diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -12,10 +12,10 @@
       "epoch": 0.0010468463752944255,
       "grad_norm": 5.5,
       "learning_rate": 5.208333333333333e-08,
-      "logits/chosen": -0.32969313859939575,
-      "logits/rejected": -0.3478139042854309,
-      "logps/chosen": -282.7626953125,
-      "logps/rejected": -210.6118621826172,
+      "logits/chosen": -0.3494967222213745,
+      "logits/rejected": -0.3728627860546112,
+      "logps/chosen": -285.8127136230469,
+      "logps/rejected": -212.7957000732422,
       "loss": 0.6931,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
@@ -27,663 +27,663 @@
       "epoch": 0.010468463752944255,
       "grad_norm": 6.0,
       "learning_rate": 5.208333333333334e-07,
-      "logits/chosen": -0.5201655030250549,
-      "logits/rejected": -0.5292737483978271,
-      "logps/chosen": -312.750244140625,
-      "logps/rejected": -275.4716491699219,
-      "loss": 0.6919,
-      "rewards/accuracies": 0.4652777910232544,
-      "rewards/chosen": 0.0034895986318588257,
-      "rewards/margins": 0.002287701005116105,
-      "rewards/rejected": 0.001201898674480617,
+      "logits/chosen": -0.5402188301086426,
+      "logits/rejected": -0.5500054955482483,
+      "logps/chosen": -315.3569030761719,
+      "logps/rejected": -278.0135192871094,
+      "loss": 0.6971,
+      "rewards/accuracies": 0.4236111044883728,
+      "rewards/chosen": -0.0025262832641601562,
+      "rewards/margins": -0.0020421938970685005,
+      "rewards/rejected": -0.0004840891051571816,
       "step": 10
     },
     {
       "epoch": 0.02093692750588851,
-      "grad_norm": 5.5,
+      "grad_norm": 5.46875,
       "learning_rate": 1.0416666666666667e-06,
-      "logits/chosen": -0.4813454747200012,
-      "logits/rejected": -0.5021841526031494,
-      "logps/chosen": -304.3456726074219,
-      "logps/rejected": -268.962158203125,
-      "loss": 0.6934,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": -0.0017953884089365602,
-      "rewards/margins": 0.0013109362917020917,
-      "rewards/rejected": -0.003106324467808008,
+      "logits/chosen": -0.5040005445480347,
+      "logits/rejected": -0.5246113538742065,
+      "logps/chosen": -306.70074462890625,
+      "logps/rejected": -271.21923828125,
+      "loss": 0.6942,
+      "rewards/accuracies": 0.5562499761581421,
+      "rewards/chosen": 0.007291222922503948,
+      "rewards/margins": 0.007475622929632664,
+      "rewards/rejected": -0.00018439842096995562,
       "step": 20
     },
     {
       "epoch": 0.031405391258832765,
-      "grad_norm": 5.78125,
+      "grad_norm": 5.9375,
       "learning_rate": 1.5625e-06,
-      "logits/chosen": -0.4880186915397644,
-      "logits/rejected": -0.4969080984592438,
-      "logps/chosen": -287.50286865234375,
-      "logps/rejected": -249.9375457763672,
-      "loss": 0.688,
+      "logits/chosen": -0.5098769664764404,
+      "logits/rejected": -0.5173646807670593,
+      "logps/chosen": -291.0880432128906,
+      "logps/rejected": -252.43124389648438,
+      "loss": 0.6896,
       "rewards/accuracies": 0.5562499761581421,
-      "rewards/chosen": 0.010473029688000679,
-      "rewards/margins": 0.01431797631084919,
-      "rewards/rejected": -0.003844945225864649,
+      "rewards/chosen": 0.007684030570089817,
+      "rewards/margins": 0.009358614683151245,
+      "rewards/rejected": -0.001674586907029152,
       "step": 30
     },
     {
       "epoch": 0.04187385501177702,
       "grad_norm": 5.375,
       "learning_rate": 2.0833333333333334e-06,
-      "logits/chosen": -0.46122869849205017,
-      "logits/rejected": -0.4958384037017822,
-      "logps/chosen": -303.1500549316406,
-      "logps/rejected": -241.998046875,
-      "loss": 0.683,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": 0.03339455649256706,
-      "rewards/margins": 0.0322079211473465,
-      "rewards/rejected": 0.0011866316199302673,
+      "logits/chosen": -0.4819061756134033,
+      "logits/rejected": -0.5171386003494263,
+      "logps/chosen": -306.039794921875,
+      "logps/rejected": -244.5795135498047,
+      "loss": 0.6816,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.026073966175317764,
+      "rewards/margins": 0.02752821519970894,
+      "rewards/rejected": -0.0014542475109919906,
       "step": 40
     },
     {
       "epoch": 0.05234231876472128,
-      "grad_norm": 4.59375,
+      "grad_norm": 4.5625,
       "learning_rate": 2.604166666666667e-06,
-      "logits/chosen": -0.5217183828353882,
-      "logits/rejected": -0.5492846369743347,
-      "logps/chosen": -302.4110412597656,
-      "logps/rejected": -280.4798278808594,
-      "loss": 0.6784,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.07218249142169952,
-      "rewards/margins": 0.04362189769744873,
-      "rewards/rejected": 0.028560593724250793,
+      "logits/chosen": -0.5450257658958435,
+      "logits/rejected": -0.5735383033752441,
+      "logps/chosen": -305.06109619140625,
+      "logps/rejected": -282.95440673828125,
+      "loss": 0.6762,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": 0.05157756805419922,
+      "rewards/margins": 0.039979297667741776,
+      "rewards/rejected": 0.011598266661167145,
       "step": 50
     },
     {
       "epoch": 0.06281078251766553,
-      "grad_norm": 5.03125,
+      "grad_norm": 5.40625,
       "learning_rate": 3.125e-06,
-      "logits/chosen": -0.5444208979606628,
-      "logits/rejected": -0.5450385212898254,
-      "logps/chosen": -288.3908996582031,
-      "logps/rejected": -252.4159393310547,
-      "loss": 0.662,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": 0.10959690809249878,
-      "rewards/margins": 0.04304002225399017,
-      "rewards/rejected": 0.0665568858385086,
+      "logits/chosen": -0.5655590891838074,
+      "logits/rejected": -0.5666534900665283,
+      "logps/chosen": -290.9037780761719,
+      "logps/rejected": -254.7635955810547,
+      "loss": 0.6644,
+      "rewards/accuracies": 0.5687500238418579,
+      "rewards/chosen": 0.08077142387628555,
+      "rewards/margins": 0.037061307579278946,
+      "rewards/rejected": 0.04371010884642601,
       "step": 60
     },
     {
       "epoch": 0.07327924627060979,
-      "grad_norm": 5.21875,
+      "grad_norm": 5.0625,
       "learning_rate": 3.6458333333333333e-06,
-      "logits/chosen": -0.5294634699821472,
-      "logits/rejected": -0.5446035265922546,
-      "logps/chosen": -284.2558288574219,
-      "logps/rejected": -261.0161437988281,
-      "loss": 0.6465,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.1622026413679123,
-      "rewards/margins": 0.1133759468793869,
-      "rewards/rejected": 0.04882669448852539,
+      "logits/chosen": -0.5553646683692932,
+      "logits/rejected": -0.5701892375946045,
+      "logps/chosen": -286.6636962890625,
+      "logps/rejected": -263.30523681640625,
+      "loss": 0.6466,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.16360881924629211,
+      "rewards/margins": 0.1136423796415329,
+      "rewards/rejected": 0.04996642470359802,
       "step": 70
     },
     {
       "epoch": 0.08374771002355404,
       "grad_norm": 4.78125,
       "learning_rate": 4.166666666666667e-06,
-      "logits/chosen": -0.46931973099708557,
-      "logits/rejected": -0.4959704279899597,
-      "logps/chosen": -285.55169677734375,
-      "logps/rejected": -271.7218017578125,
-      "loss": 0.6312,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": 0.2240675389766693,
-      "rewards/margins": 0.17874710261821747,
-      "rewards/rejected": 0.04532044008374214,
+      "logits/chosen": -0.4944031834602356,
+      "logits/rejected": -0.5216720700263977,
+      "logps/chosen": -287.98211669921875,
+      "logps/rejected": -273.87139892578125,
+      "loss": 0.6293,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.25989845395088196,
+      "rewards/margins": 0.19782108068466187,
+      "rewards/rejected": 0.06207737326622009,
       "step": 80
     },
     {
       "epoch": 0.0942161737764983,
-      "grad_norm": 5.90625,
+      "grad_norm": 5.3125,
       "learning_rate": 4.6875000000000004e-06,
-      "logits/chosen": -0.5021494626998901,
-      "logits/rejected": -0.5237849950790405,
-      "logps/chosen": -328.7523193359375,
-      "logps/rejected": -293.5586853027344,
-      "loss": 0.6197,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": 0.19337503612041473,
-      "rewards/margins": 0.22977392375469208,
-      "rewards/rejected": -0.036398887634277344,
+      "logits/chosen": -0.5244244337081909,
+      "logits/rejected": -0.5471149682998657,
+      "logps/chosen": -330.5142517089844,
+      "logps/rejected": -295.2393798828125,
+      "loss": 0.6177,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": 0.2595275342464447,
+      "rewards/margins": 0.25044435262680054,
+      "rewards/rejected": 0.009083209559321404,
       "step": 90
     },
     {
       "epoch": 0.10468463752944256,
-      "grad_norm": 5.65625,
+      "grad_norm": 5.71875,
       "learning_rate": 4.9997324926814375e-06,
-      "logits/chosen": -0.5387491583824158,
-      "logits/rejected": -0.5343273282051086,
-      "logps/chosen": -274.4442443847656,
-      "logps/rejected": -288.30035400390625,
-      "loss": 0.6297,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.221399188041687,
-      "rewards/margins": 0.2429426610469818,
-      "rewards/rejected": -0.0215434692800045,
+      "logits/chosen": -0.5673449039459229,
+      "logits/rejected": -0.5620633959770203,
+      "logps/chosen": -275.5569763183594,
+      "logps/rejected": -289.50457763671875,
+      "loss": 0.6275,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": 0.3272210955619812,
+      "rewards/margins": 0.25637245178222656,
+      "rewards/rejected": 0.07084863632917404,
       "step": 100
     },
     {
       "epoch": 0.10468463752944256,
-      "eval_logits/chosen": -0.5525645613670349,
-      "eval_logits/rejected": -0.5478540062904358,
-      "eval_logps/chosen": -288.4034118652344,
-      "eval_logps/rejected": -275.7340087890625,
-      "eval_loss": 0.6140495538711548,
-      "eval_rewards/accuracies": 0.6959999799728394,
-      "eval_rewards/chosen": 0.13576927781105042,
-      "eval_rewards/margins": 0.26342061161994934,
-      "eval_rewards/rejected": -0.1276513636112213,
-      "eval_runtime": 351.3471,
-      "eval_samples_per_second": 5.692,
-      "eval_steps_per_second": 0.356,
+      "eval_logits/chosen": -0.5618711709976196,
+      "eval_logits/rejected": -0.5581737160682678,
+      "eval_logps/chosen": -289.9532775878906,
+      "eval_logps/rejected": -276.77532958984375,
+      "eval_loss": 0.612218976020813,
+      "eval_rewards/accuracies": 0.6919999718666077,
+      "eval_rewards/chosen": 0.2593880295753479,
+      "eval_rewards/margins": 0.2692645788192749,
+      "eval_rewards/rejected": -0.00987652875483036,
+      "eval_runtime": 351.8724,
+      "eval_samples_per_second": 5.684,
+      "eval_steps_per_second": 0.355,
       "step": 100
     },
     {
       "epoch": 0.11515310128238682,
       "grad_norm": 4.25,
       "learning_rate": 4.996723692767927e-06,
-      "logits/chosen": -0.5775566101074219,
-      "logits/rejected": -0.603683590888977,
-      "logps/chosen": -287.82733154296875,
-      "logps/rejected": -275.31671142578125,
-      "loss": 0.606,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": 0.15195702016353607,
-      "rewards/margins": 0.34537211060523987,
-      "rewards/rejected": -0.1934150755405426,
+      "logits/chosen": -0.6067845225334167,
+      "logits/rejected": -0.6330805420875549,
+      "logps/chosen": -288.8458557128906,
+      "logps/rejected": -276.23870849609375,
+      "loss": 0.6078,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.27882444858551025,
+      "rewards/margins": 0.34665778279304504,
+      "rewards/rejected": -0.0678333044052124,
       "step": 110
     },
     {
       "epoch": 0.12562156503533106,
-      "grad_norm": 4.96875,
+      "grad_norm": 5.0625,
       "learning_rate": 4.9903757462135984e-06,
-      "logits/chosen": -0.5295812487602234,
-      "logits/rejected": -0.5440601110458374,
-      "logps/chosen": -260.30078125,
-      "logps/rejected": -251.58401489257812,
