,
-      "grad_norm": 3.2745267150043333,
+      "grad_norm": 3.4718163583839328,
       "learning_rate": 4.313470878825911e-05,
-      "logits/chosen": -10.0,
-      "logits/rejected": -9.125,
-      "logps/chosen": -1040.0,
-      "logps/rejected": -1368.0,
-      "loss": 0.4797,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -6.3125,
-      "rewards/margins": 1.7890625,
-      "rewards/rejected": -8.125,
+      "logits/chosen": -8.3125,
+      "logits/rejected": -7.8125,
+      "logps/chosen": -964.0,
+      "logps/rejected": -1304.0,
+      "loss": 0.4734,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -5.59375,
+      "rewards/margins": 1.84375,
+      "rewards/rejected": -7.4375,
       "step": 910
     },
     {
       "epoch": 0.9633507853403142,
-      "grad_norm": 1.9679786295179182,
+      "grad_norm": 2.300684981577585,
       "learning_rate": 4.292366194295652e-05,
-      "logits/chosen": -11.0,
-      "logits/rejected": -10.1875,
-      "logps/chosen": -940.0,
-      "logps/rejected": -1144.0,
-      "loss": 0.4708,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -5.15625,
-      "rewards/margins": 1.4921875,
-      "rewards/rejected": -6.65625,
+      "logits/chosen": -8.875,
+      "logits/rejected": -8.25,
+      "logps/chosen": -880.0,
+      "logps/rejected": -1088.0,
+      "loss": 0.4781,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -4.59375,
+      "rewards/margins": 1.5390625,
+      "rewards/rejected": -6.125,
       "step": 920
     },
     {
       "epoch": 0.9738219895287958,
-      "grad_norm": 2.467595527518976,
+      "grad_norm": 2.642183640878894,
       "learning_rate": 4.2709953421038964e-05,
-      "logits/chosen": -10.6875,
-      "logits/rejected": -9.9375,
-      "logps/chosen": -896.0,
-      "logps/rejected": -1128.0,
-      "loss": 0.457,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -5.09375,
-      "rewards/margins": 1.3515625,
-      "rewards/rejected": -6.4375,
+      "logits/chosen": -9.0,
+      "logits/rejected": -8.375,
+      "logps/chosen": -776.0,
+      "logps/rejected": -996.0,
+      "loss": 0.4454,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -3.875,
+      "rewards/margins": 1.2734375,
+      "rewards/rejected": -5.15625,
       "step": 930
     },
     {
       "epoch": 0.9842931937172775,
-      "grad_norm": 2.607984317889705,
+      "grad_norm": 2.6318167433847446,
       "learning_rate": 4.249361495837478e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -1004.0,
-      "logps/rejected": -1240.0,
-      "loss": 0.4818,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -6.0,
-      "rewards/margins": 1.421875,
-      "rewards/rejected": -7.4375,
+      "logits/chosen": -8.625,
+      "logits/rejected": -7.9375,
+      "logps/chosen": -828.0,
+      "logps/rejected": -1056.0,
+      "loss": 0.4836,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -4.28125,
+      "rewards/margins": 1.328125,
+      "rewards/rejected": -5.59375,
       "step": 940
     },
     {
       "epoch": 0.9947643979057592,
-      "grad_norm": 2.6996469835601102,
+      "grad_norm": 2.408232919315052,
       "learning_rate": 4.227467868138035e-05,
-      "logits/chosen": -10.0625,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -968.0,
-      "logps/rejected": -1248.0,
-      "loss": 0.4726,
+      "logits/chosen": -5.375,
+      "logits/rejected": -5.03125,
+      "logps/chosen": -912.0,
+      "logps/rejected": -1208.0,
+      "loss": 0.4531,
       "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -5.5,
-      "rewards/margins": 1.5078125,
-      "rewards/rejected": -7.0,
+      "rewards/chosen": -4.96875,
+      "rewards/margins": 1.65625,
+      "rewards/rejected": -6.625,
       "step": 950
     },
     {
       "epoch": 1.0052356020942408,
-      "grad_norm": 1.2576788188640022,
+      "grad_norm": 1.5778618901055894,
       "learning_rate": 4.205317710224938e-05,
-      "logits/chosen": -12.0625,
-      "logits/rejected": -11.5,
-      "logps/chosen": -880.0,
-      "logps/rejected": -1192.0,
-      "loss": 0.3033,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -4.5625,
-      "rewards/margins": 2.40625,
-      "rewards/rejected": -6.96875,
+      "logits/chosen": -6.4375,
+      "logits/rejected": -5.71875,
+      "logps/chosen": -872.0,
+      "logps/rejected": -1200.0,
+      "loss": 0.3226,
+      "rewards/accuracies": 0.8812500238418579,
+      "rewards/chosen": -4.46875,
+      "rewards/margins": 2.53125,
+      "rewards/rejected": -7.0,
       "step": 960
     },
     {
       "epoch": 1.0157068062827226,
-      "grad_norm": 0.8604580733208288,
+      "grad_norm": 0.8808161300245962,
       "learning_rate": 4.1829143114124733e-05,
-      "logits/chosen": -11.25,
-      "logits/rejected": -10.0,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -1624.0,
-      "loss": 0.1738,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.0,
-      "rewards/margins": 3.953125,
-      "rewards/rejected": -10.9375,
-      "step": 970
-    },
-    {
+      "logits/chosen": -6.125,
+      "logits/rejected": -5.4375,
+      "logps/chosen": -952.0,
+      "logps/rejected": -1456.0,
+      "loss": 0.1678,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -5.40625,
+      "rewards/margins": 3.8125,
+      "rewards/rejected": -9.25,
+      "step": 970
+    },
+    {
       "epoch": 1.0261780104712042,
-      "grad_norm": 1.174372433159922,
+      "grad_norm": 1.637090375915629,
       "learning_rate": 4.160260998621387e-05,
-      "logits/chosen": -12.5625,
-      "logits/rejected": -11.125,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -1600.0,
-      "loss": 0.1385,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.40625,
-      "rewards/margins": 4.0625,
-      "rewards/rejected": -10.4375,
+      "logits/chosen": -6.625,
+      "logits/rejected": -6.09375,
+      "logps/chosen": -1012.0,
+      "logps/rejected": -1552.0,
+      "loss": 0.1383,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -5.625,
+      "rewards/margins": 4.375,
+      "rewards/rejected": -10.0,
       "step": 980
     },
     {
       "epoch": 1.036649214659686,
-      "grad_norm": 1.300839240198588,
+      "grad_norm": 1.2880142579486806,
       "learning_rate": 4.137361135884824e-05,
-      "logits/chosen": -9.4375,
-      "logits/rejected": -8.0625,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -1720.0,
-      "loss": 0.1374,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.59375,
-      "rewards/margins": 4.1875,
-      "rewards/rejected": -11.8125,
+      "logits/chosen": -5.71875,
+      "logits/rejected": -5.0,
+      "logps/chosen": -1008.0,
+      "logps/rejected": -1568.0,
+      "loss": 0.1281,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -5.90625,
+      "rewards/margins": 4.375,
+      "rewards/rejected": -10.25,
       "step": 990
     },
     {
       "epoch": 1.0471204188481675,
-      "grad_norm": 2.0827152263073163,
+      "grad_norm": 1.7892334553697686,
       "learning_rate": 4.114218123848777e-05,
-      "logits/chosen": -6.6875,
-      "logits/rejected": -5.5625,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -1920.0,
-      "loss": 0.146,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 4.59375,
-      "rewards/rejected": -14.0625,
+      "logits/chosen": -4.71875,
+      "logits/rejected": -4.375,
+      "logps/chosen": -1000.0,
+      "logps/rejected": -1576.0,
+      "loss": 0.1407,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.0625,
+      "rewards/margins": 4.5625,
+      "rewards/rejected": -10.625,
       "step": 1000
     },
     {
       "epoch": 1.0471204188481675,
-      "eval_logits/chosen": -4.40625,
-      "eval_logits/rejected": -2.703125,
-      "eval_logps/chosen": -1384.0,
-      "eval_logps/rejected": -1440.0,
-      "eval_loss": 1.1525156497955322,
-      "eval_rewards/accuracies": 0.568359375,
-      "eval_rewards/chosen": -10.6875,
-      "eval_rewards/margins": 0.84765625,
-      "eval_rewards/rejected": -11.5,
-      "eval_runtime": 46.6558,
-      "eval_samples_per_second": 42.867,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -2.21875,
+      "eval_logits/rejected": -0.59375,
+      "eval_logps/chosen": -1128.0,
+      "eval_logps/rejected": -1192.0,
+      "eval_loss": 1.0909218788146973,
+      "eval_rewards/accuracies": 0.580078125,
+      "eval_rewards/chosen": -8.125,
+      "eval_rewards/margins": 0.9296875,
+      "eval_rewards/rejected": -9.0,
+      "eval_runtime": 47.0883,
+      "eval_samples_per_second": 42.473,
+      "eval_steps_per_second": 0.68,
       "step": 1000
     },
     {
       "epoch": 1.057591623036649,
-      "grad_norm": 3.994610680964802,
+      "grad_norm": 2.0321855371946227,
       "learning_rate": 4.0908353992670824e-05,
-      "logits/chosen": -8.375,
-      "logits/rejected": -6.875,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -1688.0,
-      "loss": 0.1548,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.28125,
+      "logits/chosen": -5.09375,
+      "logits/rejected": -4.71875,
+      "logps/chosen": -1016.0,
+      "logps/rejected": -1576.0,
+      "loss": 0.1682,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.1875,
       "rewards/margins": 4.65625,
-      "rewards/rejected": -11.9375,
+      "rewards/rejected": -10.8125,
       "step": 1010
     },
     {
       "epoch": 1.068062827225131,
-      "grad_norm": 3.698765886248804,
+      "grad_norm": 3.5321321836273967,
       "learning_rate": 4.067216434491059e-05,
-      "logits/chosen": -7.46875,
-      "logits/rejected": -6.3125,
-      "logps/chosen": -1048.0,
-      "logps/rejected": -1568.0,
-      "loss": 0.1485,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.40625,
-      "rewards/margins": 4.0625,
-      "rewards/rejected": -10.5,
+      "logits/chosen": -4.78125,
+      "logits/rejected": -4.09375,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -1640.0,
+      "loss": 0.1378,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.8125,
+      "rewards/margins": 4.34375,
+      "rewards/rejected": -11.1875,
       "step": 1020
     },
     {
       "epoch": 1.0785340314136125,
-      "grad_norm": 1.6276575619454736,
+      "grad_norm": 1.214845435762161,
       "learning_rate": 4.043364736953866e-05,
-      "logits/chosen": -6.4375,
-      "logits/rejected": -5.5625,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -1696.0,
-      "loss": 0.164,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.125,
-      "rewards/margins": 4.15625,
-      "rewards/rejected": -11.3125,
+      "logits/chosen": -4.5,
+      "logits/rejected": -3.890625,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -1784.0,
+      "loss": 0.1473,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.4375,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -12.1875,
       "step": 1030
     },
     {
       "epoch": 1.0890052356020943,
-      "grad_norm": 2.2553475006678574,
+      "grad_norm": 2.301277372132293,
       "learning_rate": 4.019283848649643e-05,
-      "logits/chosen": -9.5625,
-      "logits/rejected": -8.5,
+      "logits/chosen": -7.1875,
+      "logits/rejected": -6.28125,
       "logps/chosen": -1056.0,
-      "logps/rejected": -1552.0,
-      "loss": 0.1701,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -6.40625,
-      "rewards/margins": 3.984375,
-      "rewards/rejected": -10.375,
+      "logps/rejected": -1568.0,
+      "loss": 0.1772,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -6.375,
+      "rewards/margins": 4.125,
+      "rewards/rejected": -10.5,
       "step": 1040
     },
     {
       "epoch": 1.0994764397905759,
-      "grad_norm": 1.523265452666949,
+      "grad_norm": 2.0402015859896587,
       "learning_rate": 3.9949773456075215e-05,
-      "logits/chosen": -9.4375,
-      "logits/rejected": -8.3125,
-      "logps/chosen": -1000.0,
-      "logps/rejected": -1528.0,
-      "loss": 0.1274,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -6.25,
-      "rewards/margins": 4.09375,
-      "rewards/rejected": -10.375,
+      "logits/chosen": -6.59375,
+      "logits/rejected": -6.25,
+      "logps/chosen": -944.0,
+      "logps/rejected": -1432.0,
+      "loss": 0.1362,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -5.71875,
+      "rewards/margins": 3.734375,
+      "rewards/rejected": -9.4375,
       "step": 1050
     },
     {
       "epoch": 1.1099476439790577,
-      "grad_norm": 1.1666046345019474,
+      "grad_norm": 1.680888322977707,
       "learning_rate": 3.970448837360585e-05,
-      "logits/chosen": -9.125,
-      "logits/rejected": -7.5625,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.137,
+      "logits/chosen": -4.71875,
+      "logits/rejected": -3.9375,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -1872.0,
+      "loss": 0.1485,
       "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -8.125,
-      "rewards/margins": 4.21875,
-      "rewards/rejected": -12.3125,
+      "rewards/chosen": -9.0625,
+      "rewards/margins": 4.1875,
+      "rewards/rejected": -13.25,
       "step": 1060
     },
     {
       "epoch": 1.1204188481675392,
-      "grad_norm": 1.7821092277761532,
+      "grad_norm": 1.269913915999004,
       "learning_rate": 3.945701966409846e-05,
-      "logits/chosen": -9.8125,
-      "logits/rejected": -8.3125,
-      "logps/chosen": -1088.0,
-      "logps/rejected": -1672.0,
-      "loss": 0.1317,
+      "logits/chosen": -6.25,
+      "logits/rejected": -5.375,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -1664.0,
+      "loss": 0.126,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -6.75,
-      "rewards/margins": 4.71875,
-      "rewards/rejected": -11.4375,
+      "rewards/chosen": -6.625,
+      "rewards/margins": 4.78125,
+      "rewards/rejected": -11.375,
       "step": 1070
     },
     {
       "epoch": 1.130890052356021,
-      "grad_norm": 3.5961877150043184,
+      "grad_norm": 2.901372676292373,
       "learning_rate": 3.9207404076833374e-05,
-      "logits/chosen": -9.1875,
-      "logits/rejected": -7.6875,
-      "logps/chosen": -1192.0,
-      "logps/rejected": -1704.0,
-      "loss": 0.1348,
+      "logits/chosen": -7.625,
+      "logits/rejected": -6.34375,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -1616.0,
+      "loss": 0.1344,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -7.375,
-      "rewards/margins": 4.34375,
-      "rewards/rejected": -11.75,
+      "rewards/chosen": -6.5625,
+      "rewards/margins": 4.28125,
+      "rewards/rejected": -10.8125,
       "step": 1080
     },
     {
       "epoch": 1.1413612565445026,
-      "grad_norm": 2.039390136273228,
+      "grad_norm": 2.6514351755183356,
       "learning_rate": 3.895567867990379e-05,
-      "logits/chosen": -8.0,
-      "logits/rejected": -6.03125,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -1824.0,
-      "loss": 0.13,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -8.0625,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -12.875,
+      "logits/chosen": -6.84375,
+      "logits/rejected": -5.21875,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -1744.0,
+      "loss": 0.1542,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -7.3125,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -12.0625,
       "step": 1090
     },
     {
       "epoch": 1.1518324607329844,
-      "grad_norm": 1.5900754235751582,
+      "grad_norm": 1.5833640938680782,
       "learning_rate": 3.870188085471114e-05,
-      "logits/chosen": -7.5,
-      "logits/rejected": -5.78125,
-      "logps/chosen": -1184.0,
-      "logps/rejected": -1672.0,
-      "loss": 0.1356,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.65625,
-      "rewards/margins": 4.34375,
-      "rewards/rejected": -12.0,
+      "logits/chosen": -5.53125,
+      "logits/rejected": -4.4375,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -1608.0,
+      "loss": 0.1332,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.75,
+      "rewards/margins": 4.625,
+      "rewards/rejected": -11.375,
       "step": 1100
     },
     {
       "epoch": 1.1518324607329844,
-      "eval_logits/chosen": -5.53125,
-      "eval_logits/rejected": -3.6875,
-      "eval_logps/chosen": -1232.0,
-      "eval_logps/rejected": -1296.0,
-      "eval_loss": 0.9971562623977661,
-      "eval_rewards/accuracies": 0.609375,
-      "eval_rewards/chosen": -9.1875,
-      "eval_rewards/margins": 0.90625,
-      "eval_rewards/rejected": -10.0625,
-      "eval_runtime": 46.6258,
-      "eval_samples_per_second": 42.895,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -2.65625,
+      "eval_logits/rejected": -0.70703125,
+      "eval_logps/chosen": -1168.0,
+      "eval_logps/rejected": -1224.0,
+      "eval_loss": 1.0452343225479126,
+      "eval_rewards/accuracies": 0.5859375,
+      "eval_rewards/chosen": -8.5,
+      "eval_rewards/margins": 0.89453125,
+      "eval_rewards/rejected": -9.375,
+      "eval_runtime": 47.0824,
+      "eval_samples_per_second": 42.479,
+      "eval_steps_per_second": 0.68,
       "step": 1100
     },
     {
       "epoch": 1.162303664921466,
-      "grad_norm": 2.456500645589885,
+      "grad_norm": 2.980160708701852,
       "learning_rate": 3.844604829041395e-05,
-      "logits/chosen": -8.125,
-      "logits/rejected": -6.25,
-      "logps/chosen": -1120.0,
-      "logps/rejected": -1640.0,
-      "loss": 0.1413,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.53125,
-      "rewards/margins": 4.28125,
-      "rewards/rejected": -11.8125,
+      "logits/chosen": -5.0625,
+      "logits/rejected": -3.953125,
+      "logps/chosen": -1004.0,
+      "logps/rejected": -1520.0,
+      "loss": 0.1461,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -6.34375,
+      "rewards/margins": 4.3125,
+      "rewards/rejected": -10.625,
       "step": 1110
     },
     {
       "epoch": 1.1727748691099475,
-      "grad_norm": 2.9128742152946328,
+      "grad_norm": 2.654029006697534,
       "learning_rate": 3.818821897833095e-05,
-      "logits/chosen": -9.0625,
-      "logits/rejected": -7.75,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -1728.0,
-      "loss": 0.1384,
+      "logits/chosen": -4.59375,
+      "logits/rejected": -4.0,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -1632.0,
+      "loss": 0.151,
       "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.59375,
-      "rewards/margins": 4.625,
-      "rewards/rejected": -12.25,
+      "rewards/chosen": -7.0,
+      "rewards/margins": 4.1875,
+      "rewards/rejected": -11.1875,
       "step": 1120
     },
     {
       "epoch": 1.1832460732984293,
-      "grad_norm": 1.8525098528649633,
+      "grad_norm": 1.8983410719986766,
       "learning_rate": 3.792843120629935e-05,
-      "logits/chosen": -9.75,
-      "logits/rejected": -7.9375,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -1648.0,
-      "loss": 0.1584,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.125,
-      "rewards/margins": 4.3125,
-      "rewards/rejected": -11.4375,
+      "logits/chosen": -6.09375,
+      "logits/rejected": -4.75,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1600.0,
+      "loss": 0.1453,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.4375,
+      "rewards/margins": 4.46875,
+      "rewards/rejected": -10.875,
       "step": 1130
     },
     {
       "epoch": 1.193717277486911,
-      "grad_norm": 2.4894408923481186,
+      "grad_norm": 3.7757743609660857,
       "learning_rate": 3.766672355298907e-05,
-      "logits/chosen": -10.1875,
-      "logits/rejected": -9.3125,
-      "logps/chosen": -1096.0,
-      "logps/rejected": -1616.0,
-      "loss": 0.1484,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -7.21875,
-      "rewards/margins": 4.0625,
-      "rewards/rejected": -11.25,
+      "logits/chosen": -6.71875,
+      "logits/rejected": -6.15625,
+      "logps/chosen": -996.0,
+      "logps/rejected": -1528.0,
+      "loss": 0.1514,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -6.25,
+      "rewards/margins": 4.15625,
+      "rewards/rejected": -10.375,
       "step": 1140
     },
     {
       "epoch": 1.2041884816753927,
-      "grad_norm": 2.428943113803734,
+      "grad_norm": 2.07932423334662,
       "learning_rate": 3.740313488217373e-05,
-      "logits/chosen": -11.0,
-      "logits/rejected": -9.8125,
-      "logps/chosen": -1104.0,
-      "logps/rejected": -1672.0,
-      "loss": 0.1334,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -7.15625,
