diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -10,12 +10,12 @@
   "log_history": [
     {
       "epoch": 0.0,
-      "grad_norm": 50.348802908690914,
+      "grad_norm": 71.7172276228935,
       "learning_rate": 5.208333333333333e-09,
-      "logits/chosen": -1.8382859230041504,
-      "logits/rejected": -1.788834810256958,
-      "logps/chosen": -119.0692138671875,
-      "logps/rejected": -76.35714721679688,
+      "logits/chosen": -1.4981693029403687,
+      "logits/rejected": -1.5654948949813843,
+      "logps/chosen": -257.4665222167969,
+      "logps/rejected": -99.09321594238281,
       "loss": 0.6931,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
@@ -25,1581 +25,1581 @@
     },
     {
       "epoch": 0.01,
-      "grad_norm": 46.56256136532365,
+      "grad_norm": 72.96425106358538,
       "learning_rate": 5.208333333333333e-08,
-      "logits/chosen": -1.6623330116271973,
-      "logits/rejected": -1.5512146949768066,
-      "logps/chosen": -129.29541015625,
-      "logps/rejected": -82.82817840576172,
-      "loss": 0.6931,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": -0.00018554604321252555,
-      "rewards/margins": 0.00036421266850084066,
-      "rewards/rejected": -0.0005497586098499596,
+      "logits/chosen": -1.875927209854126,
+      "logits/rejected": -1.8456478118896484,
+      "logps/chosen": -326.4024658203125,
+      "logps/rejected": -141.96922302246094,
+      "loss": 0.6926,
+      "rewards/accuracies": 0.5138888955116272,
+      "rewards/chosen": 4.941297447658144e-05,
+      "rewards/margins": 0.001581083401106298,
+      "rewards/rejected": -0.0015316703356802464,
       "step": 10
     },
     {
       "epoch": 0.02,
-      "grad_norm": 47.9759657771044,
+      "grad_norm": 67.94798943175728,
       "learning_rate": 1.0416666666666667e-07,
-      "logits/chosen": -1.6969773769378662,
-      "logits/rejected": -1.5268709659576416,
-      "logps/chosen": -140.09909057617188,
-      "logps/rejected": -80.9607162475586,
-      "loss": 0.6866,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.013138635084033012,
-      "rewards/margins": 0.015350555069744587,
-      "rewards/rejected": -0.002211919752880931,
+      "logits/chosen": -1.8996143341064453,
+      "logits/rejected": -1.8235642910003662,
+      "logps/chosen": -323.2792663574219,
+      "logps/rejected": -135.71456909179688,
+      "loss": 0.6788,
+      "rewards/accuracies": 0.84375,
+      "rewards/chosen": 0.013530316762626171,
+      "rewards/margins": 0.030556190758943558,
+      "rewards/rejected": -0.01702587492763996,
       "step": 20
     },
     {
       "epoch": 0.03,
-      "grad_norm": 37.210502761575135,
+      "grad_norm": 50.35196678274095,
       "learning_rate": 1.5624999999999999e-07,
-      "logits/chosen": -1.7122924327850342,
-      "logits/rejected": -1.6340242624282837,
-      "logps/chosen": -119.13212585449219,
-      "logps/rejected": -84.83552551269531,
-      "loss": 0.657,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": 0.05337844043970108,
-      "rewards/margins": 0.0730198472738266,
-      "rewards/rejected": -0.01964140310883522,
+      "logits/chosen": -1.9016050100326538,
+      "logits/rejected": -1.8865772485733032,
+      "logps/chosen": -306.1601257324219,
+      "logps/rejected": -138.81069946289062,
+      "loss": 0.6075,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": 0.08184734731912613,
+      "rewards/margins": 0.19270142912864685,
+      "rewards/rejected": -0.11085411161184311,
       "step": 30
     },
     {
       "epoch": 0.04,
-      "grad_norm": 38.29612001708921,
+      "grad_norm": 24.96243157624135,
       "learning_rate": 2.0833333333333333e-07,
-      "logits/chosen": -1.8123095035552979,
-      "logits/rejected": -1.716663122177124,
-      "logps/chosen": -130.39593505859375,
-      "logps/rejected": -98.6071548461914,
-      "loss": 0.5915,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": 0.1293005496263504,
-      "rewards/margins": 0.2863259017467499,
-      "rewards/rejected": -0.15702535212039948,
+      "logits/chosen": -1.9789104461669922,
+      "logits/rejected": -2.000354766845703,
+      "logps/chosen": -269.0250549316406,
+      "logps/rejected": -152.91973876953125,
+      "loss": 0.5009,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": 0.12138372659683228,
+      "rewards/margins": 0.45498305559158325,
+      "rewards/rejected": -0.3335992991924286,
       "step": 40
     },
     {
       "epoch": 0.05,
-      "grad_norm": 25.169270167372616,
+      "grad_norm": 14.96717932132009,
       "learning_rate": 2.604166666666667e-07,
-      "logits/chosen": -1.6766433715820312,
-      "logits/rejected": -1.6306852102279663,
-      "logps/chosen": -127.97342681884766,
-      "logps/rejected": -136.70712280273438,
-      "loss": 0.484,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": 0.0137257669121027,
-      "rewards/margins": 0.6059251427650452,
-      "rewards/rejected": -0.5921992063522339,
+      "logits/chosen": -2.158332109451294,
+      "logits/rejected": -2.16444993019104,
+      "logps/chosen": -280.4437255859375,
+      "logps/rejected": -197.60487365722656,
+      "loss": 0.4282,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": 0.15608879923820496,
+      "rewards/margins": 0.867284893989563,
+      "rewards/rejected": -0.7111960053443909,
       "step": 50
     },
     {
       "epoch": 0.06,
-      "grad_norm": 26.650306729386905,
+      "grad_norm": 10.138930051409568,
       "learning_rate": 3.1249999999999997e-07,
-      "logits/chosen": -1.6139761209487915,
-      "logits/rejected": -1.610743761062622,
-      "logps/chosen": -173.83758544921875,
-      "logps/rejected": -218.09707641601562,
-      "loss": 0.3946,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -0.24665935337543488,
-      "rewards/margins": 1.0333704948425293,
-      "rewards/rejected": -1.280029535293579,
+      "logits/chosen": -2.3020527362823486,
+      "logits/rejected": -2.289559841156006,
+      "logps/chosen": -255.1080780029297,
+      "logps/rejected": -206.7790069580078,
+      "loss": 0.3883,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -0.01837761141359806,
+      "rewards/margins": 0.9224237203598022,
+      "rewards/rejected": -0.9408015012741089,
       "step": 60
     },
     {
       "epoch": 0.07,
-      "grad_norm": 32.401140738769385,
+      "grad_norm": 9.627361782454214,
       "learning_rate": 3.645833333333333e-07,
-      "logits/chosen": -1.4975354671478271,
-      "logits/rejected": -1.4718661308288574,
-      "logps/chosen": -191.6230010986328,
-      "logps/rejected": -297.1890563964844,
-      "loss": 0.3474,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -0.6469367146492004,
-      "rewards/margins": 1.586439847946167,
-      "rewards/rejected": -2.2333762645721436,
+      "logits/chosen": -2.3991897106170654,
+      "logits/rejected": -2.4038243293762207,
+      "logps/chosen": -295.4974670410156,
+      "logps/rejected": -273.0234375,
+      "loss": 0.3499,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -0.15266235172748566,
+      "rewards/margins": 1.2677773237228394,
+      "rewards/rejected": -1.4204397201538086,
       "step": 70
     },
     {
       "epoch": 0.08,
-      "grad_norm": 31.488475681745037,
+      "grad_norm": 14.313296662619264,
       "learning_rate": 4.1666666666666667e-07,
-      "logits/chosen": -1.4592185020446777,
-      "logits/rejected": -1.3416706323623657,
-      "logps/chosen": -249.4734649658203,
-      "logps/rejected": -405.06207275390625,
-      "loss": 0.2966,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -1.1794850826263428,
-      "rewards/margins": 2.0616517066955566,
-      "rewards/rejected": -3.2411365509033203,
+      "logits/chosen": -2.5073745250701904,
+      "logits/rejected": -2.553334951400757,
+      "logps/chosen": -332.49945068359375,
+      "logps/rejected": -319.8727722167969,
+      "loss": 0.3368,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -0.5055555105209351,
+      "rewards/margins": 1.468584418296814,
+      "rewards/rejected": -1.9741401672363281,
       "step": 80
     },
     {
       "epoch": 0.09,
-      "grad_norm": 34.896761047309674,
+      "grad_norm": 23.431306250567218,
       "learning_rate": 4.6874999999999996e-07,
-      "logits/chosen": -1.2935254573822021,
-      "logits/rejected": -1.215850591659546,
-      "logps/chosen": -236.42935180664062,
-      "logps/rejected": -427.88079833984375,
-      "loss": 0.2792,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -1.0514061450958252,
-      "rewards/margins": 2.4284262657165527,
-      "rewards/rejected": -3.479832410812378,
+      "logits/chosen": -2.611389636993408,
+      "logits/rejected": -2.643505334854126,
+      "logps/chosen": -355.7128601074219,
+      "logps/rejected": -361.992431640625,
+      "loss": 0.309,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -0.61676025390625,
+      "rewards/margins": 1.7021185159683228,
+      "rewards/rejected": -2.318878412246704,
       "step": 90
     },
     {
       "epoch": 0.1,
-      "grad_norm": 40.95979792917679,
+      "grad_norm": 12.356735796295311,
       "learning_rate": 4.999732492681437e-07,
-      "logits/chosen": -1.425644040107727,
-      "logits/rejected": -1.2574630975723267,
-      "logps/chosen": -236.2029571533203,
-      "logps/rejected": -470.1878967285156,
-      "loss": 0.2555,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -1.0542919635772705,
-      "rewards/margins": 2.8619871139526367,
-      "rewards/rejected": -3.916278839111328,
+      "logits/chosen": -2.621342420578003,
+      "logits/rejected": -2.6476845741271973,
+      "logps/chosen": -319.81988525390625,
+      "logps/rejected": -322.2472229003906,
+      "loss": 0.3047,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -0.6059430241584778,
+      "rewards/margins": 1.4836266040802002,
+      "rewards/rejected": -2.089569568634033,
       "step": 100
     },
     {
       "epoch": 0.1,
-      "eval_logits/chosen": -1.394241213798523,
-      "eval_logits/rejected": -1.3358198404312134,
-      "eval_logps/chosen": -800.6121215820312,
-      "eval_logps/rejected": -884.5335083007812,
-      "eval_loss": 1.4172171354293823,
-      "eval_rewards/accuracies": 0.58984375,
-      "eval_rewards/chosen": -4.888356685638428,
-      "eval_rewards/margins": 0.7817266583442688,
-      "eval_rewards/rejected": -5.670083045959473,
-      "eval_runtime": 97.8793,
-      "eval_samples_per_second": 20.433,
-      "eval_steps_per_second": 0.327,
+      "eval_logits/chosen": -2.44050931930542,
+      "eval_logits/rejected": -2.3863425254821777,
+      "eval_logps/chosen": -361.08013916015625,
+      "eval_logps/rejected": -337.77484130859375,
+      "eval_loss": 0.8551393747329712,
+      "eval_rewards/accuracies": 0.3203125,
+      "eval_rewards/chosen": -0.49303656816482544,
+      "eval_rewards/margins": -0.29053980112075806,
+      "eval_rewards/rejected": -0.20249679684638977,
+      "eval_runtime": 97.3683,
+      "eval_samples_per_second": 20.541,
+      "eval_steps_per_second": 0.329,
       "step": 100
     },
     {
       "epoch": 0.12,
-      "grad_norm": 32.26210278499666,
+      "grad_norm": 12.540254444715838,
       "learning_rate": 4.996723692767926e-07,
-      "logits/chosen": -1.300011157989502,
-      "logits/rejected": -1.1855316162109375,
-      "logps/chosen": -253.3606719970703,
-      "logps/rejected": -546.6198120117188,
-      "loss": 0.2322,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -1.3233938217163086,
-      "rewards/margins": 3.418271541595459,
-      "rewards/rejected": -4.741665840148926,
+      "logits/chosen": -2.7024824619293213,
+      "logits/rejected": -2.732452154159546,
+      "logps/chosen": -409.3314208984375,
+      "logps/rejected": -413.891357421875,
+      "loss": 0.2958,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -0.829090416431427,
+      "rewards/margins": 2.046325206756592,
+      "rewards/rejected": -2.875415325164795,
       "step": 110
     },
     {
       "epoch": 0.13,
-      "grad_norm": 47.80125591905201,
+      "grad_norm": 15.24853315289221,
       "learning_rate": 4.990375746213598e-07,
-      "logits/chosen": -1.4404656887054443,
-      "logits/rejected": -1.311632513999939,
-      "logps/chosen": -286.67169189453125,
-      "logps/rejected": -644.265625,