-      "loss": 0.5924,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.02338511124253273,
-      "rewards/margins": 0.281913697719574,
-      "rewards/rejected": -0.25852862000465393,
+      "logits/chosen": -0.5565638542175293,
+      "logits/rejected": -0.5734590291976929,
+      "logps/chosen": -260.9988098144531,
+      "logps/rejected": -252.0515594482422,
+      "loss": 0.591,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": 0.15620097517967224,
+      "rewards/margins": 0.29570046067237854,
+      "rewards/rejected": -0.1394994556903839,
       "step": 120
     },
     {
       "epoch": 0.1360900287882753,
-      "grad_norm": 5.46875,
+      "grad_norm": 5.5,
       "learning_rate": 4.980697142834315e-06,
-      "logits/chosen": -0.49980980157852173,
-      "logits/rejected": -0.5171049237251282,
-      "logps/chosen": -300.2707214355469,
-      "logps/rejected": -336.19036865234375,
-      "loss": 0.5933,
-      "rewards/accuracies": 0.65625,
-      "rewards/chosen": 0.02033509686589241,
-      "rewards/margins": 0.3162926733493805,
-      "rewards/rejected": -0.2959575653076172,
+      "logits/chosen": -0.5258967876434326,
+      "logits/rejected": -0.5430957078933716,
+      "logps/chosen": -301.23260498046875,
+      "logps/rejected": -336.9810485839844,
+      "loss": 0.5936,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": 0.17174401879310608,
+      "rewards/margins": 0.31216299533843994,
+      "rewards/rejected": -0.14041896164417267,
       "step": 130
     },
     {
       "epoch": 0.14655849254121958,
       "grad_norm": 4.75,
       "learning_rate": 4.967700826904229e-06,
-      "logits/chosen": -0.5714845061302185,
-      "logits/rejected": -0.5790292024612427,
-      "logps/chosen": -280.66845703125,
-      "logps/rejected": -273.6711120605469,
-      "loss": 0.5811,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.004461909644305706,
-      "rewards/margins": 0.407376229763031,
-      "rewards/rejected": -0.4118381440639496,
+      "logits/chosen": -0.5994604825973511,
+      "logits/rejected": -0.6112743616104126,
+      "logps/chosen": -281.85223388671875,
+      "logps/rejected": -274.6047058105469,
+      "loss": 0.5781,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.12341972440481186,
+      "rewards/margins": 0.4266698360443115,
+      "rewards/rejected": -0.30325013399124146,
       "step": 140
     },
     {
       "epoch": 0.15702695629416383,
-      "grad_norm": 4.84375,
+      "grad_norm": 4.8125,
       "learning_rate": 4.951404179843963e-06,
-      "logits/chosen": -0.5678334832191467,
-      "logits/rejected": -0.5395095944404602,
-      "logps/chosen": -306.054443359375,
-      "logps/rejected": -276.5906066894531,
-      "loss": 0.5777,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.13024172186851501,
-      "rewards/margins": 0.4546026587486267,
-      "rewards/rejected": -0.3243609070777893,
+      "logits/chosen": -0.5988300442695618,
+      "logits/rejected": -0.5718582272529602,
+      "logps/chosen": -308.1268310546875,
+      "logps/rejected": -279.3246765136719,
+      "loss": 0.5775,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": 0.22290559113025665,
+      "rewards/margins": 0.45703545212745667,
+      "rewards/rejected": -0.2341299057006836,
       "step": 150
     },
     {
       "epoch": 0.16749542004710807,
-      "grad_norm": 4.375,
+      "grad_norm": 4.5,
       "learning_rate": 4.931828996974498e-06,
-      "logits/chosen": -0.5105162858963013,
-      "logits/rejected": -0.5018196702003479,
-      "logps/chosen": -296.04840087890625,
-      "logps/rejected": -270.5509033203125,
-      "loss": 0.5478,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": 0.04980681464076042,
-      "rewards/margins": 0.5175832509994507,
-      "rewards/rejected": -0.4677763879299164,
+      "logits/chosen": -0.5415905117988586,
+      "logits/rejected": -0.5388587713241577,
+      "logps/chosen": -297.1036682128906,
+      "logps/rejected": -271.18475341796875,
+      "loss": 0.5459,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": 0.1913338154554367,
+      "rewards/margins": 0.5190017819404602,
+      "rewards/rejected": -0.3276679515838623,
       "step": 160
     },
     {
       "epoch": 0.17796388380005235,
-      "grad_norm": 4.78125,
+      "grad_norm": 4.875,
       "learning_rate": 4.909001458367867e-06,
-      "logits/chosen": -0.5939355492591858,
-      "logits/rejected": -0.5804780125617981,
-      "logps/chosen": -287.94708251953125,
-      "logps/rejected": -275.73980712890625,
-      "loss": 0.5789,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.19239839911460876,
-      "rewards/margins": 0.4369952082633972,
-      "rewards/rejected": -0.6293936967849731,
+      "logits/chosen": -0.6261523365974426,
+      "logits/rejected": -0.6151515245437622,
+      "logps/chosen": -288.1965637207031,
+      "logps/rejected": -275.8837585449219,
+      "loss": 0.5809,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": 0.018318424001336098,
+      "rewards/margins": 0.4437525272369385,
+      "rewards/rejected": -0.42543405294418335,
       "step": 170
     },
     {
       "epoch": 0.1884323475529966,
       "grad_norm": 5.15625,
       "learning_rate": 4.882952093833628e-06,
-      "logits/chosen": -0.5993765592575073,
-      "logits/rejected": -0.5707160830497742,
-      "logps/chosen": -301.9080505371094,
-      "logps/rejected": -265.2618713378906,
-      "loss": 0.555,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.19447076320648193,
-      "rewards/margins": 0.48417288064956665,
-      "rewards/rejected": -0.6786437630653381,
+      "logits/chosen": -0.6297336220741272,
+      "logits/rejected": -0.6054444313049316,
+      "logps/chosen": -303.1052551269531,
+      "logps/rejected": -265.9317932128906,
+      "loss": 0.5599,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.07898494601249695,
+      "rewards/margins": 0.4595085084438324,
+      "rewards/rejected": -0.5384935140609741,
       "step": 180
     },
     {
       "epoch": 0.19890081130594087,
-      "grad_norm": 5.90625,
+      "grad_norm": 5.875,
       "learning_rate": 4.853715742087947e-06,
-      "logits/chosen": -0.5384653806686401,
-      "logits/rejected": -0.5204354524612427,
-      "logps/chosen": -275.14886474609375,
-      "logps/rejected": -281.9067687988281,
-      "loss": 0.5706,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.033171653747558594,
-      "rewards/margins": 0.469027042388916,
-      "rewards/rejected": -0.5021986961364746,
+      "logits/chosen": -0.5653724074363708,
+      "logits/rejected": -0.5484523177146912,
+      "logps/chosen": -276.7878723144531,
+      "logps/rejected": -283.9636535644531,
+      "loss": 0.5658,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": 0.021318156272172928,
+      "rewards/margins": 0.4857168197631836,
+      "rewards/rejected": -0.46439868211746216,
       "step": 190
     },
     {
       "epoch": 0.2093692750588851,
-      "grad_norm": 6.5625,
+      "grad_norm": 6.5,
       "learning_rate": 4.821331504159906e-06,
-      "logits/chosen": -0.5440086722373962,
-      "logits/rejected": -0.5563145279884338,
-      "logps/chosen": -296.43060302734375,
-      "logps/rejected": -255.0513916015625,
-      "loss": 0.5676,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.12640251219272614,
-      "rewards/margins": 0.5171942710876465,
-      "rewards/rejected": -0.6435968279838562,
+      "logits/chosen": -0.5703824162483215,
+      "logits/rejected": -0.5851908922195435,
+      "logps/chosen": -298.11962890625,
+      "logps/rejected": -256.55633544921875,
+      "loss": 0.5726,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -0.06406483054161072,
+      "rewards/margins": 0.5046994090080261,
+      "rewards/rejected": -0.5687642097473145,
       "step": 200
     },
     {
       "epoch": 0.2093692750588851,
-      "eval_logits/chosen": -0.5116367340087891,
-      "eval_logits/rejected": -0.4945087134838104,
-      "eval_logps/chosen": -290.90509033203125,
-      "eval_logps/rejected": -281.0560302734375,
-      "eval_loss": 0.5569251179695129,
-      "eval_rewards/accuracies": 0.699999988079071,
-      "eval_rewards/chosen": -0.11436504870653152,
-      "eval_rewards/margins": 0.5454893708229065,
-      "eval_rewards/rejected": -0.6598544716835022,
-      "eval_runtime": 349.6273,
-      "eval_samples_per_second": 5.72,
+      "eval_logits/chosen": -0.5266345739364624,
+      "eval_logits/rejected": -0.510294497013092,
+      "eval_logps/chosen": -293.3344421386719,
+      "eval_logps/rejected": -283.0293273925781,
+      "eval_loss": 0.552874743938446,
+      "eval_rewards/accuracies": 0.7039999961853027,
+      "eval_rewards/chosen": -0.07872959226369858,
+      "eval_rewards/margins": 0.5565416812896729,
+      "eval_rewards/rejected": -0.6352713108062744,
+      "eval_runtime": 349.2838,
+      "eval_samples_per_second": 5.726,
       "eval_steps_per_second": 0.358,
       "step": 200
     },
     {
       "epoch": 0.21983773881182936,
-      "grad_norm": 6.15625,
+      "grad_norm": 6.21875,
       "learning_rate": 4.7858426910973435e-06,
-      "logits/chosen": -0.5827375650405884,
-      "logits/rejected": -0.5831678509712219,
-      "logps/chosen": -278.0469970703125,
-      "logps/rejected": -271.9605407714844,
-      "loss": 0.5508,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -0.025282684713602066,
-      "rewards/margins": 0.5209798216819763,
-      "rewards/rejected": -0.5462625622749329,
+      "logits/chosen": -0.6147966980934143,
+      "logits/rejected": -0.6168125867843628,
+      "logps/chosen": -280.411865234375,
+      "logps/rejected": -273.9303283691406,
+      "loss": 0.5479,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.01772790588438511,
+      "rewards/margins": 0.5174022912979126,
+      "rewards/rejected": -0.5351301431655884,
       "step": 210
     },
     {
       "epoch": 0.23030620256477363,
-      "grad_norm": 5.125,
+      "grad_norm": 5.15625,
       "learning_rate": 4.747296766042161e-06,
-      "logits/chosen": -0.5575278997421265,
-      "logits/rejected": -0.5416348576545715,
-      "logps/chosen": -317.4739990234375,
-      "logps/rejected": -269.8050231933594,
-      "loss": 0.5368,
+      "logits/chosen": -0.5869513750076294,
+      "logits/rejected": -0.5712444186210632,
+      "logps/chosen": -319.57781982421875,
+      "logps/rejected": -271.85662841796875,
+      "loss": 0.5371,
       "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.03362050652503967,
-      "rewards/margins": 0.569665789604187,
-      "rewards/rejected": -0.536045253276825,
+      "rewards/chosen": 0.07761907577514648,
+      "rewards/margins": 0.5722238421440125,
+      "rewards/rejected": -0.49460476636886597,
       "step": 220
     },
     {
       "epoch": 0.24077466631771788,
-      "grad_norm": 5.6875,
+      "grad_norm": 5.84375,
       "learning_rate": 4.705745280752586e-06,
-      "logits/chosen": -0.589636504650116,
-      "logits/rejected": -0.5510541200637817,
-      "logps/chosen": -290.55206298828125,
-      "logps/rejected": -287.7290954589844,
-      "loss": 0.5581,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.08335064351558685,
-      "rewards/margins": 0.5679007768630981,
-      "rewards/rejected": -0.6512514352798462,
+      "logits/chosen": -0.6202965974807739,
+      "logits/rejected": -0.5799211263656616,
+      "logps/chosen": -292.62506103515625,
+      "logps/rejected": -289.6524658203125,
+      "loss": 0.5593,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -0.0319809652864933,
+      "rewards/margins": 0.5654061436653137,
+      "rewards/rejected": -0.5973870754241943,
       "step": 230
     },
     {
       "epoch": 0.2512431300706621,
-      "grad_norm": 4.96875,
+      "grad_norm": 4.8125,
       "learning_rate": 4.661243806657256e-06,
-      "logits/chosen": -0.5663197636604309,
-      "logits/rejected": -0.525894045829773,
-      "logps/chosen": -299.3938903808594,
-      "logps/rejected": -263.598876953125,
-      "loss": 0.5563,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -0.1382952779531479,
-      "rewards/margins": 0.5307878255844116,
-      "rewards/rejected": -0.6690832376480103,
