diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -10,13 +10,13 @@
   "log_history": [
     {
       "epoch": 0.0010465724751439038,
-      "grad_norm": 12.735772520921401,
+      "grad_norm": 21.102116873134612,
       "learning_rate": 5.208333333333333e-09,
       "logits/chosen": -2.924262046813965,
       "logits/rejected": -2.7925047874450684,
       "logps/chosen": -380.8447570800781,
       "logps/rejected": -358.51123046875,
-      "loss": 0.6931,
+      "loss": 4.6506,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,1581 +25,1581 @@
     },
     {
       "epoch": 0.010465724751439037,
-      "grad_norm": 10.939955073523928,
+      "grad_norm": 15.822543074567085,
       "learning_rate": 5.208333333333333e-08,
-      "logits/chosen": -2.5958893299102783,
-      "logits/rejected": -2.5695536136627197,
-      "logps/chosen": -256.6239318847656,
-      "logps/rejected": -234.886962890625,
-      "loss": 0.6932,
+      "logits/chosen": -2.595761299133301,
+      "logits/rejected": -2.569227457046509,
+      "logps/chosen": -256.6064453125,
+      "logps/rejected": -234.93408203125,
+      "loss": 4.5621,
       "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.00025410810485482216,
-      "rewards/margins": 0.00034702278207987547,
-      "rewards/rejected": -9.291467722505331e-05,
+      "rewards/chosen": 0.00042897689854726195,
+      "rewards/margins": 0.0009927540086209774,
+      "rewards/rejected": -0.0005637770518660545,
       "step": 10
     },
     {
       "epoch": 0.020931449502878074,
-      "grad_norm": 12.592291847861857,
+      "grad_norm": 18.010820015079055,
       "learning_rate": 1.0416666666666667e-07,
-      "logits/chosen": -2.6138570308685303,
-      "logits/rejected": -2.5763792991638184,
-      "logps/chosen": -283.1133728027344,
-      "logps/rejected": -282.29644775390625,
-      "loss": 0.693,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": -0.00030150412931106985,
-      "rewards/margins": -8.710605470696464e-05,
-      "rewards/rejected": -0.00021439809643197805,
+      "logits/chosen": -2.613164186477661,
+      "logits/rejected": -2.5756287574768066,
+      "logps/chosen": -283.0158996582031,
+      "logps/rejected": -282.265869140625,
+      "loss": 4.4053,
+      "rewards/accuracies": 0.543749988079071,
+      "rewards/chosen": 0.0006733193295076489,
+      "rewards/margins": 0.0005819452926516533,
+      "rewards/rejected": 9.137402230408043e-05,
       "step": 20
     },
     {
       "epoch": 0.03139717425431711,
-      "grad_norm": 12.15555855567297,
+      "grad_norm": 21.44807572026145,
       "learning_rate": 1.5624999999999999e-07,
-      "logits/chosen": -2.691681385040283,
-      "logits/rejected": -2.6676158905029297,
-      "logps/chosen": -270.2156982421875,
-      "logps/rejected": -276.73724365234375,
-      "loss": 0.6923,
-      "rewards/accuracies": 0.543749988079071,
-      "rewards/chosen": 0.0006642484804615378,
-      "rewards/margins": 0.0012884512543678284,
-      "rewards/rejected": -0.0006242028321139514,
+      "logits/chosen": -2.691143035888672,
+      "logits/rejected": -2.6666667461395264,
+      "logps/chosen": -269.9042053222656,
+      "logps/rejected": -276.4795837402344,
+      "loss": 5.105,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.0037794082891196012,
+      "rewards/margins": 0.0018267262494191527,
+      "rewards/rejected": 0.0019526820397004485,
       "step": 30
     },
     {
       "epoch": 0.04186289900575615,
-      "grad_norm": 11.841392011083105,
+      "grad_norm": 17.302023991146115,
       "learning_rate": 2.0833333333333333e-07,
-      "logits/chosen": -2.6653332710266113,
-      "logits/rejected": -2.590430736541748,
-      "logps/chosen": -290.4707946777344,
-      "logps/rejected": -282.1141662597656,
-      "loss": 0.6902,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.00528017058968544,
-      "rewards/margins": 0.0057797241024672985,
-      "rewards/rejected": -0.0004995539784431458,
+      "logits/chosen": -2.6577816009521484,
+      "logits/rejected": -2.5818943977355957,
+      "logps/chosen": -288.9285888671875,
+      "logps/rejected": -280.9770202636719,
+      "loss": 4.9032,
+      "rewards/accuracies": 0.606249988079071,
+      "rewards/chosen": 0.020702064037322998,
+      "rewards/margins": 0.009830506518483162,
+      "rewards/rejected": 0.01087155845016241,
       "step": 40
     },
     {
       "epoch": 0.052328623757195186,
-      "grad_norm": 13.24083046891695,