-      "loss": 0.2299,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -1.5356190204620361,
-      "rewards/margins": 4.171111583709717,
-      "rewards/rejected": -5.706730842590332,
+      "logits/chosen": -2.661564350128174,
+      "logits/rejected": -2.7124838829040527,
+      "logps/chosen": -378.96533203125,
+      "logps/rejected": -433.1641540527344,
+      "loss": 0.2884,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -0.9071281552314758,
+      "rewards/margins": 2.1523900032043457,
+      "rewards/rejected": -3.059518337249756,
       "step": 120
     },
     {
       "epoch": 0.14,
-      "grad_norm": 103.21873941787233,
+      "grad_norm": 20.41295055769396,
       "learning_rate": 4.980697142834314e-07,
-      "logits/chosen": -1.4422317743301392,
-      "logits/rejected": -1.295238733291626,
-      "logps/chosen": -258.7148742675781,
-      "logps/rejected": -593.8055419921875,
-      "loss": 0.2288,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -1.2843689918518066,
-      "rewards/margins": 3.8400402069091797,
-      "rewards/rejected": -5.1244096755981445,
+      "logits/chosen": -2.693155527114868,
+      "logits/rejected": -2.738525390625,
+      "logps/chosen": -401.2513122558594,
+      "logps/rejected": -464.12353515625,
+      "loss": 0.2528,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -1.0387437343597412,
+      "rewards/margins": 2.259091854095459,
+      "rewards/rejected": -3.2978355884552,
       "step": 130
     },
     {
       "epoch": 0.15,
-      "grad_norm": 40.83439235144051,
+      "grad_norm": 27.736891370306594,
       "learning_rate": 4.967700826904229e-07,
-      "logits/chosen": -1.4604966640472412,
-      "logits/rejected": -1.289422631263733,
-      "logps/chosen": -291.0477600097656,
-      "logps/rejected": -753.3111572265625,
-      "loss": 0.2219,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -1.6053016185760498,
-      "rewards/margins": 5.150030612945557,
-      "rewards/rejected": -6.755332946777344,
+      "logits/chosen": -2.6829416751861572,
+      "logits/rejected": -2.743159770965576,
+      "logps/chosen": -405.8708801269531,
+      "logps/rejected": -488.6322326660156,
+      "loss": 0.2529,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -1.1785838603973389,
+      "rewards/margins": 2.3818531036376953,
+      "rewards/rejected": -3.560436964035034,
       "step": 140
     },
     {
       "epoch": 0.16,
-      "grad_norm": 35.47272090007365,
+      "grad_norm": 31.193716735224292,
       "learning_rate": 4.951404179843962e-07,
-      "logits/chosen": -1.439012885093689,
-      "logits/rejected": -1.197693109512329,
-      "logps/chosen": -317.4605407714844,
-      "logps/rejected": -829.1416015625,
-      "loss": 0.1987,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -1.6784883737564087,
-      "rewards/margins": 5.721285820007324,
-      "rewards/rejected": -7.399774074554443,
+      "logits/chosen": -2.4764091968536377,
+      "logits/rejected": -2.481703281402588,
+      "logps/chosen": -504.5975036621094,
+      "logps/rejected": -567.6092529296875,
+      "loss": 0.2358,
+      "rewards/accuracies": 0.893750011920929,
+      "rewards/chosen": -1.8161567449569702,
+      "rewards/margins": 2.5039477348327637,
+      "rewards/rejected": -4.320104598999023,
       "step": 150
     },
     {
       "epoch": 0.17,
-      "grad_norm": 41.95405531315011,
+      "grad_norm": 16.84967305265633,
       "learning_rate": 4.931828996974498e-07,
-      "logits/chosen": -1.3928442001342773,
-      "logits/rejected": -1.1220872402191162,
-      "logps/chosen": -289.9758605957031,
-      "logps/rejected": -711.052001953125,
-      "loss": 0.21,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -1.6032127141952515,
-      "rewards/margins": 4.786148548126221,
-      "rewards/rejected": -6.389361381530762,
+      "logits/chosen": -2.4284980297088623,
+      "logits/rejected": -2.4476141929626465,
+      "logps/chosen": -411.03546142578125,
+      "logps/rejected": -520.0714721679688,
+      "loss": 0.2377,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -1.3872876167297363,
+      "rewards/margins": 2.5429630279541016,
+      "rewards/rejected": -3.930250883102417,
       "step": 160
     },
     {
       "epoch": 0.18,
-      "grad_norm": 44.30975250132851,
+      "grad_norm": 25.1927910072169,
       "learning_rate": 4.909001458367866e-07,
-      "logits/chosen": -1.4432841539382935,
-      "logits/rejected": -1.237818956375122,
-      "logps/chosen": -394.95751953125,
-      "logps/rejected": -859.7473754882812,
-      "loss": 0.1885,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -2.7817580699920654,
-      "rewards/margins": 5.132468223571777,
-      "rewards/rejected": -7.914226531982422,
+      "logits/chosen": -2.2736942768096924,
+      "logits/rejected": -2.292642831802368,
+      "logps/chosen": -452.697998046875,
+      "logps/rejected": -563.0865478515625,
+      "loss": 0.2128,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -1.6114835739135742,
+      "rewards/margins": 2.781120538711548,
+      "rewards/rejected": -4.392604351043701,
       "step": 170
     },
     {
       "epoch": 0.19,
-      "grad_norm": 42.31623927536496,
+      "grad_norm": 18.107740978814483,
       "learning_rate": 4.882952093833627e-07,
-      "logits/chosen": -1.450234055519104,
-      "logits/rejected": -1.2738139629364014,
-      "logps/chosen": -342.06048583984375,
-      "logps/rejected": -785.0755004882812,
-      "loss": 0.2045,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -2.051614999771118,
-      "rewards/margins": 4.953377723693848,
-      "rewards/rejected": -7.0049920082092285,
+      "logits/chosen": -2.1593220233917236,
+      "logits/rejected": -2.1532785892486572,
+      "logps/chosen": -509.1954650878906,
+      "logps/rejected": -644.2286376953125,
+      "loss": 0.2037,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -1.7663100957870483,
+      "rewards/margins": 3.2817070484161377,
+      "rewards/rejected": -5.0480170249938965,
       "step": 180
     },
     {
       "epoch": 0.2,
-      "grad_norm": 39.855754537956905,
+      "grad_norm": 22.004607228534194,
       "learning_rate": 4.853715742087946e-07,
-      "logits/chosen": -1.3248379230499268,
-      "logits/rejected": -1.09066903591156,
-      "logps/chosen": -368.9568786621094,
-      "logps/rejected": -883.1036987304688,
-      "loss": 0.1906,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -2.428023099899292,
-      "rewards/margins": 5.649343967437744,
-      "rewards/rejected": -8.077366828918457,
+      "logits/chosen": -2.1366734504699707,
+      "logits/rejected": -2.1492080688476562,
+      "logps/chosen": -469.94976806640625,
+      "logps/rejected": -587.4523315429688,
+      "loss": 0.2059,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -1.6838188171386719,
+      "rewards/margins": 2.9172627925872803,
+      "rewards/rejected": -4.601081848144531,
       "step": 190
     },
     {
       "epoch": 0.21,
-      "grad_norm": 43.44905869134368,
+      "grad_norm": 23.839892866837005,
       "learning_rate": 4.821331504159906e-07,
-      "logits/chosen": -1.122195839881897,
-      "logits/rejected": -0.9366437196731567,
-      "logps/chosen": -330.4066467285156,
-      "logps/rejected": -748.525634765625,
-      "loss": 0.1854,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -2.157914161682129,
-      "rewards/margins": 4.577654838562012,
-      "rewards/rejected": -6.735569000244141,
+      "logits/chosen": -2.0906856060028076,
+      "logits/rejected": -2.1124696731567383,
+      "logps/chosen": -466.5874938964844,
+      "logps/rejected": -608.2426147460938,
+      "loss": 0.1861,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -1.8450790643692017,
+      "rewards/margins": 3.033769130706787,
+      "rewards/rejected": -4.878848552703857,
       "step": 200
     },
     {
       "epoch": 0.21,
-      "eval_logits/chosen": -1.1852705478668213,
-      "eval_logits/rejected": -1.108778953552246,
-      "eval_logps/chosen": -926.8517456054688,
-      "eval_logps/rejected": -1050.1199951171875,
-      "eval_loss": 1.6753935813903809,
-      "eval_rewards/accuracies": 0.62109375,
-      "eval_rewards/chosen": -6.15075159072876,
-      "eval_rewards/margins": 1.1751970052719116,
-      "eval_rewards/rejected": -7.325948715209961,
-      "eval_runtime": 97.5501,
-      "eval_samples_per_second": 20.502,
-      "eval_steps_per_second": 0.328,
+      "eval_logits/chosen": -1.5417730808258057,
+      "eval_logits/rejected": -1.5107452869415283,
+      "eval_logps/chosen": -450.2716369628906,
+      "eval_logps/rejected": -421.0933532714844,
+      "eval_loss": 1.0494639873504639,
+      "eval_rewards/accuracies": 0.38671875,
+      "eval_rewards/chosen": -1.3849513530731201,
+      "eval_rewards/margins": -0.3492693305015564,
+      "eval_rewards/rejected": -1.0356820821762085,
+      "eval_runtime": 97.2226,
+      "eval_samples_per_second": 20.571,
+      "eval_steps_per_second": 0.329,
       "step": 200
     },
     {
       "epoch": 0.22,
-      "grad_norm": 36.690007199237776,
+      "grad_norm": 21.766108201350445,
       "learning_rate": 4.785842691097342e-07,
-      "logits/chosen": -1.129880666732788,
-      "logits/rejected": -0.8643050193786621,
-      "logps/chosen": -371.864013671875,
-      "logps/rejected": -938.4737548828125,
-      "loss": 0.174,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -2.416229009628296,
-      "rewards/margins": 6.226431369781494,
-      "rewards/rejected": -8.642660140991211,
+      "logits/chosen": -1.914615273475647,
+      "logits/rejected": -1.9393634796142578,
+      "logps/chosen": -541.3153076171875,
+      "logps/rejected": -766.5374145507812,
+      "loss": 0.1892,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -2.552523374557495,
+      "rewards/margins": 3.7902703285217285,
+      "rewards/rejected": -6.342793941497803,
       "step": 210
     },
     {
       "epoch": 0.23,
-      "grad_norm": 50.37285904600086,
+      "grad_norm": 37.7465298870174,
       "learning_rate": 4.7472967660421603e-07,
-      "logits/chosen": -1.2257055044174194,
-      "logits/rejected": -0.964257538318634,
-      "logps/chosen": -339.07196044921875,
-      "logps/rejected": -897.4797973632812,
-      "loss": 0.1776,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -2.065497875213623,
-      "rewards/margins": 6.163290023803711,
-      "rewards/rejected": -8.228787422180176,
+      "logits/chosen": -2.004246234893799,
+      "logits/rejected": -2.0251355171203613,
+      "logps/chosen": -528.6552734375,
+      "logps/rejected": -743.7687377929688,
+      "loss": 0.1798,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -2.033097743988037,
+      "rewards/margins": 4.051178932189941,
+      "rewards/rejected": -6.0842766761779785,
       "step": 220
     },
     {
       "epoch": 0.24,
-      "grad_norm": 41.9657054722659,
+      "grad_norm": 37.93991066707489,
       "learning_rate": 4.705745280752585e-07,
-      "logits/chosen": -1.2483174800872803,
-      "logits/rejected": -0.9970757365226746,
-      "logps/chosen": -328.69775390625,
-      "logps/rejected": -740.644775390625,
-      "loss": 0.1898,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -1.9448823928833008,
-      "rewards/margins": 4.71053409576416,
-      "rewards/rejected": -6.655416965484619,
+      "logits/chosen": -1.9027751684188843,
+      "logits/rejected": -1.8593933582305908,
+      "logps/chosen": -495.2493591308594,
+      "logps/rejected": -688.589599609375,
+      "loss": 0.1765,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -1.9664674997329712,
+      "rewards/margins": 3.703148603439331,
+      "rewards/rejected": -5.66961669921875,
       "step": 230
     },
     {
       "epoch": 0.25,
-      "grad_norm": 44.41737576885688,
+      "grad_norm": 19.738306178674453,
       "learning_rate": 4.6612438066572555e-07,
-      "logits/chosen": -1.1642967462539673,
-      "logits/rejected": -0.9903414845466614,
-      "logps/chosen": -318.5704650878906,
-      "logps/rejected": -773.8700561523438,
-      "loss": 0.1919,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -1.9761199951171875,
-      "rewards/margins": 5.03547477722168,
-      "rewards/rejected": -7.011595249176025,
+      "logits/chosen": -1.7326488494873047,