-      "rewards/margins": 4.71875,
-      "rewards/rejected": -11.875,
+      "logits/chosen": -7.4375,
+      "logits/rejected": -6.1875,
+      "logps/chosen": -1024.0,
+      "logps/rejected": -1592.0,
+      "loss": 0.1323,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -6.40625,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -11.0625,
       "step": 1150
     },
     {
       "epoch": 1.2146596858638743,
-      "grad_norm": 1.6877001313641127,
+      "grad_norm": 2.1120074063244543,
       "learning_rate": 3.7137704336959464e-05,
-      "logits/chosen": -10.375,
-      "logits/rejected": -8.8125,
-      "logps/chosen": -1160.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.1161,
+      "logits/chosen": -5.15625,
+      "logits/rejected": -4.1875,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -1720.0,
+      "loss": 0.1376,
       "rewards/accuracies": 0.9375,
-      "rewards/chosen": -7.71875,
-      "rewards/margins": 5.0,
-      "rewards/rejected": -12.75,
+      "rewards/chosen": -7.28125,
+      "rewards/margins": 4.78125,
+      "rewards/rejected": -12.0625,
       "step": 1160
     },
     {
       "epoch": 1.225130890052356,
-      "grad_norm": 1.9072738397714215,
+      "grad_norm": 2.104542683444888,
       "learning_rate": 3.687047133397201e-05,
-      "logits/chosen": -10.25,
-      "logits/rejected": -8.6875,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -1912.0,
-      "loss": 0.1201,
+      "logits/chosen": -6.3125,
+      "logits/rejected": -5.34375,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -1736.0,
+      "loss": 0.1273,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -8.625,
-      "rewards/margins": 5.3125,
-      "rewards/rejected": -13.9375,
+      "rewards/chosen": -7.1875,
+      "rewards/margins": 5.03125,
+      "rewards/rejected": -12.1875,
       "step": 1170
     },
     {
       "epoch": 1.2356020942408377,
-      "grad_norm": 2.343319772250893,
+      "grad_norm": 1.5233310251266596,
       "learning_rate": 3.6601475557503415e-05,
-      "logits/chosen": -9.9375,
-      "logits/rejected": -8.25,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -1848.0,
-      "loss": 0.1607,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -8.375,
-      "rewards/margins": 4.84375,
-      "rewards/rejected": -13.25,
+      "logits/chosen": -5.9375,
+      "logits/rejected": -4.65625,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1584.0,
+      "loss": 0.1547,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -6.3125,
+      "rewards/margins": 4.25,
+      "rewards/rejected": -10.5625,
       "step": 1180
     },
     {
       "epoch": 1.2460732984293195,
-      "grad_norm": 2.1505219308500307,
+      "grad_norm": 1.8348060893113167,
       "learning_rate": 3.633075695361882e-05,
-      "logits/chosen": -8.75,
-      "logits/rejected": -7.03125,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -1880.0,
-      "loss": 0.1355,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -8.75,
-      "rewards/margins": 5.09375,
-      "rewards/rejected": -13.875,
+      "logits/chosen": -2.984375,
+      "logits/rejected": -2.75,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1624.0,
+      "loss": 0.1303,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.625,
+      "rewards/margins": 4.65625,
+      "rewards/rejected": -11.25,
       "step": 1190
     },
     {
       "epoch": 1.256544502617801,
-      "grad_norm": 1.1222338906202685,
+      "grad_norm": 1.653376616290843,
       "learning_rate": 3.6058355724224477e-05,
-      "logits/chosen": -10.25,
-      "logits/rejected": -8.5,
-      "logps/chosen": -1128.0,
-      "logps/rejected": -1656.0,
-      "loss": 0.1237,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.21875,
-      "rewards/margins": 4.25,
-      "rewards/rejected": -11.5,
+      "logits/chosen": -4.03125,
+      "logits/rejected": -3.125,
+      "logps/chosen": -1064.0,
+      "logps/rejected": -1608.0,
+      "loss": 0.1397,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.5625,
+      "rewards/margins": 4.4375,
+      "rewards/rejected": -11.0,
       "step": 1200
     },
     {
       "epoch": 1.256544502617801,
-      "eval_logits/chosen": -8.125,
-      "eval_logits/rejected": -6.34375,
-      "eval_logps/chosen": -1168.0,
-      "eval_logps/rejected": -1216.0,
-      "eval_loss": 0.999343752861023,
+      "eval_logits/chosen": -2.09375,
+      "eval_logits/rejected": -0.25390625,
+      "eval_logps/chosen": -1088.0,
+      "eval_logps/rejected": -1144.0,
+      "eval_loss": 1.0392968654632568,
       "eval_rewards/accuracies": 0.6015625,
-      "eval_rewards/chosen": -8.5,
-      "eval_rewards/margins": 0.76171875,
-      "eval_rewards/rejected": -9.25,
-      "eval_runtime": 46.6649,
-      "eval_samples_per_second": 42.859,
-      "eval_steps_per_second": 0.686,
+      "eval_rewards/chosen": -7.6875,
+      "eval_rewards/margins": 0.859375,
+      "eval_rewards/rejected": -8.5625,
+      "eval_runtime": 47.0786,
+      "eval_samples_per_second": 42.482,
+      "eval_steps_per_second": 0.68,
       "step": 1200
     },
     {
       "epoch": 1.2670157068062826,
-      "grad_norm": 2.313804590650624,
+      "grad_norm": 1.976876723823928,
       "learning_rate": 3.5784312321097744e-05,
-      "logits/chosen": -8.5,
-      "logits/rejected": -6.71875,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -1736.0,
-      "loss": 0.1314,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -7.8125,
-      "rewards/margins": 4.53125,
-      "rewards/rejected": -12.375,
+      "logits/chosen": -4.6875,
+      "logits/rejected": -3.78125,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1568.0,
+      "loss": 0.1431,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -6.1875,
+      "rewards/margins": 4.46875,
+      "rewards/rejected": -10.6875,
       "step": 1210
     },
     {
       "epoch": 1.2774869109947644,
-      "grad_norm": 2.9031427744429736,
+      "grad_norm": 2.4034421218737427,
       "learning_rate": 3.550866743987994e-05,
-      "logits/chosen": -7.03125,
-      "logits/rejected": -5.21875,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -1824.0,
-      "loss": 0.1504,
+      "logits/chosen": -4.15625,
+      "logits/rejected": -3.4375,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -1592.0,
+      "loss": 0.1544,
       "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -13.1875,
+      "rewards/chosen": -6.75,
+      "rewards/margins": 4.09375,
+      "rewards/rejected": -10.875,
       "step": 1220
     },
     {
       "epoch": 1.2879581151832462,
-      "grad_norm": 1.710155637141148,
+      "grad_norm": 2.1199462567109255,
       "learning_rate": 3.5231462014033024e-05,
-      "logits/chosen": -8.0625,
-      "logits/rejected": -6.0,
-      "logps/chosen": -1152.0,
-      "logps/rejected": -1680.0,
-      "loss": 0.1199,
+      "logits/chosen": -3.546875,
+      "logits/rejected": -3.0,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1560.0,
+      "loss": 0.1218,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.65625,
-      "rewards/margins": 4.59375,
-      "rewards/rejected": -12.25,
+      "rewards/chosen": -6.53125,
+      "rewards/margins": 4.46875,
+      "rewards/rejected": -11.0,
       "step": 1230
     },
     {
       "epoch": 1.2984293193717278,
-      "grad_norm": 1.6879249777383887,
+      "grad_norm": 2.442775423373829,
       "learning_rate": 3.495273720876095e-05,
-      "logits/chosen": -5.90625,
-      "logits/rejected": -4.40625,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -1840.0,
-      "loss": 0.1188,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 4.375,
-      "rewards/rejected": -13.25,
+      "logits/chosen": -1.8203125,
+      "logits/rejected": -1.078125,
+      "logps/chosen": -1328.0,
+      "logps/rejected": -1920.0,
+      "loss": 0.1275,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -9.0,
+      "rewards/margins": 5.03125,
+      "rewards/rejected": -14.0625,
       "step": 1240
     },
     {
       "epoch": 1.3089005235602094,
-      "grad_norm": 1.4402514325866238,
+      "grad_norm": 2.5747630323503783,
       "learning_rate": 3.4672534414896616e-05,
-      "logits/chosen": -8.9375,
-      "logits/rejected": -7.03125,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -1752.0,
-      "loss": 0.1329,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 4.375,
-      "rewards/rejected": -12.5625,
+      "logits/chosen": -4.65625,
+      "logits/rejected": -3.421875,
+      "logps/chosen": -1216.0,
+      "logps/rejected": -1744.0,
+      "loss": 0.1544,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -7.71875,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -12.4375,
       "step": 1250
     },
     {
       "epoch": 1.3193717277486912,
-      "grad_norm": 2.0495408205865346,
+      "grad_norm": 1.4312561985846906,
       "learning_rate": 3.439089524275526e-05,
-      "logits/chosen": -8.5,
-      "logits/rejected": -6.6875,
-      "logps/chosen": -1112.0,
-      "logps/rejected": -1608.0,
-      "loss": 0.1421,
+      "logits/chosen": -4.9375,
+      "logits/rejected": -3.8125,
+      "logps/chosen": -904.0,
+      "logps/rejected": -1376.0,
+      "loss": 0.1444,
       "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -7.25,
-      "rewards/margins": 4.125,
-      "rewards/rejected": -11.375,
+      "rewards/chosen": -5.15625,
+      "rewards/margins": 3.890625,
+      "rewards/rejected": -9.0,
       "step": 1260
     },
     {
       "epoch": 1.3298429319371727,
-      "grad_norm": 1.9371809370917994,
+      "grad_norm": 2.4550383755802114,
       "learning_rate": 3.410786151595533e-05,
-      "logits/chosen": -8.0625,
-      "logits/rejected": -6.1875,
-      "logps/chosen": -1176.0,
-      "logps/rejected": -1664.0,
-      "loss": 0.147,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.84375,
-      "rewards/margins": 4.25,
-      "rewards/rejected": -12.125,
+      "logits/chosen": -5.5625,
+      "logits/rejected": -4.53125,
+      "logps/chosen": -944.0,
+      "logps/rejected": -1416.0,
+      "loss": 0.1506,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.5625,
+      "rewards/margins": 4.03125,
+      "rewards/rejected": -9.5625,
       "step": 1270
     },
     {
       "epoch": 1.3403141361256545,
-      "grad_norm": 1.6265418274873171,
+      "grad_norm": 2.343232782392088,
       "learning_rate": 3.382347526520765e-05,
-      "logits/chosen": -7.5625,
-      "logits/rejected": -5.84375,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.1466,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -7.96875,
-      "rewards/margins": 3.765625,
-      "rewards/rejected": -11.75,
+      "logits/chosen": -4.21875,
+      "logits/rejected": -3.296875,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1664.0,
+      "loss": 0.1438,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -7.0,
+      "rewards/margins": 4.21875,
+      "rewards/rejected": -11.25,
       "step": 1280
     },
     {
       "epoch": 1.350785340314136,
-      "grad_norm": 2.276988462536782,
+      "grad_norm": 1.8958722897967832,
       "learning_rate": 3.353777872207381e-05,
-      "logits/chosen": -6.28125,
-      "logits/rejected": -4.78125,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.138,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 4.15625,
-      "rewards/rejected": -12.375,
+      "logits/chosen": -3.15625,
+      "logits/rejected": -2.578125,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -1624.0,
+      "loss": 0.1359,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -7.0625,
+      "rewards/margins": 4.375,
+      "rewards/rejected": -11.4375,
       "step": 1290
     },
     {
       "epoch": 1.3612565445026177,
-      "grad_norm": 1.647469970307563,
+      "grad_norm": 2.240998046440703,
       "learning_rate": 3.325081431269475e-05,
-      "logits/chosen": -5.28125,
-      "logits/rejected": -3.40625,
-      "logps/chosen": -1216.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.1387,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.4375,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -12.8125,
+      "logits/chosen": -4.0625,
+      "logits/rejected": -2.8125,
+      "logps/chosen": -968.0,
+      "logps/rejected": -1512.0,
+      "loss": 0.1353,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -5.9375,
+      "rewards/margins": 4.4375,
+      "rewards/rejected": -10.375,
       "step": 1300
     },
     {
       "epoch": 1.3612565445026177,
-      "eval_logits/chosen": -4.5625,
-      "eval_logits/rejected": -2.421875,
-      "eval_logps/chosen": -1408.0,
-      "eval_logps/rejected": -1488.0,
-      "eval_loss": 1.168468713760376,
-      "eval_rewards/accuracies": 0.59765625,
-      "eval_rewards/chosen": -10.875,
-      "eval_rewards/margins": 1.0625,
-      "eval_rewards/rejected": -11.9375,
-      "eval_runtime": 46.6401,
-      "eval_samples_per_second": 42.882,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -3.640625,
+      "eval_logits/rejected": -1.765625,
+      "eval_logps/chosen": -1104.0,
+      "eval_logps/rejected": -1168.0,
+      "eval_loss": 1.0451562404632568,
+      "eval_rewards/accuracies": 0.59375,
+      "eval_rewards/chosen": -7.875,
+      "eval_rewards/margins": 0.9609375,
+      "eval_rewards/rejected": -8.8125,
+      "eval_runtime": 47.0981,
+      "eval_samples_per_second": 42.465,
+      "eval_steps_per_second": 0.679,
       "step": 1300
     },
     {
       "epoch": 1.3717277486910995,
-      "grad_norm": 2.4394086197188503,
+      "grad_norm": 3.126664602745002,
       "learning_rate": 3.296262465149046e-05,
-      "logits/chosen": -6.28125,
-      "logits/rejected": -4.03125,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -2008.0,
-      "loss": 0.1324,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 5.0,
-      "rewards/rejected": -14.5,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -4.0625,
+      "logps/chosen": -1072.0,
+      "logps/rejected": -1656.0,
+      "loss": 0.1466,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.5,
+      "rewards/margins": 4.46875,
+      "rewards/rejected": -10.9375,
       "step": 1310
     },
     {
       "epoch": 1.3821989528795813,
-      "grad_norm": 2.620378844366017,
+      "grad_norm": 2.5686495470481505,
       "learning_rate": 3.2673252534831687e-05,
-      "logits/chosen": -6.0,
-      "logits/rejected": -4.09375,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -1952.0,
-      "loss": 0.1556,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 4.75,
-      "rewards/rejected": -14.5,
+      "logits/chosen": -5.0625,
+      "logits/rejected": -3.84375,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -1616.0,
+      "loss": 0.1522,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.84375,
+      "rewards/margins": 4.3125,
+      "rewards/rejected": -11.125,
       "step": 1320
     },
     {
       "epoch": 1.3926701570680629,
-      "grad_norm": 1.2886579832772507,
+      "grad_norm": 3.32148435050773,
       "learning_rate": 3.2382740934684696e-05,
-      "logits/chosen": -6.8125,
-      "logits/rejected": -4.78125,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -1848.0,
-      "loss": 0.1421,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 4.46875,
-      "rewards/rejected": -13.125,
+      "logits/chosen": -4.0625,
+      "logits/rejected": -2.875,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -1768.0,
+      "loss": 0.1445,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -7.625,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -12.3125,
       "step": 1330
     },
     {
       "epoch": 1.4031413612565444,
-      "grad_norm": 2.425757545679554,
+      "grad_norm": 1.6540822456801114,
       "learning_rate": 3.209113299222982e-05,
-      "logits/chosen": -6.53125,
-      "logits/rejected": -5.34375,
-      "logps/chosen": -1144.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.1424,
+      "logits/chosen": -4.90625,
+      "logits/rejected": -4.375,
+      "logps/chosen": -1008.0,
+      "logps/rejected": -1584.0,
+      "loss": 0.145,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -7.90625,
-      "rewards/margins": 4.28125,
-      "rewards/rejected": -12.1875,
+      "rewards/chosen": -6.5625,
+      "rewards/margins": 4.34375,
+      "rewards/rejected": -10.875,
       "step": 1340
     },
     {
       "epoch": 1.4136125654450262,
-      "grad_norm": 2.402288090297736,
+      "grad_norm": 3.066243596511335,
       "learning_rate": 3.179847201145505e-05,
-      "logits/chosen": -5.90625,
-      "logits/rejected": -4.21875,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -1768.0,
-      "loss": 0.1339,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.1875,
-      "rewards/margins": 4.1875,
-      "rewards/rejected": -12.375,
+      "logits/chosen": -4.625,
+      "logits/rejected": -3.21875,
+      "logps/chosen": -1080.0,
+      "logps/rejected": -1616.0,
+      "loss": 0.1392,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.625,
+      "rewards/margins": 4.25,
+      "rewards/rejected": -10.875,
       "step": 1350
     },
     {
       "epoch": 1.4240837696335078,
-      "grad_norm": 3.3632267314004856,
+      "grad_norm": 4.254290564614655,
       "learning_rate": 3.150480145272528e-05,
-      "logits/chosen": -6.21875,
-      "logits/rejected": -4.875,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -1808.0,
-      "loss": 0.1606,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 4.1875,
-      "rewards/rejected": -13.125,
+      "logits/chosen": -3.84375,
+      "logits/rejected": -3.09375,
+      "logps/chosen": -1160.0,
+      "logps/rejected": -1672.0,
+      "loss": 0.185,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -7.625,
+      "rewards/margins": 4.125,
+      "rewards/rejected": -11.75,
       "step": 1360
     },
     {
       "epoch": 1.4345549738219896,
-      "grad_norm": 1.2061259885036963,
+      "grad_norm": 1.6086872784198054,
       "learning_rate": 3.121016492632848e-05,
-      "logits/chosen": -7.375,
-      "logits/rejected": -5.625,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -1808.0,
-      "loss": 0.1434,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -13.0,
+      "logits/chosen": -5.59375,
+      "logits/rejected": -4.09375,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1712.0,
+      "loss": 0.1322,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.40625,
+      "rewards/margins": 4.59375,
+      "rewards/rejected": -12.0,
       "step": 1370
     },
     {
       "epoch": 1.4450261780104712,
-      "grad_norm": 2.065880760186589,
+      "grad_norm": 2.1063775523524617,
       "learning_rate": 3.091460618599951e-05,
-      "logits/chosen": -6.59375,
-      "logits/rejected": -4.1875,
-      "logps/chosen": -1256.0,
-      "logps/rejected": -1832.0,
-      "loss": 0.1191,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 4.53125,
-      "rewards/rejected": -13.0,
+      "logits/chosen": -5.4375,
+      "logits/rejected": -2.984375,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -1704.0,
+      "loss": 0.13,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -7.09375,
+      "rewards/margins": 4.625,
+      "rewards/rejected": -11.75,
       "step": 1380
     },
     {
       "epoch": 1.4554973821989527,
-      "grad_norm": 2.5237513475830076,
+      "grad_norm": 2.670761030372838,
       "learning_rate": 3.061816912242265e-05,
-      "logits/chosen": -6.15625,
-      "logits/rejected": -4.65625,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -1904.0,
-      "loss": 0.1252,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -8.875,
-      "rewards/margins": 4.65625,
-      "rewards/rejected": -13.5625,
+      "logits/chosen": -5.625,
+      "logits/rejected": -4.40625,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1688.0,
+      "loss": 0.1419,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.875,
+      "rewards/margins": 4.53125,
+      "rewards/rejected": -11.375,
       "step": 1390
     },
     {
       "epoch": 1.4659685863874345,
-      "grad_norm": 2.415003405690821,
+      "grad_norm": 1.2829725992315086,
       "learning_rate": 3.0320897756713785e-05,
-      "logits/chosen": -6.3125,
-      "logits/rejected": -4.25,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -1904.0,
-      "loss": 0.1298,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -8.75,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -13.5625,