+      "logits/chosen": -0.5910071134567261,
+      "logits/rejected": -0.5533290505409241,
+      "logps/chosen": -300.802734375,
+      "logps/rejected": -264.6828918457031,
+      "loss": 0.5557,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.033292002975940704,
+      "rewards/margins": 0.5110405087471008,
+      "rewards/rejected": -0.5443325042724609,
       "step": 240
     },
     {
       "epoch": 0.26171159382360637,
-      "grad_norm": 5.6875,
+      "grad_norm": 5.71875,
       "learning_rate": 4.613851860533367e-06,
-      "logits/chosen": -0.5768588781356812,
-      "logits/rejected": -0.5346202850341797,
-      "logps/chosen": -293.30059814453125,
-      "logps/rejected": -259.82177734375,
-      "loss": 0.5789,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.08053269237279892,
-      "rewards/margins": 0.5023098587989807,
-      "rewards/rejected": -0.5828425288200378,
+      "logits/chosen": -0.6005420684814453,
+      "logits/rejected": -0.5617684125900269,
+      "logps/chosen": -294.0511169433594,
+      "logps/rejected": -260.35443115234375,
+      "loss": 0.5825,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.09464361518621445,
+      "rewards/margins": 0.5028691291809082,
+      "rewards/rejected": -0.40822547674179077,
       "step": 250
     },
     {
       "epoch": 0.2721800575765506,
       "grad_norm": 5.6875,
       "learning_rate": 4.563632824908252e-06,
-      "logits/chosen": -0.5458053946495056,
-      "logits/rejected": -0.5105060935020447,
-      "logps/chosen": -292.6302185058594,
-      "logps/rejected": -279.1827392578125,
-      "loss": 0.5404,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": 0.12764385342597961,
-      "rewards/margins": 0.8003277778625488,
-      "rewards/rejected": -0.6726840138435364,
+      "logits/chosen": -0.5696755647659302,
+      "logits/rejected": -0.5352145433425903,
+      "logps/chosen": -293.357666015625,
+      "logps/rejected": -279.78173828125,
+      "loss": 0.5358,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": 0.2980680465698242,
+      "rewards/margins": 0.81305992603302,
+      "rewards/rejected": -0.5149918794631958,
       "step": 260
     },
     {
       "epoch": 0.2826485213294949,
-      "grad_norm": 5.78125,
+      "grad_norm": 5.59375,
       "learning_rate": 4.510653863290871e-06,
-      "logits/chosen": -0.545439600944519,
-      "logits/rejected": -0.5205134749412537,
-      "logps/chosen": -294.71783447265625,
-      "logps/rejected": -302.39599609375,
-      "loss": 0.5465,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -0.015889674425125122,
-      "rewards/margins": 0.5484660267829895,
-      "rewards/rejected": -0.564355731010437,
+      "logits/chosen": -0.5663573741912842,
+      "logits/rejected": -0.5443081259727478,
+      "logps/chosen": -296.19744873046875,
+      "logps/rejected": -303.76507568359375,
+      "loss": 0.5498,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": 0.06666271388530731,
+      "rewards/margins": 0.5538051724433899,
+      "rewards/rejected": -0.4871424734592438,
       "step": 270
     },
     {
       "epoch": 0.29311698508243916,
-      "grad_norm": 4.78125,
+      "grad_norm": 4.8125,
       "learning_rate": 4.454985830346574e-06,
-      "logits/chosen": -0.6070719957351685,
-      "logits/rejected": -0.5709980726242065,
-      "logps/chosen": -300.3115234375,
-      "logps/rejected": -283.37689208984375,
-      "loss": 0.5681,
-      "rewards/accuracies": 0.6187499761581421,
-      "rewards/chosen": -0.04909560829401016,
-      "rewards/margins": 0.496112197637558,
-      "rewards/rejected": -0.5452078580856323,
+      "logits/chosen": -0.6322006583213806,
+      "logits/rejected": -0.598243236541748,
+      "logps/chosen": -301.9215087890625,
+      "logps/rejected": -285.16412353515625,
+      "loss": 0.5696,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.03948987275362015,
+      "rewards/margins": 0.49146270751953125,
+      "rewards/rejected": -0.4519728720188141,
       "step": 280
     },
     {
       "epoch": 0.3035854488353834,
       "grad_norm": 4.65625,
       "learning_rate": 4.396703177135262e-06,
-      "logits/chosen": -0.5642537474632263,
-      "logits/rejected": -0.5349927544593811,
-      "logps/chosen": -285.5408935546875,
-      "logps/rejected": -256.77691650390625,
-      "loss": 0.5249,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.10653576999902725,
-      "rewards/margins": 0.6360087990760803,
-      "rewards/rejected": -0.5294729471206665,
+      "logits/chosen": -0.5908337831497192,
+      "logits/rejected": -0.5626215934753418,
+      "logps/chosen": -287.64849853515625,
+      "logps/rejected": -258.74005126953125,
+      "loss": 0.526,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": 0.14304542541503906,
+      "rewards/margins": 0.6406429409980774,
+      "rewards/rejected": -0.49759751558303833,
       "step": 290
     },
     {
       "epoch": 0.31405391258832765,
       "grad_norm": 5.75,
       "learning_rate": 4.335883851539693e-06,
-      "logits/chosen": -0.5834770798683167,
-      "logits/rejected": -0.544857382774353,
-      "logps/chosen": -296.65936279296875,
-      "logps/rejected": -293.7464599609375,
-      "loss": 0.5414,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.30995655059814453,
-      "rewards/margins": 0.6452124714851379,
-      "rewards/rejected": -0.9551690220832825,
+      "logits/chosen": -0.6105285882949829,
+      "logits/rejected": -0.5751320123672485,
+      "logps/chosen": -297.8828125,
+      "logps/rejected": -294.6984558105469,
+      "loss": 0.5429,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.1768026053905487,
+      "rewards/margins": 0.6462854743003845,
+      "rewards/rejected": -0.8230881690979004,
       "step": 300
     },
     {
       "epoch": 0.31405391258832765,
-      "eval_logits/chosen": -0.4775179624557495,
-      "eval_logits/rejected": -0.45395201444625854,
-      "eval_logps/chosen": -293.56982421875,
-      "eval_logps/rejected": -284.91802978515625,
-      "eval_loss": 0.5403068661689758,
+      "eval_logits/chosen": -0.4910103976726532,
+      "eval_logits/rejected": -0.4688774645328522,
+      "eval_logps/chosen": -294.2773132324219,
+      "eval_logps/rejected": -285.1317138671875,
+      "eval_loss": 0.5380240678787231,
       "eval_rewards/accuracies": 0.7260000109672546,
-      "eval_rewards/chosen": -0.38083866238594055,
-      "eval_rewards/margins": 0.6652126312255859,
-      "eval_rewards/rejected": -1.0460513830184937,
-      "eval_runtime": 349.5769,
-      "eval_samples_per_second": 5.721,
+      "eval_rewards/chosen": -0.1730155199766159,
+      "eval_rewards/margins": 0.6724972128868103,
+      "eval_rewards/rejected": -0.8455128073692322,
+      "eval_runtime": 349.3616,
+      "eval_samples_per_second": 5.725,
       "eval_steps_per_second": 0.358,
       "step": 300
     },
     {
       "epoch": 0.3245223763412719,
-      "grad_norm": 6.28125,
+      "grad_norm": 6.625,
       "learning_rate": 4.2726091940171055e-06,
-      "logits/chosen": -0.5361314415931702,
-      "logits/rejected": -0.5678786635398865,
-      "logps/chosen": -296.285400390625,
-      "logps/rejected": -342.7236022949219,
-      "loss": 0.5056,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -0.19761790335178375,
-      "rewards/margins": 0.8299154043197632,
-      "rewards/rejected": -1.0275332927703857,
+      "logits/chosen": -0.5600444078445435,
+      "logits/rejected": -0.5925548672676086,
+      "logps/chosen": -296.61083984375,
+      "logps/rejected": -342.74749755859375,
+      "loss": 0.5077,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.023765895515680313,
+      "rewards/margins": 0.7941875457763672,
+      "rewards/rejected": -0.7704216241836548,
       "step": 310
     },
     {
       "epoch": 0.33499084009421615,
-      "grad_norm": 5.21875,
+      "grad_norm": 5.15625,
       "learning_rate": 4.206963828813555e-06,
-      "logits/chosen": -0.5878556370735168,
-      "logits/rejected": -0.5477277040481567,
-      "logps/chosen": -296.2107849121094,
-      "logps/rejected": -280.1604919433594,
-      "loss": 0.524,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -0.1949264109134674,
-      "rewards/margins": 0.7900466918945312,
-      "rewards/rejected": -0.9849729537963867,
+      "logits/chosen": -0.6118437647819519,
+      "logits/rejected": -0.5762773752212524,
+      "logps/chosen": -297.16278076171875,
+      "logps/rejected": -280.4404602050781,
+      "loss": 0.531,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -0.036954768002033234,
+      "rewards/margins": 0.7416526079177856,
+      "rewards/rejected": -0.7786073684692383,
       "step": 320
     },
     {
       "epoch": 0.34545930384716045,
-      "grad_norm": 5.0625,
+      "grad_norm": 5.53125,
       "learning_rate": 4.139035550786495e-06,
-      "logits/chosen": -0.6225862503051758,
-      "logits/rejected": -0.5590678453445435,
-      "logps/chosen": -289.2728576660156,
-      "logps/rejected": -260.7643737792969,
-      "loss": 0.5217,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.2666059136390686,
-      "rewards/margins": 0.6325812339782715,
-      "rewards/rejected": -0.8991872072219849,
+      "logits/chosen": -0.6456430554389954,
+      "logits/rejected": -0.5850483179092407,
+      "logps/chosen": -290.33062744140625,
+      "logps/rejected": -261.4778137207031,
+      "loss": 0.5252,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -0.11032617092132568,
+      "rewards/margins": 0.6117701530456543,
+      "rewards/rejected": -0.7220962643623352,
       "step": 330
     },
     {
       "epoch": 0.3559277676001047,
-      "grad_norm": 5.96875,
+      "grad_norm": 6.1875,
       "learning_rate": 4.068915207986931e-06,
-      "logits/chosen": -0.5779263377189636,
-      "logits/rejected": -0.5189130902290344,
-      "logps/chosen": -298.147705078125,
-      "logps/rejected": -258.82391357421875,
-      "loss": 0.5224,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.4627823233604431,
-      "rewards/margins": 0.7694815397262573,
-      "rewards/rejected": -1.2322638034820557,
+      "logits/chosen": -0.6017712354660034,
+      "logits/rejected": -0.5405057072639465,
+      "logps/chosen": -298.4487609863281,
+      "logps/rejected": -258.83343505859375,
+      "loss": 0.5244,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -0.27785810828208923,
+      "rewards/margins": 0.7664127945899963,
+      "rewards/rejected": -1.0442708730697632,
       "step": 340
     },
     {
       "epoch": 0.36639623135304894,
-      "grad_norm": 6.21875,
+      "grad_norm": 6.25,
       "learning_rate": 3.996696580158211e-06,
-      "logits/chosen": -0.5267971158027649,
-      "logits/rejected": -0.5050491690635681,
-      "logps/chosen": -335.9749755859375,
-      "logps/rejected": -290.8840026855469,
-      "loss": 0.5206,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -0.16786900162696838,
-      "rewards/margins": 0.7010205984115601,
-      "rewards/rejected": -0.8688896894454956,
+      "logits/chosen": -0.5467532873153687,
+      "logits/rejected": -0.525019645690918,
+      "logps/chosen": -337.39007568359375,
+      "logps/rejected": -292.0143127441406,
+      "loss": 0.5204,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -0.04514966905117035,
+      "rewards/margins": 0.7053651809692383,
+      "rewards/rejected": -0.7505148649215698,
       "step": 350
     },
     {
       "epoch": 0.3768646951059932,
-      "grad_norm": 5.3125,
+      "grad_norm": 5.03125,
       "learning_rate": 3.922476253313921e-06,
-      "logits/chosen": -0.5035579800605774,
-      "logits/rejected": -0.5138710141181946,
-      "logps/chosen": -273.709716796875,
-      "logps/rejected": -296.85650634765625,
-      "loss": 0.5587,
-      "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -0.28165164589881897,
-      "rewards/margins": 0.6714268922805786,
-      "rewards/rejected": -0.9530784487724304,
+      "logits/chosen": -0.5269440412521362,
+      "logits/rejected": -0.5347733497619629,
+      "logps/chosen": -275.55755615234375,
+      "logps/rejected": -298.5218811035156,