+      "logits/rejected": -1.585479497909546,
+      "logps/chosen": -627.2045288085938,
+      "logps/rejected": -877.15234375,
+      "loss": 0.164,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -3.1001508235931396,
+      "rewards/margins": 4.346390247344971,
+      "rewards/rejected": -7.446540832519531,
       "step": 240
     },
     {
       "epoch": 0.26,
-      "grad_norm": 35.762754087270444,
+      "grad_norm": 20.99257000654717,
       "learning_rate": 4.6138518605333664e-07,
-      "logits/chosen": -1.106542706489563,
-      "logits/rejected": -0.8413636088371277,
-      "logps/chosen": -434.7865295410156,
-      "logps/rejected": -1112.9827880859375,
-      "loss": 0.1773,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -2.952258825302124,
-      "rewards/margins": 7.3310394287109375,
-      "rewards/rejected": -10.283297538757324,
+      "logits/chosen": -1.7733532190322876,
+      "logits/rejected": -1.5429089069366455,
+      "logps/chosen": -534.9285888671875,
+      "logps/rejected": -766.8218383789062,
+      "loss": 0.1779,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -2.3250820636749268,
+      "rewards/margins": 4.1490068435668945,
+      "rewards/rejected": -6.4740891456604,
       "step": 250
     },
     {
       "epoch": 0.27,
-      "grad_norm": 35.92606291569474,
+      "grad_norm": 24.636105883207083,
       "learning_rate": 4.5636328249082514e-07,
-      "logits/chosen": -0.869143009185791,
-      "logits/rejected": -0.5741620063781738,
-      "logps/chosen": -467.0962829589844,
-      "logps/rejected": -1112.403564453125,
-      "loss": 0.162,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -3.3054141998291016,
-      "rewards/margins": 7.030813694000244,
-      "rewards/rejected": -10.336227416992188,
+      "logits/chosen": -1.7743794918060303,
+      "logits/rejected": -1.6784734725952148,
+      "logps/chosen": -502.83929443359375,
+      "logps/rejected": -686.4469604492188,
+      "loss": 0.1666,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -2.020374298095703,
+      "rewards/margins": 3.5390701293945312,
+      "rewards/rejected": -5.559444427490234,
       "step": 260
     },
     {
       "epoch": 0.28,
-      "grad_norm": 53.50950070995327,
+      "grad_norm": 26.442049627508165,
       "learning_rate": 4.510653863290871e-07,
-      "logits/chosen": -0.9721959829330444,
-      "logits/rejected": -0.6126078367233276,
-      "logps/chosen": -439.41436767578125,
-      "logps/rejected": -1052.373291015625,
-      "loss": 0.1814,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -2.9460256099700928,
-      "rewards/margins": 6.7303972244262695,
-      "rewards/rejected": -9.676422119140625,
+      "logits/chosen": -1.8089313507080078,
+      "logits/rejected": -1.504288673400879,
+      "logps/chosen": -553.7681884765625,
+      "logps/rejected": -773.63671875,
+      "loss": 0.163,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.3301241397857666,
+      "rewards/margins": 4.031432628631592,
+      "rewards/rejected": -6.361557483673096,
       "step": 270
     },
     {
       "epoch": 0.29,
-      "grad_norm": 39.535885045091504,
+      "grad_norm": 21.752850388248913,
       "learning_rate": 4.4549858303465737e-07,
-      "logits/chosen": -1.0113680362701416,
-      "logits/rejected": -0.6861897706985474,
-      "logps/chosen": -475.51251220703125,
-      "logps/rejected": -1126.9036865234375,
-      "loss": 0.1758,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -3.39562726020813,
-      "rewards/margins": 7.080619812011719,
-      "rewards/rejected": -10.47624683380127,
+      "logits/chosen": -1.792066216468811,
+      "logits/rejected": -1.5596303939819336,
+      "logps/chosen": -521.0010986328125,
+      "logps/rejected": -791.414794921875,
+      "loss": 0.1514,
+      "rewards/accuracies": 0.918749988079071,
+      "rewards/chosen": -2.1834633350372314,
+      "rewards/margins": 4.415928363800049,
+      "rewards/rejected": -6.599390983581543,
       "step": 280
     },
     {
       "epoch": 0.3,
-      "grad_norm": 27.716897646836834,
+      "grad_norm": 21.62051159722618,
       "learning_rate": 4.396703177135261e-07,
-      "logits/chosen": -0.9551790952682495,
-      "logits/rejected": -0.6880333423614502,
-      "logps/chosen": -410.09857177734375,
-      "logps/rejected": -904.8678588867188,
-      "loss": 0.1956,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -2.7216389179229736,
-      "rewards/margins": 5.489422798156738,
-      "rewards/rejected": -8.21106243133545,
+      "logits/chosen": -1.573697805404663,
+      "logits/rejected": -1.1857482194900513,
+      "logps/chosen": -611.6795654296875,
+      "logps/rejected": -855.2744140625,
+      "loss": 0.1493,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -2.884495258331299,
+      "rewards/margins": 4.296299934387207,
+      "rewards/rejected": -7.180795192718506,
       "step": 290
     },
     {
       "epoch": 0.31,
-      "grad_norm": 39.15676652312449,
+      "grad_norm": 16.812667194402408,
       "learning_rate": 4.335883851539693e-07,
-      "logits/chosen": -1.0925099849700928,
-      "logits/rejected": -0.763215184211731,
-      "logps/chosen": -293.86553955078125,
-      "logps/rejected": -759.0989990234375,
-      "loss": 0.1799,
+      "logits/chosen": -1.738490343093872,
+      "logits/rejected": -1.7563718557357788,
+      "logps/chosen": -494.66253662109375,
+      "logps/rejected": -731.6812744140625,
+      "loss": 0.1608,
       "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -1.6122547388076782,
-      "rewards/margins": 5.217433452606201,
-      "rewards/rejected": -6.82968807220459,
+      "rewards/chosen": -1.9538425207138062,
+      "rewards/margins": 3.9912962913513184,
+      "rewards/rejected": -5.945138931274414,
       "step": 300
     },
     {
       "epoch": 0.31,
-      "eval_logits/chosen": -1.1109943389892578,
-      "eval_logits/rejected": -1.0192608833312988,
-      "eval_logps/chosen": -903.3418579101562,
-      "eval_logps/rejected": -1015.4615478515625,
-      "eval_loss": 1.5589935779571533,
-      "eval_rewards/accuracies": 0.59765625,
-      "eval_rewards/chosen": -5.915654182434082,
-      "eval_rewards/margins": 1.0637093782424927,
-      "eval_rewards/rejected": -6.979363441467285,
-      "eval_runtime": 97.4807,
-      "eval_samples_per_second": 20.517,
-      "eval_steps_per_second": 0.328,
+      "eval_logits/chosen": -1.4367032051086426,
+      "eval_logits/rejected": -1.4021562337875366,
+      "eval_logps/chosen": -454.9446105957031,
+      "eval_logps/rejected": -422.9683532714844,
+      "eval_loss": 1.0910040140151978,
+      "eval_rewards/accuracies": 0.39453125,
+      "eval_rewards/chosen": -1.4316813945770264,
+      "eval_rewards/margins": -0.37724918127059937,
+      "eval_rewards/rejected": -1.0544321537017822,
+      "eval_runtime": 97.2015,
+      "eval_samples_per_second": 20.576,
+      "eval_steps_per_second": 0.329,
       "step": 300
     },
     {
       "epoch": 0.32,
-      "grad_norm": 31.761072657280515,
+      "grad_norm": 23.495796172679743,
       "learning_rate": 4.272609194017105e-07,
-      "logits/chosen": -1.161056637763977,
-      "logits/rejected": -0.8523654937744141,
-      "logps/chosen": -387.35833740234375,
-      "logps/rejected": -977.5528564453125,
-      "loss": 0.1806,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -2.436070442199707,
-      "rewards/margins": 6.559242248535156,
-      "rewards/rejected": -8.99531364440918,
+      "logits/chosen": -1.7330036163330078,
+      "logits/rejected": -1.6350265741348267,
+      "logps/chosen": -519.0903930664062,
+      "logps/rejected": -720.5287475585938,
+      "loss": 0.1552,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -1.7445989847183228,
+      "rewards/margins": 4.112780570983887,
+      "rewards/rejected": -5.857378959655762,
       "step": 310
     },
     {
       "epoch": 0.33,
-      "grad_norm": 36.03487432587838,
+      "grad_norm": 16.298880424996614,
       "learning_rate": 4.2069638288135547e-07,
-      "logits/chosen": -0.8736389875411987,
-      "logits/rejected": -0.6802955865859985,
-      "logps/chosen": -391.59564208984375,
-      "logps/rejected": -1047.556884765625,
-      "loss": 0.1698,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -2.664311170578003,
-      "rewards/margins": 7.013079643249512,
-      "rewards/rejected": -9.677392959594727,
+      "logits/chosen": -1.7347389459609985,
+      "logits/rejected": -1.7288787364959717,
+      "logps/chosen": -510.26483154296875,
+      "logps/rejected": -713.36572265625,
+      "loss": 0.157,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -1.7514524459838867,
+      "rewards/margins": 3.96923828125,
+      "rewards/rejected": -5.720690727233887,
       "step": 320
     },
     {
       "epoch": 0.35,
-      "grad_norm": 31.572371493306342,
+      "grad_norm": 19.47125814373079,
       "learning_rate": 4.139035550786494e-07,
-      "logits/chosen": -1.1484278440475464,
-      "logits/rejected": -0.7356959581375122,
-      "logps/chosen": -385.1824645996094,
-      "logps/rejected": -1041.6712646484375,
-      "loss": 0.1467,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -2.5319342613220215,
-      "rewards/margins": 7.115927219390869,
-      "rewards/rejected": -9.647860527038574,
+      "logits/chosen": -1.1601016521453857,
+      "logits/rejected": -0.8797602653503418,
+      "logps/chosen": -555.39990234375,
+      "logps/rejected": -796.0065307617188,
+      "loss": 0.1561,
+      "rewards/accuracies": 0.887499988079071,
+      "rewards/chosen": -2.3634750843048096,
+      "rewards/margins": 4.329623222351074,
+      "rewards/rejected": -6.693098545074463,
       "step": 330
     },
     {
       "epoch": 0.36,
-      "grad_norm": 30.493158942186962,
+      "grad_norm": 28.613677806731356,
       "learning_rate": 4.0689152079869306e-07,
-      "logits/chosen": -1.111301064491272,
-      "logits/rejected": -0.7459093332290649,
-      "logps/chosen": -419.9774475097656,
-      "logps/rejected": -1145.664306640625,
-      "loss": 0.1589,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -2.8832128047943115,
-      "rewards/margins": 7.796321868896484,
-      "rewards/rejected": -10.679533958435059,
+      "logits/chosen": -1.2572317123413086,
+      "logits/rejected": -0.9607254862785339,
+      "logps/chosen": -606.7630615234375,
+      "logps/rejected": -846.2305908203125,
+      "loss": 0.1416,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -2.5160038471221924,
+      "rewards/margins": 4.53468132019043,
+      "rewards/rejected": -7.050685882568359,
       "step": 340
     },
     {
       "epoch": 0.37,
-      "grad_norm": 35.73091881437617,
+      "grad_norm": 17.707474904024725,
       "learning_rate": 3.99669658015821e-07,
-      "logits/chosen": -0.9885336756706238,
-      "logits/rejected": -0.6827311515808105,
-      "logps/chosen": -417.18621826171875,
-      "logps/rejected": -1244.3375244140625,
-      "loss": 0.1607,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -2.9024839401245117,
-      "rewards/margins": 8.765107154846191,
-      "rewards/rejected": -11.667591094970703,
+      "logits/chosen": -1.3649070262908936,
+      "logits/rejected": -0.9824110269546509,
+      "logps/chosen": -558.0640869140625,
+      "logps/rejected": -775.701171875,
+      "loss": 0.1563,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -2.48624849319458,
+      "rewards/margins": 4.034348964691162,
+      "rewards/rejected": -6.5205979347229,
       "step": 350
     },
     {
       "epoch": 0.38,
-      "grad_norm": 26.448274353945774,
+      "grad_norm": 17.329304967511433,
       "learning_rate": 3.92247625331392e-07,
-      "logits/chosen": -0.9021160006523132,
-      "logits/rejected": -0.46636825799942017,
-      "logps/chosen": -396.3219299316406,
-      "logps/rejected": -1072.203857421875,
-      "loss": 0.168,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -2.622156858444214,
-      "rewards/margins": 7.3311333656311035,
-      "rewards/rejected": -9.953289985656738,
+      "logits/chosen": -1.5912964344024658,
+      "logits/rejected": -1.5140416622161865,
+      "logps/chosen": -473.7013244628906,
+      "logps/rejected": -705.4163818359375,
+      "loss": 0.1391,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -2.0327391624450684,