+      "logits/chosen": -5.40625,
+      "logits/rejected": -3.890625,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1616.0,
+      "loss": 0.122,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.28125,
+      "rewards/margins": 4.40625,
+      "rewards/rejected": -10.6875,
       "step": 1400
     },
     {
       "epoch": 1.4659685863874345,
-      "eval_logits/chosen": -4.9375,
-      "eval_logits/rejected": -2.984375,
-      "eval_logps/chosen": -1368.0,
-      "eval_logps/rejected": -1440.0,
-      "eval_loss": 1.0589531660079956,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -10.4375,
-      "eval_rewards/margins": 1.03125,
-      "eval_rewards/rejected": -11.5,
-      "eval_runtime": 46.6488,
-      "eval_samples_per_second": 42.874,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -2.1875,
+      "eval_logits/rejected": -0.2578125,
+      "eval_logps/chosen": -1232.0,
+      "eval_logps/rejected": -1304.0,
+      "eval_loss": 1.1135936975479126,
+      "eval_rewards/accuracies": 0.595703125,
+      "eval_rewards/chosen": -9.125,
+      "eval_rewards/margins": 0.9765625,
+      "eval_rewards/rejected": -10.125,
+      "eval_runtime": 47.0922,
+      "eval_samples_per_second": 42.47,
+      "eval_steps_per_second": 0.68,
       "step": 1400
     },
     {
       "epoch": 1.4764397905759163,
-      "grad_norm": 2.616377633595522,
+      "grad_norm": 2.185815234278631,
       "learning_rate": 3.002283623388332e-05,
-      "logits/chosen": -6.8125,
-      "logits/rejected": -5.8125,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.1286,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 4.40625,
-      "rewards/rejected": -13.125,
+      "logits/chosen": -3.96875,
+      "logits/rejected": -3.359375,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1696.0,
+      "loss": 0.1221,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -7.8125,
+      "rewards/margins": 4.4375,
+      "rewards/rejected": -12.25,
       "step": 1410
     },
     {
       "epoch": 1.486910994764398,
-      "grad_norm": 2.704853003531505,
+      "grad_norm": 1.949989533928198,
       "learning_rate": 2.9724028816280507e-05,
-      "logits/chosen": -7.0625,
-      "logits/rejected": -5.96875,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -1848.0,
-      "loss": 0.1211,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -8.875,
+      "logits/chosen": -3.53125,
+      "logits/rejected": -3.0,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -1752.0,
+      "loss": 0.1342,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -7.90625,
       "rewards/margins": 4.53125,
-      "rewards/rejected": -13.4375,
+      "rewards/rejected": -12.4375,
       "step": 1420
     },
     {
       "epoch": 1.4973821989528795,
-      "grad_norm": 3.0357510036280972,
+      "grad_norm": 2.1022475919934607,
       "learning_rate": 2.9424519877020524e-05,
-      "logits/chosen": -9.0625,
-      "logits/rejected": -7.59375,
-      "logps/chosen": -1328.0,
-      "logps/rejected": -1952.0,
-      "loss": 0.1298,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -13.625,
+      "logits/chosen": -5.90625,
+      "logits/rejected": -4.71875,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1784.0,
+      "loss": 0.1324,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.125,
+      "rewards/margins": 4.84375,
+      "rewards/rejected": -11.9375,
       "step": 1430
     },
     {
       "epoch": 1.5078534031413613,
-      "grad_norm": 2.362095273903202,
+      "grad_norm": 2.044694779870266,
       "learning_rate": 2.912435389339504e-05,
-      "logits/chosen": -8.875,
-      "logits/rejected": -6.65625,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.1319,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 4.75,
-      "rewards/rejected": -12.9375,
+      "logits/chosen": -4.53125,
+      "logits/rejected": -2.375,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1656.0,
+      "loss": 0.1287,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -7.03125,
+      "rewards/margins": 4.875,
+      "rewards/rejected": -11.9375,
       "step": 1440
     },
     {
       "epoch": 1.518324607329843,
-      "grad_norm": 2.863166032070748,
+      "grad_norm": 2.843656129833261,
       "learning_rate": 2.8823575440267226e-05,
-      "logits/chosen": -7.59375,
-      "logits/rejected": -6.03125,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -1920.0,
-      "loss": 0.144,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -9.5,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -14.1875,
+      "logits/chosen": -4.5625,
+      "logits/rejected": -3.65625,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -1688.0,
+      "loss": 0.1423,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -7.5,
+      "rewards/margins": 4.3125,
+      "rewards/rejected": -11.8125,
       "step": 1450
     },
     {
       "epoch": 1.5287958115183247,
-      "grad_norm": 2.0692666473338495,
+      "grad_norm": 2.014724300339082,
       "learning_rate": 2.8522229183452465e-05,
-      "logits/chosen": -7.15625,
-      "logits/rejected": -5.65625,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -1824.0,
-      "loss": 0.1936,
+      "logits/chosen": -3.828125,
+      "logits/rejected": -2.75,
+      "logps/chosen": -1160.0,
+      "logps/rejected": -1688.0,
+      "loss": 0.2039,
       "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 4.34375,
-      "rewards/rejected": -13.3125,
+      "rewards/chosen": -7.40625,
+      "rewards/margins": 4.5,
+      "rewards/rejected": -11.9375,
       "step": 1460
     },
     {
       "epoch": 1.5392670157068062,
-      "grad_norm": 2.455707497935021,
+      "grad_norm": 3.477381031764582,
       "learning_rate": 2.8220359873085373e-05,
-      "logits/chosen": -7.8125,
-      "logits/rejected": -6.5625,
-      "logps/chosen": -1200.0,
-      "logps/rejected": -1720.0,
-      "loss": 0.1291,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -7.90625,
-      "rewards/margins": 4.21875,
-      "rewards/rejected": -12.125,
+      "logits/chosen": -2.9375,
+      "logits/rejected": -2.390625,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -1672.0,
+      "loss": 0.1214,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.0625,
+      "rewards/margins": 4.59375,
+      "rewards/rejected": -11.625,
       "step": 1470
     },
     {
       "epoch": 1.5497382198952878,
-      "grad_norm": 2.2263010110276724,
+      "grad_norm": 2.6855263103832954,
       "learning_rate": 2.7918012336974388e-05,
-      "logits/chosen": -6.3125,
-      "logits/rejected": -4.40625,
-      "logps/chosen": -1232.0,
-      "logps/rejected": -1712.0,
-      "loss": 0.1175,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 4.46875,
-      "rewards/rejected": -12.75,
+      "logits/chosen": -2.4375,
+      "logits/rejected": -1.453125,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -1544.0,
+      "loss": 0.1827,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -6.90625,
+      "rewards/margins": 4.1875,
+      "rewards/rejected": -11.125,
       "step": 1480
     },
     {
       "epoch": 1.5602094240837696,
-      "grad_norm": 2.464334639841183,
+      "grad_norm": 2.8303591673095116,
       "learning_rate": 2.7615231473944815e-05,
-      "logits/chosen": -5.28125,
-      "logits/rejected": -3.9375,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -1840.0,
-      "loss": 0.1282,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -9.25,
-      "rewards/margins": 4.46875,
-      "rewards/rejected": -13.75,
+      "logits/chosen": -1.0078125,
+      "logits/rejected": -0.5390625,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -1704.0,
+      "loss": 0.1261,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.59375,
+      "rewards/margins": 4.8125,
+      "rewards/rejected": -12.375,
       "step": 1490
     },
     {
       "epoch": 1.5706806282722514,
-      "grad_norm": 2.8500507853350356,
+      "grad_norm": 3.3112692355741826,
       "learning_rate": 2.7312062247171327e-05,
-      "logits/chosen": -6.65625,
-      "logits/rejected": -5.0625,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -1792.0,
-      "loss": 0.1284,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 4.3125,
-      "rewards/rejected": -13.125,
+      "logits/chosen": -0.62109375,
+      "logits/rejected": -0.08203125,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1712.0,
+      "loss": 0.1192,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -7.8125,
+      "rewards/margins": 4.5,
+      "rewards/rejected": -12.3125,
       "step": 1500
     },
     {
       "epoch": 1.5706806282722514,
-      "eval_logits/chosen": -3.984375,
-      "eval_logits/rejected": -1.9765625,
-      "eval_logps/chosen": -1432.0,
-      "eval_logps/rejected": -1512.0,
-      "eval_loss": 1.1093593835830688,
-      "eval_rewards/accuracies": 0.619140625,
-      "eval_rewards/chosen": -11.125,
-      "eval_rewards/margins": 1.1171875,
-      "eval_rewards/rejected": -12.25,
-      "eval_runtime": 46.6551,
-      "eval_samples_per_second": 42.868,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": 1.1953125,
+      "eval_logits/rejected": 3.03125,
+      "eval_logps/chosen": -1296.0,
+      "eval_logps/rejected": -1376.0,
+      "eval_loss": 1.1467187404632568,
+      "eval_rewards/accuracies": 0.59375,
+      "eval_rewards/chosen": -9.75,
+      "eval_rewards/margins": 1.171875,
+      "eval_rewards/rejected": -10.9375,
+      "eval_runtime": 47.0921,
+      "eval_samples_per_second": 42.47,
+      "eval_steps_per_second": 0.68,
       "step": 1500
     },
     {
       "epoch": 1.581151832460733,
-      "grad_norm": 1.8267481969777848,
+      "grad_norm": 2.3557398850960665,
       "learning_rate": 2.7008549677500876e-05,
-      "logits/chosen": -7.375,
-      "logits/rejected": -5.59375,
-      "logps/chosen": -1400.0,
-      "logps/rejected": -1968.0,
-      "loss": 0.1052,
+      "logits/chosen": -3.359375,
+      "logits/rejected": -2.171875,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -1736.0,
+      "loss": 0.1255,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -9.625,
+      "rewards/chosen": -7.40625,
       "rewards/margins": 4.6875,
-      "rewards/rejected": -14.375,
+      "rewards/rejected": -12.0625,
       "step": 1510
     },
     {
       "epoch": 1.5916230366492146,
-      "grad_norm": 1.4554245461129063,
+      "grad_norm": 1.3066792494889292,
       "learning_rate": 2.670473883676709e-05,
-      "logits/chosen": -5.65625,
-      "logits/rejected": -3.96875,
-      "logps/chosen": -1416.0,
-      "logps/rejected": -2080.0,
-      "loss": 0.0955,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -10.0625,
-      "rewards/margins": 5.34375,
-      "rewards/rejected": -15.375,
+      "logits/chosen": -2.265625,
+      "logits/rejected": -1.5546875,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1760.0,
+      "loss": 0.0918,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.875,
+      "rewards/margins": 5.28125,
+      "rewards/rejected": -12.1875,
       "step": 1520
     },
     {
       "epoch": 1.6020942408376964,
-      "grad_norm": 1.1347279717376941,
+      "grad_norm": 2.041389023202545,
       "learning_rate": 2.6400674841097074e-05,
-      "logits/chosen": -5.5625,
-      "logits/rejected": -3.78125,
-      "logps/chosen": -1528.0,
-      "logps/rejected": -2080.0,
-      "loss": 0.1202,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -11.0625,
-      "rewards/margins": 5.0625,
-      "rewards/rejected": -16.125,
+      "logits/chosen": -2.125,
+      "logits/rejected": -1.1953125,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -1888.0,
+      "loss": 0.1302,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -9.1875,
+      "rewards/margins": 4.96875,
+      "rewards/rejected": -14.1875,
       "step": 1530
     },
     {
       "epoch": 1.6125654450261782,
-      "grad_norm": 3.1201807061357383,
+      "grad_norm": 2.4932473780808575,
       "learning_rate": 2.609640284421158e-05,
-      "logits/chosen": -6.21875,
-      "logits/rejected": -4.09375,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -1960.0,
-      "loss": 0.1446,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -9.8125,
-      "rewards/margins": 4.84375,
-      "rewards/rejected": -14.6875,
+      "logits/chosen": -2.953125,
+      "logits/rejected": -1.5625,
+      "logps/chosen": -1208.0,
+      "logps/rejected": -1784.0,
+      "loss": 0.1329,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.0,
+      "rewards/margins": 4.875,
+      "rewards/rejected": -12.875,
       "step": 1540
     },
     {
       "epoch": 1.6230366492146597,
-      "grad_norm": 2.301582229102563,
+      "grad_norm": 2.1722646498577713,
       "learning_rate": 2.5791968030719726e-05,
-      "logits/chosen": -7.03125,
-      "logits/rejected": -5.40625,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -1856.0,
-      "loss": 0.1321,
+      "logits/chosen": -4.25,
+      "logits/rejected": -3.421875,
+      "logps/chosen": -1112.0,
+      "logps/rejected": -1680.0,
+      "loss": 0.1202,
       "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 4.4375,
-      "rewards/rejected": -12.8125,
+      "rewards/chosen": -6.65625,
+      "rewards/margins": 4.34375,
+      "rewards/rejected": -11.0,
       "step": 1550
     },
     {
       "epoch": 1.6335078534031413,
-      "grad_norm": 2.991400879098044,
+      "grad_norm": 1.424684372442472,
       "learning_rate": 2.548741560940902e-05,
-      "logits/chosen": -7.125,
-      "logits/rejected": -5.9375,
-      "logps/chosen": -1136.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.1464,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -7.40625,
-      "rewards/margins": 4.46875,
-      "rewards/rejected": -11.875,
+      "logits/chosen": -4.21875,
+      "logits/rejected": -3.78125,
+      "logps/chosen": -1040.0,
+      "logps/rejected": -1688.0,
+      "loss": 0.1409,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -6.46875,
+      "rewards/margins": 4.6875,
+      "rewards/rejected": -11.125,
       "step": 1560
     },
     {
       "epoch": 1.6439790575916229,
-      "grad_norm": 0.9647598047441381,
+      "grad_norm": 1.5809581402678632,
       "learning_rate": 2.5182790806531787e-05,
-      "logits/chosen": -5.5,
-      "logits/rejected": -4.1875,
-      "logps/chosen": -1248.0,
-      "logps/rejected": -1840.0,
-      "loss": 0.1271,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 4.75,
-      "rewards/rejected": -13.0,
+      "logits/chosen": -2.5625,
+      "logits/rejected": -2.1875,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -1800.0,
+      "loss": 0.1436,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -7.71875,
+      "rewards/margins": 4.875,
+      "rewards/rejected": -12.5625,
       "step": 1570
     },
     {
       "epoch": 1.6544502617801047,
-      "grad_norm": 1.5707480530886089,
+      "grad_norm": 1.4065472006426647,
       "learning_rate": 2.4878138859089074e-05,
-      "logits/chosen": -5.65625,
-      "logits/rejected": -4.1875,
-      "logps/chosen": -1264.0,
-      "logps/rejected": -1840.0,
-      "loss": 0.1319,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -8.4375,
-      "rewards/margins": 4.71875,
-      "rewards/rejected": -13.125,
+      "logits/chosen": -3.28125,
+      "logits/rejected": -2.296875,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1744.0,
+      "loss": 0.1181,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -7.28125,
+      "rewards/margins": 4.90625,
+      "rewards/rejected": -12.1875,
       "step": 1580
     },
     {
       "epoch": 1.6649214659685865,
-      "grad_norm": 1.9017764943290447,
+      "grad_norm": 1.667002744945226,
       "learning_rate": 2.4573505008112924e-05,
-      "logits/chosen": -5.9375,
-      "logits/rejected": -4.625,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -1872.0,
-      "loss": 0.1167,
+      "logits/chosen": -3.890625,
+      "logits/rejected": -2.9375,
+      "logps/chosen": -1152.0,
+      "logps/rejected": -1752.0,
+      "loss": 0.0995,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 4.71875,
-      "rewards/rejected": -13.625,
+      "rewards/chosen": -7.46875,
+      "rewards/margins": 4.875,
+      "rewards/rejected": -12.375,
       "step": 1590
     },
     {
       "epoch": 1.675392670157068,
-      "grad_norm": 2.0162735021915306,
+      "grad_norm": 2.716072870381967,
       "learning_rate": 2.4268934491948032e-05,
-      "logits/chosen": -6.34375,
-      "logits/rejected": -5.15625,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -1824.0,
-      "loss": 0.1084,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -13.1875,
+      "logits/chosen": -3.234375,
+      "logits/rejected": -2.59375,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -1848.0,
+      "loss": 0.12,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -8.1875,
+      "rewards/margins": 5.1875,
+      "rewards/rejected": -13.375,
       "step": 1600
     },
     {
       "epoch": 1.675392670157068,
-      "eval_logits/chosen": -4.1875,
-      "eval_logits/rejected": -2.28125,
-      "eval_logps/chosen": -1400.0,
-      "eval_logps/rejected": -1480.0,
-      "eval_loss": 1.1327344179153442,
-      "eval_rewards/accuracies": 0.623046875,
-      "eval_rewards/chosen": -10.8125,
-      "eval_rewards/margins": 1.1328125,
-      "eval_rewards/rejected": -11.9375,
-      "eval_runtime": 46.6466,
-      "eval_samples_per_second": 42.876,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -0.79296875,
+      "eval_logits/rejected": 1.203125,
+      "eval_logps/chosen": -1296.0,
+      "eval_logps/rejected": -1392.0,
+      "eval_loss": 1.234218716621399,
+      "eval_rewards/accuracies": 0.607421875,
+      "eval_rewards/chosen": -9.75,
+      "eval_rewards/margins": 1.265625,
+      "eval_rewards/rejected": -11.0625,
+      "eval_runtime": 47.097,
+      "eval_samples_per_second": 42.466,
+      "eval_steps_per_second": 0.679,
       "step": 1600
     },
     {
       "epoch": 1.6858638743455496,
-      "grad_norm": 2.1873137827769047,
+      "grad_norm": 2.788552327078761,
       "learning_rate": 2.396447253953385e-05,
-      "logits/chosen": -8.0625,
-      "logits/rejected": -6.21875,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -1856.0,
-      "loss": 0.1316,
+      "logits/chosen": -4.75,
+      "logits/rejected": -3.671875,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -1776.0,
+      "loss": 0.1376,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -8.75,
-      "rewards/margins": 4.71875,
-      "rewards/rejected": -13.4375,
+      "rewards/chosen": -7.53125,
+      "rewards/margins": 5.0625,
+      "rewards/rejected": -12.625,
       "step": 1610
     },
     {
       "epoch": 1.6963350785340314,
-      "grad_norm": 2.207454990900936,
+      "grad_norm": 2.4507855582026026,
       "learning_rate": 2.3660164363687998e-05,
-      "logits/chosen": -7.71875,
-      "logits/rejected": -6.375,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -1936.0,
-      "loss": 0.1212,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -8.625,
-      "rewards/margins": 4.8125,
-      "rewards/rejected": -13.4375,
+      "logits/chosen": -3.65625,
+      "logits/rejected": -3.0,
+      "logps/chosen": -1192.0,
+      "logps/rejected": -1800.0,
+      "loss": 0.1148,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -7.3125,
+      "rewards/margins": 4.875,
+      "rewards/rejected": -12.1875,
       "step": 1620
     },
     {
       "epoch": 1.7068062827225132,
-      "grad_norm": 2.6039061147405396,
+      "grad_norm": 2.808161182289165,
       "learning_rate": 2.33560551543922e-05,
-      "logits/chosen": -7.5,
-      "logits/rejected": -5.875,
-      "logps/chosen": -1344.0,
-      "logps/rejected": -1840.0,
-      "loss": 0.132,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 4.34375,
-      "rewards/rejected": -13.25,
+      "logits/chosen": -4.25,
+      "logits/rejected": -3.40625,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -1584.0,
+      "loss": 0.143,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -6.6875,
+      "rewards/margins": 4.03125,
+      "rewards/rejected": -10.75,
       "step": 1630
     },
     {
       "epoch": 1.7172774869109948,
-      "grad_norm": 2.205201817015934,
+      "grad_norm": 2.302324739002916,