+      "loss": 0.5549,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.23508167266845703,
+      "rewards/margins": 0.6767452955245972,
+      "rewards/rejected": -0.9118270874023438,
       "step": 360
     },
     {
       "epoch": 0.38733315885893743,
-      "grad_norm": 5.875,
+      "grad_norm": 5.78125,
       "learning_rate": 3.846353490562664e-06,
-      "logits/chosen": -0.5329638123512268,
-      "logits/rejected": -0.5389483571052551,
-      "logps/chosen": -288.07745361328125,
-      "logps/rejected": -262.96124267578125,
-      "loss": 0.5038,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -0.05612688139081001,
-      "rewards/margins": 0.8792274594306946,
-      "rewards/rejected": -0.9353543519973755,
+      "logits/chosen": -0.5508509278297424,
+      "logits/rejected": -0.5576980113983154,
+      "logps/chosen": -290.8457946777344,
+      "logps/rejected": -265.32647705078125,
+      "loss": 0.508,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.06573277711868286,
+      "rewards/margins": 0.8556879758834839,
+      "rewards/rejected": -0.9214208722114563,
       "step": 370
     },
     {
       "epoch": 0.39780162261188173,
-      "grad_norm": 5.25,
+      "grad_norm": 5.28125,
       "learning_rate": 3.768430099352445e-06,
-      "logits/chosen": -0.5536895990371704,
-      "logits/rejected": -0.5606611371040344,
-      "logps/chosen": -304.82574462890625,
-      "logps/rejected": -277.5342712402344,
-      "loss": 0.5089,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -0.04281007498502731,
-      "rewards/margins": 0.9195000529289246,
-      "rewards/rejected": -0.9623101353645325,
+      "logits/chosen": -0.5762262344360352,
+      "logits/rejected": -0.5799992084503174,
+      "logps/chosen": -306.72760009765625,
+      "logps/rejected": -279.5985107421875,
+      "loss": 0.5075,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -0.010313669219613075,
+      "rewards/margins": 0.9469502568244934,
+      "rewards/rejected": -0.9572639465332031,
       "step": 380
     },
     {
       "epoch": 0.408270086364826,
-      "grad_norm": 6.5,
+      "grad_norm": 6.40625,
       "learning_rate": 3.6888102953122307e-06,
-      "logits/chosen": -0.5975215435028076,
-      "logits/rejected": -0.5678552389144897,
-      "logps/chosen": -261.8155212402344,
-      "logps/rejected": -262.86151123046875,
-      "loss": 0.5554,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.049895767122507095,
-      "rewards/margins": 0.72125244140625,
-      "rewards/rejected": -0.7711483240127563,
+      "logits/chosen": -0.6278566122055054,
+      "logits/rejected": -0.593431830406189,
+      "logps/chosen": -262.35601806640625,
+      "logps/rejected": -263.1413879394531,
+      "loss": 0.5558,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.06627392768859863,
+      "rewards/margins": 0.7061600089073181,
+      "rewards/rejected": -0.639886200428009,
       "step": 390
     },
     {
       "epoch": 0.4187385501177702,
-      "grad_norm": 5.40625,
+      "grad_norm": 5.21875,
       "learning_rate": 3.607600562872785e-06,
-      "logits/chosen": -0.5524119138717651,
-      "logits/rejected": -0.5214421153068542,
-      "logps/chosen": -284.4479675292969,
-      "logps/rejected": -274.91790771484375,
-      "loss": 0.5124,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.08701231330633163,
-      "rewards/margins": 0.7274686694145203,
-      "rewards/rejected": -0.8144810795783997,
+      "logits/chosen": -0.5806395411491394,
+      "logits/rejected": -0.5490812659263611,
+      "logps/chosen": -284.69036865234375,
+      "logps/rejected": -275.1679992675781,
+      "loss": 0.5054,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": 0.10525357723236084,
+      "rewards/margins": 0.7453383803367615,
+      "rewards/rejected": -0.6400848031044006,
       "step": 400
     },
     {
       "epoch": 0.4187385501177702,
-      "eval_logits/chosen": -0.45163384079933167,
-      "eval_logits/rejected": -0.42426639795303345,
-      "eval_logps/chosen": -292.09857177734375,
-      "eval_logps/rejected": -284.3532409667969,
-      "eval_loss": 0.5341187715530396,
-      "eval_rewards/accuracies": 0.7039999961853027,
-      "eval_rewards/chosen": -0.23370328545570374,
-      "eval_rewards/margins": 0.7558707594871521,
-      "eval_rewards/rejected": -0.9895740151405334,
-      "eval_runtime": 349.3488,
-      "eval_samples_per_second": 5.725,
+      "eval_logits/chosen": -0.45346179604530334,
+      "eval_logits/rejected": -0.42607197165489197,
+      "eval_logps/chosen": -293.4172668457031,
+      "eval_logps/rejected": -285.14593505859375,
+      "eval_loss": 0.5331673622131348,
+      "eval_rewards/accuracies": 0.7239999771118164,
+      "eval_rewards/chosen": -0.08701039105653763,
+      "eval_rewards/margins": 0.7599251866340637,
+      "eval_rewards/rejected": -0.8469354510307312,
+      "eval_runtime": 349.2048,
+      "eval_samples_per_second": 5.727,
       "eval_steps_per_second": 0.358,
       "step": 400
     },
@@ -691,915 +691,915 @@
       "epoch": 0.42920701387071447,
       "grad_norm": 5.1875,
       "learning_rate": 3.5249095128531863e-06,
-      "logits/chosen": -0.5735105872154236,
-      "logits/rejected": -0.5243286490440369,
-      "logps/chosen": -278.07611083984375,
-      "logps/rejected": -275.5641784667969,
-      "loss": 0.5147,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -0.19002243876457214,
-      "rewards/margins": 0.7818821668624878,
-      "rewards/rejected": -0.9719046354293823,
+      "logits/chosen": -0.6033052206039429,
+      "logits/rejected": -0.5547356605529785,
+      "logps/chosen": -279.104248046875,
+      "logps/rejected": -276.4415588378906,
+      "loss": 0.5154,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -0.09319617599248886,
+      "rewards/margins": 0.780913770198822,
+      "rewards/rejected": -0.8741100430488586,
       "step": 410
     },
     {
       "epoch": 0.4396754776236587,
-      "grad_norm": 5.75,
+      "grad_norm": 5.9375,
       "learning_rate": 3.4408477372034743e-06,
-      "logits/chosen": -0.5684117078781128,
-      "logits/rejected": -0.5453870296478271,
-      "logps/chosen": -308.50128173828125,
-      "logps/rejected": -296.2371826171875,
-      "loss": 0.5608,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -0.22137188911437988,
-      "rewards/margins": 0.63337242603302,
-      "rewards/rejected": -0.8547442555427551,
+      "logits/chosen": -0.5911046266555786,
+      "logits/rejected": -0.5682761669158936,
+      "logps/chosen": -309.4401550292969,
+      "logps/rejected": -297.2247009277344,
+      "loss": 0.5621,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -0.05225636810064316,
+      "rewards/margins": 0.636865496635437,
+      "rewards/rejected": -0.6891219615936279,
       "step": 420
     },
     {
       "epoch": 0.45014394137660296,
-      "grad_norm": 6.125,
+      "grad_norm": 6.34375,
       "learning_rate": 3.355527661097728e-06,
-      "logits/chosen": -0.5637370347976685,
-      "logits/rejected": -0.5657171010971069,
-      "logps/chosen": -280.50177001953125,
-      "logps/rejected": -281.47467041015625,
-      "loss": 0.5229,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.39310961961746216,
-      "rewards/margins": 0.6110748648643494,
-      "rewards/rejected": -1.0041844844818115,
+      "logits/chosen": -0.5924620032310486,
+      "logits/rejected": -0.595079243183136,
+      "logps/chosen": -280.9834899902344,
+      "logps/rejected": -281.77105712890625,
+      "loss": 0.5226,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.19577598571777344,
+      "rewards/margins": 0.608073353767395,
+      "rewards/rejected": -0.8038493990898132,
       "step": 430
     },
     {
       "epoch": 0.46061240512954726,
-      "grad_norm": 5.53125,
+      "grad_norm": 5.5625,
       "learning_rate": 3.269063392575352e-06,
-      "logits/chosen": -0.5321542024612427,
-      "logits/rejected": -0.5234124660491943,
-      "logps/chosen": -328.39447021484375,
-      "logps/rejected": -307.01556396484375,
-      "loss": 0.5135,
+      "logits/chosen": -0.5576502680778503,
+      "logits/rejected": -0.5509397983551025,
+      "logps/chosen": -328.8077087402344,
+      "logps/rejected": -306.9642333984375,
+      "loss": 0.5141,
       "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.2074432671070099,
-      "rewards/margins": 0.7093298435211182,
-      "rewards/rejected": -0.9167732000350952,
+      "rewards/chosen": -0.0049751936458051205,
+      "rewards/margins": 0.7285463213920593,
+      "rewards/rejected": -0.733521580696106,
       "step": 440
     },
     {
       "epoch": 0.4710808688824915,
-      "grad_norm": 6.625,
+      "grad_norm": 6.4375,
       "learning_rate": 3.181570569931697e-06,
-      "logits/chosen": -0.5547568202018738,
-      "logits/rejected": -0.5429819822311401,
-      "logps/chosen": -286.77447509765625,
-      "logps/rejected": -282.605224609375,
-      "loss": 0.5,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -0.41850847005844116,
-      "rewards/margins": 0.6984038352966309,
-      "rewards/rejected": -1.1169124841690063,
+      "logits/chosen": -0.5819513201713562,
+      "logits/rejected": -0.5696187019348145,
+      "logps/chosen": -287.6014709472656,
+      "logps/rejected": -283.2913513183594,
+      "loss": 0.5033,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.24654550850391388,
+      "rewards/margins": 0.6834465265274048,
+      "rewards/rejected": -0.9299919009208679,
       "step": 450
     },
     {
       "epoch": 0.48154933263543576,
-      "grad_norm": 5.125,
+      "grad_norm": 4.8125,
       "learning_rate": 3.09316620706208e-06,
-      "logits/chosen": -0.5107685327529907,
-      "logits/rejected": -0.5204188227653503,
-      "logps/chosen": -307.8090515136719,
-      "logps/rejected": -288.7105407714844,
-      "loss": 0.4984,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -0.21866807341575623,
-      "rewards/margins": 0.888043999671936,
-      "rewards/rejected": -1.106711983680725,
+      "logits/chosen": -0.5356208086013794,
+      "logits/rejected": -0.5477502346038818,
+      "logps/chosen": -308.5799560546875,
+      "logps/rejected": -289.5464172363281,
+      "loss": 0.493,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -0.07389375567436218,
+      "rewards/margins": 0.9217998385429382,
+      "rewards/rejected": -0.995693564414978,
       "step": 460
     },
     {
       "epoch": 0.49201779638838,
       "grad_norm": 5.3125,
       "learning_rate": 3.0039685369660785e-06,
-      "logits/chosen": -0.5120314359664917,
-      "logits/rejected": -0.46821707487106323,
-      "logps/chosen": -281.59722900390625,
-      "logps/rejected": -266.38134765625,
-      "loss": 0.5331,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -0.03815505653619766,
-      "rewards/margins": 0.8301553726196289,
-      "rewards/rejected": -0.8683103322982788,
+      "logits/chosen": -0.5380591154098511,
+      "logits/rejected": -0.49667781591415405,
+      "logps/chosen": -283.11297607421875,
+      "logps/rejected": -267.4956359863281,
+      "loss": 0.5366,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.035861968994140625,
+      "rewards/margins": 0.8068122863769531,
+      "rewards/rejected": -0.7709503173828125,
       "step": 470
     },
     {
       "epoch": 0.5024862601413242,
-      "grad_norm": 5.53125,
+      "grad_norm": 5.375,
       "learning_rate": 2.91409685362137e-06,
-      "logits/chosen": -0.5207866430282593,
-      "logits/rejected": -0.5130727887153625,
-      "logps/chosen": -278.73504638671875,
-      "logps/rejected": -276.3402404785156,
-      "loss": 0.5098,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.15987971425056458,
-      "rewards/margins": 0.769777774810791,
-      "rewards/rejected": -0.9296576380729675,
+      "logits/chosen": -0.5487444996833801,
+      "logits/rejected": -0.5391730666160583,
+      "logps/chosen": -280.1451110839844,
+      "logps/rejected": -277.56793212890625,
+      "loss": 0.5096,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.06115647405385971,
+      "rewards/margins": 0.7818087339401245,
+      "rewards/rejected": -0.8429651260375977,
       "step": 480