+      "grad_norm": 22.46337927130885,
       "learning_rate": 2.604166666666667e-07,
-      "logits/chosen": -2.6722922325134277,
-      "logits/rejected": -2.588569164276123,
-      "logps/chosen": -266.1282958984375,
-      "logps/rejected": -236.51809692382812,
-      "loss": 0.6859,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.014677075669169426,
-      "rewards/margins": 0.013995910063385963,
-      "rewards/rejected": 0.0006811673520132899,
+      "logits/chosen": -2.6507585048675537,
+      "logits/rejected": -2.5627222061157227,
+      "logps/chosen": -263.1905212402344,
+      "logps/rejected": -234.9305419921875,
+      "loss": 4.8274,
+      "rewards/accuracies": 0.606249988079071,
+      "rewards/chosen": 0.044054824858903885,
+      "rewards/margins": 0.02749818004667759,
+      "rewards/rejected": 0.016556641086935997,
       "step": 50
     },
     {
       "epoch": 0.06279434850863422,
-      "grad_norm": 11.856942211355204,
+      "grad_norm": 18.98737987603255,
       "learning_rate": 3.1249999999999997e-07,
-      "logits/chosen": -2.627357244491577,
-      "logits/rejected": -2.591691493988037,
-      "logps/chosen": -299.61175537109375,
-      "logps/rejected": -274.43731689453125,
-      "loss": 0.6791,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": 0.044125042855739594,
-      "rewards/margins": 0.030969683080911636,
-      "rewards/rejected": 0.013155360706150532,
+      "logits/chosen": -2.5976526737213135,
+      "logits/rejected": -2.5587098598480225,
+      "logps/chosen": -299.9574890136719,
+      "logps/rejected": -276.1783142089844,
+      "loss": 4.5279,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": 0.040667824447155,
+      "rewards/margins": 0.04492232948541641,
+      "rewards/rejected": -0.004254504106938839,
       "step": 60
     },
     {
       "epoch": 0.07326007326007326,
-      "grad_norm": 11.929671055235316,
+      "grad_norm": 20.501382800234886,
       "learning_rate": 3.645833333333333e-07,
-      "logits/chosen": -2.5305747985839844,
-      "logits/rejected": -2.523099422454834,
-      "logps/chosen": -257.43914794921875,
-      "logps/rejected": -262.67779541015625,
-      "loss": 0.6702,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": 0.021997224539518356,
-      "rewards/margins": 0.06655967980623245,
-      "rewards/rejected": -0.044562458992004395,
+      "logits/chosen": -2.469130039215088,
+      "logits/rejected": -2.452857732772827,
+      "logps/chosen": -265.96978759765625,
+      "logps/rejected": -271.6788330078125,
+      "loss": 4.6703,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.0633089542388916,
+      "rewards/margins": 0.07126398384571075,
+      "rewards/rejected": -0.13457295298576355,
       "step": 70
     },
     {
       "epoch": 0.0837257980115123,
-      "grad_norm": 13.864057015926953,
+      "grad_norm": 25.49997843488533,
       "learning_rate": 4.1666666666666667e-07,
-      "logits/chosen": -2.5468287467956543,
-      "logits/rejected": -2.474470376968384,
-      "logps/chosen": -274.47784423828125,
-      "logps/rejected": -261.062255859375,
-      "loss": 0.6503,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": 0.01564904674887657,
-      "rewards/margins": 0.11314449459314346,
-      "rewards/rejected": -0.097495436668396,
+      "logits/chosen": -2.4551777839660645,
+      "logits/rejected": -2.3624327182769775,
+      "logps/chosen": -285.5320739746094,
+      "logps/rejected": -276.4596252441406,
+      "loss": 4.5605,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.09489366412162781,
+      "rewards/margins": 0.15657536685466766,
+      "rewards/rejected": -0.2514690160751343,
       "step": 80
     },
     {
       "epoch": 0.09419152276295134,
-      "grad_norm": 15.875862157352557,
+      "grad_norm": 30.61647338954573,
       "learning_rate": 4.6874999999999996e-07,
-      "logits/chosen": -2.488557815551758,
-      "logits/rejected": -2.4535651206970215,
-      "logps/chosen": -261.57366943359375,
-      "logps/rejected": -278.4251403808594,
-      "loss": 0.642,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -0.059757404029369354,
-      "rewards/margins": 0.07676272094249725,
-      "rewards/rejected": -0.1365201324224472,
+      "logits/chosen": -2.3756256103515625,
+      "logits/rejected": -2.332918882369995,
+      "logps/chosen": -277.46014404296875,
+      "logps/rejected": -290.0049743652344,
+      "loss": 4.1231,