       "learning_rate": 2.3052190072081493e-05,
-      "logits/chosen": -7.28125,
-      "logits/rejected": -5.4375,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -1816.0,
-      "loss": 0.121,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 4.71875,
-      "rewards/rejected": -13.375,
+      "logits/chosen": -4.9375,
+      "logits/rejected": -3.75,
+      "logps/chosen": -1004.0,
+      "logps/rejected": -1496.0,
+      "loss": 0.13,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -5.84375,
+      "rewards/margins": 4.375,
+      "rewards/rejected": -10.1875,
       "step": 1640
     },
     {
       "epoch": 1.7277486910994764,
-      "grad_norm": 1.964845684748554,
+      "grad_norm": 0.9358322568851414,
       "learning_rate": 2.2748614240937864e-05,
-      "logits/chosen": -6.125,
-      "logits/rejected": -4.3125,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -1880.0,
-      "loss": 0.1177,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 4.78125,
-      "rewards/rejected": -13.5625,
+      "logits/chosen": -3.859375,
+      "logits/rejected": -2.765625,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1624.0,
+      "loss": 0.1017,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.53125,
+      "rewards/margins": 4.4375,
+      "rewards/rejected": -11.0,
       "step": 1650
     },
     {
       "epoch": 1.738219895287958,
-      "grad_norm": 1.958350267089358,
+      "grad_norm": 3.0624446415790967,
       "learning_rate": 2.2445372742189334e-05,
-      "logits/chosen": -6.375,
-      "logits/rejected": -4.71875,
-      "logps/chosen": -1376.0,
-      "logps/rejected": -2032.0,
-      "loss": 0.1178,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -9.5625,
+      "logits/chosen": -2.703125,
+      "logits/rejected": -2.125,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -1840.0,
+      "loss": 0.119,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -7.65625,
       "rewards/margins": 5.15625,
-      "rewards/rejected": -14.75,
+      "rewards/rejected": -12.8125,
       "step": 1660
     },
     {
       "epoch": 1.7486910994764397,
-      "grad_norm": 2.4190744561338517,
+      "grad_norm": 2.5795420674566394,
       "learning_rate": 2.214251060741528e-05,
-      "logits/chosen": -6.40625,
-      "logits/rejected": -5.03125,
-      "logps/chosen": -1440.0,
-      "logps/rejected": -2144.0,
-      "loss": 0.1162,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -10.0625,
-      "rewards/margins": 5.8125,
-      "rewards/rejected": -15.875,
+      "logits/chosen": -2.75,
+      "logits/rejected": -2.078125,
+      "logps/chosen": -1224.0,
+      "logps/rejected": -1896.0,
+      "loss": 0.113,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -7.90625,
+      "rewards/margins": 5.4375,
+      "rewards/rejected": -13.3125,
       "step": 1670
     },
     {
       "epoch": 1.7591623036649215,
-      "grad_norm": 3.1819690697281726,
+      "grad_norm": 2.9365838164360247,
       "learning_rate": 2.1840072811859302e-05,
-      "logits/chosen": -6.09375,
-      "logits/rejected": -4.9375,
-      "logps/chosen": -1416.0,
-      "logps/rejected": -1968.0,
-      "loss": 0.147,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -9.9375,
-      "rewards/margins": 4.75,
-      "rewards/rejected": -14.6875,
+      "logits/chosen": -2.703125,
+      "logits/rejected": -2.03125,
+      "logps/chosen": -1168.0,
+      "logps/rejected": -1752.0,
+      "loss": 0.1233,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -7.4375,
+      "rewards/margins": 5.15625,
+      "rewards/rejected": -12.625,
       "step": 1680
     },
     {
       "epoch": 1.7696335078534031,
-      "grad_norm": 1.4767462518688441,
+      "grad_norm": 1.6487021235353991,
       "learning_rate": 2.1538104267750287e-05,
-      "logits/chosen": -5.9375,
-      "logits/rejected": -4.75,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -1768.0,
-      "loss": 0.1076,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 4.3125,
-      "rewards/rejected": -12.8125,
+      "logits/chosen": -1.3515625,
+      "logits/rejected": -0.796875,
+      "logps/chosen": -1088.0,
+      "logps/rejected": -1640.0,
+      "loss": 0.1242,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -7.03125,
+      "rewards/margins": 4.5,
+      "rewards/rejected": -11.5,
       "step": 1690
     },
     {
       "epoch": 1.7801047120418847,
-      "grad_norm": 1.6346180738709883,
+      "grad_norm": 1.4050520433214713,
       "learning_rate": 2.1236649817632952e-05,
-      "logits/chosen": -6.28125,
-      "logits/rejected": -5.21875,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -1760.0,
-      "loss": 0.1028,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 4.25,
-      "rewards/rejected": -12.75,
+      "logits/chosen": -2.515625,
+      "logits/rejected": -2.546875,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1592.0,
+      "loss": 0.0994,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -6.875,
+      "rewards/margins": 4.1875,
+      "rewards/rejected": -11.0625,
       "step": 1700
     },
     {
       "epoch": 1.7801047120418847,
-      "eval_logits/chosen": -3.625,
-      "eval_logits/rejected": -1.703125,
-      "eval_logps/chosen": -1424.0,
-      "eval_logps/rejected": -1512.0,
-      "eval_loss": 1.1308437585830688,
-      "eval_rewards/accuracies": 0.623046875,
-      "eval_rewards/chosen": -11.0,
-      "eval_rewards/margins": 1.234375,
-      "eval_rewards/rejected": -12.25,
-      "eval_runtime": 46.6437,
-      "eval_samples_per_second": 42.878,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": 0.150390625,
+      "eval_logits/rejected": 1.921875,
+      "eval_logps/chosen": -1256.0,
+      "eval_logps/rejected": -1352.0,
+      "eval_loss": 1.1657187938690186,
+      "eval_rewards/accuracies": 0.6171875,
+      "eval_rewards/chosen": -9.4375,
+      "eval_rewards/margins": 1.2265625,
+      "eval_rewards/rejected": -10.625,
+      "eval_runtime": 47.0171,
+      "eval_samples_per_second": 42.538,
+      "eval_steps_per_second": 0.681,
       "step": 1700
     },
     {
       "epoch": 1.7905759162303665,
-      "grad_norm": 2.2977809419919315,
+      "grad_norm": 2.0957695765984257,
       "learning_rate": 2.093575422770872e-05,
-      "logits/chosen": -4.90625,
-      "logits/rejected": -3.78125,
-      "logps/chosen": -1320.0,
-      "logps/rejected": -1920.0,
-      "loss": 0.1092,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -9.4375,
-      "rewards/margins": 5.0625,
-      "rewards/rejected": -14.5,
+      "logits/chosen": -1.2734375,
+      "logits/rejected": -1.0625,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -1768.0,
+      "loss": 0.1023,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.65625,
+      "rewards/margins": 5.25,
+      "rewards/rejected": -12.875,
       "step": 1710
     },
     {
       "epoch": 1.8010471204188483,
-      "grad_norm": 3.2986929081463185,
+      "grad_norm": 2.969497642326565,
       "learning_rate": 2.0635462181187827e-05,
-      "logits/chosen": -5.09375,
-      "logits/rejected": -3.765625,
-      "logps/chosen": -1512.0,
-      "logps/rejected": -2272.0,
-      "loss": 0.1379,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -10.9375,
-      "rewards/margins": 5.71875,
-      "rewards/rejected": -16.625,
+      "logits/chosen": -2.078125,
+      "logits/rejected": -1.5390625,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -2096.0,
+      "loss": 0.1332,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -9.25,
+      "rewards/margins": 5.75,
+      "rewards/rejected": -15.0,
       "step": 1720
     },
     {
       "epoch": 1.8115183246073299,
-      "grad_norm": 2.3041951640716314,
+      "grad_norm": 2.706907534704252,
       "learning_rate": 2.0335818271653917e-05,
-      "logits/chosen": -5.0,
-      "logits/rejected": -3.78125,
-      "logps/chosen": -1392.0,
-      "logps/rejected": -1928.0,
-      "loss": 0.1347,
+      "logits/chosen": -2.4375,
+      "logits/rejected": -1.7421875,
+      "logps/chosen": -1144.0,
+      "logps/rejected": -1704.0,
+      "loss": 0.1238,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -9.75,
-      "rewards/margins": 4.5,
-      "rewards/rejected": -14.25,
+      "rewards/chosen": -7.28125,
+      "rewards/margins": 4.6875,
+      "rewards/rejected": -12.0,
       "step": 1730
     },
     {
       "epoch": 1.8219895287958114,
-      "grad_norm": 1.882100776860651,
+      "grad_norm": 3.8817653735861977,
       "learning_rate": 2.0036866996441815e-05,
-      "logits/chosen": -6.78125,
-      "logits/rejected": -4.75,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -1816.0,
-      "loss": 0.1259,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -8.6875,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -13.3125,
+      "logits/chosen": -3.421875,
+      "logits/rejected": -2.3125,
+      "logps/chosen": -1048.0,
+      "logps/rejected": -1640.0,
+      "loss": 0.1168,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -6.4375,
+      "rewards/margins": 5.125,
+      "rewards/rejected": -11.5625,
       "step": 1740
     },
     {
       "epoch": 1.8324607329842932,
-      "grad_norm": 2.876926509381833,
+      "grad_norm": 2.91389051522039,
       "learning_rate": 1.9738652750029677e-05,
-      "logits/chosen": -5.75,
-      "logits/rejected": -4.21875,
-      "logps/chosen": -1336.0,
-      "logps/rejected": -1880.0,
-      "loss": 0.1126,
+      "logits/chosen": -1.984375,
+      "logits/rejected": -0.95703125,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -1712.0,
+      "loss": 0.1123,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 4.5625,
-      "rewards/rejected": -13.625,
+      "rewards/chosen": -7.03125,
+      "rewards/margins": 4.96875,
+      "rewards/rejected": -12.0,
       "step": 1750
     },
     {
       "epoch": 1.8429319371727748,
-      "grad_norm": 3.190643645172407,
+      "grad_norm": 2.562428000429967,
       "learning_rate": 1.944121981744631e-05,
-      "logits/chosen": -6.53125,
-      "logits/rejected": -5.09375,
-      "logps/chosen": -1416.0,
-      "logps/rejected": -1984.0,
-      "loss": 0.1348,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -9.5625,
-      "rewards/margins": 4.6875,
-      "rewards/rejected": -14.25,
+      "logits/chosen": -2.828125,
+      "logits/rejected": -1.953125,
+      "logps/chosen": -1184.0,
+      "logps/rejected": -1768.0,
+      "loss": 0.1378,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -7.3125,
+      "rewards/margins": 4.8125,
+      "rewards/rejected": -12.125,
       "step": 1760
     },
     {
       "epoch": 1.8534031413612566,
-      "grad_norm": 2.0355047401087405,
+      "grad_norm": 1.8525894005603285,
       "learning_rate": 1.9144612367694887e-05,
-      "logits/chosen": -6.84375,
-      "logits/rejected": -5.15625,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -1912.0,
-      "loss": 0.1113,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -8.8125,
-      "rewards/margins": 4.71875,
-      "rewards/rejected": -13.5625,
+      "logits/chosen": -1.8984375,
+      "logits/rejected": -1.1640625,
+      "logps/chosen": -1096.0,
+      "logps/rejected": -1744.0,
+      "loss": 0.1121,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -6.875,
+      "rewards/margins": 4.96875,
+      "rewards/rejected": -11.875,
       "step": 1770
     },
     {
       "epoch": 1.8638743455497382,
-      "grad_norm": 3.0485713847419023,
+      "grad_norm": 2.7543114660925383,
       "learning_rate": 1.8848874447193803e-05,
-      "logits/chosen": -5.6875,
-      "logits/rejected": -4.5,
-      "logps/chosen": -1288.0,
-      "logps/rejected": -1808.0,
-      "loss": 0.1335,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 4.5625,
-      "rewards/rejected": -13.5625,
+      "logits/chosen": -1.453125,
+      "logits/rejected": -0.81640625,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -1640.0,
+      "loss": 0.1106,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -7.28125,
+      "rewards/margins": 4.625,
+      "rewards/rejected": -11.875,
       "step": 1780
     },
     {
       "epoch": 1.8743455497382198,
-      "grad_norm": 2.3203017320605106,
+      "grad_norm": 3.1323838649256595,
       "learning_rate": 1.855404997323571e-05,
-      "logits/chosen": -6.53125,
-      "logits/rejected": -5.0625,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -1896.0,
-      "loss": 0.1254,
+      "logits/chosen": -1.3984375,
+      "logits/rejected": -0.97265625,
+      "logps/chosen": -1200.0,
+      "logps/rejected": -1784.0,
+      "loss": 0.1389,
       "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -8.9375,
+      "rewards/chosen": -7.84375,
       "rewards/margins": 4.75,
-      "rewards/rejected": -13.6875,
+      "rewards/rejected": -12.5625,
       "step": 1790
     },
     {
       "epoch": 1.8848167539267016,
-      "grad_norm": 1.939657098205315,
+      "grad_norm": 1.7535037123372952,
       "learning_rate": 1.82601827274658e-05,
-      "logits/chosen": -7.40625,
-      "logits/rejected": -5.6875,
-      "logps/chosen": -1168.0,
-      "logps/rejected": -1704.0,
-      "loss": 0.1227,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -7.84375,
-      "rewards/margins": 4.59375,
-      "rewards/rejected": -12.4375,
+      "logits/chosen": -2.75,
+      "logits/rejected": -2.21875,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -1608.0,
+      "loss": 0.1305,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -6.71875,
+      "rewards/margins": 4.71875,
+      "rewards/rejected": -11.4375,
       "step": 1800
     },
     {
       "epoch": 1.8848167539267016,
-      "eval_logits/chosen": -4.375,
-      "eval_logits/rejected": -2.515625,
-      "eval_logps/chosen": -1264.0,
-      "eval_logps/rejected": -1336.0,
-      "eval_loss": 1.0561875104904175,
-      "eval_rewards/accuracies": 0.611328125,
-      "eval_rewards/chosen": -9.5,
-      "eval_rewards/margins": 0.99609375,
-      "eval_rewards/rejected": -10.5,
-      "eval_runtime": 46.6378,
-      "eval_samples_per_second": 42.884,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": -0.44921875,
+      "eval_logits/rejected": 1.34375,
+      "eval_logps/chosen": -1112.0,
+      "eval_logps/rejected": -1176.0,
+      "eval_loss": 1.0662031173706055,
+      "eval_rewards/accuracies": 0.580078125,
+      "eval_rewards/chosen": -7.9375,
+      "eval_rewards/margins": 0.96484375,
+      "eval_rewards/rejected": -8.875,
+      "eval_runtime": 47.1062,
+      "eval_samples_per_second": 42.457,
+      "eval_steps_per_second": 0.679,
       "step": 1800
     },
     {
       "epoch": 1.8952879581151834,
-      "grad_norm": 3.307246733433114,
+      "grad_norm": 3.6509572274513666,
       "learning_rate": 1.7967316349380213e-05,
-      "logits/chosen": -6.1875,
-      "logits/rejected": -4.90625,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -1752.0,
-      "loss": 0.1175,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -8.25,
-      "rewards/margins": 4.46875,
-      "rewards/rejected": -12.75,
+      "logits/chosen": -3.25,
+      "logits/rejected": -2.65625,
+      "logps/chosen": -1020.0,
+      "logps/rejected": -1544.0,
+      "loss": 0.1121,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -6.375,
+      "rewards/margins": 4.3125,
+      "rewards/rejected": -10.6875,
       "step": 1810
     },
     {
       "epoch": 1.905759162303665,
-      "grad_norm": 2.556689093178943,
+      "grad_norm": 3.2949368379432973,
       "learning_rate": 1.7675494329845516e-05,
-      "logits/chosen": -5.09375,
-      "logits/rejected": -3.453125,
-      "logps/chosen": -1304.0,
-      "logps/rejected": -1872.0,
-      "loss": 0.1214,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 4.78125,
-      "rewards/rejected": -13.8125,
+      "logits/chosen": -1.671875,
+      "logits/rejected": -0.62109375,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -1680.0,
+      "loss": 0.1198,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -7.15625,
+      "rewards/margins": 4.75,
+      "rewards/rejected": -11.9375,
       "step": 1820
     },
     {
       "epoch": 1.9162303664921465,
-      "grad_norm": 1.5700886379294252,
+      "grad_norm": 1.2881759321461188,
       "learning_rate": 1.738476000464026e-05,
-      "logits/chosen": -4.46875,
-      "logits/rejected": -3.09375,
-      "logps/chosen": -1360.0,
-      "logps/rejected": -2000.0,
-      "loss": 0.1272,
+      "logits/chosen": -0.6171875,
+      "logits/rejected": -0.03076171875,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -1920.0,
+      "loss": 0.1236,
       "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -9.5625,
-      "rewards/margins": 5.1875,
-      "rewards/rejected": -14.75,
+      "rewards/chosen": -8.4375,
+      "rewards/margins": 5.4375,
+      "rewards/rejected": -13.875,
       "step": 1830
     },
     {
       "epoch": 1.9267015706806283,
-      "grad_norm": 1.641614792050765,
+      "grad_norm": 1.415482103371834,
       "learning_rate": 1.7095156548019646e-05,
-      "logits/chosen": -5.21875,
-      "logits/rejected": -4.09375,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -1992.0,
-      "loss": 0.0992,
-      "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -9.25,
-      "rewards/margins": 5.21875,
-      "rewards/rejected": -14.5,
+      "logits/chosen": -1.6953125,
+      "logits/rejected": -1.5078125,
+      "logps/chosen": -1240.0,
+      "logps/rejected": -1936.0,
+      "loss": 0.1013,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -8.5625,
+      "rewards/margins": 5.34375,
+      "rewards/rejected": -13.875,
       "step": 1840
     },
     {
       "epoch": 1.93717277486911,
-      "grad_norm": 2.2020116638544858,
+      "grad_norm": 2.1595786386792026,
       "learning_rate": 1.6806726966304064e-05,
-      "logits/chosen": -6.15625,
-      "logits/rejected": -4.75,
-      "logps/chosen": -1312.0,
-      "logps/rejected": -1880.0,
-      "loss": 0.1283,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -8.75,
-      "rewards/margins": 4.875,
-      "rewards/rejected": -13.625,
+      "logits/chosen": -3.0625,
+      "logits/rejected": -2.28125,
+      "logps/chosen": -1248.0,
+      "logps/rejected": -1832.0,
+      "loss": 0.1317,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -8.125,
+      "rewards/margins": 5.03125,
+      "rewards/rejected": -13.125,
       "step": 1850
     },
     {
       "epoch": 1.9476439790575917,
-      "grad_norm": 3.6806777570651836,
+      "grad_norm": 4.977505849038182,
       "learning_rate": 1.6519514091492625e-05,
-      "logits/chosen": -6.1875,
-      "logits/rejected": -4.65625,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -1784.0,
-      "loss": 0.1485,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -8.6875,
+      "logits/chosen": -2.890625,
+      "logits/rejected": -2.140625,
+      "logps/chosen": -1128.0,
+      "logps/rejected": -1672.0,
+      "loss": 0.1685,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -7.5625,
       "rewards/margins": 4.4375,
-      "rewards/rejected": -13.125,
+      "rewards/rejected": -12.0,
       "step": 1860
     },
     {
       "epoch": 1.9581151832460733,
-      "grad_norm": 2.4683894236598647,
+      "grad_norm": 2.0289069166448437,
       "learning_rate": 1.6233560574902627e-05,
-      "logits/chosen": -6.21875,
-      "logits/rejected": -4.875,
-      "logps/chosen": -1208.0,
-      "logps/rejected": -1792.0,
-      "loss": 0.1229,
-      "rewards/accuracies": 0.949999988079071,
-      "rewards/chosen": -8.5,
-      "rewards/margins": 4.71875,
-      "rewards/rejected": -13.1875,
+      "logits/chosen": -2.046875,
+      "logits/rejected": -1.7734375,
+      "logps/chosen": -1104.0,
+      "logps/rejected": -1680.0,
+      "loss": 0.1239,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -7.5,
+      "rewards/margins": 4.59375,
+      "rewards/rejected": -12.125,
       "step": 1870
     },
     {
       "epoch": 1.9685863874345548,
-      "grad_norm": 1.8432384006404061,
+      "grad_norm": 1.8696097188199352,
       "learning_rate": 1.5948908880835753e-05,
-      "logits/chosen": -6.1875,
-      "logits/rejected": -4.4375,
-      "logps/chosen": -1240.0,
-      "logps/rejected": -1792.0,
-      "loss": 0.1205,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -8.375,
-      "rewards/margins": 4.875,
-      "rewards/rejected": -13.25,