     },
     {
       "epoch": 0.5129547238942685,
-      "grad_norm": 4.9375,
+      "grad_norm": 4.8125,
       "learning_rate": 2.8236713524386085e-06,
-      "logits/chosen": -0.5813151597976685,
-      "logits/rejected": -0.5460097193717957,
-      "logps/chosen": -278.7942810058594,
-      "logps/rejected": -256.22735595703125,
-      "loss": 0.498,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -0.02987014129757881,
-      "rewards/margins": 0.7739948034286499,
-      "rewards/rejected": -0.8038649559020996,
+      "logits/chosen": -0.6103276014328003,
+      "logits/rejected": -0.5744163393974304,
+      "logps/chosen": -279.46942138671875,
+      "logps/rejected": -257.06500244140625,
+      "loss": 0.5051,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.14644916355609894,
+      "rewards/margins": 0.8112751841545105,
+      "rewards/rejected": -0.6648260354995728,
       "step": 490
     },
     {
       "epoch": 0.5234231876472127,
-      "grad_norm": 5.59375,
+      "grad_norm": 5.5625,
       "learning_rate": 2.7328129695107205e-06,
-      "logits/chosen": -0.4917908310890198,
-      "logits/rejected": -0.5057145357131958,
-      "logps/chosen": -264.3317565917969,
-      "logps/rejected": -273.6480712890625,
-      "loss": 0.5529,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -0.05067148059606552,
-      "rewards/margins": 0.9685169458389282,
-      "rewards/rejected": -1.0191885232925415,
+      "logits/chosen": -0.5192793607711792,
+      "logits/rejected": -0.5326711535453796,
+      "logps/chosen": -264.8558349609375,
+      "logps/rejected": -274.16455078125,
+      "loss": 0.5508,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": 0.14748027920722961,
+      "rewards/margins": 0.9845119714736938,
+      "rewards/rejected": -0.8370317220687866,
       "step": 500
     },
     {
       "epoch": 0.5234231876472127,
-      "eval_logits/chosen": -0.4290330708026886,
-      "eval_logits/rejected": -0.3994762897491455,
-      "eval_logps/chosen": -291.93804931640625,
-      "eval_logps/rejected": -284.49481201171875,
-      "eval_loss": 0.5259639620780945,
-      "eval_rewards/accuracies": 0.7239999771118164,
-      "eval_rewards/chosen": -0.2176663726568222,
-      "eval_rewards/margins": 0.7860648036003113,
-      "eval_rewards/rejected": -1.0037312507629395,
-      "eval_runtime": 349.4608,
-      "eval_samples_per_second": 5.723,
+      "eval_logits/chosen": -0.43347951769828796,
+      "eval_logits/rejected": -0.4044938385486603,
+      "eval_logps/chosen": -292.7540283203125,
+      "eval_logps/rejected": -284.7646179199219,
+      "eval_loss": 0.5267207622528076,
+      "eval_rewards/accuracies": 0.7179999947547913,
+      "eval_rewards/chosen": -0.02068748138844967,
+      "eval_rewards/margins": 0.7881156206130981,
+      "eval_rewards/rejected": -0.8088030219078064,
+      "eval_runtime": 349.1516,
+      "eval_samples_per_second": 5.728,
       "eval_steps_per_second": 0.358,
       "step": 500
     },
     {
       "epoch": 0.533891651400157,
-      "grad_norm": 4.75,
+      "grad_norm": 4.65625,
       "learning_rate": 2.641643219871597e-06,
-      "logits/chosen": -0.5275259017944336,
-      "logits/rejected": -0.49047979712486267,
-      "logps/chosen": -313.4814147949219,
-      "logps/rejected": -297.6827697753906,
-      "loss": 0.5085,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.3123074471950531,
-      "rewards/margins": 0.7204921841621399,
-      "rewards/rejected": -1.0327996015548706,
+      "logits/chosen": -0.5523602962493896,
+      "logits/rejected": -0.5135177969932556,
+      "logps/chosen": -313.77227783203125,
+      "logps/rejected": -297.81756591796875,
+      "loss": 0.5053,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.082377590239048,
+      "rewards/margins": 0.7289600968360901,
+      "rewards/rejected": -0.8113376498222351,
       "step": 510
     },
     {
       "epoch": 0.5443601151531012,
-      "grad_norm": 6.90625,
+      "grad_norm": 6.84375,
       "learning_rate": 2.5502840349805074e-06,
-      "logits/chosen": -0.5200182199478149,
-      "logits/rejected": -0.5098154544830322,
-      "logps/chosen": -310.73944091796875,
-      "logps/rejected": -297.84326171875,
-      "loss": 0.5471,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.353024423122406,
-      "rewards/margins": 0.8153821229934692,
-      "rewards/rejected": -1.1684064865112305,
+      "logits/chosen": -0.5441479086875916,
+      "logits/rejected": -0.5340668559074402,
+      "logps/chosen": -309.962158203125,
+      "logps/rejected": -297.2099304199219,
+      "loss": 0.5476,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.005710450001060963,
+      "rewards/margins": 0.8396749496459961,
+      "rewards/rejected": -0.8339645266532898,
       "step": 520
     },
     {
       "epoch": 0.5548285789060455,
-      "grad_norm": 5.03125,
+      "grad_norm": 5.125,
       "learning_rate": 2.4588575996495797e-06,
-      "logits/chosen": -0.47494420409202576,
-      "logits/rejected": -0.48031002283096313,
-      "logps/chosen": -272.15692138671875,
-      "logps/rejected": -262.4747009277344,
-      "loss": 0.5205,
+      "logits/chosen": -0.4979037344455719,
+      "logits/rejected": -0.5032162666320801,
+      "logps/chosen": -272.52459716796875,
+      "logps/rejected": -262.5303039550781,
+      "loss": 0.5183,
       "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -0.41294246912002563,
-      "rewards/margins": 0.8126422166824341,
-      "rewards/rejected": -1.225584626197815,
+      "rewards/chosen": -0.2069997489452362,
+      "rewards/margins": 0.838523268699646,
+      "rewards/rejected": -1.0455230474472046,
       "step": 530
     },
     {
       "epoch": 0.5652970426589898,
-      "grad_norm": 5.53125,
+      "grad_norm": 5.59375,
       "learning_rate": 2.367486188632446e-06,
-      "logits/chosen": -0.5167185068130493,
-      "logits/rejected": -0.514365017414093,
-      "logps/chosen": -285.01080322265625,
-      "logps/rejected": -324.41302490234375,
-      "loss": 0.5084,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.18025778234004974,
-      "rewards/margins": 0.8430238962173462,
-      "rewards/rejected": -1.0232816934585571,
+      "logits/chosen": -0.5421893000602722,
+      "logits/rejected": -0.539043128490448,
+      "logps/chosen": -286.2185363769531,
+      "logps/rejected": -325.55206298828125,
+      "loss": 0.5085,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -0.05336964130401611,
+      "rewards/margins": 0.8456710577011108,
+      "rewards/rejected": -0.8990408182144165,
       "step": 540
     },
     {
       "epoch": 0.575765506411934,
-      "grad_norm": 4.1875,
+      "grad_norm": 4.34375,
       "learning_rate": 2.276292003092593e-06,
-      "logits/chosen": -0.5328477621078491,
-      "logits/rejected": -0.514901340007782,
-      "logps/chosen": -257.83953857421875,
-      "logps/rejected": -265.4601135253906,
-      "loss": 0.4871,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -0.24479135870933533,
-      "rewards/margins": 0.8982963562011719,
-      "rewards/rejected": -1.1430877447128296,
+      "logits/chosen": -0.5617792010307312,
+      "logits/rejected": -0.5426347851753235,
+      "logps/chosen": -257.9263610839844,
+      "logps/rejected": -265.53057861328125,
+      "loss": 0.486,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -0.004675072617828846,
+      "rewards/margins": 0.9126423597335815,
+      "rewards/rejected": -0.9173175096511841,
       "step": 550
     },
     {
       "epoch": 0.5862339701648783,
-      "grad_norm": 4.875,
+      "grad_norm": 4.84375,
       "learning_rate": 2.1853970071701415e-06,
-      "logits/chosen": -0.5274088978767395,
-      "logits/rejected": -0.4985182285308838,
-      "logps/chosen": -278.2223815917969,
-      "logps/rejected": -279.3916931152344,
-      "loss": 0.5206,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -0.12979185581207275,
-      "rewards/margins": 0.8115113973617554,
-      "rewards/rejected": -0.9413032531738281,
+      "logits/chosen": -0.5527303218841553,
+      "logits/rejected": -0.5213286876678467,
+      "logps/chosen": -279.15924072265625,
+      "logps/rejected": -280.146484375,
+      "loss": 0.5179,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.01912788115441799,
+      "rewards/margins": 0.8087967038154602,
+      "rewards/rejected": -0.7896687388420105,
       "step": 560
     },
     {
       "epoch": 0.5967024339178225,
-      "grad_norm": 5.84375,
+      "grad_norm": 5.71875,
       "learning_rate": 2.0949227648656194e-06,
-      "logits/chosen": -0.5541775822639465,
-      "logits/rejected": -0.5307848453521729,
-      "logps/chosen": -294.7757263183594,
-      "logps/rejected": -261.8687438964844,
-      "loss": 0.5405,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -0.2242761105298996,
-      "rewards/margins": 0.8702728152275085,
-      "rewards/rejected": -1.0945489406585693,
+      "logits/chosen": -0.5781675577163696,
+      "logits/rejected": -0.5574057102203369,
+      "logps/chosen": -295.38726806640625,
+      "logps/rejected": -262.20684814453125,
+      "loss": 0.5412,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.023471500724554062,
+      "rewards/margins": 0.8612861633300781,
+      "rewards/rejected": -0.8847578167915344,
       "step": 570
     },
     {
       "epoch": 0.6071708976707668,
-      "grad_norm": 5.78125,
+      "grad_norm": 5.8125,
       "learning_rate": 2.00499027745888e-06,
-      "logits/chosen": -0.5282704830169678,
-      "logits/rejected": -0.5080757141113281,
-      "logps/chosen": -299.1280212402344,
-      "logps/rejected": -297.7095031738281,
-      "loss": 0.5509,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -0.1550193578004837,
-      "rewards/margins": 0.7501784563064575,
-      "rewards/rejected": -0.9051979184150696,
+      "logits/chosen": -0.5529965162277222,
+      "logits/rejected": -0.5318363308906555,
+      "logps/chosen": -299.06695556640625,
+      "logps/rejected": -297.82098388671875,
+      "loss": 0.5449,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": 0.07610292732715607,
+      "rewards/margins": 0.7717056274414062,
+      "rewards/rejected": -0.6956026554107666,
       "step": 580
     },
     {
       "epoch": 0.6176393614237111,
-      "grad_norm": 5.65625,
+      "grad_norm": 5.625,
       "learning_rate": 1.915719821680624e-06,
-      "logits/chosen": -0.5544322729110718,
-      "logits/rejected": -0.5079981088638306,
-      "logps/chosen": -285.83575439453125,
-      "logps/rejected": -281.488037109375,
-      "loss": 0.5198,
+      "logits/chosen": -0.5771445035934448,
+      "logits/rejected": -0.5279287099838257,
+      "logps/chosen": -286.1835021972656,
+      "logps/rejected": -281.80828857421875,
+      "loss": 0.5206,
       "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": 0.1644931137561798,
-      "rewards/margins": 0.8637059330940247,
-      "rewards/rejected": -0.6992128491401672,
+      "rewards/chosen": 0.37562283873558044,
+      "rewards/margins": 0.8612510561943054,
+      "rewards/rejected": -0.485628217458725,
       "step": 590
     },
     {
       "epoch": 0.6281078251766553,
-      "grad_norm": 5.34375,
+      "grad_norm": 5.75,
       "learning_rate": 1.8272307888529276e-06,
-      "logits/chosen": -0.47845038771629333,
-      "logits/rejected": -0.44052690267562866,
-      "logps/chosen": -261.1684265136719,
-      "logps/rejected": -279.1067810058594,
-      "loss": 0.53,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -0.013519972562789917,
-      "rewards/margins": 0.8593659400939941,
-      "rewards/rejected": -0.8728858828544617,
+      "logits/chosen": -0.5050173997879028,
+      "logits/rejected": -0.4642249643802643,
+      "logps/chosen": -262.1390380859375,
+      "logps/rejected": -278.6227111816406,
+      "loss": 0.5338,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.2233264446258545,
+      "rewards/margins": 0.8637221455574036,
+      "rewards/rejected": -0.6403957009315491,
       "step": 600
     },
     {
       "epoch": 0.6281078251766553,
-      "eval_logits/chosen": -0.4317292273044586,
-      "eval_logits/rejected": -0.4027506709098816,