+      "rewards/accuracies": 0.5562499761581421,
+      "rewards/chosen": -0.21862252056598663,
+      "rewards/margins": 0.033695660531520844,
+      "rewards/rejected": -0.25231820344924927,
       "step": 90
     },
     {
       "epoch": 0.10465724751439037,
-      "grad_norm": 13.065702068285603,
+      "grad_norm": 38.124561793065574,
       "learning_rate": 4.999732492681437e-07,
-      "logits/chosen": -2.4728801250457764,
-      "logits/rejected": -2.390676498413086,
-      "logps/chosen": -304.7428894042969,
-      "logps/rejected": -304.18231201171875,
-      "loss": 0.6267,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -0.0491873137652874,
-      "rewards/margins": 0.19260282814502716,
-      "rewards/rejected": -0.24179014563560486,
+      "logits/chosen": -2.332035779953003,
+      "logits/rejected": -2.2253689765930176,
+      "logps/chosen": -314.4341125488281,
+      "logps/rejected": -317.18695068359375,
+      "loss": 4.5854,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.1460995227098465,
+      "rewards/margins": 0.22573721408843994,
+      "rewards/rejected": -0.37183672189712524,
       "step": 100
     },
     {
       "epoch": 0.10465724751439037,
-      "eval_logits/chosen": -2.4988505840301514,
-      "eval_logits/rejected": -2.424436569213867,
-      "eval_logps/chosen": -280.5349426269531,
-      "eval_logps/rejected": -274.64080810546875,
-      "eval_loss": 0.6375061273574829,
-      "eval_rewards/accuracies": 0.7063491940498352,
-      "eval_rewards/chosen": 0.014330551959574223,
-      "eval_rewards/margins": 0.1586705446243286,
-      "eval_rewards/rejected": -0.14433999359607697,
-      "eval_runtime": 175.7842,
-      "eval_samples_per_second": 11.378,
-      "eval_steps_per_second": 0.358,
+      "eval_logits/chosen": -2.2812609672546387,
+      "eval_logits/rejected": -2.192293167114258,
+      "eval_logps/chosen": -309.1551818847656,
+      "eval_logps/rejected": -310.1242370605469,
+      "eval_loss": 4.381103515625,
+      "eval_rewards/accuracies": 0.648809552192688,
+      "eval_rewards/chosen": -0.2718724012374878,
+      "eval_rewards/margins": 0.2273014634847641,
+      "eval_rewards/rejected": -0.4991738498210907,
+      "eval_runtime": 176.2372,
+      "eval_samples_per_second": 11.348,
+      "eval_steps_per_second": 0.357,
       "step": 100
     },
     {
       "epoch": 0.1151229722658294,
-      "grad_norm": 18.15644072773089,
+      "grad_norm": 47.336977780094564,
       "learning_rate": 4.996723692767926e-07,
-      "logits/chosen": -2.5023550987243652,
-      "logits/rejected": -2.432821273803711,
-      "logps/chosen": -249.1396484375,
-      "logps/rejected": -264.470458984375,
-      "loss": 0.6297,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -0.07685011625289917,
-      "rewards/margins": 0.19513027369976044,
-      "rewards/rejected": -0.2719804048538208,
+      "logits/chosen": -2.0436112880706787,
+      "logits/rejected": -1.9534924030303955,
+      "logps/chosen": -310.6973571777344,
+      "logps/rejected": -324.1681823730469,
+      "loss": 3.758,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -0.6924275755882263,
+      "rewards/margins": 0.17653007805347443,
+      "rewards/rejected": -0.8689576387405396,
       "step": 110
     },
     {
       "epoch": 0.12558869701726844,
-      "grad_norm": 22.023309150811986,
+      "grad_norm": 109.43376131471078,
       "learning_rate": 4.990375746213598e-07,
-      "logits/chosen": -2.3616247177124023,
-      "logits/rejected": -2.281648874282837,
-      "logps/chosen": -285.5514831542969,
-      "logps/rejected": -339.8910217285156,
-      "loss": 0.5961,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -0.2297312319278717,
-      "rewards/margins": 0.2847149968147278,
-      "rewards/rejected": -0.5144462585449219,
+      "logits/chosen": -0.08515436947345734,
+      "logits/rejected": 0.34949326515197754,
+      "logps/chosen": -343.26495361328125,
+      "logps/rejected": -412.98577880859375,
+      "loss": 4.0333,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": -0.8068662881851196,
+      "rewards/margins": 0.438527911901474,
+      "rewards/rejected": -1.2453943490982056,
       "step": 120
     },
     {
       "epoch": 0.1360544217687075,
-      "grad_norm": 19.678835809215887,
+      "grad_norm": 95.04671304091885,
       "learning_rate": 4.980697142834314e-07,
-      "logits/chosen": -2.283674478530884,
-      "logits/rejected": -2.192523956298828,