+      "logits/chosen": -1.7265625,
+      "logits/rejected": -0.984375,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -1696.0,
+      "loss": 0.1196,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -7.375,
+      "rewards/margins": 4.90625,
+      "rewards/rejected": -12.3125,
       "step": 1880
     },
     {
       "epoch": 1.9790575916230366,
-      "grad_norm": 2.8078625046549783,
+      "grad_norm": 1.7499419294576035,
       "learning_rate": 1.5665601280272124e-05,
-      "logits/chosen": -6.09375,
-      "logits/rejected": -4.40625,
-      "logps/chosen": -1296.0,
-      "logps/rejected": -1776.0,
-      "loss": 0.1267,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 4.1875,
-      "rewards/rejected": -13.1875,
+      "logits/chosen": -1.8671875,
+      "logits/rejected": -0.734375,
+      "logps/chosen": -1232.0,
+      "logps/rejected": -1752.0,
+      "loss": 0.1238,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -8.375,
+      "rewards/margins": 4.59375,
+      "rewards/rejected": -12.9375,
       "step": 1890
     },
     {
       "epoch": 1.9895287958115184,
-      "grad_norm": 2.930621737497199,
+      "grad_norm": 3.355068222272933,
       "learning_rate": 1.538367984459298e-05,
-      "logits/chosen": -5.09375,
-      "logits/rejected": -3.25,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -1872.0,
-      "loss": 0.1119,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -9.0,
-      "rewards/margins": 5.0625,
-      "rewards/rejected": -14.0625,
+      "logits/chosen": -0.58984375,
+      "logits/rejected": 0.2001953125,
+      "logps/chosen": -1176.0,
+      "logps/rejected": -1776.0,
+      "loss": 0.1205,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -7.96875,
+      "rewards/margins": 5.1875,
+      "rewards/rejected": -13.125,
       "step": 1900
     },
     {
       "epoch": 1.9895287958115184,
-      "eval_logits/chosen": -3.109375,
-      "eval_logits/rejected": -1.2734375,
-      "eval_logps/chosen": -1448.0,
-      "eval_logps/rejected": -1544.0,
-      "eval_loss": 1.1632968187332153,
-      "eval_rewards/accuracies": 0.599609375,
-      "eval_rewards/chosen": -11.3125,
-      "eval_rewards/margins": 1.234375,
-      "eval_rewards/rejected": -12.5625,
-      "eval_runtime": 46.6424,
-      "eval_samples_per_second": 42.879,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": 0.84765625,
+      "eval_logits/rejected": 2.5625,
+      "eval_logps/chosen": -1296.0,
+      "eval_logps/rejected": -1384.0,
+      "eval_loss": 1.169265627861023,
+      "eval_rewards/accuracies": 0.583984375,
+      "eval_rewards/chosen": -9.8125,
+      "eval_rewards/margins": 1.1328125,
+      "eval_rewards/rejected": -10.9375,
+      "eval_runtime": 47.1112,
+      "eval_samples_per_second": 42.453,
+      "eval_steps_per_second": 0.679,
       "step": 1900
     },
     {
       "epoch": 2.0,
-      "grad_norm": 1.6684214980595502,
+      "grad_norm": 1.4265551523217805,
       "learning_rate": 1.5103186439333133e-05,
-      "logits/chosen": -4.59375,
-      "logits/rejected": -3.15625,
-      "logps/chosen": -1272.0,
-      "logps/rejected": -1952.0,
-      "loss": 0.1129,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -8.9375,
-      "rewards/margins": 5.3125,
-      "rewards/rejected": -14.25,
+      "logits/chosen": -1.234375,
+      "logits/rejected": -0.91796875,
+      "logps/chosen": -1120.0,
+      "logps/rejected": -1824.0,
+      "loss": 0.1166,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -7.4375,
+      "rewards/margins": 5.5625,
+      "rewards/rejected": -13.0,
       "step": 1910
     },
     {
       "epoch": 2.0104712041884816,
-      "grad_norm": 0.13112525115152174,
+      "grad_norm": 0.1558878005903202,
       "learning_rate": 1.4824162717963829e-05,
-      "logits/chosen": -4.90625,
-      "logits/rejected": -2.140625,
-      "logps/chosen": -1224.0,
-      "logps/rejected": -1944.0,
-      "loss": 0.0193,
+      "logits/chosen": -2.046875,
+      "logits/rejected": -0.640625,
+      "logps/chosen": -1056.0,
+      "logps/rejected": -1744.0,
+      "loss": 0.0198,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -8.3125,
-      "rewards/margins": 6.65625,
-      "rewards/rejected": -14.9375,
+      "rewards/chosen": -6.65625,
+      "rewards/margins": 6.28125,
+      "rewards/rejected": -12.9375,
       "step": 1920
     },
     {
       "epoch": 2.020942408376963,
-      "grad_norm": 1.189860298324291,
+      "grad_norm": 0.11973742901260617,
       "learning_rate": 1.4546650115707192e-05,
-      "logits/chosen": -4.03125,
-      "logits/rejected": -2.1875,
-      "logps/chosen": -1280.0,
-      "logps/rejected": -2096.0,
-      "loss": 0.0188,
+      "logits/chosen": -2.15625,
+      "logits/rejected": -1.4140625,
+      "logps/chosen": -1136.0,
+      "logps/rejected": -1912.0,
+      "loss": 0.0173,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.0625,
-      "rewards/margins": 7.21875,
-      "rewards/rejected": -16.25,
+      "rewards/chosen": -7.53125,
+      "rewards/margins": 6.90625,
+      "rewards/rejected": -14.4375,
       "step": 1930
     },
     {
       "epoch": 2.031413612565445,
-      "grad_norm": 0.14484999474340313,
+      "grad_norm": 0.37575907796650043,
       "learning_rate": 1.4270689843383111e-05,
-      "logits/chosen": -3.0,
-      "logits/rejected": -0.71875,
-      "logps/chosen": -1368.0,
-      "logps/rejected": -2192.0,
-      "loss": 0.0242,
+      "logits/chosen": -0.71484375,
+      "logits/rejected": 0.330078125,
+      "logps/chosen": -1272.0,
+      "logps/rejected": -2112.0,
+      "loss": 0.0217,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -9.5625,
-      "rewards/margins": 7.28125,
-      "rewards/rejected": -16.875,
+      "rewards/chosen": -8.625,
+      "rewards/margins": 7.3125,
+      "rewards/rejected": -15.9375,
       "step": 1940
     },
     {
       "epoch": 2.0418848167539267,
-      "grad_norm": 1.6217742970691786,
+      "grad_norm": 2.4737543691223167,
       "learning_rate": 1.3996322881289347e-05,
-      "logits/chosen": -3.453125,
-      "logits/rejected": -1.0,
-      "logps/chosen": -1384.0,
-      "logps/rejected": -2256.0,
-      "loss": 0.0259,
+      "logits/chosen": -1.0859375,
+      "logits/rejected": 0.2265625,
+      "logps/chosen": -1304.0,
+      "logps/rejected": -2240.0,
+      "loss": 0.0301,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -10.125,
-      "rewards/margins": 7.59375,
-      "rewards/rejected": -17.75,
+      "rewards/chosen": -9.375,
+      "rewards/margins": 8.1875,
+      "rewards/rejected": -17.5,
       "step": 1950
     },
     {
       "epoch": 2.0523560209424083,
-      "grad_norm": 0.11021242979116014,
+      "grad_norm": 0.07046893422043043,
       "learning_rate": 1.3723589973115963e-05,
-      "logits/chosen": -2.515625,
-      "logits/rejected": -0.51171875,
-      "logps/chosen": -1424.0,
-      "logps/rejected": -2272.0,
-      "loss": 0.0179,
+      "logits/chosen": -0.234375,
+      "logits/rejected": 0.64453125,
+      "logps/chosen": -1320.0,
+      "logps/rejected": -2224.0,
+      "loss": 0.0175,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.5,
-      "rewards/margins": 7.65625,
-      "rewards/rejected": -18.125,
+      "rewards/chosen": -9.4375,
+      "rewards/margins": 8.3125,
+      "rewards/rejected": -17.75,
       "step": 1960
     },
     {
       "epoch": 2.06282722513089,
-      "grad_norm": 0.5628366758197821,
+      "grad_norm": 0.3476925796558248,
       "learning_rate": 1.3452531619894837e-05,
-      "logits/chosen": -2.375,
-      "logits/rejected": -0.68359375,
-      "logps/chosen": -1472.0,
-      "logps/rejected": -2448.0,
-      "loss": 0.0157,
+      "logits/chosen": 0.4609375,
+      "logits/rejected": 0.71484375,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.0172,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -10.875,
-      "rewards/margins": 8.1875,
-      "rewards/rejected": -19.0,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 8.5625,
+      "rewards/rejected": -18.875,
       "step": 1970
     },
     {
       "epoch": 2.073298429319372,
-      "grad_norm": 0.2900776842492875,
+      "grad_norm": 0.49193736319365683,
       "learning_rate": 1.3183188073985245e-05,
-      "logits/chosen": -3.265625,
-      "logits/rejected": -1.1640625,
-      "logps/chosen": -1496.0,
-      "logps/rejected": -2512.0,
-      "loss": 0.0232,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -20.0,
+      "logits/chosen": -0.40234375,
+      "logits/rejected": 0.357421875,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.0188,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.0625,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -19.125,
       "step": 1980
     },
     {
       "epoch": 2.0837696335078535,
-      "grad_norm": 0.26368985361388025,
+      "grad_norm": 0.10462393480862757,
       "learning_rate": 1.2915599333096351e-05,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -1.75,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2384.0,
-      "loss": 0.0176,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.0,
-      "rewards/margins": 7.8125,
-      "rewards/rejected": -18.875,
+      "logits/chosen": -0.8671875,
+      "logits/rejected": -0.388671875,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -2352.0,
+      "loss": 0.0162,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 8.25,
+      "rewards/rejected": -18.5,
       "step": 1990
     },
     {
       "epoch": 2.094240837696335,
-      "grad_norm": 0.23935043680853396,
+      "grad_norm": 0.28462723328541073,
       "learning_rate": 1.264980513434752e-05,
-      "logits/chosen": -3.390625,
-      "logits/rejected": -1.1640625,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2416.0,
-      "loss": 0.0166,
+      "logits/chosen": -0.388671875,
+      "logits/rejected": 0.83984375,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2496.0,
+      "loss": 0.0161,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -10.625,
-      "rewards/margins": 8.125,
-      "rewards/rejected": -18.75,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -19.625,
       "step": 2000
     },
     {
       "epoch": 2.094240837696335,
-      "eval_logits/chosen": 0.054931640625,
-      "eval_logits/rejected": 1.9375,
-      "eval_logps/chosen": -1744.0,
+      "eval_logits/chosen": 2.84375,
+      "eval_logits/rejected": 4.625,
+      "eval_logps/chosen": -1736.0,
       "eval_logps/rejected": -1864.0,
-      "eval_loss": 1.5514531135559082,
-      "eval_rewards/accuracies": 0.583984375,
-      "eval_rewards/chosen": -14.25,
-      "eval_rewards/margins": 1.5234375,
+      "eval_loss": 1.7203906774520874,
+      "eval_rewards/accuracies": 0.591796875,
+      "eval_rewards/chosen": -14.1875,
+      "eval_rewards/margins": 1.5625,
       "eval_rewards/rejected": -15.75,
-      "eval_runtime": 46.6515,
-      "eval_samples_per_second": 42.871,
-      "eval_steps_per_second": 0.686,
+      "eval_runtime": 47.0893,
+      "eval_samples_per_second": 42.473,
+      "eval_steps_per_second": 0.68,
       "step": 2000
     },
     {
       "epoch": 2.1047120418848166,
-      "grad_norm": 0.18125247281437176,
+      "grad_norm": 0.43064574642558345,
       "learning_rate": 1.2385844948367324e-05,
-      "logits/chosen": -2.828125,
-      "logits/rejected": -0.875,
-      "logps/chosen": -1512.0,
-      "logps/rejected": -2336.0,
-      "loss": 0.0341,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -11.4375,
-      "rewards/margins": 7.625,
-      "rewards/rejected": -19.0,
+      "logits/chosen": -0.072265625,
+      "logits/rejected": 0.83984375,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -2352.0,
+      "loss": 0.0265,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -10.75,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -19.25,
       "step": 2010
     },
     {
       "epoch": 2.115183246073298,
-      "grad_norm": 0.29229470864014834,
+      "grad_norm": 0.11341181702809795,
       "learning_rate": 1.2123757973432114e-05,
-      "logits/chosen": -3.4375,
-      "logits/rejected": -1.4296875,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2336.0,
-      "loss": 0.0148,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -10.9375,
-      "rewards/margins": 7.625,
+      "logits/chosen": -0.73046875,
+      "logits/rejected": 0.458984375,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2352.0,
+      "loss": 0.0184,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -10.5,
+      "rewards/margins": 8.125,
       "rewards/rejected": -18.625,
       "step": 2020
     },
     {
       "epoch": 2.1256544502617802,
-      "grad_norm": 0.30795366200482666,
+      "grad_norm": 0.16304768581546453,
       "learning_rate": 1.1863583129645106e-05,
-      "logits/chosen": -3.078125,
-      "logits/rejected": -1.1953125,
-      "logps/chosen": -1488.0,
-      "logps/rejected": -2448.0,
-      "loss": 0.0222,
-      "rewards/accuracies": 0.96875,
-      "rewards/chosen": -10.9375,
-      "rewards/margins": 8.4375,
-      "rewards/rejected": -19.375,
+      "logits/chosen": -0.154296875,
+      "logits/rejected": 0.4765625,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -2400.0,
+      "loss": 0.0227,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 8.625,
+      "rewards/rejected": -18.875,
       "step": 2030
     },
     {
       "epoch": 2.136125654450262,
-      "grad_norm": 0.23652450702189734,
+      "grad_norm": 0.5673517577548601,
       "learning_rate": 1.1605359053156607e-05,
-      "logits/chosen": -3.234375,
-      "logits/rejected": -1.0703125,
-      "logps/chosen": -1584.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0144,
+      "logits/chosen": -0.76953125,
+      "logits/rejected": 0.41015625,
+      "logps/chosen": -1480.0,
+      "logps/rejected": -2496.0,
+      "loss": 0.017,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 8.5,
-      "rewards/rejected": -20.125,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 8.875,
+      "rewards/rejected": -19.5,
       "step": 2040
     },
     {
       "epoch": 2.1465968586387434,
-      "grad_norm": 0.30147902289963274,
+      "grad_norm": 0.23647477849108278,
       "learning_rate": 1.1349124090426644e-05,
-      "logits/chosen": -2.9375,
-      "logits/rejected": -0.65625,
-      "logps/chosen": -1544.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0206,
+      "logits/chosen": -0.54296875,
+      "logits/rejected": 0.703125,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0153,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.375,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -20.625,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -19.625,
       "step": 2050
     },
     {
       "epoch": 2.157068062827225,
-      "grad_norm": 0.5500534888475488,
+      "grad_norm": 0.13432733743096906,
       "learning_rate": 1.1094916292530406e-05,
-      "logits/chosen": -2.8125,
-      "logits/rejected": -1.25,
-      "logps/chosen": -1480.0,
-      "logps/rejected": -2416.0,
-      "loss": 0.0158,
+      "logits/chosen": -0.1015625,
+      "logits/rejected": 0.55859375,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.0198,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.625,
-      "rewards/margins": 8.375,
-      "rewards/rejected": -19.0,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -19.25,
       "step": 2060
     },
     {
       "epoch": 2.167539267015707,
-      "grad_norm": 0.2543616840160018,
+      "grad_norm": 0.6383658370553249,
       "learning_rate": 1.0842773409507623e-05,
-      "logits/chosen": -3.046875,
-      "logits/rejected": -0.6796875,
-      "logps/chosen": -1432.0,
-      "logps/rejected": -2384.0,
-      "loss": 0.0189,
+      "logits/chosen": 0.228515625,
+      "logits/rejected": 1.125,
+      "logps/chosen": -1360.0,
+      "logps/rejected": -2352.0,
+      "loss": 0.0243,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.625,
-      "rewards/margins": 8.4375,
-      "rewards/rejected": -19.125,
+      "rewards/chosen": -9.9375,
+      "rewards/margins": 8.75,
+      "rewards/rejected": -18.75,
       "step": 2070
     },
     {
       "epoch": 2.1780104712041886,
-      "grad_norm": 1.547753200569628,
+      "grad_norm": 0.6941250172518184,
       "learning_rate": 1.0592732884756754e-05,
-      "logits/chosen": -3.265625,
-      "logits/rejected": -1.0390625,
-      "logps/chosen": -1632.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0167,
+      "logits/chosen": -0.890625,
+      "logits/rejected": -0.08056640625,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.013,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.125,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -21.375,
+      "rewards/chosen": -10.8125,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -19.875,
       "step": 2080
     },
     {
       "epoch": 2.18848167539267,
-      "grad_norm": 0.09366111619278192,
+      "grad_norm": 0.16763126153838873,
       "learning_rate": 1.0344831849474507e-05,
-      "logits/chosen": -2.40625,
-      "logits/rejected": -0.12255859375,
-      "logps/chosen": -1680.0,
-      "logps/rejected": -2672.0,
+      "logits/chosen": -0.23828125,
+      "logits/rejected": 0.76953125,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -2480.0,
       "loss": 0.0181,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -12.625,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -21.625,
+      "rewards/chosen": -10.8125,
+      "rewards/margins": 8.75,
+      "rewards/rejected": -19.625,
       "step": 2090
     },
     {
       "epoch": 2.1989528795811517,
-      "grad_norm": 0.2173202996713778,
+      "grad_norm": 0.9027101090553931,
       "learning_rate": 1.0099107117141879e-05,
-      "logits/chosen": -2.796875,
-      "logits/rejected": -0.6640625,
-      "logps/chosen": -1536.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0122,
+      "logits/chosen": 0.2265625,
+      "logits/rejected": 0.93359375,
+      "logps/chosen": -1352.0,
+      "logps/rejected": -2352.0,
+      "loss": 0.0152,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 8.75,
-      "rewards/rejected": -20.375,
+      "rewards/chosen": -9.6875,
+      "rewards/margins": 8.875,
+      "rewards/rejected": -18.625,
       "step": 2100
     },
     {
       "epoch": 2.1989528795811517,
-      "eval_logits/chosen": -0.412109375,
-      "eval_logits/rejected": 1.5703125,
-      "eval_logps/chosen": -1800.0,
-      "eval_logps/rejected": -1936.0,
-      "eval_loss": 1.669374942779541,
-      "eval_rewards/accuracies": 0.609375,
-      "eval_rewards/chosen": -14.8125,
-      "eval_rewards/margins": 1.6484375,
-      "eval_rewards/rejected": -16.5,
-      "eval_runtime": 46.6466,
-      "eval_samples_per_second": 42.876,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": 2.328125,
+      "eval_logits/rejected": 4.0625,
+      "eval_logps/chosen": -1624.0,
+      "eval_logps/rejected": -1760.0,
+      "eval_loss": 1.5836405754089355,
+      "eval_rewards/accuracies": 0.591796875,
+      "eval_rewards/chosen": -13.125,
+      "eval_rewards/margins": 1.59375,
+      "eval_rewards/rejected": -14.6875,
+      "eval_runtime": 47.107,
+      "eval_samples_per_second": 42.457,
+      "eval_steps_per_second": 0.679,
       "step": 2100
     },
     {
       "epoch": 2.2094240837696333,
-      "grad_norm": 0.5829882131071631,
+      "grad_norm": 0.7589391273413818,
       "learning_rate": 9.855595178057334e-06,
-      "logits/chosen": -3.421875,
-      "logits/rejected": -1.4453125,
-      "logps/chosen": -1568.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0152,
+      "logits/chosen": -0.62890625,
+      "logits/rejected": 0.048583984375,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.015,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.4375,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -20.625,
+      "rewards/chosen": -9.9375,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -18.875,
       "step": 2110
     },
     {
       "epoch": 2.2198952879581153,
-      "grad_norm": 0.2424013055067862,
+      "grad_norm": 0.08411781196723762,
       "learning_rate": 9.614332193917949e-06,
-      "logits/chosen": -3.53125,
-      "logits/rejected": -0.76171875,
-      "logps/chosen": -1696.0,
-      "logps/rejected": -2720.0,
-      "loss": 0.0168,
+      "logits/chosen": -0.2412109375,