-      "eval_logps/chosen": -290.4488830566406,
-      "eval_logps/rejected": -283.0403137207031,
-      "eval_loss": 0.5244117379188538,
-      "eval_rewards/accuracies": 0.7200000286102295,
-      "eval_rewards/chosen": -0.06874838471412659,
-      "eval_rewards/margins": 0.7895320653915405,
-      "eval_rewards/rejected": -0.8582804203033447,
-      "eval_runtime": 349.48,
-      "eval_samples_per_second": 5.723,
+      "eval_logits/chosen": -0.4303784966468811,
+      "eval_logits/rejected": -0.40024831891059875,
+      "eval_logps/chosen": -290.5658874511719,
+      "eval_logps/rejected": -282.5771179199219,
+      "eval_loss": 0.5262989401817322,
+      "eval_rewards/accuracies": 0.7300000190734863,
+      "eval_rewards/chosen": 0.19812585413455963,
+      "eval_rewards/margins": 0.788177490234375,
+      "eval_rewards/rejected": -0.5900516510009766,
+      "eval_runtime": 349.0523,
+      "eval_samples_per_second": 5.73,
       "eval_steps_per_second": 0.358,
       "step": 600
     },
     {
       "epoch": 0.6385762889295996,
-      "grad_norm": 4.625,
+      "grad_norm": 4.53125,
       "learning_rate": 1.739641525213929e-06,
-      "logits/chosen": -0.510543942451477,
-      "logits/rejected": -0.5007991194725037,
-      "logps/chosen": -264.67572021484375,
-      "logps/rejected": -270.05438232421875,
-      "loss": 0.4957,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -0.04250194877386093,
-      "rewards/margins": 0.8683522343635559,
-      "rewards/rejected": -0.9108540415763855,
+      "logits/chosen": -0.5362275838851929,
+      "logits/rejected": -0.5250522494316101,
+      "logps/chosen": -264.38677978515625,
+      "logps/rejected": -269.70172119140625,
+      "loss": 0.4947,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": 0.2012002021074295,
+      "rewards/margins": 0.8771921396255493,
+      "rewards/rejected": -0.675991952419281,
       "step": 610
     },
     {
       "epoch": 0.6490447526825438,
-      "grad_norm": 4.5625,
+      "grad_norm": 4.65625,
       "learning_rate": 1.6530691736402317e-06,
-      "logits/chosen": -0.5225728750228882,
-      "logits/rejected": -0.49780869483947754,
-      "logps/chosen": -291.7324523925781,
-      "logps/rejected": -280.8247985839844,
-      "loss": 0.5072,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -0.15649111568927765,
-      "rewards/margins": 0.8370261192321777,
-      "rewards/rejected": -0.9935172200202942,
+      "logits/chosen": -0.547565221786499,
+      "logits/rejected": -0.5210016965866089,
+      "logps/chosen": -292.267822265625,
+      "logps/rejected": -281.25054931640625,
+      "loss": 0.5076,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": 0.06052354723215103,
+      "rewards/margins": 0.8467457890510559,
+      "rewards/rejected": -0.7862221598625183,
       "step": 620
     },
     {
       "epoch": 0.6595132164354881,
-      "grad_norm": 5.34375,
+      "grad_norm": 5.5625,
       "learning_rate": 1.5676295169786864e-06,
-      "logits/chosen": -0.5341150164604187,
-      "logits/rejected": -0.4966842532157898,
-      "logps/chosen": -283.0377502441406,
-      "logps/rejected": -271.4356994628906,
-      "loss": 0.5294,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.335578590631485,
-      "rewards/margins": 0.8176633715629578,
-      "rewards/rejected": -1.1532419919967651,
+      "logits/chosen": -0.5603654980659485,
+      "logits/rejected": -0.5221408605575562,
+      "logps/chosen": -287.28192138671875,
+      "logps/rejected": -272.3338317871094,
+      "loss": 0.5266,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.08671960234642029,
+      "rewards/margins": 0.8218274116516113,
+      "rewards/rejected": -0.908547043800354,
       "step": 630
     },
     {
       "epoch": 0.6699816801884323,
-      "grad_norm": 5.25,
+      "grad_norm": 5.40625,
       "learning_rate": 1.4834368231970922e-06,
-      "logits/chosen": -0.5592583417892456,
-      "logits/rejected": -0.5109054446220398,
-      "logps/chosen": -287.51068115234375,
-      "logps/rejected": -272.8183288574219,
-      "loss": 0.5,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.3240619897842407,
-      "rewards/margins": 0.7095167636871338,
-      "rewards/rejected": -1.0335787534713745,
+      "logits/chosen": -0.5882564783096313,
+      "logits/rejected": -0.5383986234664917,
+      "logps/chosen": -288.31964111328125,
+      "logps/rejected": -273.548583984375,
+      "loss": 0.5013,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.17558792233467102,
+      "rewards/margins": 0.7076612710952759,
+      "rewards/rejected": -0.8832491040229797,
       "step": 640
     },
     {
       "epoch": 0.6804501439413766,
-      "grad_norm": 5.28125,
+      "grad_norm": 5.3125,
       "learning_rate": 1.4006036925609245e-06,
-      "logits/chosen": -0.5300047993659973,
-      "logits/rejected": -0.49794530868530273,
-      "logps/chosen": -300.1281433105469,
-      "logps/rejected": -249.81936645507812,
-      "loss": 0.5413,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.32234618067741394,
-      "rewards/margins": 0.8122032880783081,
-      "rewards/rejected": -1.1345494985580444,
+      "logits/chosen": -0.5553715825080872,
+      "logits/rejected": -0.5215333104133606,
+      "logps/chosen": -301.14434814453125,
+      "logps/rejected": -250.81289672851562,
+      "loss": 0.539,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.25257399678230286,
+      "rewards/margins": 0.822566032409668,
+      "rewards/rejected": -1.0751399993896484,
       "step": 650
     },
     {
       "epoch": 0.6909186076943209,
-      "grad_norm": 5.15625,
+      "grad_norm": 5.1875,
       "learning_rate": 1.3192409070404582e-06,
-      "logits/chosen": -0.550862193107605,
-      "logits/rejected": -0.532629668712616,
-      "logps/chosen": -303.74505615234375,
-      "logps/rejected": -305.6463928222656,
-      "loss": 0.5143,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -0.20528188347816467,
-      "rewards/margins": 0.7565289735794067,
-      "rewards/rejected": -0.9618108868598938,
+      "logits/chosen": -0.5745340585708618,
+      "logits/rejected": -0.551754355430603,
+      "logps/chosen": -305.2149658203125,
+      "logps/rejected": -307.12396240234375,
+      "loss": 0.5097,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.0797838568687439,
+      "rewards/margins": 0.7875758409500122,
+      "rewards/rejected": -0.8673597574234009,
       "step": 660
     },
     {
       "epoch": 0.7013870714472651,
-      "grad_norm": 6.59375,
+      "grad_norm": 6.6875,
       "learning_rate": 1.2394572821496953e-06,
-      "logits/chosen": -0.5544015169143677,
-      "logits/rejected": -0.5221393704414368,
-      "logps/chosen": -277.2763977050781,
-      "logps/rejected": -258.49676513671875,
-      "loss": 0.5181,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.20379216969013214,
-      "rewards/margins": 0.7780588269233704,
-      "rewards/rejected": -0.9818509817123413,
+      "logits/chosen": -0.5750129818916321,
+      "logits/rejected": -0.5432807207107544,
+      "logps/chosen": -278.31634521484375,
+      "logps/rejected": -259.3590087890625,
+      "loss": 0.5169,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.12945926189422607,
+      "rewards/margins": 0.7823888659477234,
+      "rewards/rejected": -0.9118480682373047,
       "step": 670
     },
     {
       "epoch": 0.7118555352002094,
-      "grad_norm": 5.09375,
+      "grad_norm": 4.9375,
       "learning_rate": 1.1613595214152713e-06,
-      "logits/chosen": -0.5822625160217285,
-      "logits/rejected": -0.540503203868866,
-      "logps/chosen": -286.454833984375,
-      "logps/rejected": -275.0696105957031,
-      "loss": 0.5043,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -0.32862305641174316,
-      "rewards/margins": 0.7822999954223633,
-      "rewards/rejected": -1.1109230518341064,
+      "logits/chosen": -0.6039143800735474,
+      "logits/rejected": -0.5614811182022095,
+      "logps/chosen": -288.0231628417969,
+      "logps/rejected": -276.5443420410156,
+      "loss": 0.5014,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.20713257789611816,
+      "rewards/margins": 0.7958000302314758,
+      "rewards/rejected": -1.0029326677322388,
       "step": 680
     },
     {
       "epoch": 0.7223239989531536,
-      "grad_norm": 4.3125,
+      "grad_norm": 4.1875,
       "learning_rate": 1.0850520736699362e-06,
-      "logits/chosen": -0.5231102705001831,
-      "logits/rejected": -0.49847880005836487,
-      "logps/chosen": -340.5146484375,
-      "logps/rejected": -315.625,
-      "loss": 0.5348,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -0.29148069024086,
-      "rewards/margins": 0.8814651370048523,
-      "rewards/rejected": -1.1729459762573242,
+      "logits/chosen": -0.5435994863510132,
+      "logits/rejected": -0.5199321508407593,
+      "logps/chosen": -342.0343933105469,
+      "logps/rejected": -316.83929443359375,
+      "loss": 0.5367,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.19384647905826569,
+      "rewards/margins": 0.8790246248245239,
+      "rewards/rejected": -1.072871208190918,
       "step": 690
     },
     {
       "epoch": 0.7327924627060979,
-      "grad_norm": 5.625,
+      "grad_norm": 5.53125,
       "learning_rate": 1.0106369933615043e-06,
-      "logits/chosen": -0.5652047991752625,
-      "logits/rejected": -0.5295856595039368,
-      "logps/chosen": -315.13458251953125,
-      "logps/rejected": -262.6495361328125,
-      "loss": 0.5028,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -0.3104090690612793,
-      "rewards/margins": 0.7127448320388794,
-      "rewards/rejected": -1.0231537818908691,
+      "logits/chosen": -0.5901396870613098,
+      "logits/rejected": -0.551986813545227,
+      "logps/chosen": -316.48187255859375,
+      "logps/rejected": -263.9020080566406,
+      "loss": 0.5064,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.20547008514404297,
+      "rewards/margins": 0.7208881378173828,
+      "rewards/rejected": -0.9263582229614258,
       "step": 700
     },
     {
       "epoch": 0.7327924627060979,
-      "eval_logits/chosen": -0.417924702167511,
-      "eval_logits/rejected": -0.38744810223579407,
-      "eval_logps/chosen": -293.1184387207031,
-      "eval_logps/rejected": -285.8177185058594,
-      "eval_loss": 0.5190439224243164,
-      "eval_rewards/accuracies": 0.7319999933242798,
-      "eval_rewards/chosen": -0.33567702770233154,
-      "eval_rewards/margins": 0.8003441095352173,
-      "eval_rewards/rejected": -1.1360211372375488,
-      "eval_runtime": 349.5304,
-      "eval_samples_per_second": 5.722,
+      "eval_logits/chosen": -0.4080447256565094,
+      "eval_logits/rejected": -0.37605100870132446,
+      "eval_logps/chosen": -294.5545959472656,
+      "eval_logps/rejected": -286.7521057128906,
+      "eval_loss": 0.5174716114997864,
+      "eval_rewards/accuracies": 0.7300000190734863,
+      "eval_rewards/chosen": -0.20074398815631866,
+      "eval_rewards/margins": 0.8068088889122009,
+      "eval_rewards/rejected": -1.0075528621673584,
+      "eval_runtime": 349.1622,
+      "eval_samples_per_second": 5.728,
       "eval_steps_per_second": 0.358,
       "step": 700
     },
     {
       "epoch": 0.7432609264590422,
-      "grad_norm": 6.1875,
+      "grad_norm": 6.03125,
       "learning_rate": 9.382138040640714e-07,
-      "logits/chosen": -0.5737341046333313,
-      "logits/rejected": -0.5333853363990784,
-      "logps/chosen": -264.9908752441406,
-      "logps/rejected": -278.7557067871094,
-      "loss": 0.5569,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.3085269331932068,
-      "rewards/margins": 0.818738579750061,
-      "rewards/rejected": -1.1272655725479126,
+      "logits/chosen": -0.5988644361495972,
+      "logits/rejected": -0.5565031170845032,
+      "logps/chosen": -265.79888916015625,
+      "logps/rejected": -279.4545593261719,
+      "loss": 0.5573,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": -0.14814189076423645,
+      "rewards/margins": 0.8269087076187134,
+      "rewards/rejected": -0.9750504493713379,
       "step": 710
     },
     {