+      "rewards/margins": 3.767451047897339,
+      "rewards/rejected": -5.800190448760986,
       "step": 360
     },
     {
       "epoch": 0.39,
-      "grad_norm": 26.57225676522975,
+      "grad_norm": 21.039638710046997,
       "learning_rate": 3.846353490562664e-07,
-      "logits/chosen": -0.7423251867294312,
-      "logits/rejected": -0.3315241038799286,
-      "logps/chosen": -403.2605285644531,
-      "logps/rejected": -1123.38623046875,
-      "loss": 0.1524,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -2.6948583126068115,
-      "rewards/margins": 7.7378740310668945,
-      "rewards/rejected": -10.432731628417969,
+      "logits/chosen": -1.3319923877716064,
+      "logits/rejected": -1.090010404586792,
+      "logps/chosen": -520.6598510742188,
+      "logps/rejected": -788.8143310546875,
+      "loss": 0.1403,
+      "rewards/accuracies": 0.8687499761581421,
+      "rewards/chosen": -2.3732516765594482,
+      "rewards/margins": 4.287914276123047,
+      "rewards/rejected": -6.661165714263916,
       "step": 370
     },
     {
       "epoch": 0.4,
-      "grad_norm": 31.253155533057107,
+      "grad_norm": 19.41739908042072,
       "learning_rate": 3.768430099352445e-07,
-      "logits/chosen": -0.6284547448158264,
-      "logits/rejected": -0.2557411789894104,
-      "logps/chosen": -438.34454345703125,
-      "logps/rejected": -1187.157958984375,
-      "loss": 0.161,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -3.113786220550537,
-      "rewards/margins": 7.986392974853516,
-      "rewards/rejected": -11.100178718566895,
+      "logits/chosen": -1.2049062252044678,
+      "logits/rejected": -0.9995657205581665,
+      "logps/chosen": -531.4547119140625,
+      "logps/rejected": -863.2029418945312,
+      "loss": 0.1359,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.5220770835876465,
+      "rewards/margins": 4.725777626037598,
+      "rewards/rejected": -7.247855186462402,
       "step": 380
     },
     {
       "epoch": 0.41,
-      "grad_norm": 30.747112252887895,
+      "grad_norm": 30.846420024935966,
       "learning_rate": 3.6888102953122304e-07,
-      "logits/chosen": -0.888546347618103,
-      "logits/rejected": -0.5933178663253784,
-      "logps/chosen": -410.141357421875,
-      "logps/rejected": -1144.3177490234375,
-      "loss": 0.1578,
-      "rewards/accuracies": 0.8687499761581421,
-      "rewards/chosen": -2.877619981765747,
-      "rewards/margins": 7.820855617523193,
-      "rewards/rejected": -10.69847583770752,
+      "logits/chosen": -1.4293053150177002,
+      "logits/rejected": -1.129652976989746,
+      "logps/chosen": -509.592529296875,
+      "logps/rejected": -834.6383056640625,
+      "loss": 0.1217,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -2.1002399921417236,
+      "rewards/margins": 4.87478494644165,
+      "rewards/rejected": -6.975025177001953,
       "step": 390
     },
     {
       "epoch": 0.42,
-      "grad_norm": 46.52274946683974,
+      "grad_norm": 26.99256302878051,
       "learning_rate": 3.607600562872785e-07,
-      "logits/chosen": -0.6739174127578735,
-      "logits/rejected": -0.4188503324985504,
-      "logps/chosen": -447.11083984375,
-      "logps/rejected": -1131.9605712890625,
-      "loss": 0.1679,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -3.236891508102417,
-      "rewards/margins": 7.345515251159668,
-      "rewards/rejected": -10.582406997680664,
+      "logits/chosen": -1.3965580463409424,
+      "logits/rejected": -0.9727011919021606,
+      "logps/chosen": -591.9832763671875,
+      "logps/rejected": -926.43701171875,
+      "loss": 0.1368,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -2.8865580558776855,
+      "rewards/margins": 5.026923656463623,
+      "rewards/rejected": -7.913481712341309,
       "step": 400
     },
     {
       "epoch": 0.42,
-      "eval_logits/chosen": -0.7096253037452698,
-      "eval_logits/rejected": -0.5753335356712341,
-      "eval_logps/chosen": -1096.810791015625,
-      "eval_logps/rejected": -1238.125244140625,
-      "eval_loss": 2.102952241897583,
-      "eval_rewards/accuracies": 0.609375,
-      "eval_rewards/chosen": -7.850342750549316,
-      "eval_rewards/margins": 1.355657696723938,
-      "eval_rewards/rejected": -9.206000328063965,
-      "eval_runtime": 97.4138,
-      "eval_samples_per_second": 20.531,
-      "eval_steps_per_second": 0.328,
+      "eval_logits/chosen": -1.053754210472107,
+      "eval_logits/rejected": -1.013121485710144,
+      "eval_logps/chosen": -520.169921875,
+      "eval_logps/rejected": -479.64556884765625,
+      "eval_loss": 1.3009716272354126,
+      "eval_rewards/accuracies": 0.41015625,
+      "eval_rewards/chosen": -2.0839340686798096,
+      "eval_rewards/margins": -0.46272987127304077,
+      "eval_rewards/rejected": -1.6212042570114136,
+      "eval_runtime": 97.208,
+      "eval_samples_per_second": 20.574,
+      "eval_steps_per_second": 0.329,
       "step": 400
     },
     {
       "epoch": 0.43,
-      "grad_norm": 39.27703884155879,
+      "grad_norm": 18.264964860581,
       "learning_rate": 3.5249095128531856e-07,
-      "logits/chosen": -0.6762361526489258,
-      "logits/rejected": -0.2780495882034302,
-      "logps/chosen": -492.8828125,
-      "logps/rejected": -1225.805419921875,
-      "loss": 0.1705,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -3.5533318519592285,
-      "rewards/margins": 7.891423225402832,
-      "rewards/rejected": -11.444755554199219,
+      "logits/chosen": -1.6575086116790771,
+      "logits/rejected": -1.4926228523254395,
+      "logps/chosen": -505.78448486328125,
+      "logps/rejected": -805.1259155273438,
+      "loss": 0.151,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -2.203605890274048,
+      "rewards/margins": 4.460430145263672,
+      "rewards/rejected": -6.664034843444824,
       "step": 410
     },
     {
       "epoch": 0.44,
-      "grad_norm": 29.296123751306272,
+      "grad_norm": 19.09047976589624,
       "learning_rate": 3.4408477372034736e-07,
-      "logits/chosen": -0.6263514757156372,
-      "logits/rejected": -0.36908912658691406,
-      "logps/chosen": -436.956298828125,
-      "logps/rejected": -1203.846923828125,
-      "loss": 0.1542,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -3.0062294006347656,
-      "rewards/margins": 8.188032150268555,
-      "rewards/rejected": -11.19426155090332,
+      "logits/chosen": -1.6263633966445923,
+      "logits/rejected": -1.4956812858581543,
+      "logps/chosen": -479.7705993652344,
+      "logps/rejected": -766.1605224609375,
+      "loss": 0.1448,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -2.093071937561035,
+      "rewards/margins": 4.394181728363037,
+      "rewards/rejected": -6.4872541427612305,
       "step": 420
     },
     {
       "epoch": 0.45,
-      "grad_norm": 38.53939686623868,
+      "grad_norm": 26.836781627013515,
       "learning_rate": 3.3555276610977276e-07,
-      "logits/chosen": -0.7492274045944214,
-      "logits/rejected": -0.3440536856651306,
-      "logps/chosen": -440.6358337402344,
-      "logps/rejected": -1202.001953125,
-      "loss": 0.1624,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -3.0986926555633545,
-      "rewards/margins": 8.180191040039062,
-      "rewards/rejected": -11.278883934020996,
+      "logits/chosen": -1.662540078163147,
+      "logits/rejected": -1.4674957990646362,
+      "logps/chosen": -527.6272583007812,
+      "logps/rejected": -827.05224609375,
+      "loss": 0.1477,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -2.207035541534424,
+      "rewards/margins": 4.785927772521973,
+      "rewards/rejected": -6.9929633140563965,
       "step": 430
     },
     {
       "epoch": 0.46,
-      "grad_norm": 44.81056957165417,
+      "grad_norm": 24.71697296764681,
       "learning_rate": 3.269063392575352e-07,
-      "logits/chosen": -0.7927631139755249,
-      "logits/rejected": -0.4414951801300049,
-      "logps/chosen": -404.3497009277344,
-      "logps/rejected": -1111.9443359375,
-      "loss": 0.1883,
-      "rewards/accuracies": 0.862500011920929,
-      "rewards/chosen": -2.7665016651153564,
-      "rewards/margins": 7.598056793212891,
-      "rewards/rejected": -10.364558219909668,
+      "logits/chosen": -1.682995080947876,
+      "logits/rejected": -1.625732660293579,
+      "logps/chosen": -521.5386352539062,
+      "logps/rejected": -861.8421630859375,
+      "loss": 0.1431,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.1410486698150635,
+      "rewards/margins": 5.073235511779785,
+      "rewards/rejected": -7.214284420013428,
       "step": 440
     },
     {
       "epoch": 0.47,
-      "grad_norm": 46.720798060289724,
+      "grad_norm": 35.22220664866878,
       "learning_rate": 3.1815705699316964e-07,
-      "logits/chosen": -0.7731117010116577,
-      "logits/rejected": -0.4938638210296631,
-      "logps/chosen": -400.8499755859375,
-      "logps/rejected": -1148.82177734375,
-      "loss": 0.1656,
-      "rewards/accuracies": 0.84375,
-      "rewards/chosen": -2.6965749263763428,
-      "rewards/margins": 8.003564834594727,
-      "rewards/rejected": -10.700139045715332,
+      "logits/chosen": -1.578249216079712,
+      "logits/rejected": -1.4307564496994019,
+      "logps/chosen": -618.6419067382812,
+      "logps/rejected": -917.3260498046875,
+      "loss": 0.1288,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -3.068641185760498,
+      "rewards/margins": 4.769894599914551,
+      "rewards/rejected": -7.838536262512207,
       "step": 450
     },
     {
       "epoch": 0.48,
-      "grad_norm": 35.01981360851454,
+      "grad_norm": 15.368374650378705,
       "learning_rate": 3.0931662070620794e-07,
-      "logits/chosen": -0.787259578704834,
-      "logits/rejected": -0.5048869848251343,
-      "logps/chosen": -389.9725036621094,
-      "logps/rejected": -1115.496337890625,
-      "loss": 0.1644,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -2.5107271671295166,
-      "rewards/margins": 7.763253688812256,
-      "rewards/rejected": -10.273981094360352,
+      "logits/chosen": -1.8674280643463135,
+      "logits/rejected": -1.8419768810272217,
+      "logps/chosen": -574.4186401367188,
+      "logps/rejected": -899.6975708007812,
+      "loss": 0.1317,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -2.381072998046875,
+      "rewards/margins": 5.151331424713135,
+      "rewards/rejected": -7.53240442276001,
       "step": 460
     },
     {
       "epoch": 0.49,
-      "grad_norm": 32.88620144282501,
+      "grad_norm": 32.45699366987169,
       "learning_rate": 3.003968536966078e-07,
-      "logits/chosen": -0.8425847887992859,
-      "logits/rejected": -0.507870614528656,
-      "logps/chosen": -377.3370666503906,
-      "logps/rejected": -1063.682861328125,
-      "loss": 0.1581,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -2.452817440032959,
-      "rewards/margins": 7.454998970031738,
-      "rewards/rejected": -9.907815933227539,
+      "logits/chosen": -1.779637098312378,
+      "logits/rejected": -1.7716691493988037,
+      "logps/chosen": -548.851806640625,
+      "logps/rejected": -895.9444580078125,
+      "loss": 0.1467,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -2.4585914611816406,
+      "rewards/margins": 5.207429885864258,
+      "rewards/rejected": -7.666021823883057,
       "step": 470
     },
     {
       "epoch": 0.5,
-      "grad_norm": 55.426706045689585,
+      "grad_norm": 20.390767179507876,
       "learning_rate": 2.9140968536213693e-07,
-      "logits/chosen": -0.915787398815155,
-      "logits/rejected": -0.6263198852539062,
-      "logps/chosen": -401.2126770019531,
-      "logps/rejected": -1090.992431640625,
-      "loss": 0.176,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -2.7542808055877686,
-      "rewards/margins": 7.400801181793213,
-      "rewards/rejected": -10.155081748962402,
+      "logits/chosen": -1.919952630996704,
+      "logits/rejected": -1.9492056369781494,
+      "logps/chosen": -536.4271240234375,
+      "logps/rejected": -884.7716064453125,
+      "loss": 0.1285,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -2.535160541534424,
+      "rewards/margins": 5.057857990264893,
+      "rewards/rejected": -7.593018531799316,
       "step": 480
     },
     {
       "epoch": 0.51,