+      "logits/rejected": 0.5859375,
+      "logps/chosen": -1528.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0194,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -12.5625,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -21.875,
+      "rewards/chosen": -10.9375,
+      "rewards/margins": 8.9375,
+      "rewards/rejected": -19.875,
       "step": 2120
     },
     {
       "epoch": 2.230366492146597,
-      "grad_norm": 0.27028475654174455,
+      "grad_norm": 0.41233621808486143,
       "learning_rate": 9.375353992449385e-06,
-      "logits/chosen": -3.34375,
-      "logits/rejected": -1.5078125,
-      "logps/chosen": -1640.0,
-      "logps/rejected": -2704.0,
-      "loss": 0.0127,
+      "logits/chosen": -0.625,
+      "logits/rejected": -0.1669921875,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.0151,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -12.4375,
-      "rewards/margins": 9.5,
-      "rewards/rejected": -22.0,
+      "rewards/chosen": -10.75,
+      "rewards/margins": 8.875,
+      "rewards/rejected": -19.625,
       "step": 2130
     },
     {
       "epoch": 2.2408376963350785,
-      "grad_norm": 0.8848690452153692,
+      "grad_norm": 0.7203027483383831,
       "learning_rate": 9.138696062085442e-06,
-      "logits/chosen": -3.203125,
-      "logits/rejected": -0.94140625,
-      "logps/chosen": -1616.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.025,
+      "logits/chosen": -0.359375,
+      "logits/rejected": 0.3984375,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.0149,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.1875,
-      "rewards/margins": 9.0625,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -10.375,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -19.5,
       "step": 2140
     },
     {
       "epoch": 2.25130890052356,
-      "grad_norm": 0.4601370478678084,
+      "grad_norm": 0.13038083252395144,
       "learning_rate": 8.904393546698007e-06,
-      "logits/chosen": -2.96875,
-      "logits/rejected": -1.234375,
-      "logps/chosen": -1608.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.016,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -12.0,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -21.125,
+      "logits/chosen": -0.67578125,
+      "logits/rejected": -0.1015625,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2576.0,
+      "loss": 0.0178,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -11.375,
+      "rewards/margins": 9.25,
+      "rewards/rejected": -20.625,
       "step": 2150
     },
     {
       "epoch": 2.261780104712042,
-      "grad_norm": 0.2803650810497035,
+      "grad_norm": 0.36178107880695465,
       "learning_rate": 8.672481240378142e-06,
-      "logits/chosen": -2.40625,
-      "logits/rejected": -0.302734375,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0238,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.5625,
-      "rewards/margins": 8.375,
-      "rewards/rejected": -19.875,
+      "logits/chosen": 0.234375,
+      "logits/rejected": 1.09375,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0224,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.625,
+      "rewards/margins": 8.9375,
+      "rewards/rejected": -19.625,
       "step": 2160
     },
     {
       "epoch": 2.2722513089005236,
-      "grad_norm": 0.3295121660857614,
+      "grad_norm": 0.2868947377558203,
       "learning_rate": 8.44299358226919e-06,
-      "logits/chosen": -2.328125,
-      "logits/rejected": 0.2734375,
-      "logps/chosen": -1592.0,
-      "logps/rejected": -2544.0,
-      "loss": 0.0257,
-      "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -12.0,
-      "rewards/margins": 8.625,
-      "rewards/rejected": -20.625,
+      "logits/chosen": 0.2431640625,
+      "logits/rejected": 1.609375,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0204,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -11.5,
+      "rewards/margins": 8.875,
+      "rewards/rejected": -20.375,
       "step": 2170
     },
     {
       "epoch": 2.282722513089005,
-      "grad_norm": 0.35860568378176194,
+      "grad_norm": 0.16423230582158319,
       "learning_rate": 8.215964651452457e-06,
-      "logits/chosen": -2.265625,
-      "logits/rejected": 0.0341796875,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0143,
+      "logits/chosen": 0.412109375,
+      "logits/rejected": 1.4765625,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0128,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -11.8125,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -20.625,
+      "rewards/chosen": -11.375,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -20.875,
       "step": 2180
     },
     {
       "epoch": 2.2931937172774868,
-      "grad_norm": 0.15758545486146067,
+      "grad_norm": 0.1777372600100184,
       "learning_rate": 7.991428161886502e-06,
-      "logits/chosen": -2.546875,
-      "logits/rejected": -0.43359375,
-      "logps/chosen": -1632.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.013,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.0,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -21.25,
+      "logits/chosen": 0.25,
+      "logits/rejected": 0.99609375,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0138,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -11.125,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -20.875,
       "step": 2190
     },
     {
       "epoch": 2.303664921465969,
-      "grad_norm": 0.3540106891584038,
+      "grad_norm": 0.08093567931258469,
       "learning_rate": 7.76941745740061e-06,
-      "logits/chosen": -2.578125,
-      "logits/rejected": -0.56640625,
-      "logps/chosen": -1712.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.0158,
+      "logits/chosen": -0.53515625,
+      "logits/rejected": 0.431640625,
+      "logps/chosen": -1576.0,
+      "logps/rejected": -2640.0,
+      "loss": 0.0138,
       "rewards/accuracies": 0.96875,
-      "rewards/chosen": -12.9375,
+      "rewards/chosen": -11.5,
       "rewards/margins": 9.4375,
-      "rewards/rejected": -22.375,
+      "rewards/rejected": -21.0,
       "step": 2200
     },
     {
       "epoch": 2.303664921465969,
-      "eval_logits/chosen": 0.64453125,
-      "eval_logits/rejected": 2.515625,
-      "eval_logps/chosen": -1904.0,
-      "eval_logps/rejected": -2040.0,
-      "eval_loss": 1.791171908378601,
-      "eval_rewards/accuracies": 0.603515625,
-      "eval_rewards/chosen": -15.8125,
-      "eval_rewards/margins": 1.6875,
-      "eval_rewards/rejected": -17.5,
-      "eval_runtime": 46.6495,
-      "eval_samples_per_second": 42.873,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": 2.8125,
+      "eval_logits/rejected": 4.5625,
+      "eval_logps/chosen": -1728.0,
+      "eval_logps/rejected": -1856.0,
+      "eval_loss": 1.7811719179153442,
+      "eval_rewards/accuracies": 0.587890625,
+      "eval_rewards/chosen": -14.125,
+      "eval_rewards/margins": 1.6015625,
+      "eval_rewards/rejected": -15.6875,
+      "eval_runtime": 47.1151,
+      "eval_samples_per_second": 42.449,
+      "eval_steps_per_second": 0.679,
       "step": 2200
     },
     {
       "epoch": 2.3141361256544504,
-      "grad_norm": 0.2401128865298814,
+      "grad_norm": 0.05537550847015806,
       "learning_rate": 7.549965506743123e-06,
-      "logits/chosen": -1.9140625,
-      "logits/rejected": -0.13671875,
-      "logps/chosen": -1624.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0231,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.375,
-      "rewards/margins": 8.75,
-      "rewards/rejected": -21.125,
+      "logits/chosen": 0.2470703125,
+      "logits/rejected": 0.9765625,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2384.0,
+      "loss": 0.0226,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -10.875,
+      "rewards/margins": 8.5625,
+      "rewards/rejected": -19.375,
       "step": 2210
     },
     {
       "epoch": 2.324607329842932,
-      "grad_norm": 0.0730586595349473,
+      "grad_norm": 0.33951267065745216,
       "learning_rate": 7.333104898685631e-06,
-      "logits/chosen": -2.625,
-      "logits/rejected": -0.7890625,
-      "logps/chosen": -1560.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0183,
+      "logits/chosen": -0.7578125,
+      "logits/rejected": 0.057373046875,
+      "logps/chosen": -1424.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0218,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 8.75,
-      "rewards/rejected": -20.375,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 8.9375,
+      "rewards/rejected": -19.25,
       "step": 2220
     },
     {
       "epoch": 2.3350785340314135,
-      "grad_norm": 0.1684457172881231,
+      "grad_norm": 0.26553766275724483,
       "learning_rate": 7.118867837183518e-06,
-      "logits/chosen": -3.265625,
-      "logits/rejected": -0.96484375,
-      "logps/chosen": -1592.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0184,
+      "logits/chosen": -1.484375,
+      "logits/rejected": -0.09423828125,
+      "logps/chosen": -1504.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0156,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.1875,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -20.5,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 9.4375,
+      "rewards/rejected": -19.75,
       "step": 2230
     },
     {
       "epoch": 2.345549738219895,
-      "grad_norm": 1.0109715083282962,
+      "grad_norm": 0.20959716156360483,
       "learning_rate": 6.9072861365936045e-06,
-      "logits/chosen": -3.34375,
-      "logits/rejected": -1.3671875,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0151,
+      "logits/chosen": -1.0,
+      "logits/rejected": -0.451171875,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2576.0,
+      "loss": 0.0168,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.375,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -20.375,
+      "rewards/chosen": -10.75,
+      "rewards/margins": 8.875,
+      "rewards/rejected": -19.625,
       "step": 2240
     },
     {
       "epoch": 2.356020942408377,
-      "grad_norm": 0.8886833189304637,
+      "grad_norm": 1.111880569888357,
       "learning_rate": 6.6983912169497005e-06,
-      "logits/chosen": -2.515625,
-      "logits/rejected": -0.3359375,
-      "logps/chosen": -1544.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0192,
+      "logits/chosen": -0.625,
+      "logits/rejected": 0.5546875,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.0184,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.0625,
-      "rewards/margins": 8.75,
-      "rewards/rejected": -19.75,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -19.0,
       "step": 2250
     },
     {
       "epoch": 2.3664921465968587,
-      "grad_norm": 0.13708274798450046,
+      "grad_norm": 0.1438874455745137,
       "learning_rate": 6.49221409929677e-06,
-      "logits/chosen": -1.359375,
-      "logits/rejected": 0.69921875,
-      "logps/chosen": -1512.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0113,
+      "logits/chosen": 0.421875,
+      "logits/rejected": 1.375,
+      "logps/chosen": -1384.0,
+      "logps/rejected": -2368.0,
+      "loss": 0.0121,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.3125,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -20.25,
+      "rewards/chosen": -10.0,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -18.875,
       "step": 2260
     },
     {
       "epoch": 2.3769633507853403,
-      "grad_norm": 1.8834904381998978,
+      "grad_norm": 0.42465253279236465,
       "learning_rate": 6.2887854010842075e-06,
-      "logits/chosen": -1.875,
-      "logits/rejected": -0.62890625,
-      "logps/chosen": -1560.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0152,
+      "logits/chosen": -0.1494140625,
+      "logits/rejected": 0.1083984375,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0141,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -11.6875,
-      "rewards/margins": 9.0625,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -19.625,
       "step": 2270
     },
     {
       "epoch": 2.387434554973822,
-      "grad_norm": 0.33036446598709845,
+      "grad_norm": 0.15618632105590635,
       "learning_rate": 6.0881353316191385e-06,
-      "logits/chosen": -2.3125,
-      "logits/rejected": -0.8046875,
-      "logps/chosen": -1608.0,
-      "logps/rejected": -2544.0,
-      "loss": 0.0151,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.875,
-      "rewards/margins": 8.5625,
-      "rewards/rejected": -20.375,
+      "logits/chosen": -0.72265625,
+      "logits/rejected": -0.06884765625,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.015,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -10.6875,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -19.75,
       "step": 2280
     },
     {
       "epoch": 2.3979057591623034,
-      "grad_norm": 0.5536309572727743,
+      "grad_norm": 0.20886121996596013,
       "learning_rate": 5.890293687580381e-06,
-      "logits/chosen": -1.8671875,
-      "logits/rejected": 0.2060546875,
-      "logps/chosen": -1616.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0174,
+      "logits/chosen": 0.107421875,
+      "logits/rejected": 1.2578125,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0178,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.0,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -10.6875,
+      "rewards/margins": 8.875,
+      "rewards/rejected": -19.5,
       "step": 2290
     },
     {
       "epoch": 2.4083769633507854,
-      "grad_norm": 0.10988173112717037,
+      "grad_norm": 0.13028602403715359,
       "learning_rate": 5.695289848593532e-06,
-      "logits/chosen": -2.125,
-      "logits/rejected": -0.9609375,
-      "logps/chosen": -1632.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0165,
+      "logits/chosen": -0.87890625,
+      "logits/rejected": -0.361328125,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.0157,
       "rewards/accuracies": 0.9624999761581421,
-      "rewards/chosen": -12.25,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -10.6875,
+      "rewards/margins": 8.6875,
+      "rewards/rejected": -19.375,
       "step": 2300
     },
     {
       "epoch": 2.4083769633507854,
-      "eval_logits/chosen": 1.1171875,
-      "eval_logits/rejected": 2.921875,
-      "eval_logps/chosen": -1896.0,
-      "eval_logps/rejected": -2024.0,
-      "eval_loss": 1.781656265258789,
-      "eval_rewards/accuracies": 0.59375,
-      "eval_rewards/chosen": -15.8125,
-      "eval_rewards/margins": 1.5703125,
-      "eval_rewards/rejected": -17.375,
-      "eval_runtime": 46.6487,
-      "eval_samples_per_second": 42.874,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": 2.8125,
+      "eval_logits/rejected": 4.5,
+      "eval_logps/chosen": -1736.0,
+      "eval_logps/rejected": -1864.0,
+      "eval_loss": 1.7641562223434448,
+      "eval_rewards/accuracies": 0.58203125,
+      "eval_rewards/chosen": -14.1875,
+      "eval_rewards/margins": 1.5546875,
+      "eval_rewards/rejected": -15.75,
+      "eval_runtime": 47.1322,
+      "eval_samples_per_second": 42.434,
+      "eval_steps_per_second": 0.679,
       "step": 2300
     },
     {
       "epoch": 2.418848167539267,
-      "grad_norm": 0.14844038313112592,
+      "grad_norm": 0.13763289503523016,
       "learning_rate": 5.50315277286812e-06,
-      "logits/chosen": -2.359375,
-      "logits/rejected": -0.61328125,
-      "logps/chosen": -1608.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0191,
+      "logits/chosen": -0.375,
+      "logits/rejected": 0.251953125,
+      "logps/chosen": -1456.0,
+      "logps/rejected": -2384.0,
+      "loss": 0.0158,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.375,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -10.875,
+      "rewards/margins": 8.5625,
+      "rewards/rejected": -19.5,
       "step": 2310
     },
     {
       "epoch": 2.4293193717277486,
-      "grad_norm": 0.43936779227361,
+      "grad_norm": 1.1892839510746438,
       "learning_rate": 5.313910992897281e-06,
-      "logits/chosen": -1.3125,
-      "logits/rejected": 0.5234375,
-      "logps/chosen": -1544.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0154,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.5,
+      "logits/chosen": 0.6875,
+      "logits/rejected": 1.46875,
+      "logps/chosen": -1416.0,
+      "logps/rejected": -2480.0,
+      "loss": 0.0134,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -10.1875,
       "rewards/margins": 9.25,
-      "rewards/rejected": -20.75,
+      "rewards/rejected": -19.5,
       "step": 2320
     },
     {
       "epoch": 2.4397905759162306,
-      "grad_norm": 1.0196548089785136,
+      "grad_norm": 0.26145833355035725,
       "learning_rate": 5.127592611220658e-06,
-      "logits/chosen": -2.40625,
-      "logits/rejected": -0.34765625,
-      "logps/chosen": -1664.0,
-      "logps/rejected": -2752.0,
-      "loss": 0.0263,
+      "logits/chosen": -0.80859375,
+      "logits/rejected": 0.46484375,
+      "logps/chosen": -1504.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0268,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -12.375,
+      "rewards/chosen": -10.75,
       "rewards/margins": 9.6875,
-      "rewards/rejected": -22.0,
+      "rewards/rejected": -20.375,
       "step": 2330
     },
     {
       "epoch": 2.450261780104712,
-      "grad_norm": 0.4218483763721506,
+      "grad_norm": 0.42699763590668244,
       "learning_rate": 4.944225296251159e-06,
-      "logits/chosen": -2.125,
-      "logits/rejected": -0.58984375,
-      "logps/chosen": -1528.0,
-      "logps/rejected": -2544.0,
-      "loss": 0.0127,
+      "logits/chosen": -0.26171875,
+      "logits/rejected": 0.291015625,
+      "logps/chosen": -1344.0,
+      "logps/rejected": -2336.0,
+      "loss": 0.0143,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.75,
-      "rewards/margins": 9.0625,
-      "rewards/rejected": -20.875,
+      "rewards/chosen": -9.9375,
+      "rewards/margins": 8.8125,
+      "rewards/rejected": -18.75,
       "step": 2340
     },
     {
       "epoch": 2.4607329842931938,
-      "grad_norm": 3.0025945375518996,
+      "grad_norm": 0.46330307002324184,
       "learning_rate": 4.763836278166172e-06,
-      "logits/chosen": -2.078125,
-      "logits/rejected": 0.07470703125,
-      "logps/chosen": -1584.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0187,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.8125,
-      "rewards/margins": 8.375,
-      "rewards/rejected": -20.125,
+      "logits/chosen": 0.030029296875,
+      "logits/rejected": 1.2890625,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -2336.0,
+      "loss": 0.0114,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.3125,
+      "rewards/margins": 8.5625,
+      "rewards/rejected": -18.875,
       "step": 2350
     },
     {
       "epoch": 2.4712041884816753,
-      "grad_norm": 0.4114665112435162,
+      "grad_norm": 0.07996374807985139,
       "learning_rate": 4.586452344863876e-06,
-      "logits/chosen": -2.46875,
-      "logits/rejected": -0.7578125,
-      "logps/chosen": -1584.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0152,
+      "logits/chosen": -0.0634765625,
+      "logits/rejected": 0.57421875,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0145,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -11.75,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -20.5,
+      "rewards/chosen": -10.75,
+      "rewards/margins": 8.875,
+      "rewards/rejected": -19.625,
       "step": 2360
     },
     {
       "epoch": 2.481675392670157,
-      "grad_norm": 0.7002478176984285,
+      "grad_norm": 0.4929328901597208,
       "learning_rate": 4.412099837985192e-06,
-      "logits/chosen": -2.265625,
-      "logits/rejected": -0.08154296875,
-      "logps/chosen": -1624.0,
-      "logps/rejected": -2736.0,
+      "logits/chosen": 0.0279541015625,
+      "logits/rejected": 1.1875,
+      "logps/chosen": -1504.0,
+      "logps/rejected": -2656.0,
       "loss": 0.014,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.0,
-      "rewards/margins": 9.5,
-      "rewards/rejected": -21.5,
+      "rewards/chosen": -10.75,
+      "rewards/margins": 9.9375,
+      "rewards/rejected": -20.75,
       "step": 2370
     },
     {
       "epoch": 2.492146596858639,