       "epoch": 0.7537293902119864,
-      "grad_norm": 5.34375,
+      "grad_norm": 5.3125,
       "learning_rate": 8.678793653740633e-07,
-      "logits/chosen": -0.4978073537349701,
-      "logits/rejected": -0.49744582176208496,
-      "logps/chosen": -263.29010009765625,
-      "logps/rejected": -263.02618408203125,
-      "loss": 0.5204,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -0.2587752938270569,
-      "rewards/margins": 0.8211803436279297,
-      "rewards/rejected": -1.0799555778503418,
+      "logits/chosen": -0.5240329504013062,
+      "logits/rejected": -0.522124171257019,
+      "logps/chosen": -264.15338134765625,
+      "logps/rejected": -263.60552978515625,
+      "loss": 0.5183,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -0.0942729264497757,
+      "rewards/margins": 0.8182271122932434,
+      "rewards/rejected": -0.9125000834465027,
       "step": 720
     },
     {
       "epoch": 0.7641978539649307,
-      "grad_norm": 4.75,
+      "grad_norm": 4.6875,
       "learning_rate": 7.997277433690984e-07,
-      "logits/chosen": -0.5180292725563049,
-      "logits/rejected": -0.47244367003440857,
-      "logps/chosen": -301.2723693847656,
-      "logps/rejected": -287.05584716796875,
-      "loss": 0.5153,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -0.28081485629081726,
-      "rewards/margins": 0.7739724516868591,
-      "rewards/rejected": -1.054787278175354,
+      "logits/chosen": -0.5428770780563354,
+      "logits/rejected": -0.4947339594364166,
+      "logps/chosen": -302.0629577636719,
+      "logps/rejected": -287.492431640625,
+      "loss": 0.5134,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -0.12038590759038925,
+      "rewards/margins": 0.7666383981704712,
+      "rewards/rejected": -0.8870244026184082,
       "step": 730
     },
     {
       "epoch": 0.7746663177178749,
-      "grad_norm": 4.65625,
+      "grad_norm": 4.6875,
       "learning_rate": 7.338500848029603e-07,
-      "logits/chosen": -0.47029241919517517,
-      "logits/rejected": -0.48912668228149414,
-      "logps/chosen": -290.9686584472656,
-      "logps/rejected": -274.376953125,
+      "logits/chosen": -0.49188748002052307,
+      "logits/rejected": -0.5088191628456116,
+      "logps/chosen": -292.02117919921875,
+      "logps/rejected": -275.2965393066406,
       "loss": 0.4939,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -0.0831630602478981,
-      "rewards/margins": 0.7876149415969849,
-      "rewards/rejected": -0.8707780838012695,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.023575307801365852,
+      "rewards/margins": 0.7873858213424683,
+      "rewards/rejected": -0.7638105154037476,
       "step": 740
     },
     {
       "epoch": 0.7851347814708192,
-      "grad_norm": 5.09375,
+      "grad_norm": 5.0625,
       "learning_rate": 6.70334495204884e-07,
-      "logits/chosen": -0.5157723426818848,
-      "logits/rejected": -0.49868693947792053,
-      "logps/chosen": -323.7619323730469,
-      "logps/rejected": -285.82916259765625,
-      "loss": 0.5023,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -0.11948978900909424,
-      "rewards/margins": 0.8135945200920105,
-      "rewards/rejected": -0.9330843687057495,
+      "logits/chosen": -0.5374957919120789,
+      "logits/rejected": -0.516038179397583,
+      "logps/chosen": -324.4206848144531,
+      "logps/rejected": -286.3302917480469,
+      "loss": 0.5036,
+      "rewards/accuracies": 0.78125,
+      "rewards/chosen": 0.0536157488822937,
+      "rewards/margins": 0.7997394800186157,
+      "rewards/rejected": -0.746123731136322,
       "step": 750
     },
     {
       "epoch": 0.7956032452237635,
-      "grad_norm": 5.46875,
+      "grad_norm": 5.34375,
       "learning_rate": 6.092659210462232e-07,
-      "logits/chosen": -0.5377079248428345,
-      "logits/rejected": -0.5193135142326355,
-      "logps/chosen": -268.98455810546875,
-      "logps/rejected": -268.32025146484375,
-      "loss": 0.5455,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.2502976357936859,
-      "rewards/margins": 0.698966920375824,
-      "rewards/rejected": -0.949264645576477,
+      "logits/chosen": -0.5625780820846558,
+      "logits/rejected": -0.5423771142959595,
+      "logps/chosen": -269.4684143066406,
+      "logps/rejected": -268.89886474609375,
+      "loss": 0.5496,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.04108331725001335,
+      "rewards/margins": 0.7257132530212402,
+      "rewards/rejected": -0.7667967081069946,
       "step": 760
     },
     {
       "epoch": 0.8060717089767077,
-      "grad_norm": 5.28125,
+      "grad_norm": 5.6875,
       "learning_rate": 5.507260361320738e-07,
-      "logits/chosen": -0.5280566215515137,
-      "logits/rejected": -0.5302770137786865,
-      "logps/chosen": -284.72637939453125,
-      "logps/rejected": -279.0828552246094,
-      "loss": 0.5093,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.12712079286575317,
-      "rewards/margins": 0.6862561106681824,
-      "rewards/rejected": -0.8133770227432251,
+      "logits/chosen": -0.5479773283004761,
+      "logits/rejected": -0.5508543252944946,
+      "logps/chosen": -285.4389953613281,
+      "logps/rejected": -279.5918884277344,
+      "loss": 0.5066,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.08319403976202011,
+      "rewards/margins": 0.6797414422035217,
+      "rewards/rejected": -0.5965473651885986,
       "step": 770
     },
     {
       "epoch": 0.816540172729652,
       "grad_norm": 5.21875,
       "learning_rate": 4.947931323697983e-07,
-      "logits/chosen": -0.5104556083679199,
-      "logits/rejected": -0.48463624715805054,
-      "logps/chosen": -286.16925048828125,
-      "logps/rejected": -279.46295166015625,
-      "loss": 0.5278,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -0.2269764244556427,
-      "rewards/margins": 0.6587584018707275,
-      "rewards/rejected": -0.8857349157333374,
+      "logits/chosen": -0.5364648103713989,
+      "logits/rejected": -0.5092785954475403,
+      "logps/chosen": -286.8380432128906,
+      "logps/rejected": -279.98956298828125,
+      "loss": 0.522,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -0.04336369037628174,
+      "rewards/margins": 0.6655139327049255,
+      "rewards/rejected": -0.7088775634765625,
       "step": 780
     },
     {
       "epoch": 0.8270086364825961,
-      "grad_norm": 5.375,
+      "grad_norm": 5.4375,
       "learning_rate": 4.4154201506053985e-07,
-      "logits/chosen": -0.5380119681358337,
-      "logits/rejected": -0.5271115303039551,
-      "logps/chosen": -300.27313232421875,
-      "logps/rejected": -263.89984130859375,
-      "loss": 0.511,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -0.15483184158802032,
-      "rewards/margins": 0.8044347763061523,
-      "rewards/rejected": -0.9592665433883667,
+      "logits/chosen": -0.5611253976821899,
+      "logits/rejected": -0.5496819019317627,
+      "logps/chosen": -301.17742919921875,
+      "logps/rejected": -264.5899658203125,
+      "loss": 0.5154,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": -0.043927911669015884,
+      "rewards/margins": 0.7878991365432739,
+      "rewards/rejected": -0.8318269848823547,
       "step": 790
     },
     {
       "epoch": 0.8374771002355405,
-      "grad_norm": 4.84375,
+      "grad_norm": 4.875,
       "learning_rate": 3.910439028537638e-07,
-      "logits/chosen": -0.5491563677787781,
-      "logits/rejected": -0.5051991939544678,
-      "logps/chosen": -347.644287109375,
-      "logps/rejected": -302.0996398925781,
-      "loss": 0.5347,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.08171938359737396,
-      "rewards/margins": 0.7037029266357422,
-      "rewards/rejected": -0.7854223251342773,
+      "logits/chosen": -0.5692980289459229,
+      "logits/rejected": -0.5227521657943726,
+      "logps/chosen": -348.53631591796875,
+      "logps/rejected": -302.92352294921875,
+      "loss": 0.5349,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.09300284087657928,
+      "rewards/margins": 0.6956671476364136,
+      "rewards/rejected": -0.6026642322540283,
       "step": 800
     },
     {
       "epoch": 0.8374771002355405,
-      "eval_logits/chosen": -0.42248398065567017,
-      "eval_logits/rejected": -0.39241039752960205,
-      "eval_logps/chosen": -291.1650085449219,
-      "eval_logps/rejected": -283.87603759765625,
-      "eval_loss": 0.5191403031349182,
-      "eval_rewards/accuracies": 0.7319999933242798,
-      "eval_rewards/chosen": -0.14035792648792267,
-      "eval_rewards/margins": 0.8014954924583435,
-      "eval_rewards/rejected": -0.9418535232543945,
-      "eval_runtime": 349.5564,
-      "eval_samples_per_second": 5.722,
+      "eval_logits/chosen": -0.4160803556442261,
+      "eval_logits/rejected": -0.3852657377719879,
+      "eval_logps/chosen": -292.3984069824219,
+      "eval_logps/rejected": -284.57269287109375,
+      "eval_loss": 0.5197243690490723,
+      "eval_rewards/accuracies": 0.7200000286102295,
+      "eval_rewards/chosen": 0.014875907450914383,
+      "eval_rewards/margins": 0.8044845461845398,
+      "eval_rewards/rejected": -0.7896087169647217,
+      "eval_runtime": 349.1752,
+      "eval_samples_per_second": 5.728,
       "eval_steps_per_second": 0.358,
       "step": 800
     },
     {
       "epoch": 0.8479455639884846,
-      "grad_norm": 4.6875,
+      "grad_norm": 4.8125,
       "learning_rate": 3.4336633249862084e-07,
-      "logits/chosen": -0.5698053240776062,
-      "logits/rejected": -0.5044723153114319,
-      "logps/chosen": -319.55126953125,
-      "logps/rejected": -288.7781066894531,
-      "loss": 0.5021,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -0.17804962396621704,
-      "rewards/margins": 0.8044875860214233,
-      "rewards/rejected": -0.9825371503829956,
+      "logits/chosen": -0.5947756767272949,
+      "logits/rejected": -0.5246752500534058,
+      "logps/chosen": -320.2247009277344,
+      "logps/rejected": -289.26373291015625,
+      "loss": 0.5012,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": 0.009494239464402199,
+      "rewards/margins": 0.8274568319320679,
+      "rewards/rejected": -0.8179625272750854,
       "step": 810
     },
     {
       "epoch": 0.8584140277414289,
-      "grad_norm": 5.25,
+      "grad_norm": 5.09375,
       "learning_rate": 2.98573068519539e-07,
-      "logits/chosen": -0.5468612909317017,
-      "logits/rejected": -0.5319810509681702,
-      "logps/chosen": -307.5137939453125,
-      "logps/rejected": -293.56561279296875,
-      "loss": 0.5358,
+      "logits/chosen": -0.570965588092804,
+      "logits/rejected": -0.5559743046760559,
+      "logps/chosen": -308.2550048828125,
+      "logps/rejected": -294.05462646484375,
+      "loss": 0.5287,
       "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.15299567580223083,
-      "rewards/margins": 0.7571539282798767,
-      "rewards/rejected": -0.9101495742797852,
+      "rewards/chosen": 0.03936102241277695,
+      "rewards/margins": 0.7610028982162476,
+      "rewards/rejected": -0.7216418981552124,
       "step": 820
     },
     {
       "epoch": 0.8688824914943732,
-      "grad_norm": 5.375,
+      "grad_norm": 5.53125,
       "learning_rate": 2.5672401793681854e-07,
-      "logits/chosen": -0.565108597278595,
-      "logits/rejected": -0.5492611527442932,
-      "logps/chosen": -274.61260986328125,
-      "logps/rejected": -269.25439453125,
-      "loss": 0.5048,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -0.17146927118301392,
-      "rewards/margins": 0.8092424273490906,
-      "rewards/rejected": -0.9807117581367493,
+      "logits/chosen": -0.5910287499427795,
+      "logits/rejected": -0.5737439393997192,
+      "logps/chosen": -275.55743408203125,
+      "logps/rejected": -270.0027770996094,
+      "loss": 0.5101,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.0218903087079525,
+      "rewards/margins": 0.816811203956604,
+      "rewards/rejected": -0.8387015461921692,
       "step": 830
     },
     {
       "epoch": 0.8793509552473174,
-      "grad_norm": 5.1875,
+      "grad_norm": 4.9375,
       "learning_rate": 2.178751501463036e-07,
-      "logits/chosen": -0.533309280872345,