-      "logps/chosen": -294.853515625,
-      "logps/rejected": -303.66986083984375,
-      "loss": 0.5883,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -0.1358092576265335,
-      "rewards/margins": 0.3562476634979248,
-      "rewards/rejected": -0.4920569360256195,
+      "logits/chosen": 0.396954745054245,
+      "logits/rejected": 1.0232269763946533,
+      "logps/chosen": -406.28521728515625,
+      "logps/rejected": -430.10760498046875,
+      "loss": 4.2005,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -1.2501262426376343,
+      "rewards/margins": 0.5063079595565796,
+      "rewards/rejected": -1.7564342021942139,
       "step": 130
     },
     {
       "epoch": 0.14652014652014653,
-      "grad_norm": 18.045058125702276,
+      "grad_norm": 144.39035434160894,
       "learning_rate": 4.967700826904229e-07,
-      "logits/chosen": -2.1272222995758057,
-      "logits/rejected": -2.050865888595581,
-      "logps/chosen": -321.29254150390625,
-      "logps/rejected": -360.5655517578125,
-      "loss": 0.5733,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -0.3020511567592621,
-      "rewards/margins": 0.41385501623153687,
-      "rewards/rejected": -0.7159062027931213,
+      "logits/chosen": -0.1560676395893097,
+      "logits/rejected": 0.6105406880378723,
+      "logps/chosen": -416.2538146972656,
+      "logps/rejected": -463.2472229003906,
+      "loss": 3.7876,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -1.2516638040542603,
+      "rewards/margins": 0.49105915427207947,
+      "rewards/rejected": -1.7427231073379517,
       "step": 140
     },
     {
       "epoch": 0.15698587127158556,
-      "grad_norm": 23.733353319304864,
+      "grad_norm": 125.21681673589694,
       "learning_rate": 4.951404179843962e-07,
-      "logits/chosen": -1.8601700067520142,
-      "logits/rejected": -1.6911884546279907,
-      "logps/chosen": -328.0176696777344,
-      "logps/rejected": -331.93365478515625,
-      "loss": 0.5724,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.40121254324913025,
-      "rewards/margins": 0.44814401865005493,
-      "rewards/rejected": -0.849356472492218,
+      "logits/chosen": 2.0407581329345703,
+      "logits/rejected": 2.8481547832489014,
+      "logps/chosen": -510.521484375,
+      "logps/rejected": -534.6341552734375,
+      "loss": 3.898,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -2.226250648498535,
+      "rewards/margins": 0.6501102447509766,
+      "rewards/rejected": -2.876361131668091,
       "step": 150
     },
     {
       "epoch": 0.1674515960230246,
-      "grad_norm": 28.013542721553446,
+      "grad_norm": 66.88313091855639,
       "learning_rate": 4.931828996974498e-07,
-      "logits/chosen": -0.7096320390701294,
-      "logits/rejected": -0.25428909063339233,
-      "logps/chosen": -341.81439208984375,
-      "logps/rejected": -363.79217529296875,
-      "loss": 0.5414,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -0.46001678705215454,
-      "rewards/margins": 0.6181891560554504,
-      "rewards/rejected": -1.0782058238983154,
+      "logits/chosen": 2.163175106048584,
+      "logits/rejected": 3.5420451164245605,
+      "logps/chosen": -585.4688720703125,
+      "logps/rejected": -635.2697143554688,
+      "loss": 3.9393,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -2.89656138420105,
+      "rewards/margins": 0.8964195251464844,
+      "rewards/rejected": -3.792980909347534,
       "step": 160
     },
     {
       "epoch": 0.17791732077446362,
-      "grad_norm": 23.342052779259543,
+      "grad_norm": 188.98325062900707,
       "learning_rate": 4.909001458367866e-07,
-      "logits/chosen": -0.7341230511665344,
-      "logits/rejected": -0.3530420660972595,
-      "logps/chosen": -352.09222412109375,
-      "logps/rejected": -376.5685119628906,
-      "loss": 0.5581,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.7362300753593445,
-      "rewards/margins": 0.6209839582443237,
-      "rewards/rejected": -1.3572142124176025,
+      "logits/chosen": 0.49319368600845337,
+      "logits/rejected": 1.3766599893569946,
+      "logps/chosen": -599.5331420898438,
+      "logps/rejected": -654.1383056640625,
+      "loss": 3.9922,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -3.210639476776123,
+      "rewards/margins": 0.922272801399231,
+      "rewards/rejected": -4.132911682128906,
       "step": 170
     },
     {
       "epoch": 0.18838304552590268,