-      "grad_norm": 0.2414253837920038,
+      "grad_norm": 0.18742205047534752,
       "learning_rate": 4.240804649002089e-06,
-      "logits/chosen": -2.703125,
-      "logits/rejected": -0.921875,
-      "logps/chosen": -1472.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0241,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -20.25,
+      "logits/chosen": 0.006500244140625,
+      "logits/rejected": 0.5,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.0276,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -10.25,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -19.375,
       "step": 2380
     },
     {
       "epoch": 2.5026178010471205,
-      "grad_norm": 0.4707372790690564,
+      "grad_norm": 0.5502702109382654,
       "learning_rate": 4.072592215372598e-06,
-      "logits/chosen": -3.125,
-      "logits/rejected": -1.1875,
-      "logps/chosen": -1656.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0132,
+      "logits/chosen": -0.97265625,
+      "logits/rejected": -0.0311279296875,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0158,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.125,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 9.25,
+      "rewards/rejected": -20.25,
       "step": 2390
     },
     {
       "epoch": 2.513089005235602,
-      "grad_norm": 0.6792175752753643,
+      "grad_norm": 0.13593228148382305,
       "learning_rate": 3.907487516763389e-06,
-      "logits/chosen": -2.171875,
-      "logits/rejected": -0.388671875,
-      "logps/chosen": -1592.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0169,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.9375,
-      "rewards/margins": 8.6875,
-      "rewards/rejected": -20.625,
+      "logits/chosen": 0.275390625,
+      "logits/rejected": 1.0703125,
+      "logps/chosen": -1480.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.0147,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -10.875,
+      "rewards/margins": 8.5,
+      "rewards/rejected": -19.375,
       "step": 2400
     },
     {
       "epoch": 2.513089005235602,
-      "eval_logits/chosen": 0.68359375,
-      "eval_logits/rejected": 2.515625,
-      "eval_logps/chosen": -1848.0,
-      "eval_logps/rejected": -1976.0,
-      "eval_loss": 1.7279062271118164,
-      "eval_rewards/accuracies": 0.59765625,
-      "eval_rewards/chosen": -15.25,
-      "eval_rewards/margins": 1.5625,
-      "eval_rewards/rejected": -16.875,
-      "eval_runtime": 46.657,
-      "eval_samples_per_second": 42.866,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": 3.09375,
+      "eval_logits/rejected": 4.8125,
+      "eval_logps/chosen": -1768.0,
+      "eval_logps/rejected": -1912.0,
+      "eval_loss": 1.8053125143051147,
+      "eval_rewards/accuracies": 0.59375,
+      "eval_rewards/chosen": -14.5,
+      "eval_rewards/margins": 1.6953125,
+      "eval_rewards/rejected": -16.25,
+      "eval_runtime": 47.1167,
+      "eval_samples_per_second": 42.448,
+      "eval_steps_per_second": 0.679,
       "step": 2400
     },
     {
       "epoch": 2.5235602094240837,
-      "grad_norm": 0.5585669264048738,
+      "grad_norm": 0.5828898304061597,
       "learning_rate": 3.745515071340286e-06,
-      "logits/chosen": -2.328125,
-      "logits/rejected": 0.220703125,
-      "logps/chosen": -1616.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0181,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.9375,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -21.0,
+      "logits/chosen": 0.322265625,
+      "logits/rejected": 1.5859375,
+      "logps/chosen": -1520.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0166,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.9375,
+      "rewards/margins": 9.6875,
+      "rewards/rejected": -20.625,
       "step": 2410
     },
     {
       "epoch": 2.5340314136125652,
-      "grad_norm": 0.07493111846576456,
+      "grad_norm": 0.14007168048596014,
       "learning_rate": 3.586698932127236e-06,
-      "logits/chosen": -3.125,
-      "logits/rejected": -0.76953125,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2432.0,
-      "loss": 0.0074,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.0,
-      "rewards/margins": 8.6875,
-      "rewards/rejected": -19.625,
+      "logits/chosen": -0.72265625,
+      "logits/rejected": 0.65234375,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0086,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.375,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -19.75,
       "step": 2420
     },
     {
       "epoch": 2.5445026178010473,
-      "grad_norm": 0.17910523942496434,
+      "grad_norm": 0.19732825682118527,
       "learning_rate": 3.4310626834344744e-06,
-      "logits/chosen": -3.15625,
-      "logits/rejected": -1.5234375,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0162,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.8125,
-      "rewards/margins": 8.6875,
-      "rewards/rejected": -20.5,
+      "logits/chosen": -0.96875,
+      "logits/rejected": -0.26171875,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2576.0,
+      "loss": 0.0116,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 9.5625,
+      "rewards/rejected": -20.75,
       "step": 2430
     },
     {
       "epoch": 2.554973821989529,
-      "grad_norm": 0.06575843343697423,
+      "grad_norm": 0.09927419608842018,
       "learning_rate": 3.2786294373562343e-06,
-      "logits/chosen": -2.625,
-      "logits/rejected": -0.369140625,
-      "logps/chosen": -1568.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0084,
+      "logits/chosen": 0.185546875,
+      "logits/rejected": 1.2265625,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0068,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.5625,
-      "rewards/margins": 8.375,
-      "rewards/rejected": -19.875,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -20.25,
       "step": 2440
     },
     {
       "epoch": 2.5654450261780104,
-      "grad_norm": 0.45193754215926607,
+      "grad_norm": 0.4306125267974309,
       "learning_rate": 3.1294218303385526e-06,
-      "logits/chosen": -2.453125,
-      "logits/rejected": -0.30859375,
-      "logps/chosen": -1624.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0153,
+      "logits/chosen": -0.1923828125,
+      "logits/rejected": 0.984375,
+      "logps/chosen": -1600.0,
+      "logps/rejected": -2688.0,
+      "loss": 0.0127,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.875,
-      "rewards/margins": 8.5,
-      "rewards/rejected": -20.375,
+      "rewards/chosen": -11.625,
+      "rewards/margins": 9.6875,
+      "rewards/rejected": -21.25,
       "step": 2450
     },
     {
       "epoch": 2.5759162303664924,
-      "grad_norm": 0.8057461914654177,
+      "grad_norm": 0.3710613895687642,
       "learning_rate": 2.9834620198177483e-06,
-      "logits/chosen": -2.25,
-      "logits/rejected": -0.890625,
-      "logps/chosen": -1608.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0173,
+      "logits/chosen": -0.11083984375,
+      "logits/rejected": 0.302734375,
+      "logps/chosen": -1584.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0191,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.125,
-      "rewards/margins": 8.625,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -11.8125,
+      "rewards/margins": 9.4375,
+      "rewards/rejected": -21.25,
       "step": 2460
     },
     {
       "epoch": 2.5863874345549736,
-      "grad_norm": 0.2431117081850439,
+      "grad_norm": 1.9576916809674352,
       "learning_rate": 2.840771680930068e-06,
-      "logits/chosen": -2.40625,
-      "logits/rejected": -0.8671875,
-      "logps/chosen": -1696.0,
-      "logps/rejected": -2800.0,
-      "loss": 0.0143,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -12.625,
-      "rewards/margins": 9.5,
-      "rewards/rejected": -22.125,
+      "logits/chosen": 0.0228271484375,
+      "logits/rejected": 0.375,
+      "logps/chosen": -1656.0,
+      "logps/rejected": -2816.0,
+      "loss": 0.0135,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -12.3125,
+      "rewards/margins": 9.9375,
+      "rewards/rejected": -22.25,
       "step": 2470
     },
     {
       "epoch": 2.5968586387434556,
-      "grad_norm": 0.24821789781009196,
+      "grad_norm": 0.45417686300499405,
       "learning_rate": 2.701372003292868e-06,
-      "logits/chosen": -1.890625,
-      "logits/rejected": -0.2490234375,
-      "logps/chosen": -1544.0,
+      "logits/chosen": 0.8046875,
+      "logits/rejected": 1.2421875,
+      "logps/chosen": -1512.0,
       "logps/rejected": -2576.0,
-      "loss": 0.0133,
+      "loss": 0.0141,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.5625,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -20.625,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 9.5625,
+      "rewards/rejected": -20.75,
       "step": 2480
     },
     {
       "epoch": 2.607329842931937,
-      "grad_norm": 0.2847839899181593,
+      "grad_norm": 0.888750828862411,
       "learning_rate": 2.56528368785795e-06,
-      "logits/chosen": -2.71875,
-      "logits/rejected": -0.080078125,
-      "logps/chosen": -1552.0,
+      "logits/chosen": -0.111328125,
+      "logits/rejected": 1.3359375,
+      "logps/chosen": -1512.0,
       "logps/rejected": -2496.0,
-      "loss": 0.0127,
+      "loss": 0.0131,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.3125,
-      "rewards/margins": 9.0625,
+      "rewards/chosen": -10.9375,
+      "rewards/margins": 9.5,
       "rewards/rejected": -20.375,
       "step": 2490
     },
     {
       "epoch": 2.6178010471204187,
-      "grad_norm": 1.908192432045549,
+      "grad_norm": 0.24134271131946747,
       "learning_rate": 2.432526943837471e-06,
-      "logits/chosen": -2.296875,
-      "logits/rejected": 0.412109375,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0189,
+      "logits/chosen": 0.8203125,
+      "logits/rejected": 2.0,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0161,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.3125,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -20.625,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 10.1875,
+      "rewards/rejected": -21.25,
       "step": 2500
     },
     {
       "epoch": 2.6178010471204187,
-      "eval_logits/chosen": 0.83203125,
-      "eval_logits/rejected": 2.671875,
-      "eval_logps/chosen": -1856.0,
+      "eval_logits/chosen": 3.15625,
+      "eval_logits/rejected": 4.875,
+      "eval_logps/chosen": -1848.0,
       "eval_logps/rejected": -1992.0,
-      "eval_loss": 1.7442500591278076,
-      "eval_rewards/accuracies": 0.6015625,
-      "eval_rewards/chosen": -15.375,
-      "eval_rewards/margins": 1.640625,
+      "eval_loss": 1.9171562194824219,
+      "eval_rewards/accuracies": 0.58984375,
+      "eval_rewards/chosen": -15.25,
+      "eval_rewards/margins": 1.75,
       "eval_rewards/rejected": -17.0,
-      "eval_runtime": 46.6662,
-      "eval_samples_per_second": 42.858,
-      "eval_steps_per_second": 0.686,
+      "eval_runtime": 47.1269,
+      "eval_samples_per_second": 42.439,
+      "eval_steps_per_second": 0.679,
       "step": 2500
     },
     {
       "epoch": 2.6282722513089007,
-      "grad_norm": 0.07655568789515692,
+      "grad_norm": 0.3052902141891288,
       "learning_rate": 2.3031214857028446e-06,
-      "logits/chosen": -1.8203125,
-      "logits/rejected": 0.0888671875,
-      "logps/chosen": -1584.0,
+      "logits/chosen": 0.6484375,
+      "logits/rejected": 1.53125,
+      "logps/chosen": -1552.0,
       "logps/rejected": -2592.0,
-      "loss": 0.0153,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 9.375,
+      "loss": 0.0173,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 9.75,
       "rewards/rejected": -21.0,
       "step": 2510
     },
     {
       "epoch": 2.6387434554973823,
-      "grad_norm": 0.1288903261198696,
+      "grad_norm": 0.3410854599302668,
       "learning_rate": 2.177086530257139e-06,
-      "logits/chosen": -2.421875,
-      "logits/rejected": -0.0693359375,
-      "logps/chosen": -1552.0,
+      "logits/chosen": 0.357421875,
+      "logits/rejected": 1.3359375,
+      "logps/chosen": -1504.0,
       "logps/rejected": -2512.0,
-      "loss": 0.0177,
+      "loss": 0.0161,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.625,
-      "rewards/margins": 8.9375,
+      "rewards/chosen": -11.125,
+      "rewards/margins": 9.5,
       "rewards/rejected": -20.625,
       "step": 2520
     },
     {
       "epoch": 2.649214659685864,
-      "grad_norm": 0.2777580455626625,
+      "grad_norm": 0.7507252053960208,
       "learning_rate": 2.054440793781354e-06,
-      "logits/chosen": -2.796875,
-      "logits/rejected": -0.412109375,
-      "logps/chosen": -1528.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0198,
+      "logits/chosen": -0.2734375,
+      "logits/rejected": 0.93359375,
+      "logps/chosen": -1456.0,
+      "logps/rejected": -2576.0,
+      "loss": 0.0184,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.25,
-      "rewards/margins": 9.5625,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 10.1875,
       "rewards/rejected": -20.75,
       "step": 2530
     },
     {
       "epoch": 2.6596858638743455,
-      "grad_norm": 0.06996163806870984,
+      "grad_norm": 0.06587722371656514,
       "learning_rate": 1.9352024892550587e-06,
-      "logits/chosen": -1.8671875,
-      "logits/rejected": -0.34765625,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.0141,
+      "logits/chosen": -0.01080322265625,
+      "logits/rejected": 0.7109375,
+      "logps/chosen": -1528.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0171,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.6875,
-      "rewards/margins": 9.3125,
-      "rewards/rejected": -21.0,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 9.625,
+      "rewards/rejected": -20.875,
       "step": 2540
     },
     {
       "epoch": 2.670157068062827,
-      "grad_norm": 0.3134344838095471,
+      "grad_norm": 0.2079156149407264,
       "learning_rate": 1.8193893236517379e-06,
-      "logits/chosen": -2.6875,
-      "logits/rejected": -0.330078125,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2560.0,
-      "loss": 0.0157,
+      "logits/chosen": -0.6328125,
+      "logits/rejected": 0.72265625,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0134,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.75,
-      "rewards/margins": 9.375,
-      "rewards/rejected": -21.125,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 9.5625,
+      "rewards/rejected": -20.75,
       "step": 2550
     },
     {
       "epoch": 2.680628272251309,
-      "grad_norm": 0.2679659535142388,
+      "grad_norm": 0.38034464094621967,
       "learning_rate": 1.7070184953092956e-06,
-      "logits/chosen": -2.765625,
-      "logits/rejected": -0.80859375,
-      "logps/chosen": -1512.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0156,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 8.6875,
-      "rewards/rejected": -19.75,
+      "logits/chosen": -0.45703125,
+      "logits/rejected": 0.451171875,
+      "logps/chosen": -1440.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0128,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -10.375,
+      "rewards/margins": 9.75,
+      "rewards/rejected": -20.125,
       "step": 2560
     },
     {
       "epoch": 2.6910994764397906,
-      "grad_norm": 0.5343281926367408,
+      "grad_norm": 0.8267473052453628,
       "learning_rate": 1.5981066913760918e-06,
-      "logits/chosen": -1.6875,
-      "logits/rejected": 0.08642578125,
-      "logps/chosen": -1448.0,
-      "logps/rejected": -2464.0,
-      "loss": 0.0104,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -10.75,
-      "rewards/margins": 8.75,
-      "rewards/rejected": -19.5,
+      "logits/chosen": 0.80078125,
+      "logits/rejected": 1.3359375,
+      "logps/chosen": -1392.0,
+      "logps/rejected": -2448.0,
+      "loss": 0.0125,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.1875,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -19.25,
       "step": 2570
     },
     {
       "epoch": 2.701570680628272,
-      "grad_norm": 0.7772208052418603,
+      "grad_norm": 0.13489609463032673,
       "learning_rate": 1.492670085332884e-06,
-      "logits/chosen": -2.46875,
-      "logits/rejected": -0.70703125,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0195,
+      "logits/chosen": -0.19140625,
+      "logits/rejected": 0.470703125,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2624.0,
+      "loss": 0.0211,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.9375,
-      "rewards/margins": 9.0,
-      "rewards/rejected": -21.0,
+      "rewards/chosen": -11.25,
+      "rewards/margins": 9.625,
+      "rewards/rejected": -20.875,
       "step": 2580
     },
     {
       "epoch": 2.712041884816754,
-      "grad_norm": 0.3909040077870737,
+      "grad_norm": 0.2603801564005355,
       "learning_rate": 1.3907243345910787e-06,
-      "logits/chosen": -2.0,
-      "logits/rejected": 0.33984375,
-      "logps/chosen": -1616.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0157,
+      "logits/chosen": 0.294921875,
+      "logits/rejected": 1.5,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2608.0,
+      "loss": 0.015,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.6875,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -20.875,
+      "rewards/chosen": -10.875,
+      "rewards/margins": 9.8125,
+      "rewards/rejected": -20.625,
       "step": 2590
     },
     {
       "epoch": 2.7225130890052354,
-      "grad_norm": 0.1069725799306176,
+      "grad_norm": 0.3964085440309203,
       "learning_rate": 1.292284578167538e-06,
-      "logits/chosen": -2.609375,
-      "logits/rejected": -0.91015625,
-      "logps/chosen": -1512.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0134,
+      "logits/chosen": -0.49609375,
+      "logits/rejected": 0.12060546875,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.014,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.25,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -20.375,
+      "rewards/chosen": -10.625,
+      "rewards/margins": 9.5,
+      "rewards/rejected": -20.125,
       "step": 2600
     },
     {
       "epoch": 2.7225130890052354,
-      "eval_logits/chosen": 0.7109375,
-      "eval_logits/rejected": 2.546875,
-      "eval_logps/chosen": -1816.0,
-      "eval_logps/rejected": -1952.0,
-      "eval_loss": 1.68540620803833,
-      "eval_rewards/accuracies": 0.58984375,
-      "eval_rewards/chosen": -15.0,
-      "eval_rewards/margins": 1.6171875,
-      "eval_rewards/rejected": -16.625,
-      "eval_runtime": 46.6645,
-      "eval_samples_per_second": 42.859,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": 2.890625,
+      "eval_logits/rejected": 4.625,
+      "eval_logps/chosen": -1760.0,
+      "eval_logps/rejected": -1896.0,
+      "eval_loss": 1.8166249990463257,
+      "eval_rewards/accuracies": 0.595703125,
+      "eval_rewards/chosen": -14.375,
+      "eval_rewards/margins": 1.6953125,
+      "eval_rewards/rejected": -16.125,
+      "eval_runtime": 47.1183,
+      "eval_samples_per_second": 42.446,
+      "eval_steps_per_second": 0.679,
       "step": 2600
     },
     {
       "epoch": 2.7329842931937174,
-      "grad_norm": 0.308940796134209,
+      "grad_norm": 0.45541759875785576,
       "learning_rate": 1.1973654344364925e-06,
-      "logits/chosen": -2.4375,
-      "logits/rejected": -1.09375,
-      "logps/chosen": -1584.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0134,
+      "logits/chosen": -0.259765625,
+      "logits/rejected": 0.10400390625,
+      "logps/chosen": -1480.0,
+      "logps/rejected": -2544.0,
+      "loss": 0.0116,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.4375,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -20.375,
+      "rewards/chosen": -10.375,
+      "rewards/margins": 9.4375,
+      "rewards/rejected": -19.875,
       "step": 2610
     },
     {
       "epoch": 2.743455497382199,
-      "grad_norm": 0.19657020267705813,
+      "grad_norm": 0.15144544785356837,
       "learning_rate": 1.1059809989586583e-06,
-      "logits/chosen": -1.75,
-      "logits/rejected": -0.453125,
-      "logps/chosen": -1536.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0227,
+      "logits/chosen": 0.63671875,