-      "grad_norm": 22.799610213421833,
+      "grad_norm": 23.067489784018093,
       "learning_rate": 2.823671352438608e-07,
-      "logits/chosen": -0.8439705967903137,
-      "logits/rejected": -0.6034272313117981,
-      "logps/chosen": -354.22772216796875,
-      "logps/rejected": -956.9412841796875,
-      "loss": 0.1624,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -2.377206802368164,
-      "rewards/margins": 6.480790615081787,
-      "rewards/rejected": -8.85799789428711,
+      "logits/chosen": -1.9916563034057617,
+      "logits/rejected": -2.021616220474243,
+      "logps/chosen": -518.0189208984375,
+      "logps/rejected": -824.4202880859375,
+      "loss": 0.1367,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -2.1029324531555176,
+      "rewards/margins": 4.754405498504639,
+      "rewards/rejected": -6.857337951660156,
       "step": 490
     },
     {
       "epoch": 0.52,
-      "grad_norm": 39.88720282221281,
+      "grad_norm": 25.320584902838384,
       "learning_rate": 2.73281296951072e-07,
-      "logits/chosen": -1.0180326700210571,
-      "logits/rejected": -0.7310226559638977,
-      "logps/chosen": -385.81866455078125,
-      "logps/rejected": -1014.95458984375,
-      "loss": 0.1693,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -2.5694196224212646,
-      "rewards/margins": 6.7894439697265625,
-      "rewards/rejected": -9.358863830566406,
+      "logits/chosen": -1.8573198318481445,
+      "logits/rejected": -1.8858684301376343,
+      "logps/chosen": -541.9103393554688,
+      "logps/rejected": -806.6531372070312,
+      "loss": 0.1364,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.396160840988159,
+      "rewards/margins": 4.32682466506958,
+      "rewards/rejected": -6.722985744476318,
       "step": 500
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -0.987293541431427,
-      "eval_logits/rejected": -0.8597699403762817,
-      "eval_logps/chosen": -945.861083984375,
-      "eval_logps/rejected": -1084.707763671875,
-      "eval_loss": 1.6562931537628174,
-      "eval_rewards/accuracies": 0.625,
-      "eval_rewards/chosen": -6.340845108032227,
-      "eval_rewards/margins": 1.3309805393218994,
-      "eval_rewards/rejected": -7.671825885772705,
-      "eval_runtime": 97.6234,
-      "eval_samples_per_second": 20.487,
-      "eval_steps_per_second": 0.328,
+      "eval_logits/chosen": -1.6466461420059204,
+      "eval_logits/rejected": -1.6090116500854492,
+      "eval_logps/chosen": -470.0933532714844,
+      "eval_logps/rejected": -430.8614196777344,
+      "eval_loss": 1.1773370504379272,
+      "eval_rewards/accuracies": 0.37109375,
+      "eval_rewards/chosen": -1.5831681489944458,
+      "eval_rewards/margins": -0.44980528950691223,
+      "eval_rewards/rejected": -1.133362889289856,
+      "eval_runtime": 97.142,
+      "eval_samples_per_second": 20.588,
+      "eval_steps_per_second": 0.329,
       "step": 500
     },
     {
       "epoch": 0.53,
-      "grad_norm": 36.2723513365496,
+      "grad_norm": 22.389152396843905,
       "learning_rate": 2.641643219871597e-07,
-      "logits/chosen": -1.0296061038970947,
-      "logits/rejected": -0.6009940505027771,
-      "logps/chosen": -367.12713623046875,
-      "logps/rejected": -957.3023681640625,
-      "loss": 0.1488,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -2.196636199951172,
-      "rewards/margins": 6.570919990539551,
-      "rewards/rejected": -8.767557144165039,
+      "logits/chosen": -1.9270741939544678,
+      "logits/rejected": -1.8253233432769775,
+      "logps/chosen": -544.8509521484375,
+      "logps/rejected": -834.5119018554688,
+      "loss": 0.1273,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.5420873165130615,
+      "rewards/margins": 4.583073616027832,
+      "rewards/rejected": -7.125161170959473,
       "step": 510
     },
     {
       "epoch": 0.54,
-      "grad_norm": 41.00764971947524,
+      "grad_norm": 29.91148010858245,
       "learning_rate": 2.550284034980507e-07,
-      "logits/chosen": -0.9317744970321655,
-      "logits/rejected": -0.4665905833244324,
-      "logps/chosen": -389.6575012207031,
-      "logps/rejected": -1017.7044067382812,
-      "loss": 0.1731,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -2.4715576171875,
-      "rewards/margins": 6.838615417480469,
-      "rewards/rejected": -9.310173034667969,
+      "logits/chosen": -1.8461729288101196,
+      "logits/rejected": -1.9183883666992188,
+      "logps/chosen": -536.2467041015625,
+      "logps/rejected": -846.56689453125,
+      "loss": 0.1301,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.3210761547088623,
+      "rewards/margins": 4.834765911102295,
+      "rewards/rejected": -7.1558427810668945,
       "step": 520
     },
     {
       "epoch": 0.55,
-      "grad_norm": 24.760088451093488,
+      "grad_norm": 24.244708458502263,
       "learning_rate": 2.4588575996495794e-07,
-      "logits/chosen": -1.0027564764022827,
-      "logits/rejected": -0.5636709928512573,
-      "logps/chosen": -361.3106384277344,
-      "logps/rejected": -1032.2296142578125,
-      "loss": 0.1409,
-      "rewards/accuracies": 0.925000011920929,
-      "rewards/chosen": -2.317450761795044,
-      "rewards/margins": 7.240323066711426,
-      "rewards/rejected": -9.557774543762207,
+      "logits/chosen": -1.80156672000885,
+      "logits/rejected": -1.815596580505371,
+      "logps/chosen": -494.86798095703125,
+      "logps/rejected": -798.2820434570312,
+      "loss": 0.1237,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -2.3462131023406982,
+      "rewards/margins": 4.387810230255127,
+      "rewards/rejected": -6.7340240478515625,
       "step": 530
     },
     {
       "epoch": 0.57,
-      "grad_norm": 29.45673267405302,
+      "grad_norm": 19.910786436303987,
       "learning_rate": 2.367486188632446e-07,
-      "logits/chosen": -0.6549164652824402,
-      "logits/rejected": -0.2623385787010193,
-      "logps/chosen": -400.9425048828125,
-      "logps/rejected": -1065.4869384765625,
-      "loss": 0.1447,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -2.686591863632202,
-      "rewards/margins": 7.183202266693115,
-      "rewards/rejected": -9.869793891906738,
+      "logits/chosen": -1.9172916412353516,
+      "logits/rejected": -1.9127076864242554,
+      "logps/chosen": -525.8649291992188,
+      "logps/rejected": -845.7303466796875,
+      "loss": 0.1138,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -2.059016704559326,
+      "rewards/margins": 4.951017379760742,
+      "rewards/rejected": -7.010035037994385,
       "step": 540
     },
     {
       "epoch": 0.58,
-      "grad_norm": 28.112202270911702,
+      "grad_norm": 26.12771211985874,
       "learning_rate": 2.276292003092593e-07,
-      "logits/chosen": -0.7201762199401855,
-      "logits/rejected": -0.20686273276805878,
-      "logps/chosen": -414.7852478027344,
-      "logps/rejected": -1124.4974365234375,
-      "loss": 0.154,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -2.9022722244262695,
-      "rewards/margins": 7.623268127441406,
-      "rewards/rejected": -10.525540351867676,
+      "logits/chosen": -1.8340644836425781,
+      "logits/rejected": -1.8532531261444092,
+      "logps/chosen": -547.7066650390625,
+      "logps/rejected": -828.9231567382812,
+      "loss": 0.1309,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -2.3669962882995605,
+      "rewards/margins": 4.561534404754639,
+      "rewards/rejected": -6.928530216217041,
       "step": 550
     },
     {
       "epoch": 0.59,
-      "grad_norm": 39.46687358645063,
+      "grad_norm": 23.903525728545898,
       "learning_rate": 2.185397007170141e-07,
-      "logits/chosen": -0.40733662247657776,
-      "logits/rejected": 0.02392803505063057,
-      "logps/chosen": -366.5719299316406,
-      "logps/rejected": -931.62841796875,
-      "loss": 0.1679,
-      "rewards/accuracies": 0.856249988079071,
-      "rewards/chosen": -2.599259853363037,
-      "rewards/margins": 6.089108467102051,
-      "rewards/rejected": -8.68836784362793,
+      "logits/chosen": -1.7198997735977173,
+      "logits/rejected": -1.5677220821380615,
+      "logps/chosen": -525.4856567382812,
+      "logps/rejected": -855.2352294921875,
+      "loss": 0.1157,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -2.4956746101379395,
+      "rewards/margins": 4.861161708831787,
+      "rewards/rejected": -7.356836795806885,
       "step": 560
     },
     {
       "epoch": 0.6,
-      "grad_norm": 45.19451118355015,
+      "grad_norm": 24.886700178362616,
       "learning_rate": 2.094922764865619e-07,
-      "logits/chosen": -0.6102726459503174,
-      "logits/rejected": -0.25020402669906616,
-      "logps/chosen": -418.000732421875,
-      "logps/rejected": -1118.455322265625,
-      "loss": 0.1527,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -2.813910722732544,
-      "rewards/margins": 7.487911224365234,
-      "rewards/rejected": -10.3018217086792,
+      "logits/chosen": -1.7758210897445679,
+      "logits/rejected": -1.836538314819336,
+      "logps/chosen": -506.49053955078125,
+      "logps/rejected": -846.2745971679688,
+      "loss": 0.1366,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -2.2135753631591797,
+      "rewards/margins": 4.973353385925293,
+      "rewards/rejected": -7.186928749084473,
       "step": 570
     },
     {
       "epoch": 0.61,
-      "grad_norm": 27.296281473923543,
+      "grad_norm": 21.212825758656418,
       "learning_rate": 2.0049902774588797e-07,
-      "logits/chosen": -0.5131772756576538,
-      "logits/rejected": -0.07118790596723557,
-      "logps/chosen": -358.7304992675781,
-      "logps/rejected": -981.1790161132812,
-      "loss": 0.1379,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -2.248889923095703,
-      "rewards/margins": 6.8233489990234375,
-      "rewards/rejected": -9.072239875793457,
+      "logits/chosen": -1.7731355428695679,
+      "logits/rejected": -1.777611494064331,
+      "logps/chosen": -581.9010009765625,
+      "logps/rejected": -871.5733642578125,
+      "loss": 0.123,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -2.585883378982544,
+      "rewards/margins": 4.742143154144287,
+      "rewards/rejected": -7.32802677154541,
       "step": 580
     },
     {
       "epoch": 0.62,
-      "grad_norm": 28.400791802256467,
+      "grad_norm": 24.501331083650943,
       "learning_rate": 1.9157198216806238e-07,
-      "logits/chosen": -0.5218924283981323,
-      "logits/rejected": -0.11522813141345978,
-      "logps/chosen": -385.2206115722656,
-      "logps/rejected": -1059.663330078125,
-      "loss": 0.1526,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -2.5596184730529785,
-      "rewards/margins": 7.243247032165527,
-      "rewards/rejected": -9.802865028381348,
+      "logits/chosen": -1.7602859735488892,
+      "logits/rejected": -1.84162175655365,
+      "logps/chosen": -452.37420654296875,
+      "logps/rejected": -707.836181640625,
+      "loss": 0.1305,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -2.025653600692749,
+      "rewards/margins": 3.844447374343872,
+      "rewards/rejected": -5.870100975036621,
       "step": 590
     },
     {
       "epoch": 0.63,
-      "grad_norm": 36.01204411609108,
+      "grad_norm": 31.082842710792463,
       "learning_rate": 1.8272307888529274e-07,
-      "logits/chosen": -0.6575510501861572,
-      "logits/rejected": -0.13078925013542175,
-      "logps/chosen": -393.07891845703125,
-      "logps/rejected": -1083.641357421875,
-      "loss": 0.1609,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -2.5931639671325684,
-      "rewards/margins": 7.458055019378662,
-      "rewards/rejected": -10.05121898651123,
+      "logits/chosen": -1.5975382328033447,
+      "logits/rejected": -1.4109210968017578,
+      "logps/chosen": -539.9990234375,
+      "logps/rejected": -849.1795654296875,
+      "loss": 0.1223,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -2.722317934036255,
+      "rewards/margins": 4.622250556945801,
+      "rewards/rejected": -7.344568729400635,
       "step": 600
     },
     {
       "epoch": 0.63,
-      "eval_logits/chosen": -0.616359293460846,
-      "eval_logits/rejected": -0.45154044032096863,
-      "eval_logps/chosen": -959.7227172851562,
-      "eval_logps/rejected": -1097.447998046875,