-      "logits/rejected": -0.5187536478042603,
-      "logps/chosen": -313.8269958496094,
-      "logps/rejected": -307.17987060546875,
-      "loss": 0.517,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -0.20717601478099823,
-      "rewards/margins": 0.7723677158355713,
-      "rewards/rejected": -0.9795438051223755,
+      "logits/chosen": -0.5559400916099548,
+      "logits/rejected": -0.5404902696609497,
+      "logps/chosen": -315.4130859375,
+      "logps/rejected": -308.131591796875,
+      "loss": 0.5091,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.04114828258752823,
+      "rewards/margins": 0.7911642789840698,
+      "rewards/rejected": -0.8323125839233398,
       "step": 840
     },
     {
       "epoch": 0.8898194190002617,
       "grad_norm": 5.4375,
       "learning_rate": 1.820784220652766e-07,
-      "logits/chosen": -0.5832973122596741,
-      "logits/rejected": -0.5479104518890381,
-      "logps/chosen": -345.9014587402344,
-      "logps/rejected": -279.53985595703125,
-      "loss": 0.5083,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.0073674083687365055,
-      "rewards/margins": 0.875013530254364,
-      "rewards/rejected": -0.8823810815811157,
+      "logits/chosen": -0.6067087054252625,
+      "logits/rejected": -0.5707641839981079,
+      "logps/chosen": -347.02484130859375,
+      "logps/rejected": -280.36297607421875,
+      "loss": 0.5052,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.15417176485061646,
+      "rewards/margins": 0.8600176572799683,
+      "rewards/rejected": -0.7058458924293518,
       "step": 850
     },
     {
       "epoch": 0.9002878827532059,
-      "grad_norm": 5.375,
+      "grad_norm": 5.46875,
       "learning_rate": 1.4938170864468636e-07,
-      "logits/chosen": -0.5060287117958069,
-      "logits/rejected": -0.4896390438079834,
-      "logps/chosen": -289.37042236328125,
-      "logps/rejected": -268.93487548828125,
-      "loss": 0.4806,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -0.11767210066318512,
-      "rewards/margins": 0.9038281440734863,
-      "rewards/rejected": -1.0215001106262207,
+      "logits/chosen": -0.5292789936065674,
+      "logits/rejected": -0.509742021560669,
+      "logps/chosen": -290.58807373046875,
+      "logps/rejected": -269.5320739746094,
+      "loss": 0.4873,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": 0.0017050117021426558,
+      "rewards/margins": 0.8745778799057007,
+      "rewards/rejected": -0.8728728294372559,
       "step": 860
     },
     {
       "epoch": 0.9107563465061502,
-      "grad_norm": 5.53125,
+      "grad_norm": 5.625,
       "learning_rate": 1.1982873884064466e-07,
-      "logits/chosen": -0.4754942059516907,
-      "logits/rejected": -0.4765293598175049,
-      "logps/chosen": -287.2554931640625,
-      "logps/rejected": -276.70794677734375,
-      "loss": 0.5263,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.16617272794246674,
-      "rewards/margins": 0.6353666186332703,
-      "rewards/rejected": -0.801539421081543,
+      "logits/chosen": -0.49922800064086914,
+      "logits/rejected": -0.4975685477256775,
+      "logps/chosen": -288.2568054199219,
+      "logps/rejected": -277.89642333984375,
+      "loss": 0.5241,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.032976943999528885,
+      "rewards/margins": 0.6596282720565796,
+      "rewards/rejected": -0.6926053166389465,
       "step": 870
     },
     {
       "epoch": 0.9212248102590945,
-      "grad_norm": 5.75,
+      "grad_norm": 5.78125,
       "learning_rate": 9.345903713082305e-08,
-      "logits/chosen": -0.5580819845199585,
-      "logits/rejected": -0.5499614477157593,
-      "logps/chosen": -315.15478515625,
-      "logps/rejected": -280.8697814941406,
-      "loss": 0.5463,
-      "rewards/accuracies": 0.625,
-      "rewards/chosen": -0.1284041702747345,
-      "rewards/margins": 0.6124891042709351,
-      "rewards/rejected": -0.7408932447433472,
+      "logits/chosen": -0.58247309923172,
+      "logits/rejected": -0.574453592300415,
+      "logps/chosen": -315.4455871582031,
+      "logps/rejected": -281.5334167480469,
+      "loss": 0.5391,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": 0.010342386551201344,
+      "rewards/margins": 0.6320812702178955,
+      "rewards/rejected": -0.621738851070404,
       "step": 880
     },
     {
       "epoch": 0.9316932740120387,
-      "grad_norm": 5.125,
+      "grad_norm": 5.0625,
       "learning_rate": 7.030787065396866e-08,
-      "logits/chosen": -0.5307958722114563,
-      "logits/rejected": -0.49479198455810547,
-      "logps/chosen": -318.7244567871094,
-      "logps/rejected": -293.06219482421875,
-      "loss": 0.5198,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.11201741546392441,
-      "rewards/margins": 0.724387526512146,
-      "rewards/rejected": -0.8364049792289734,
+      "logits/chosen": -0.5529004335403442,
+      "logits/rejected": -0.5129944682121277,
+      "logps/chosen": -319.9122314453125,
+      "logps/rejected": -294.0019836425781,
+      "loss": 0.5227,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.045671284198760986,
+      "rewards/margins": 0.709865391254425,
+      "rewards/rejected": -0.6641942262649536,
       "step": 890
     },
     {
       "epoch": 0.942161737764983,
-      "grad_norm": 4.96875,
+      "grad_norm": 5.09375,
       "learning_rate": 5.0406202043228604e-08,
-      "logits/chosen": -0.541701078414917,
-      "logits/rejected": -0.5375146865844727,
-      "logps/chosen": -333.6339111328125,
-      "logps/rejected": -274.9059753417969,
-      "loss": 0.4783,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -0.0002611428499221802,
-      "rewards/margins": 0.9774513244628906,
-      "rewards/rejected": -0.9777124524116516,
+      "logits/chosen": -0.5645865201950073,
+      "logits/rejected": -0.5604526400566101,
+      "logps/chosen": -334.4140930175781,
+      "logps/rejected": -275.6079406738281,
+      "loss": 0.4775,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.1654743105173111,
+      "rewards/margins": 0.9833230972290039,
+      "rewards/rejected": -0.817848801612854,
       "step": 900
     },
     {
       "epoch": 0.942161737764983,
-      "eval_logits/chosen": -0.4219192862510681,
-      "eval_logits/rejected": -0.391748309135437,
-      "eval_logps/chosen": -291.15997314453125,
-      "eval_logps/rejected": -283.916259765625,
-      "eval_loss": 0.5189567804336548,
+      "eval_logits/chosen": -0.4151337742805481,
+      "eval_logits/rejected": -0.3841632902622223,
+      "eval_logps/chosen": -292.3968200683594,
+      "eval_logps/rejected": -284.6649169921875,
+      "eval_loss": 0.5181210041046143,
       "eval_rewards/accuracies": 0.7260000109672546,
-      "eval_rewards/chosen": -0.13985459506511688,
-      "eval_rewards/margins": 0.8060198426246643,
-      "eval_rewards/rejected": -0.9458745121955872,
-      "eval_runtime": 349.4658,
-      "eval_samples_per_second": 5.723,
+      "eval_rewards/chosen": 0.01503300666809082,
+      "eval_rewards/margins": 0.8138652443885803,
+      "eval_rewards/rejected": -0.7988322973251343,
+      "eval_runtime": 349.0965,
+      "eval_samples_per_second": 5.729,
       "eval_steps_per_second": 0.358,
       "step": 900
     },
     {
       "epoch": 0.9526302015179272,
-      "grad_norm": 5.15625,
+      "grad_norm": 5.0625,
       "learning_rate": 3.378064801637687e-08,
-      "logits/chosen": -0.5818526148796082,
-      "logits/rejected": -0.5350463390350342,
-      "logps/chosen": -314.6712341308594,
-      "logps/rejected": -313.63116455078125,
-      "loss": 0.5208,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -0.047819893807172775,
-      "rewards/margins": 0.7616057991981506,
-      "rewards/rejected": -0.8094256520271301,
+      "logits/chosen": -0.6055993437767029,
+      "logits/rejected": -0.557610034942627,
+      "logps/chosen": -315.5115966796875,
+      "logps/rejected": -314.4410400390625,
+      "loss": 0.5209,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.11909689754247665,
+      "rewards/margins": 0.7703868746757507,
+      "rewards/rejected": -0.6512900590896606,
       "step": 910
     },
     {
       "epoch": 0.9630986652708715,
       "grad_norm": 4.6875,
       "learning_rate": 2.0453443778310766e-08,
-      "logits/chosen": -0.5121539831161499,
-      "logits/rejected": -0.48242586851119995,
-      "logps/chosen": -329.2786865234375,
-      "logps/rejected": -304.5957336425781,
-      "loss": 0.5356,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -0.06458105146884918,
-      "rewards/margins": 0.869833767414093,
-      "rewards/rejected": -0.9344147443771362,
+      "logits/chosen": -0.5314048528671265,
+      "logits/rejected": -0.49928751587867737,
+      "logps/chosen": -330.26824951171875,
+      "logps/rejected": -305.7253723144531,
+      "loss": 0.533,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.07462009787559509,
+      "rewards/margins": 0.8954026103019714,
+      "rewards/rejected": -0.820782482624054,
       "step": 920
     },
     {
       "epoch": 0.9735671290238157,
       "grad_norm": 4.78125,
       "learning_rate": 1.0442413283435759e-08,
-      "logits/chosen": -0.509577751159668,
-      "logits/rejected": -0.47219425439834595,
-      "logps/chosen": -317.72943115234375,
-      "logps/rejected": -278.4623718261719,
-      "loss": 0.5143,
+      "logits/chosen": -0.5311998724937439,
+      "logits/rejected": -0.49351367354393005,
+      "logps/chosen": -318.3543701171875,
+      "logps/rejected": -278.8912048339844,
+      "loss": 0.514,
       "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -0.05142117291688919,
-      "rewards/margins": 1.1160078048706055,
-      "rewards/rejected": -1.167428970336914,
+      "rewards/chosen": 0.13066162168979645,
+      "rewards/margins": 1.1091994047164917,
+      "rewards/rejected": -0.9785377383232117,
       "step": 930
     },
     {
       "epoch": 0.98403559277676,
-      "grad_norm": 4.1875,
+      "grad_norm": 4.28125,
       "learning_rate": 3.760945397705828e-09,
-      "logits/chosen": -0.5354495048522949,
-      "logits/rejected": -0.4856652617454529,
-      "logps/chosen": -290.7127380371094,
-      "logps/rejected": -260.788330078125,
-      "loss": 0.487,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": 0.03208203241229057,
-      "rewards/margins": 1.0025131702423096,
-      "rewards/rejected": -0.9704310297966003,
+      "logits/chosen": -0.5610886812210083,
+      "logits/rejected": -0.5083616971969604,
+      "logps/chosen": -291.5441589355469,
+      "logps/rejected": -261.2994079589844,
+      "loss": 0.4851,
+      "rewards/accuracies": 0.8187500238418579,
+      "rewards/chosen": 0.18183191120624542,
+      "rewards/margins": 1.0001600980758667,
+      "rewards/rejected": -0.8183282613754272,
       "step": 940
     },
     {
       "epoch": 0.9945040565297043,
-      "grad_norm": 4.6875,
+      "grad_norm": 4.875,
       "learning_rate": 4.1797599220405605e-10,
-      "logits/chosen": -0.5512427687644958,
-      "logits/rejected": -0.5348908305168152,
-      "logps/chosen": -295.21319580078125,
-      "logps/rejected": -278.82257080078125,
-      "loss": 0.5053,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -0.0824163556098938,
-      "rewards/margins": 0.8457223176956177,
-      "rewards/rejected": -0.9281387329101562,
+      "logits/chosen": -0.576126217842102,
+      "logits/rejected": -0.5575291514396667,
+      "logps/chosen": -295.9472961425781,
+      "logps/rejected": -279.5827331542969,
+      "loss": 0.5083,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.11896193027496338,
+      "rewards/margins": 0.8570976257324219,
+      "rewards/rejected": -0.7381356358528137,
       "step": 950
     },
     {
       "epoch": 0.9997382884061764,
       "step": 955,
       "total_flos": 0.0,
-      "train_loss": 0.543355998318857,
-      "train_runtime": 19333.2554,
-      "train_samples_per_second": 3.162,
-      "train_steps_per_second": 0.049
+      "train_loss": 0.5431846207973221,
+      "train_runtime": 19104.9346,
+      "train_samples_per_second": 3.2,
+      "train_steps_per_second": 0.05
     }
   ],
   "logging_steps": 10,