-      "grad_norm": 28.21709060740068,
+      "grad_norm": 320.6202106283321,
       "learning_rate": 4.882952093833627e-07,
-      "logits/chosen": -0.876774787902832,
-      "logits/rejected": -0.6448796391487122,
-      "logps/chosen": -335.0474548339844,
-      "logps/rejected": -404.52728271484375,
-      "loss": 0.5274,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -0.9509947896003723,
-      "rewards/margins": 0.6264718174934387,
-      "rewards/rejected": -1.5774667263031006,
+      "logits/chosen": 0.6820823550224304,
+      "logits/rejected": 1.588409185409546,
+      "logps/chosen": -1040.5491943359375,
+      "logps/rejected": -1233.1207275390625,
+      "loss": 3.36,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -8.006011009216309,
+      "rewards/margins": 1.8573882579803467,
+      "rewards/rejected": -9.86340045928955,
       "step": 180
     },
     {
       "epoch": 0.1988487702773417,
-      "grad_norm": 40.43719634141344,
+      "grad_norm": 157.79546381015746,
       "learning_rate": 4.853715742087946e-07,
-      "logits/chosen": -0.12615634500980377,
-      "logits/rejected": 0.2832840085029602,
-      "logps/chosen": -344.6008605957031,
-      "logps/rejected": -427.69354248046875,
-      "loss": 0.5309,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.0862140655517578,
-      "rewards/margins": 0.6919594407081604,
-      "rewards/rejected": -1.7781736850738525,
+      "logits/chosen": 3.3087031841278076,
+      "logits/rejected": 4.11985445022583,
+      "logps/chosen": -1690.8167724609375,
+      "logps/rejected": -1890.6634521484375,
+      "loss": 2.6799,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -14.548372268676758,
+      "rewards/margins": 1.8595011234283447,
+      "rewards/rejected": -16.407875061035156,
       "step": 190
     },
     {
       "epoch": 0.20931449502878074,
-      "grad_norm": 24.48228039656553,
+      "grad_norm": 178.97245767319544,
       "learning_rate": 4.821331504159906e-07,
-      "logits/chosen": -1.134807825088501,
-      "logits/rejected": -0.7409021258354187,
-      "logps/chosen": -399.9931640625,
-      "logps/rejected": -406.24249267578125,
-      "loss": 0.578,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.893362820148468,
-      "rewards/margins": 0.5979009866714478,
-      "rewards/rejected": -1.49126398563385,
+      "logits/chosen": 0.3337511122226715,
+      "logits/rejected": 1.9961885213851929,
+      "logps/chosen": -1578.712158203125,
+      "logps/rejected": -1801.65625,
+      "loss": 2.6464,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -12.680551528930664,
+      "rewards/margins": 2.764849901199341,
+      "rewards/rejected": -15.445402145385742,
       "step": 200
     },
     {
       "epoch": 0.20931449502878074,
-      "eval_logits/chosen": -1.7000207901000977,
-      "eval_logits/rejected": -1.3956148624420166,
-      "eval_logps/chosen": -357.6234130859375,
-      "eval_logps/rejected": -397.53143310546875,
-      "eval_loss": 0.5543879270553589,
-      "eval_rewards/accuracies": 0.7480158805847168,
-      "eval_rewards/chosen": -0.7565548419952393,
-      "eval_rewards/margins": 0.616691529750824,
-      "eval_rewards/rejected": -1.3732463121414185,
-      "eval_runtime": 175.6227,
-      "eval_samples_per_second": 11.388,
-      "eval_steps_per_second": 0.359,
+      "eval_logits/chosen": -0.35622134804725647,
+      "eval_logits/rejected": 0.6981890797615051,
+      "eval_logps/chosen": -1244.43603515625,
+      "eval_logps/rejected": -1423.3580322265625,
+      "eval_loss": 2.606262683868408,
+      "eval_rewards/accuracies": 0.625,
+      "eval_rewards/chosen": -9.624680519104004,
+      "eval_rewards/margins": 2.0068302154541016,
+      "eval_rewards/rejected": -11.631510734558105,
+      "eval_runtime": 177.3795,
+      "eval_samples_per_second": 11.275,
+      "eval_steps_per_second": 0.355,
       "step": 200
     },
     {
       "epoch": 0.21978021978021978,
-      "grad_norm": 33.0816384461788,
+      "grad_norm": 221.39959720400535,
       "learning_rate": 4.785842691097342e-07,
-      "logits/chosen": -1.2056543827056885,
-      "logits/rejected": -0.7488024830818176,
-      "logps/chosen": -379.0580139160156,
-      "logps/rejected": -430.2911682128906,
-      "loss": 0.5639,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -0.9624747037887573,
-      "rewards/margins": 0.6178582906723022,
-      "rewards/rejected": -1.5803329944610596,
+      "logits/chosen": 0.43124809861183167,