-      "eval_loss": 1.6818441152572632,
-      "eval_rewards/accuracies": 0.62109375,
-      "eval_rewards/chosen": -6.479461669921875,
-      "eval_rewards/margins": 1.3197669982910156,
-      "eval_rewards/rejected": -7.799228668212891,
-      "eval_runtime": 97.4083,
-      "eval_samples_per_second": 20.532,
+      "eval_logits/chosen": -1.1879667043685913,
+      "eval_logits/rejected": -1.1541048288345337,
+      "eval_logps/chosen": -541.48828125,
+      "eval_logps/rejected": -500.4929504394531,
+      "eval_loss": 1.3205512762069702,
+      "eval_rewards/accuracies": 0.4140625,
+      "eval_rewards/chosen": -2.2971181869506836,
+      "eval_rewards/margins": -0.46744003891944885,
+      "eval_rewards/rejected": -1.829677939414978,
+      "eval_runtime": 97.2287,
+      "eval_samples_per_second": 20.57,
       "eval_steps_per_second": 0.329,
       "step": 600
     },
     {
       "epoch": 0.64,
-      "grad_norm": 25.93993016029263,
+      "grad_norm": 25.74237035768959,
       "learning_rate": 1.7396415252139288e-07,
-      "logits/chosen": -0.9318272471427917,
-      "logits/rejected": -0.5557786226272583,
-      "logps/chosen": -372.4070129394531,
-      "logps/rejected": -1039.4412841796875,
-      "loss": 0.1523,
+      "logits/chosen": -1.5792185068130493,
+      "logits/rejected": -1.6271593570709229,
+      "logps/chosen": -576.2874755859375,
+      "logps/rejected": -917.5999755859375,
+      "loss": 0.1228,
       "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -2.3948473930358887,
-      "rewards/margins": 7.164391994476318,
-      "rewards/rejected": -9.559239387512207,
+      "rewards/chosen": -2.802475929260254,
+      "rewards/margins": 5.048992156982422,
+      "rewards/rejected": -7.851468563079834,
       "step": 610
     },
     {
       "epoch": 0.65,
-      "grad_norm": 43.56570182026958,
+      "grad_norm": 29.519623894853613,
       "learning_rate": 1.6530691736402316e-07,
-      "logits/chosen": -0.6166467070579529,
-      "logits/rejected": -0.15916576981544495,
-      "logps/chosen": -382.71112060546875,
-      "logps/rejected": -1044.164794921875,
-      "loss": 0.1514,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -2.561720848083496,
-      "rewards/margins": 7.1611833572387695,
-      "rewards/rejected": -9.722905158996582,
+      "logits/chosen": -1.6389156579971313,
+      "logits/rejected": -1.6191425323486328,
+      "logps/chosen": -530.8707275390625,
+      "logps/rejected": -890.0936279296875,
+      "loss": 0.1254,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.3094658851623535,
+      "rewards/margins": 5.2504377365112305,
+      "rewards/rejected": -7.559903144836426,
       "step": 620
     },
     {
       "epoch": 0.66,
-      "grad_norm": 28.928080380488158,
+      "grad_norm": 50.42769593704326,
       "learning_rate": 1.5676295169786864e-07,
-      "logits/chosen": -0.5803557634353638,
-      "logits/rejected": -0.2489413022994995,
-      "logps/chosen": -383.3107604980469,
-      "logps/rejected": -997.8870239257812,
-      "loss": 0.164,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -2.578981876373291,
-      "rewards/margins": 6.61221170425415,
-      "rewards/rejected": -9.191193580627441,
+      "logits/chosen": -1.7326772212982178,
+      "logits/rejected": -1.5138975381851196,
+      "logps/chosen": -561.8060302734375,
+      "logps/rejected": -941.2083740234375,
+      "loss": 0.1236,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -2.4162166118621826,
+      "rewards/margins": 5.7043328285217285,
+      "rewards/rejected": -8.120549201965332,
       "step": 630
     },
     {
       "epoch": 0.67,
-      "grad_norm": 20.471962287027644,
+      "grad_norm": 27.38270007193824,
       "learning_rate": 1.483436823197092e-07,
-      "logits/chosen": -0.6895365118980408,
-      "logits/rejected": -0.1990533173084259,
-      "logps/chosen": -362.12237548828125,
-      "logps/rejected": -1061.60400390625,
-      "loss": 0.1475,
-      "rewards/accuracies": 0.956250011920929,
-      "rewards/chosen": -2.2892990112304688,
-      "rewards/margins": 7.556910514831543,
-      "rewards/rejected": -9.846209526062012,
+      "logits/chosen": -1.5837154388427734,
+      "logits/rejected": -1.4245226383209229,
+      "logps/chosen": -562.3756103515625,
+      "logps/rejected": -949.3406982421875,
+      "loss": 0.1163,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -3.040762186050415,
+      "rewards/margins": 5.183028697967529,
+      "rewards/rejected": -8.223790168762207,
       "step": 640
     },
     {
       "epoch": 0.68,
-      "grad_norm": 26.665412290672624,
+      "grad_norm": 35.27601401211646,
       "learning_rate": 1.4006036925609243e-07,
-      "logits/chosen": -0.6053385734558105,
-      "logits/rejected": -0.19510070979595184,
-      "logps/chosen": -381.46514892578125,
-      "logps/rejected": -1129.877685546875,
-      "loss": 0.1432,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -2.4206106662750244,
-      "rewards/margins": 7.994618892669678,
-      "rewards/rejected": -10.415228843688965,
+      "logits/chosen": -1.7516834735870361,
+      "logits/rejected": -1.7616602182388306,
+      "logps/chosen": -558.3771362304688,
+      "logps/rejected": -879.00341796875,
+      "loss": 0.1139,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -2.68957257270813,
+      "rewards/margins": 4.818731784820557,
+      "rewards/rejected": -7.508304595947266,
       "step": 650
     },
     {
       "epoch": 0.69,
-      "grad_norm": 33.292289854748574,
+      "grad_norm": 22.77394327537456,
       "learning_rate": 1.319240907040458e-07,
-      "logits/chosen": -0.5973755717277527,
-      "logits/rejected": -0.16952477395534515,
-      "logps/chosen": -410.3275451660156,
-      "logps/rejected": -1123.840087890625,
-      "loss": 0.1388,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -2.748767852783203,
-      "rewards/margins": 7.701995849609375,
-      "rewards/rejected": -10.450764656066895,
+      "logits/chosen": -1.7446712255477905,
+      "logits/rejected": -1.6567370891571045,
+      "logps/chosen": -559.461669921875,
+      "logps/rejected": -941.3621826171875,
+      "loss": 0.1117,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -2.522426128387451,
+      "rewards/margins": 5.564127445220947,
+      "rewards/rejected": -8.086553573608398,
       "step": 660
     },
     {
       "epoch": 0.7,
-      "grad_norm": 30.896865825928337,
+      "grad_norm": 30.36694985683218,
       "learning_rate": 1.239457282149695e-07,
-      "logits/chosen": -0.5566490292549133,
-      "logits/rejected": -0.23316040635108948,
-      "logps/chosen": -428.1212463378906,
-      "logps/rejected": -1241.2476806640625,
-      "loss": 0.1475,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -3.0948407649993896,
-      "rewards/margins": 8.58587646484375,
-      "rewards/rejected": -11.680717468261719,
+      "logits/chosen": -1.4815702438354492,
+      "logits/rejected": -1.3673474788665771,
+      "logps/chosen": -548.4630737304688,
+      "logps/rejected": -930.783203125,
+      "loss": 0.1092,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -2.9377596378326416,
+      "rewards/margins": 5.2314453125,
+      "rewards/rejected": -8.169205665588379,
       "step": 670
     },
     {
       "epoch": 0.71,
-      "grad_norm": 35.84504872343951,
+      "grad_norm": 28.757425831930576,
       "learning_rate": 1.1613595214152711e-07,
-      "logits/chosen": -0.5821831822395325,
-      "logits/rejected": -0.07458965480327606,
-      "logps/chosen": -399.4475402832031,
-      "logps/rejected": -1166.2537841796875,
-      "loss": 0.1518,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -2.7263970375061035,
-      "rewards/margins": 8.200857162475586,
-      "rewards/rejected": -10.927252769470215,
+      "logits/chosen": -1.5986506938934326,
+      "logits/rejected": -1.5326309204101562,
+      "logps/chosen": -513.4288940429688,
+      "logps/rejected": -861.1954345703125,
+      "loss": 0.1105,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -2.518094062805176,
+      "rewards/margins": 4.961692810058594,
+      "rewards/rejected": -7.4797868728637695,
       "step": 680
     },
     {
       "epoch": 0.72,
-      "grad_norm": 33.84684500335279,
+      "grad_norm": 23.459702688906162,
       "learning_rate": 1.0850520736699362e-07,
-      "logits/chosen": -0.5655028223991394,
-      "logits/rejected": -0.15712787210941315,
-      "logps/chosen": -377.57208251953125,
-      "logps/rejected": -1100.81884765625,
-      "loss": 0.1454,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -2.5599417686462402,
-      "rewards/margins": 7.693644046783447,
-      "rewards/rejected": -10.253584861755371,
+      "logits/chosen": -1.6289697885513306,
+      "logits/rejected": -1.527553677558899,
+      "logps/chosen": -537.8997192382812,
+      "logps/rejected": -878.1143798828125,
+      "loss": 0.113,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -2.5290098190307617,
+      "rewards/margins": 4.979551315307617,
+      "rewards/rejected": -7.508561611175537,
       "step": 690
     },
     {
       "epoch": 0.73,
-      "grad_norm": 47.388927709457874,
+      "grad_norm": 25.531163769236304,
       "learning_rate": 1.0106369933615042e-07,
-      "logits/chosen": -0.2697560787200928,
-      "logits/rejected": 0.21001645922660828,
-      "logps/chosen": -389.8284606933594,
-      "logps/rejected": -1081.4793701171875,
-      "loss": 0.1559,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -2.6655356884002686,
-      "rewards/margins": 7.382843971252441,
-      "rewards/rejected": -10.048379898071289,
+      "logits/chosen": -1.5353871583938599,
+      "logits/rejected": -1.3200981616973877,
+      "logps/chosen": -662.8204956054688,
+      "logps/rejected": -1062.7938232421875,
+      "loss": 0.0971,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -3.5107734203338623,
+      "rewards/margins": 5.729162693023682,
+      "rewards/rejected": -9.239934921264648,
       "step": 700
     },
     {
       "epoch": 0.73,
-      "eval_logits/chosen": -0.5852116346359253,
-      "eval_logits/rejected": -0.4166191518306732,
-      "eval_logps/chosen": -1046.6217041015625,
-      "eval_logps/rejected": -1197.0731201171875,
-      "eval_loss": 1.9277653694152832,
-      "eval_rewards/accuracies": 0.61328125,
-      "eval_rewards/chosen": -7.348450660705566,
-      "eval_rewards/margins": 1.447028398513794,
-      "eval_rewards/rejected": -8.795478820800781,
-      "eval_runtime": 97.4242,
-      "eval_samples_per_second": 20.529,
-      "eval_steps_per_second": 0.328,
+      "eval_logits/chosen": -0.9711934924125671,
+      "eval_logits/rejected": -0.9391850829124451,
+      "eval_logps/chosen": -577.3128051757812,
+      "eval_logps/rejected": -533.4666748046875,
+      "eval_loss": 1.463817834854126,
+      "eval_rewards/accuracies": 0.390625,
+      "eval_rewards/chosen": -2.6553633213043213,
+      "eval_rewards/margins": -0.4959481358528137,
+      "eval_rewards/rejected": -2.1594154834747314,
+      "eval_runtime": 97.1589,
+      "eval_samples_per_second": 20.585,
+      "eval_steps_per_second": 0.329,
       "step": 700
     },
     {
       "epoch": 0.74,
-      "grad_norm": 21.54892454805239,
+      "grad_norm": 43.118837668077326,
       "learning_rate": 9.382138040640714e-08,
-      "logits/chosen": -0.6082885265350342,
-      "logits/rejected": 0.03385084122419357,
-      "logps/chosen": -389.99542236328125,
-      "logps/rejected": -1186.9339599609375,
-      "loss": 0.1358,
-      "rewards/accuracies": 0.9312499761581421,
-      "rewards/chosen": -2.5867698192596436,
-      "rewards/margins": 8.532285690307617,
-      "rewards/rejected": -11.11905574798584,
+      "logits/chosen": -1.6447147130966187,
+      "logits/rejected": -1.4446865320205688,
+      "logps/chosen": -644.3197021484375,
+      "logps/rejected": -1070.079345703125,
+      "loss": 0.11,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -3.044210910797119,
+      "rewards/margins": 6.259350776672363,
+      "rewards/rejected": -9.303561210632324,
       "step": 710
     },
     {
       "epoch": 0.75,
-      "grad_norm": 45.770131366856944,
+      "grad_norm": 32.87349411722058,
       "learning_rate": 8.678793653740632e-08,
-      "logits/chosen": -0.45069313049316406,