+      "logits/rejected": 0.9609375,
+      "logps/chosen": -1480.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.014,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.3125,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -20.5,
+      "rewards/chosen": -10.6875,
+      "rewards/margins": 9.3125,
+      "rewards/rejected": -20.0,
       "step": 2620
     },
     {
       "epoch": 2.7539267015706805,
-      "grad_norm": 3.03559659464108,
+      "grad_norm": 1.1743488713271935,
       "learning_rate": 1.018144842388044e-06,
-      "logits/chosen": -1.9609375,
-      "logits/rejected": 0.484375,
-      "logps/chosen": -1584.0,
+      "logits/chosen": 0.59375,
+      "logits/rejected": 1.65625,
+      "logps/chosen": -1496.0,
       "logps/rejected": -2640.0,
-      "loss": 0.0191,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.4375,
-      "rewards/margins": 8.875,
+      "loss": 0.0151,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.625,
+      "rewards/margins": 9.6875,
       "rewards/rejected": -20.25,
       "step": 2630
     },
     {
       "epoch": 2.7643979057591626,
-      "grad_norm": 0.1622247348983575,
+      "grad_norm": 0.28330996294350125,
       "learning_rate": 9.338700084567109e-07,
-      "logits/chosen": -2.359375,
-      "logits/rejected": -0.44140625,
-      "logps/chosen": -1552.0,
-      "logps/rejected": -2640.0,
-      "loss": 0.0129,
+      "logits/chosen": 0.099609375,
+      "logits/rejected": 0.921875,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0138,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.5,
-      "rewards/margins": 9.375,
-      "rewards/rejected": -20.875,
+      "rewards/chosen": -10.6875,
+      "rewards/margins": 9.6875,
+      "rewards/rejected": -20.375,
       "step": 2640
     },
     {
       "epoch": 2.774869109947644,
-      "grad_norm": 0.6038268939541903,
+      "grad_norm": 0.9128132421954135,
       "learning_rate": 8.531690120377606e-07,
-      "logits/chosen": -2.390625,
-      "logits/rejected": -0.76171875,
-      "logps/chosen": -1704.0,
-      "logps/rejected": -2768.0,
-      "loss": 0.0141,
+      "logits/chosen": -0.3984375,
+      "logits/rejected": 0.4765625,
+      "logps/chosen": -1664.0,
+      "logps/rejected": -2736.0,
+      "loss": 0.0177,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.5,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -21.75,
+      "rewards/chosen": -12.0625,
+      "rewards/margins": 9.375,
+      "rewards/rejected": -21.5,
       "step": 2650
     },
     {
       "epoch": 2.7853403141361257,
-      "grad_norm": 0.38710210344185003,
+      "grad_norm": 0.47741993579545966,
       "learning_rate": 7.760538372868636e-07,
-      "logits/chosen": -2.78125,
-      "logits/rejected": -0.5078125,
-      "logps/chosen": -1552.0,
-      "logps/rejected": -2528.0,
-      "loss": 0.0164,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.5,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -20.5,
+      "logits/chosen": -0.361328125,
+      "logits/rejected": 0.8828125,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2512.0,
+      "loss": 0.0172,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -10.625,
+      "rewards/margins": 9.625,
+      "rewards/rejected": -20.25,
       "step": 2660
     },
     {
       "epoch": 2.7958115183246073,
-      "grad_norm": 0.182754981587112,
+      "grad_norm": 0.30693549369263295,
       "learning_rate": 7.025359358626166e-07,
-      "logits/chosen": -2.375,
-      "logits/rejected": 0.12109375,
-      "logps/chosen": -1640.0,
-      "logps/rejected": -2624.0,
-      "loss": 0.0114,
+      "logits/chosen": 0.08642578125,
+      "logits/rejected": 1.78125,
+      "logps/chosen": -1544.0,
+      "logps/rejected": -2592.0,
+      "loss": 0.0121,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -12.0,
-      "rewards/margins": 9.1875,
-      "rewards/rejected": -21.25,
+      "rewards/chosen": -11.0625,
+      "rewards/margins": 9.875,
+      "rewards/rejected": -21.0,
       "step": 2670
     },
     {
       "epoch": 2.806282722513089,
-      "grad_norm": 0.4829630471064681,
+      "grad_norm": 0.4424994691812759,
       "learning_rate": 6.326262252259463e-07,
-      "logits/chosen": -2.015625,
-      "logits/rejected": 0.50390625,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2544.0,
-      "loss": 0.0173,
+      "logits/chosen": 0.271484375,
+      "logits/rejected": 1.84375,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2496.0,
+      "loss": 0.0163,
       "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.5,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -20.625,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 9.625,
+      "rewards/rejected": -20.125,
       "step": 2680
     },
     {
       "epoch": 2.816753926701571,
-      "grad_norm": 0.19443469229353988,
+      "grad_norm": 0.21704077642021244,
       "learning_rate": 5.66335087018871e-07,
-      "logits/chosen": -2.921875,
-      "logits/rejected": -0.384765625,
-      "logps/chosen": -1568.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0072,
+      "logits/chosen": -0.9765625,
+      "logits/rejected": 0.62890625,
+      "logps/chosen": -1496.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0078,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.3125,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -20.25,
+      "rewards/chosen": -10.625,
+      "rewards/margins": 9.3125,
+      "rewards/rejected": -20.0,
       "step": 2690
     },
     {
       "epoch": 2.8272251308900525,
-      "grad_norm": 0.1635176444004718,
+      "grad_norm": 0.21690823474208473,
       "learning_rate": 5.036723655228226e-07,
-      "logits/chosen": -2.453125,
-      "logits/rejected": -0.515625,
-      "logps/chosen": -1496.0,
-      "logps/rejected": -2448.0,
-      "loss": 0.0161,
+      "logits/chosen": 0.2421875,
+      "logits/rejected": 0.79296875,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -2400.0,
+      "loss": 0.0166,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.25,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -20.0,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 9.125,
+      "rewards/rejected": -19.625,
       "step": 2700
     },
     {
       "epoch": 2.8272251308900525,
-      "eval_logits/chosen": 0.68359375,
-      "eval_logits/rejected": 2.53125,
-      "eval_logps/chosen": -1824.0,
-      "eval_logps/rejected": -1960.0,
-      "eval_loss": 1.6882343292236328,
-      "eval_rewards/accuracies": 0.587890625,
-      "eval_rewards/chosen": -15.0625,
-      "eval_rewards/margins": 1.625,
-      "eval_rewards/rejected": -16.625,
-      "eval_runtime": 46.6551,
-      "eval_samples_per_second": 42.868,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": 3.0,
+      "eval_logits/rejected": 4.71875,
+      "eval_logps/chosen": -1776.0,
+      "eval_logps/rejected": -1920.0,
+      "eval_loss": 1.828531265258789,
+      "eval_rewards/accuracies": 0.59375,
+      "eval_rewards/chosen": -14.625,
+      "eval_rewards/margins": 1.7109375,
+      "eval_rewards/rejected": -16.375,
+      "eval_runtime": 47.1398,
+      "eval_samples_per_second": 42.427,
+      "eval_steps_per_second": 0.679,
       "step": 2700
     },
     {
       "epoch": 2.837696335078534,
-      "grad_norm": 0.2163692953663021,
+      "grad_norm": 0.5960415700111674,
       "learning_rate": 4.446473661967432e-07,
-      "logits/chosen": -3.078125,
-      "logits/rejected": -0.8125,
-      "logps/chosen": -1544.0,
-      "logps/rejected": -2464.0,
-      "loss": 0.0145,
+      "logits/chosen": -0.5703125,
+      "logits/rejected": 1.0546875,
+      "logps/chosen": -1472.0,
+      "logps/rejected": -2496.0,
+      "loss": 0.016,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.375,
-      "rewards/margins": 8.8125,
-      "rewards/rejected": -20.125,
+      "rewards/chosen": -10.625,
+      "rewards/margins": 9.75,
+      "rewards/rejected": -20.375,
       "step": 2710
     },
     {
       "epoch": 2.8481675392670156,
-      "grad_norm": 0.27303054886309563,
+      "grad_norm": 0.8082759308562784,
       "learning_rate": 3.892688542952411e-07,
-      "logits/chosen": -1.9140625,
-      "logits/rejected": 0.2060546875,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.0186,
-      "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 8.4375,
-      "rewards/rejected": -19.625,
+      "logits/chosen": 0.8515625,
+      "logits/rejected": 1.765625,
+      "logps/chosen": -1432.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.015,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -19.5,
       "step": 2720
     },
     {
       "epoch": 2.858638743455497,
-      "grad_norm": 1.9303658172652438,
+      "grad_norm": 0.48586361654791427,
       "learning_rate": 3.3754505356694e-07,
-      "logits/chosen": -3.0625,
-      "logits/rejected": -1.6171875,
-      "logps/chosen": -1600.0,
-      "logps/rejected": -2608.0,
-      "loss": 0.017,
+      "logits/chosen": -1.0,
+      "logits/rejected": -0.439453125,
+      "logps/chosen": -1536.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.0162,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.8125,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -11.1875,
+      "rewards/margins": 9.0625,
+      "rewards/rejected": -20.25,
       "step": 2730
     },
     {
       "epoch": 2.869109947643979,
-      "grad_norm": 0.3577776032079035,
+      "grad_norm": 0.45455049740146347,
       "learning_rate": 2.8948364503322276e-07,
-      "logits/chosen": -2.234375,
-      "logits/rejected": -0.443359375,
-      "logps/chosen": -1504.0,
-      "logps/rejected": -2480.0,
-      "loss": 0.017,
+      "logits/chosen": 0.267578125,
+      "logits/rejected": 1.15625,
+      "logps/chosen": -1448.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.0199,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.1875,
-      "rewards/margins": 8.875,
-      "rewards/rejected": -20.0,
+      "rewards/chosen": -10.625,
+      "rewards/margins": 8.9375,
+      "rewards/rejected": -19.5,
       "step": 2740
     },
     {
       "epoch": 2.8795811518324608,
-      "grad_norm": 0.24443735291517754,
+      "grad_norm": 0.30116199091235635,
       "learning_rate": 2.4509176584762706e-07,
-      "logits/chosen": -2.421875,
-      "logits/rejected": -0.18359375,
-      "logps/chosen": -1528.0,
-      "logps/rejected": -2448.0,
-      "loss": 0.0127,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.3125,
-      "rewards/margins": 8.4375,
-      "rewards/rejected": -19.75,
+      "logits/chosen": -0.0537109375,
+      "logits/rejected": 1.015625,
+      "logps/chosen": -1488.0,
+      "logps/rejected": -2416.0,
+      "loss": 0.0144,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -10.8125,
+      "rewards/margins": 8.5625,
+      "rewards/rejected": -19.375,
       "step": 2750
     },
     {
       "epoch": 2.8900523560209423,
-      "grad_norm": 0.4972489167452459,
+      "grad_norm": 0.6876848106455995,
       "learning_rate": 2.0437600823595694e-07,
-      "logits/chosen": -2.703125,
-      "logits/rejected": -0.69140625,
-      "logps/chosen": -1624.0,
+      "logits/chosen": -0.515625,
+      "logits/rejected": 0.5390625,
+      "logps/chosen": -1536.0,
       "logps/rejected": -2576.0,
-      "loss": 0.0148,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.75,
-      "rewards/margins": 8.5,
-      "rewards/rejected": -20.25,
+      "loss": 0.018,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.875,
+      "rewards/margins": 9.4375,
+      "rewards/rejected": -20.375,
       "step": 2760
     },
     {
       "epoch": 2.900523560209424,
-      "grad_norm": 0.601126967554501,
+      "grad_norm": 0.43336586182577924,
       "learning_rate": 1.6734241851733e-07,
-      "logits/chosen": -2.9375,
-      "logits/rejected": -0.39453125,
-      "logps/chosen": -1648.0,
-      "logps/rejected": -2656.0,
-      "loss": 0.0144,
-      "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.875,
-      "rewards/margins": 9.625,
+      "logits/chosen": -0.40234375,
+      "logits/rejected": 0.81640625,
+      "logps/chosen": -1608.0,
+      "logps/rejected": -2672.0,
+      "loss": 0.0138,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -11.5,
+      "rewards/margins": 10.0625,
       "rewards/rejected": -21.5,
       "step": 2770
     },
     {
       "epoch": 2.9109947643979055,
-      "grad_norm": 0.07989631772018826,
+      "grad_norm": 0.7112967772105524,
       "learning_rate": 1.339964962062984e-07,
-      "logits/chosen": -2.671875,
-      "logits/rejected": -1.2890625,
-      "logps/chosen": -1528.0,
+      "logits/chosen": -0.353515625,
+      "logits/rejected": 0.11962890625,
+      "logps/chosen": -1464.0,
       "logps/rejected": -2576.0,
-      "loss": 0.0093,
+      "loss": 0.0099,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.5625,
-      "rewards/margins": 9.125,
-      "rewards/rejected": -20.75,
+      "rewards/chosen": -10.9375,
+      "rewards/margins": 9.75,
+      "rewards/rejected": -20.625,
       "step": 2780
     },
     {
       "epoch": 2.9214659685863875,
-      "grad_norm": 0.2927355531444079,
+      "grad_norm": 0.22959668951509873,
       "learning_rate": 1.0434319319617136e-07,
-      "logits/chosen": -1.9609375,
-      "logits/rejected": 0.06591796875,
-      "logps/chosen": -1528.0,
-      "logps/rejected": -2496.0,
-      "loss": 0.0161,
+      "logits/chosen": 0.12353515625,
+      "logits/rejected": 1.28125,
+      "logps/chosen": -1464.0,
+      "logps/rejected": -2464.0,
+      "loss": 0.0165,
       "rewards/accuracies": 0.981249988079071,
-      "rewards/chosen": -11.375,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -20.25,
+      "rewards/chosen": -10.75,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -20.0,
       "step": 2790
     },
     {
       "epoch": 2.931937172774869,
-      "grad_norm": 0.4611553014168555,
+      "grad_norm": 0.14997153620154965,
       "learning_rate": 7.838691302365363e-08,
-      "logits/chosen": -2.390625,
-      "logits/rejected": -0.494140625,
-      "logps/chosen": -1520.0,
-      "logps/rejected": -2592.0,
-      "loss": 0.0165,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 9.25,
-      "rewards/rejected": -20.375,
+      "logits/chosen": -0.034912109375,
+      "logits/rejected": 0.76953125,
+      "logps/chosen": -1464.0,
+      "logps/rejected": -2560.0,
+      "loss": 0.018,
+      "rewards/accuracies": 0.9937499761581421,
+      "rewards/chosen": -10.5625,
+      "rewards/margins": 9.4375,
+      "rewards/rejected": -20.0,
       "step": 2800
     },
     {
       "epoch": 2.931937172774869,
-      "eval_logits/chosen": 0.64453125,
-      "eval_logits/rejected": 2.484375,
-      "eval_logps/chosen": -1824.0,
-      "eval_logps/rejected": -1952.0,
-      "eval_loss": 1.6896874904632568,
-      "eval_rewards/accuracies": 0.591796875,
-      "eval_rewards/chosen": -15.0625,
-      "eval_rewards/margins": 1.6171875,
-      "eval_rewards/rejected": -16.625,
-      "eval_runtime": 46.6465,
-      "eval_samples_per_second": 42.876,
-      "eval_steps_per_second": 0.686,
+      "eval_logits/chosen": 3.0,
+      "eval_logits/rejected": 4.75,
+      "eval_logps/chosen": -1784.0,
+      "eval_logps/rejected": -1928.0,
+      "eval_loss": 1.8392812013626099,
+      "eval_rewards/accuracies": 0.59375,
+      "eval_rewards/chosen": -14.6875,
+      "eval_rewards/margins": 1.7265625,
+      "eval_rewards/rejected": -16.375,
+      "eval_runtime": 47.1275,
+      "eval_samples_per_second": 42.438,
+      "eval_steps_per_second": 0.679,
       "step": 2800
     },
     {
       "epoch": 2.9424083769633507,
-      "grad_norm": 0.8843762593296421,
+      "grad_norm": 0.8766862568421802,
       "learning_rate": 5.613151021490459e-08,
-      "logits/chosen": -1.8984375,
-      "logits/rejected": -0.65234375,
-      "logps/chosen": -1456.0,
+      "logits/chosen": 0.64453125,
+      "logits/rejected": 0.890625,
+      "logps/chosen": -1392.0,
       "logps/rejected": -2416.0,
-      "loss": 0.0149,
+      "loss": 0.0122,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.125,
-      "rewards/margins": 8.375,
+      "rewards/chosen": -10.4375,
+      "rewards/margins": 9.0,
       "rewards/rejected": -19.5,
       "step": 2810
     },
     {
       "epoch": 2.9528795811518327,
-      "grad_norm": 0.26602996563802084,
+      "grad_norm": 0.23242088757088739,
       "learning_rate": 3.7580289713159965e-08,
-      "logits/chosen": -2.140625,
-      "logits/rejected": -0.4375,
-      "logps/chosen": -1616.0,
-      "logps/rejected": -2624.0,
+      "logits/chosen": 0.4453125,
+      "logits/rejected": 0.95703125,
+      "logps/chosen": -1512.0,
+      "logps/rejected": -2672.0,
       "loss": 0.0157,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.875,
-      "rewards/margins": 8.9375,
-      "rewards/rejected": -20.875,
+      "rewards/chosen": -10.9375,
+      "rewards/margins": 10.3125,
+      "rewards/rejected": -21.25,
       "step": 2820
     },
     {
       "epoch": 2.9633507853403143,
-      "grad_norm": 0.16297411803369016,
+      "grad_norm": 0.17213464030772163,
       "learning_rate": 2.2736006387927277e-08,
-      "logits/chosen": -2.765625,
-      "logits/rejected": -0.77734375,
-      "logps/chosen": -1576.0,
-      "logps/rejected": -2576.0,
-      "loss": 0.0115,
-      "rewards/accuracies": 0.987500011920929,
-      "rewards/chosen": -11.5625,
-      "rewards/margins": 9.0625,
-      "rewards/rejected": -20.625,
+      "logits/chosen": -0.546875,
+      "logits/rejected": 0.43359375,
+      "logps/chosen": -1528.0,
+      "logps/rejected": -2528.0,
+      "loss": 0.0126,
+      "rewards/accuracies": 0.981249988079071,
+      "rewards/chosen": -11.125,
+      "rewards/margins": 9.0,
+      "rewards/rejected": -20.125,
       "step": 2830
     },
     {
       "epoch": 2.973821989528796,
-      "grad_norm": 0.23526111657034204,
+      "grad_norm": 0.2626097785705042,
       "learning_rate": 1.1600864625893604e-08,
-      "logits/chosen": -2.609375,
-      "logits/rejected": -0.4921875,
-      "logps/chosen": -1544.0,
+      "logits/chosen": -0.2177734375,
+      "logits/rejected": 0.7890625,
+      "logps/chosen": -1488.0,
       "logps/rejected": -2528.0,
-      "loss": 0.0117,
+      "loss": 0.0142,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.25,
-      "rewards/margins": 9.0,
+      "rewards/chosen": -10.6875,
+      "rewards/margins": 9.625,
       "rewards/rejected": -20.25,
       "step": 2840
     },
     {
       "epoch": 2.9842931937172774,
-      "grad_norm": 0.22464297127570537,
+      "grad_norm": 0.5263341430750755,
       "learning_rate": 4.176518003573549e-09,
-      "logits/chosen": -2.453125,
-      "logits/rejected": -0.609375,
-      "logps/chosen": -1576.0,
+      "logits/chosen": -0.1591796875,
+      "logits/rejected": 0.78515625,
+      "logps/chosen": -1512.0,
       "logps/rejected": -2560.0,
-      "loss": 0.019,
+      "loss": 0.0227,
       "rewards/accuracies": 0.9937499761581421,
-      "rewards/chosen": -11.75,
-      "rewards/margins": 8.625,
+      "rewards/chosen": -11.125,
+      "rewards/margins": 9.3125,
       "rewards/rejected": -20.375,
       "step": 2850
     },
     {
       "epoch": 2.994764397905759,
-      "grad_norm": 0.5260568319393822,
+      "grad_norm": 1.1592955705049761,
       "learning_rate": 4.640690417528903e-10,
-      "logits/chosen": -2.90625,
-      "logits/rejected": -0.890625,
-      "logps/chosen": -1528.0,
-      "logps/rejected": -2464.0,
-      "loss": 0.0162,
+      "logits/chosen": -0.462890625,
+      "logits/rejected": 0.462890625,
+      "logps/chosen": -1464.0,
+      "logps/rejected": -2432.0,
+      "loss": 0.0175,
       "rewards/accuracies": 0.9750000238418579,
-      "rewards/chosen": -11.5,
-      "rewards/margins": 8.75,
-      "rewards/rejected": -20.25,
+      "rewards/chosen": -10.8125,
+      "rewards/margins": 9.1875,
+      "rewards/rejected": -20.0,
       "step": 2860
     },
     {
       "epoch": 3.0,
       "step": 2865,
       "total_flos": 0.0,
-      "train_loss": 0.22519320912178067,
-      "train_runtime": 12281.0925,
-      "train_samples_per_second": 14.93,
-      "train_steps_per_second": 0.233
+      "train_loss": 0.22500331487643158,
+      "train_runtime": 12392.8387,
+      "train_samples_per_second": 14.795,
+      "train_steps_per_second": 0.231
     }
   ],
   "logging_steps": 10,