+      "logits/rejected": 1.6196168661117554,
+      "logps/chosen": -1394.329345703125,
+      "logps/rejected": -1612.8701171875,
+      "loss": 2.2192,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -11.115188598632812,
+      "rewards/margins": 2.29093337059021,
+      "rewards/rejected": -13.406122207641602,
       "step": 210
     },
     {
       "epoch": 0.2302459445316588,
-      "grad_norm": 33.16914134802434,
+      "grad_norm": 107.97254065213261,
       "learning_rate": 4.7472967660421603e-07,
-      "logits/chosen": -0.9634332656860352,
-      "logits/rejected": -0.056523989886045456,
-      "logps/chosen": -388.16888427734375,
-      "logps/rejected": -407.8970642089844,
-      "loss": 0.5222,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.0096079111099243,
-      "rewards/margins": 0.7075966000556946,
-      "rewards/rejected": -1.7172044515609741,
+      "logits/chosen": 0.5400440096855164,
+      "logits/rejected": 1.9760030508041382,
+      "logps/chosen": -1507.001220703125,
+      "logps/rejected": -1713.616455078125,
+      "loss": 2.018,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -12.197932243347168,
+      "rewards/margins": 2.5764663219451904,
+      "rewards/rejected": -14.774396896362305,
       "step": 220
     },
     {
       "epoch": 0.24071166928309787,
-      "grad_norm": 35.571946638008065,
+      "grad_norm": 217.88193736039008,
       "learning_rate": 4.705745280752585e-07,
-      "logits/chosen": -0.3167392611503601,
-      "logits/rejected": 0.18852970004081726,
-      "logps/chosen": -397.49847412109375,
-      "logps/rejected": -440.3253479003906,
-      "loss": 0.5468,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -1.0877156257629395,
-      "rewards/margins": 0.6301809549331665,
-      "rewards/rejected": -1.7178964614868164,
+      "logits/chosen": 1.4225207567214966,
+      "logits/rejected": 2.4756038188934326,
+      "logps/chosen": -1726.320068359375,
+      "logps/rejected": -2005.7041015625,
+      "loss": 1.9719,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -14.375930786132812,
+      "rewards/margins": 2.995753526687622,
+      "rewards/rejected": -17.37168312072754,
       "step": 230
     },
     {
       "epoch": 0.25117739403453687,
-      "grad_norm": 32.242793295380665,
+      "grad_norm": 109.77258728949327,
       "learning_rate": 4.6612438066572555e-07,
-      "logits/chosen": -0.004159653093665838,
-      "logits/rejected": 0.8073694109916687,
-      "logps/chosen": -364.86566162109375,
-      "logps/rejected": -433.24072265625,
-      "loss": 0.5162,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -0.9729146957397461,
-      "rewards/margins": 0.8177644610404968,
-      "rewards/rejected": -1.7906792163848877,
+      "logits/chosen": 2.2113587856292725,
+      "logits/rejected": 3.125591993331909,
+      "logps/chosen": -1894.770751953125,
+      "logps/rejected": -2110.86376953125,
+      "loss": 1.9847,
+      "rewards/accuracies": 0.59375,
+      "rewards/chosen": -16.27196502685547,
+      "rewards/margins": 2.294943332672119,
+      "rewards/rejected": -18.56690788269043,
       "step": 240
     },
     {
       "epoch": 0.2616431187859759,
-      "grad_norm": 29.7046569499327,
+      "grad_norm": 276.53415343052893,
       "learning_rate": 4.6138518605333664e-07,
-      "logits/chosen": -0.3281826674938202,
-      "logits/rejected": 0.29218000173568726,
-      "logps/chosen": -366.93865966796875,
-      "logps/rejected": -417.32666015625,
-      "loss": 0.5058,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.8915438652038574,
-      "rewards/margins": 0.7059476375579834,
-      "rewards/rejected": -1.5974915027618408,
+      "logits/chosen": 1.203977108001709,
+      "logits/rejected": 1.9225616455078125,
+      "logps/chosen": -1561.0047607421875,
+      "logps/rejected": -1763.075439453125,
+      "loss": 2.257,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -12.832204818725586,
+      "rewards/margins": 2.222775936126709,
+      "rewards/rejected": -15.05497932434082,
       "step": 250
     },
     {
       "epoch": 0.272108843537415,
-      "grad_norm": 26.07189300308382,
+      "grad_norm": 159.14963627253198,
       "learning_rate": 4.5636328249082514e-07,
-      "logits/chosen": -0.7655261158943176,
-      "logits/rejected": 0.05788875371217728,
-      "logps/chosen": -377.3932800292969,
-      "logps/rejected": -431.24603271484375,