-      "logits/rejected": -0.01569805108010769,
-      "logps/chosen": -418.41015625,
-      "logps/rejected": -1100.5294189453125,
-      "loss": 0.1518,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -2.9045753479003906,
-      "rewards/margins": 7.348256587982178,
-      "rewards/rejected": -10.252832412719727,
+      "logits/chosen": -1.7681348323822021,
+      "logits/rejected": -1.6581518650054932,
+      "logps/chosen": -622.3300170898438,
+      "logps/rejected": -1050.2943115234375,
+      "loss": 0.1069,
+      "rewards/accuracies": 0.987500011920929,
+      "rewards/chosen": -2.925344944000244,
+      "rewards/margins": 6.121405601501465,
+      "rewards/rejected": -9.04675006866455,
       "step": 720
     },
     {
       "epoch": 0.76,
-      "grad_norm": 26.24794912329421,
+      "grad_norm": 36.992204810711705,
       "learning_rate": 7.997277433690983e-08,
-      "logits/chosen": -0.6921511292457581,
-      "logits/rejected": -0.1756693720817566,
-      "logps/chosen": -402.31463623046875,
-      "logps/rejected": -1098.74951171875,
-      "loss": 0.1545,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -2.674762725830078,
-      "rewards/margins": 7.446639060974121,
-      "rewards/rejected": -10.1214017868042,
+      "logits/chosen": -1.623159646987915,
+      "logits/rejected": -1.4811782836914062,
+      "logps/chosen": -612.2554931640625,
+      "logps/rejected": -995.8507690429688,
+      "loss": 0.118,
+      "rewards/accuracies": 0.925000011920929,
+      "rewards/chosen": -3.1586315631866455,
+      "rewards/margins": 5.526102066040039,
+      "rewards/rejected": -8.684733390808105,
       "step": 730
     },
     {
       "epoch": 0.77,
-      "grad_norm": 42.21104117559166,
+      "grad_norm": 31.945634867150687,
       "learning_rate": 7.338500848029602e-08,
-      "logits/chosen": -0.5230437517166138,
-      "logits/rejected": -0.08038869500160217,
-      "logps/chosen": -374.2379150390625,
-      "logps/rejected": -1173.2691650390625,
-      "loss": 0.1524,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -2.3602004051208496,
-      "rewards/margins": 8.614425659179688,
-      "rewards/rejected": -10.974625587463379,
+      "logits/chosen": -1.51155686378479,
+      "logits/rejected": -1.349172592163086,
+      "logps/chosen": -575.31103515625,
+      "logps/rejected": -1048.4927978515625,
+      "loss": 0.1135,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -3.0343711376190186,
+      "rewards/margins": 6.1768388748168945,
+      "rewards/rejected": -9.211209297180176,
       "step": 740
     },
     {
       "epoch": 0.78,
-      "grad_norm": 45.35711154178482,
+      "grad_norm": 26.030459327741653,
       "learning_rate": 6.70334495204884e-08,
-      "logits/chosen": -0.4201095998287201,
-      "logits/rejected": 0.112357497215271,
-      "logps/chosen": -399.91766357421875,
-      "logps/rejected": -1149.41357421875,
-      "loss": 0.1425,
-      "rewards/accuracies": 0.918749988079071,
-      "rewards/chosen": -2.750382900238037,
-      "rewards/margins": 8.02344799041748,
-      "rewards/rejected": -10.773832321166992,
+      "logits/chosen": -1.6075376272201538,
+      "logits/rejected": -1.2717111110687256,
+      "logps/chosen": -635.0384521484375,
+      "logps/rejected": -1016.2355346679688,
+      "loss": 0.0957,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -3.0112128257751465,
+      "rewards/margins": 5.901360511779785,
+      "rewards/rejected": -8.912572860717773,
       "step": 750
     },
     {
       "epoch": 0.8,
-      "grad_norm": 26.89122562285564,
+      "grad_norm": 34.627238331382905,
       "learning_rate": 6.092659210462231e-08,
-      "logits/chosen": -0.6679720878601074,
-      "logits/rejected": -0.1197366863489151,
-      "logps/chosen": -417.52996826171875,
-      "logps/rejected": -1188.0615234375,
-      "loss": 0.1345,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -2.673654079437256,
-      "rewards/margins": 8.336530685424805,
-      "rewards/rejected": -11.010185241699219,
+      "logits/chosen": -1.448529839515686,
+      "logits/rejected": -1.4409586191177368,
+      "logps/chosen": -625.353271484375,
+      "logps/rejected": -1024.951416015625,
+      "loss": 0.1095,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -3.2976653575897217,
+      "rewards/margins": 5.690510272979736,
+      "rewards/rejected": -8.988176345825195,
       "step": 760
     },
     {
       "epoch": 0.81,
-      "grad_norm": 34.7731803926411,
+      "grad_norm": 29.968791403480516,
       "learning_rate": 5.507260361320737e-08,
-      "logits/chosen": -0.6739757657051086,
-      "logits/rejected": -0.05404149740934372,
-      "logps/chosen": -402.13238525390625,
-      "logps/rejected": -1078.5826416015625,
-      "loss": 0.1516,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -2.7083559036254883,
-      "rewards/margins": 7.324588775634766,
-      "rewards/rejected": -10.032943725585938,
+      "logits/chosen": -1.5798556804656982,
+      "logits/rejected": -1.5097671747207642,
+      "logps/chosen": -600.1854248046875,
+      "logps/rejected": -997.0427856445312,
+      "loss": 0.1033,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -3.1436641216278076,
+      "rewards/margins": 5.432265281677246,
+      "rewards/rejected": -8.575929641723633,
       "step": 770
     },
     {
       "epoch": 0.82,
-      "grad_norm": 25.35847618120489,
+      "grad_norm": 23.15982358200479,
       "learning_rate": 4.947931323697982e-08,
-      "logits/chosen": -0.5962761044502258,
-      "logits/rejected": -0.05018671602010727,
-      "logps/chosen": -392.66363525390625,
-      "logps/rejected": -1066.8021240234375,
-      "loss": 0.1551,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -2.6160168647766113,
-      "rewards/margins": 7.251564025878906,
-      "rewards/rejected": -9.86758041381836,
+      "logits/chosen": -1.4408910274505615,
+      "logits/rejected": -1.3072535991668701,
+      "logps/chosen": -556.1472778320312,
+      "logps/rejected": -968.1648559570312,
+      "loss": 0.0969,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -3.2009377479553223,
+      "rewards/margins": 5.246421813964844,
+      "rewards/rejected": -8.447359085083008,
       "step": 780
     },
     {
       "epoch": 0.83,
-      "grad_norm": 27.140797859113846,
+      "grad_norm": 23.663576108460163,
       "learning_rate": 4.415420150605398e-08,
-      "logits/chosen": -0.6300855278968811,
-      "logits/rejected": -0.1746017038822174,
-      "logps/chosen": -400.11431884765625,
-      "logps/rejected": -1069.177001953125,
-      "loss": 0.1616,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -2.7479443550109863,
-      "rewards/margins": 7.171761989593506,
-      "rewards/rejected": -9.919707298278809,
+      "logits/chosen": -1.5656832456588745,
+      "logits/rejected": -1.3935534954071045,
+      "logps/chosen": -601.1109008789062,
+      "logps/rejected": -999.2574462890625,
+      "loss": 0.1023,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -3.033841609954834,
+      "rewards/margins": 5.674118995666504,
+      "rewards/rejected": -8.70796012878418,
       "step": 790
     },
     {
       "epoch": 0.84,
-      "grad_norm": 20.145953031877735,
+      "grad_norm": 23.462966354133016,
       "learning_rate": 3.9104390285376374e-08,
-      "logits/chosen": -0.44818106293678284,
-      "logits/rejected": -0.035945743322372437,
-      "logps/chosen": -396.0670471191406,
-      "logps/rejected": -1147.34619140625,
-      "loss": 0.1433,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -2.6665611267089844,
-      "rewards/margins": 8.064103126525879,
-      "rewards/rejected": -10.730664253234863,
+      "logits/chosen": -1.466312289237976,
+      "logits/rejected": -1.2542184591293335,
+      "logps/chosen": -619.60009765625,
+      "logps/rejected": -1041.6527099609375,
+      "loss": 0.1035,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -3.135887861251831,
+      "rewards/margins": 5.890833854675293,
+      "rewards/rejected": -9.026721954345703,
       "step": 800
     },
     {
       "epoch": 0.84,
-      "eval_logits/chosen": -0.6744760870933533,
-      "eval_logits/rejected": -0.5140590667724609,
-      "eval_logps/chosen": -1026.7318115234375,
-      "eval_logps/rejected": -1180.040283203125,
-      "eval_loss": 1.9049861431121826,
-      "eval_rewards/accuracies": 0.6171875,
-      "eval_rewards/chosen": -7.149553298950195,
-      "eval_rewards/margins": 1.4755982160568237,
-      "eval_rewards/rejected": -8.625151634216309,
-      "eval_runtime": 97.364,
-      "eval_samples_per_second": 20.541,
+      "eval_logits/chosen": -0.9231772422790527,
+      "eval_logits/rejected": -0.8901588916778564,
+      "eval_logps/chosen": -569.3816528320312,
+      "eval_logps/rejected": -532.9067993164062,
+      "eval_loss": 1.4475353956222534,
+      "eval_rewards/accuracies": 0.39453125,
+      "eval_rewards/chosen": -2.5760514736175537,
+      "eval_rewards/margins": -0.42223480343818665,
+      "eval_rewards/rejected": -2.1538166999816895,
+      "eval_runtime": 97.1382,
+      "eval_samples_per_second": 20.589,
       "eval_steps_per_second": 0.329,
       "step": 800
     },
     {
       "epoch": 0.85,
-      "grad_norm": 21.539181167828147,
+      "grad_norm": 30.508507386848592,
       "learning_rate": 3.433663324986208e-08,
-      "logits/chosen": -0.5423880815505981,
-      "logits/rejected": -0.08916589617729187,
-      "logps/chosen": -378.41876220703125,
-      "logps/rejected": -1105.344970703125,
-      "loss": 0.156,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -2.467062473297119,
-      "rewards/margins": 7.7889909744262695,
-      "rewards/rejected": -10.256052017211914,
+      "logits/chosen": -1.4120627641677856,
+      "logits/rejected": -1.3580305576324463,
+      "logps/chosen": -642.4114990234375,
+      "logps/rejected": -1046.66357421875,
+      "loss": 0.1086,
+      "rewards/accuracies": 0.90625,
+      "rewards/chosen": -3.5207436084747314,
+      "rewards/margins": 5.544968605041504,
+      "rewards/rejected": -9.065712928771973,
       "step": 810
     },
     {
       "epoch": 0.86,
-      "grad_norm": 25.60967931137277,
+      "grad_norm": 15.450308092025814,
       "learning_rate": 2.9857306851953897e-08,
-      "logits/chosen": -0.5412378907203674,
-      "logits/rejected": -0.05382275581359863,
-      "logps/chosen": -373.0191345214844,
-      "logps/rejected": -1065.0948486328125,
-      "loss": 0.1524,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -2.4381911754608154,
-      "rewards/margins": 7.450464725494385,
-      "rewards/rejected": -9.888655662536621,
+      "logits/chosen": -1.5075926780700684,
+      "logits/rejected": -1.286818504333496,
+      "logps/chosen": -577.96533203125,
+      "logps/rejected": -1041.372802734375,
+      "loss": 0.0844,
+      "rewards/accuracies": 0.9624999761581421,
+      "rewards/chosen": -2.9870734214782715,
+      "rewards/margins": 6.117280960083008,
+      "rewards/rejected": -9.104352951049805,
       "step": 820
     },
     {
       "epoch": 0.87,
-      "grad_norm": 49.176125880409266,
+      "grad_norm": 32.428690976585585,
       "learning_rate": 2.567240179368185e-08,
-      "logits/chosen": -0.5491518378257751,
-      "logits/rejected": 0.00046962351188994944,
-      "logps/chosen": -376.9706115722656,
-      "logps/rejected": -1090.8355712890625,
-      "loss": 0.1429,
-      "rewards/accuracies": 0.8812500238418579,
-      "rewards/chosen": -2.460528612136841,
-      "rewards/margins": 7.668572902679443,
-      "rewards/rejected": -10.129101753234863,
+      "logits/chosen": -1.5189392566680908,
+      "logits/rejected": -1.3501880168914795,
+      "logps/chosen": -603.621337890625,
+      "logps/rejected": -1007.1226806640625,
+      "loss": 0.0982,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -3.046630620956421,
+      "rewards/margins": 5.773746967315674,
+      "rewards/rejected": -8.820378303527832,
       "step": 830
     },
     {
       "epoch": 0.88,
-      "grad_norm": 37.134305931663185,
+      "grad_norm": 35.07586716910879,
       "learning_rate": 2.1787515014630357e-08,
-      "logits/chosen": -0.8858461380004883,
-      "logits/rejected": -0.30200493335723877,
-      "logps/chosen": -372.07769775390625,
-      "logps/rejected": -1134.0087890625,
-      "loss": 0.1405,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -2.2380013465881348,
-      "rewards/margins": 8.35414981842041,
-      "rewards/rejected": -10.592150688171387,
+      "logits/chosen": -1.406670331954956,
+      "logits/rejected": -1.1134653091430664,
+      "logps/chosen": -552.0911254882812,
+      "logps/rejected": -1012.5986328125,
+      "loss": 0.1141,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -2.7988593578338623,
+      "rewards/margins": 6.124402046203613,
+      "rewards/rejected": -8.923261642456055,
       "step": 840
     },
     {
       "epoch": 0.89,
-      "grad_norm": 44.332126805939915,
+      "grad_norm": 35.12000542720853,
       "learning_rate": 1.820784220652766e-08,
-      "logits/chosen": -0.44481903314590454,
-      "logits/rejected": -0.00442737340927124,
-      "logps/chosen": -393.5993957519531,
-      "logps/rejected": -1024.645751953125,
-      "loss": 0.1641,
-      "rewards/accuracies": 0.8500000238418579,
-      "rewards/chosen": -2.714064121246338,
-      "rewards/margins": 6.799878120422363,
-      "rewards/rejected": -9.513941764831543,
+      "logits/chosen": -1.5011646747589111,
+      "logits/rejected": -1.4681178331375122,
+      "logps/chosen": -629.8143310546875,
+      "logps/rejected": -1053.095703125,
+      "loss": 0.0947,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -3.13895583152771,
+      "rewards/margins": 5.925333499908447,
+      "rewards/rejected": -9.064289093017578,
       "step": 850
     },
     {
       "epoch": 0.9,
-      "grad_norm": 27.781110281391676,
+      "grad_norm": 37.505156397985395,
       "learning_rate": 1.4938170864468636e-08,
-      "logits/chosen": -0.713314414024353,
-      "logits/rejected": -0.28505033254623413,
-      "logps/chosen": -372.9283447265625,
-      "logps/rejected": -1127.688232421875,
-      "loss": 0.151,
-      "rewards/accuracies": 0.9437500238418579,
-      "rewards/chosen": -2.429081678390503,
-      "rewards/margins": 8.078619003295898,
-      "rewards/rejected": -10.507699966430664,
+      "logits/chosen": -1.5276473760604858,
+      "logits/rejected": -1.3087430000305176,
+      "logps/chosen": -611.999267578125,
+      "logps/rejected": -982.8528442382812,
+      "loss": 0.1089,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -3.328166961669922,
+      "rewards/margins": 5.248373031616211,
+      "rewards/rejected": -8.576539993286133,
       "step": 860
     },
     {
       "epoch": 0.91,
-      "grad_norm": 40.33670958727325,
+      "grad_norm": 33.59787382725122,
       "learning_rate": 1.1982873884064465e-08,
-      "logits/chosen": -0.642200767993927,
-      "logits/rejected": -0.2615371346473694,
-      "logps/chosen": -382.0066223144531,
-      "logps/rejected": -1205.7874755859375,
-      "loss": 0.1425,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -2.4844939708709717,
-      "rewards/margins": 8.8142728805542,
-      "rewards/rejected": -11.298765182495117,
+      "logits/chosen": -1.5041674375534058,
+      "logits/rejected": -1.186408281326294,
+      "logps/chosen": -543.9867553710938,
+      "logps/rejected": -960.0665893554688,
+      "loss": 0.1023,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -2.6006431579589844,
+      "rewards/margins": 5.71505069732666,
+      "rewards/rejected": -8.315693855285645,
       "step": 870
     },
     {
       "epoch": 0.92,
-      "grad_norm": 41.06397869005653,
+      "grad_norm": 24.198197782455043,
       "learning_rate": 9.345903713082304e-09,
-      "logits/chosen": -0.6282533407211304,
-      "logits/rejected": -0.19878247380256653,
-      "logps/chosen": -400.76605224609375,
-      "logps/rejected": -1277.481689453125,
-      "loss": 0.1417,
-      "rewards/accuracies": 0.9375,
-      "rewards/chosen": -2.5621111392974854,
-      "rewards/margins": 9.376588821411133,
-      "rewards/rejected": -11.938699722290039,
+      "logits/chosen": -1.5898497104644775,
+      "logits/rejected": -1.4362118244171143,
+      "logps/chosen": -581.5143432617188,
+      "logps/rejected": -1014.4615478515625,
+      "loss": 0.0837,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -2.9443726539611816,
+      "rewards/margins": 5.875336647033691,
+      "rewards/rejected": -8.819709777832031,
       "step": 880
     },
     {
       "epoch": 0.93,
-      "grad_norm": 59.83522561090434,
+      "grad_norm": 32.4653792997969,
       "learning_rate": 7.030787065396865e-09,
-      "logits/chosen": -0.5553107857704163,
-      "logits/rejected": -0.15107165277004242,
-      "logps/chosen": -375.3813781738281,
-      "logps/rejected": -1131.446533203125,
-      "loss": 0.1561,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -2.4943184852600098,
-      "rewards/margins": 8.074764251708984,
-      "rewards/rejected": -10.56908130645752,
+      "logits/chosen": -1.5886590480804443,
+      "logits/rejected": -1.2676339149475098,
+      "logps/chosen": -612.6450805664062,
+      "logps/rejected": -982.3916015625,
+      "loss": 0.1056,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -3.1008591651916504,
+      "rewards/margins": 5.476729393005371,
+      "rewards/rejected": -8.57758903503418,
       "step": 890
     },
     {
       "epoch": 0.94,
-      "grad_norm": 20.123097535281502,
+      "grad_norm": 28.89242092867656,
       "learning_rate": 5.04062020432286e-09,
-      "logits/chosen": -0.5520753860473633,
-      "logits/rejected": -0.19162021577358246,
-      "logps/chosen": -359.1523742675781,
-      "logps/rejected": -1050.653076171875,
-      "loss": 0.1479,
-      "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -2.378551959991455,
-      "rewards/margins": 7.357401371002197,
-      "rewards/rejected": -9.735953330993652,
+      "logits/chosen": -1.5399090051651,
+      "logits/rejected": -1.42655611038208,
+      "logps/chosen": -595.1466064453125,
+      "logps/rejected": -1033.9515380859375,
+      "loss": 0.088,
+      "rewards/accuracies": 0.9437500238418579,
+      "rewards/chosen": -2.967661142349243,
+      "rewards/margins": 6.027240753173828,
+      "rewards/rejected": -8.994901657104492,
       "step": 900
     },
     {
       "epoch": 0.94,
-      "eval_logits/chosen": -0.7198767066001892,
-      "eval_logits/rejected": -0.5643453598022461,
-      "eval_logps/chosen": -1010.4668579101562,
-      "eval_logps/rejected": -1164.5386962890625,
-      "eval_loss": 1.8978877067565918,
-      "eval_rewards/accuracies": 0.609375,
-      "eval_rewards/chosen": -6.986903190612793,
-      "eval_rewards/margins": 1.483232855796814,
-      "eval_rewards/rejected": -8.470136642456055,
-      "eval_runtime": 97.3726,
-      "eval_samples_per_second": 20.54,
+      "eval_logits/chosen": -1.1238834857940674,
+      "eval_logits/rejected": -1.0823395252227783,
+      "eval_logps/chosen": -554.9180297851562,
+      "eval_logps/rejected": -517.7515869140625,
+      "eval_loss": 1.3947275876998901,
+      "eval_rewards/accuracies": 0.38671875,
+      "eval_rewards/chosen": -2.43141508102417,
+      "eval_rewards/margins": -0.4291508197784424,
+      "eval_rewards/rejected": -2.0022642612457275,
+      "eval_runtime": 97.126,
+      "eval_samples_per_second": 20.592,
       "eval_steps_per_second": 0.329,
       "step": 900
     },
     {
       "epoch": 0.95,
-      "grad_norm": 27.829801809728036,
+      "grad_norm": 26.806730695970433,
       "learning_rate": 3.3780648016376866e-09,
-      "logits/chosen": -0.5426167249679565,
-      "logits/rejected": -0.20517143607139587,
-      "logps/chosen": -367.1676330566406,
-      "logps/rejected": -1091.75390625,
-      "loss": 0.1338,
-      "rewards/accuracies": 0.9125000238418579,
-      "rewards/chosen": -2.459343910217285,
-      "rewards/margins": 7.6989898681640625,
-      "rewards/rejected": -10.158334732055664,
+      "logits/chosen": -1.658847451210022,
+      "logits/rejected": -1.3762922286987305,
+      "logps/chosen": -603.6043701171875,
+      "logps/rejected": -1017.3605346679688,
+      "loss": 0.0898,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.103484630584717,
+      "rewards/margins": 5.833575248718262,
+      "rewards/rejected": -8.93705940246582,
       "step": 910
     },
     {
       "epoch": 0.96,
-      "grad_norm": 28.33961947218242,
+      "grad_norm": 15.136819931212546,
       "learning_rate": 2.0453443778310766e-09,
-      "logits/chosen": -0.6683856844902039,
-      "logits/rejected": -0.24923817813396454,
-      "logps/chosen": -397.66192626953125,
-      "logps/rejected": -1160.3267822265625,
-      "loss": 0.1607,
-      "rewards/accuracies": 0.887499988079071,
-      "rewards/chosen": -2.5613341331481934,
-      "rewards/margins": 8.2199125289917,
-      "rewards/rejected": -10.781246185302734,
+      "logits/chosen": -1.627918004989624,
+      "logits/rejected": -1.4132583141326904,
+      "logps/chosen": -563.746337890625,
+      "logps/rejected": -1013.9581909179688,
+      "loss": 0.0936,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -2.8101205825805664,
+      "rewards/margins": 6.073071479797363,
+      "rewards/rejected": -8.88319206237793,
       "step": 920
     },
     {
       "epoch": 0.97,
-      "grad_norm": 31.245120668993923,
+      "grad_norm": 36.0925338348187,
       "learning_rate": 1.0442413283435758e-09,
-      "logits/chosen": -0.5892433524131775,
-      "logits/rejected": -0.16971439123153687,
-      "logps/chosen": -382.60394287109375,
-      "logps/rejected": -1098.789306640625,
-      "loss": 0.1559,
-      "rewards/accuracies": 0.893750011920929,
-      "rewards/chosen": -2.5811736583709717,
-      "rewards/margins": 7.667700290679932,
-      "rewards/rejected": -10.248873710632324,
+      "logits/chosen": -1.5613000392913818,
+      "logits/rejected": -1.3666173219680786,
+      "logps/chosen": -583.2642822265625,
+      "logps/rejected": -1070.8175048828125,
+      "loss": 0.0947,
+      "rewards/accuracies": 0.96875,
+      "rewards/chosen": -3.103260040283203,
+      "rewards/margins": 6.282073020935059,
+      "rewards/rejected": -9.385333061218262,
       "step": 930
     },
     {
       "epoch": 0.98,
-      "grad_norm": 34.32124556738926,
+      "grad_norm": 30.829530604495584,
       "learning_rate": 3.760945397705828e-10,
-      "logits/chosen": -0.6785596013069153,
-      "logits/rejected": -0.16107648611068726,
-      "logps/chosen": -382.91876220703125,
-      "logps/rejected": -1063.984375,
-      "loss": 0.152,
-      "rewards/accuracies": 0.875,
-      "rewards/chosen": -2.6222147941589355,
-      "rewards/margins": 7.256742000579834,
-      "rewards/rejected": -9.878957748413086,
+      "logits/chosen": -1.6098768711090088,
+      "logits/rejected": -1.4137991666793823,
+      "logps/chosen": -652.3986206054688,
+      "logps/rejected": -1045.8623046875,
+      "loss": 0.1041,
+      "rewards/accuracies": 0.949999988079071,
+      "rewards/chosen": -3.067474365234375,
+      "rewards/margins": 6.041169166564941,
+      "rewards/rejected": -9.108643531799316,
       "step": 940
     },
     {
       "epoch": 0.99,
-      "grad_norm": 33.81539431852237,
+      "grad_norm": 28.95439781379653,
       "learning_rate": 4.17975992204056e-11,
-      "logits/chosen": -0.4908576011657715,
-      "logits/rejected": -0.08039870113134384,
-      "logps/chosen": -358.55047607421875,
-      "logps/rejected": -1048.7579345703125,
-      "loss": 0.1541,
-      "rewards/accuracies": 0.90625,
-      "rewards/chosen": -2.3112599849700928,
-      "rewards/margins": 7.406725883483887,
-      "rewards/rejected": -9.717985153198242,
+      "logits/chosen": -1.6131422519683838,
+      "logits/rejected": -1.4869133234024048,
+      "logps/chosen": -605.0281372070312,
+      "logps/rejected": -1031.686279296875,
+      "loss": 0.0954,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": -3.1735126972198486,
+      "rewards/margins": 5.8094305992126465,
+      "rewards/rejected": -8.982942581176758,
       "step": 950
     },
     {
       "epoch": 1.0,
       "step": 955,
       "total_flos": 0.0,
-      "train_loss": 0.1961859940234279,
-      "train_runtime": 15468.9338,
-      "train_samples_per_second": 7.904,
-      "train_steps_per_second": 0.062
+      "train_loss": 0.05187356284775659,
+      "train_runtime": 7314.4586,
+      "train_samples_per_second": 16.716,
+      "train_steps_per_second": 0.131
     }
   ],
   "logging_steps": 10,