-      "loss": 0.5333,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -0.8842679262161255,
-      "rewards/margins": 0.6056793928146362,
-      "rewards/rejected": -1.4899474382400513,
+      "logits/chosen": 1.134037733078003,
+      "logits/rejected": 2.1568219661712646,
+      "logps/chosen": -1608.8623046875,
+      "logps/rejected": -1763.599853515625,
+      "loss": 2.2606,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -13.198956489562988,
+      "rewards/margins": 1.6145280599594116,
+      "rewards/rejected": -14.813486099243164,
       "step": 260
     },
     {
       "epoch": 0.282574568288854,
-      "grad_norm": 27.48585519105196,
+      "grad_norm": 199.45417630865836,
       "learning_rate": 4.510653863290871e-07,
-      "logits/chosen": -0.8150612711906433,
-      "logits/rejected": -0.19155649840831757,
-      "logps/chosen": -376.1461486816406,
-      "logps/rejected": -399.7369689941406,
-      "loss": 0.5074,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -0.7649620175361633,
-      "rewards/margins": 0.6968373656272888,
-      "rewards/rejected": -1.4617992639541626,
+      "logits/chosen": 0.3547247350215912,
+      "logits/rejected": 1.2751286029815674,
+      "logps/chosen": -1781.0726318359375,
+      "logps/rejected": -2089.05615234375,
+      "loss": 1.7211,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -14.814226150512695,
+      "rewards/margins": 3.540767192840576,
+      "rewards/rejected": -18.354991912841797,
       "step": 270
     },
     {
       "epoch": 0.29304029304029305,
-      "grad_norm": 28.38677400170819,
+      "grad_norm": 162.5497817330968,
       "learning_rate": 4.4549858303465737e-07,
-      "logits/chosen": 0.3504285514354706,
-      "logits/rejected": 0.9512192606925964,
-      "logps/chosen": -415.9444274902344,
-      "logps/rejected": -479.0673828125,
-      "loss": 0.5416,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -1.1970031261444092,
-      "rewards/margins": 0.7609006762504578,
-      "rewards/rejected": -1.9579038619995117,
+      "logits/chosen": 0.21130748093128204,
+      "logits/rejected": 1.2269564867019653,
+      "logps/chosen": -1743.0787353515625,
+      "logps/rejected": -2033.669921875,
+      "loss": 1.9445,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -14.468345642089844,
+      "rewards/margins": 3.0355849266052246,
+      "rewards/rejected": -17.50392723083496,
       "step": 280
     },
     {
       "epoch": 0.3035060177917321,
-      "grad_norm": 32.744553637286586,
+      "grad_norm": 307.15808847538113,
       "learning_rate": 4.396703177135261e-07,
-      "logits/chosen": -0.09853874146938324,
-      "logits/rejected": 0.8018854260444641,
-      "logps/chosen": -399.70263671875,
-      "logps/rejected": -441.5960388183594,
-      "loss": 0.5186,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -1.269182801246643,
-      "rewards/margins": 0.7766303420066833,
-      "rewards/rejected": -2.0458133220672607,
+      "logits/chosen": 0.7419403791427612,
+      "logits/rejected": 1.9202260971069336,
+      "logps/chosen": -1948.1787109375,
+      "logps/rejected": -2273.5205078125,
+      "loss": 1.9864,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -16.753948211669922,
+      "rewards/margins": 3.6111111640930176,
+      "rewards/rejected": -20.36505699157715,
       "step": 290
     },
     {
       "epoch": 0.3139717425431711,
-      "grad_norm": 36.30101198384572,
+      "grad_norm": 90.00202577382801,
       "learning_rate": 4.335883851539693e-07,
-      "logits/chosen": -1.0064457654953003,
-      "logits/rejected": -0.4348925054073334,
-      "logps/chosen": -373.8060302734375,
-      "logps/rejected": -438.51513671875,
-      "loss": 0.5301,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -1.0478389263153076,
-      "rewards/margins": 0.6600475907325745,
-      "rewards/rejected": -1.7078864574432373,
+      "logits/chosen": 0.30849236249923706,
+      "logits/rejected": 1.1072229146957397,
+      "logps/chosen": -1431.3275146484375,
+      "logps/rejected": -1653.4029541015625,
+      "loss": 1.9069,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -11.623054504394531,
+      "rewards/margins": 2.2337088584899902,
+      "rewards/rejected": -13.856762886047363,
       "step": 300
     },
     {
       "epoch": 0.3139717425431711,
-      "eval_logits/chosen": -0.7774365544319153,
-      "eval_logits/rejected": -0.07988195866346359,
-      "eval_logps/chosen": -377.41851806640625,
-      "eval_logps/rejected": -439.006103515625,
-      "eval_loss": 0.519775927066803,
-      "eval_rewards/accuracies": 0.7757936716079712,
-      "eval_rewards/chosen": -0.9545055627822876,
-      "eval_rewards/margins": 0.8334872126579285,
-      "eval_rewards/rejected": -1.7879927158355713,
-      "eval_runtime": 175.6202,
-      "eval_samples_per_second": 11.388,
-      "eval_steps_per_second": 0.359,
+      "eval_logits/chosen": 0.45899611711502075,
+      "eval_logits/rejected": 1.5569082498550415,
+      "eval_logps/chosen": -1266.6490478515625,
+      "eval_logps/rejected": -1452.7674560546875,
+      "eval_loss": 2.262396812438965,
+      "eval_rewards/accuracies": 0.6329365372657776,
+      "eval_rewards/chosen": -9.846811294555664,
+      "eval_rewards/margins": 2.0787949562072754,
+      "eval_rewards/rejected": -11.925606727600098,
+      "eval_runtime": 176.5188,
+      "eval_samples_per_second": 11.33,
+      "eval_steps_per_second": 0.357,
       "step": 300
     },
     {
       "epoch": 0.32443746729461015,
-      "grad_norm": 34.23825403742458,
+      "grad_norm": 177.8700388917398,
       "learning_rate": 4.272609194017105e-07,
-      "logits/chosen": -0.798404335975647,
-      "logits/rejected": 0.40799885988235474,
-      "logps/chosen": -376.5071716308594,
-      "logps/rejected": -409.09600830078125,
-      "loss": 0.5113,
+      "logits/chosen": 0.647371768951416,
+      "logits/rejected": 2.9104599952697754,
+      "logps/chosen": -1395.496826171875,
+      "logps/rejected": -1711.9573974609375,
+      "loss": 2.3095,
       "rewards/accuracies": 0.75,
-      "rewards/chosen": -0.9274541735649109,
-      "rewards/margins": 0.8360295295715332,
-      "rewards/rejected": -1.7634836435317993,
+      "rewards/chosen": -11.117349624633789,
+      "rewards/margins": 3.674748182296753,
+      "rewards/rejected": -14.792098999023438,
       "step": 310
     },
     {
       "epoch": 0.3349031920460492,
-      "grad_norm": 39.68212351207402,
+      "grad_norm": 180.92515200199898,
       "learning_rate": 4.2069638288135547e-07,
-      "logits/chosen": 0.02141362428665161,
-      "logits/rejected": 0.8004009127616882,
-      "logps/chosen": -371.86358642578125,
-      "logps/rejected": -460.26971435546875,
-      "loss": 0.5256,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -1.1884684562683105,
-      "rewards/margins": 0.9082374572753906,
-      "rewards/rejected": -2.096705913543701,
+      "logits/chosen": 0.9543863534927368,
+      "logits/rejected": 1.7447538375854492,
+      "logps/chosen": -1926.299560546875,
+      "logps/rejected": -2217.88037109375,
+      "loss": 2.1724,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -16.73282814025879,
+      "rewards/margins": 2.939984083175659,
+      "rewards/rejected": -19.672813415527344,
       "step": 320
     },
     {
       "epoch": 0.3453689167974882,
-      "grad_norm": 37.95982237233792,
+      "grad_norm": 145.6894284610869,
       "learning_rate": 4.139035550786494e-07,
-      "logits/chosen": 0.24720951914787292,
-      "logits/rejected": 0.8929317593574524,
-      "logps/chosen": -380.25689697265625,
-      "logps/rejected": -435.5292053222656,
-      "loss": 0.5049,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -1.2391283512115479,
-      "rewards/margins": 0.795093297958374,
-      "rewards/rejected": -2.034221649169922,
+      "logits/chosen": -0.039321091026067734,
+      "logits/rejected": 0.5018073320388794,
+      "logps/chosen": -1734.091796875,
+      "logps/rejected": -1908.339111328125,
+      "loss": 1.716,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -14.77747917175293,
+      "rewards/margins": 1.9848415851593018,
+      "rewards/rejected": -16.76232147216797,
       "step": 330
     },
     {
       "epoch": 0.35583464154892724,
-      "grad_norm": 31.44829466131635,
+      "grad_norm": 183.78050890033984,
       "learning_rate": 4.0689152079869306e-07,
-      "logits/chosen": 0.19869890809059143,
-      "logits/rejected": 0.7656279802322388,
-      "logps/chosen": -384.75897216796875,
-      "logps/rejected": -426.62274169921875,
-      "loss": 0.5341,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -1.1527179479599,
-      "rewards/margins": 0.7536069750785828,
-      "rewards/rejected": -1.9063247442245483,
+      "logits/chosen": -0.5724295377731323,
+      "logits/rejected": 0.023262571543455124,
+      "logps/chosen": -1660.732177734375,
+      "logps/rejected": -1876.1025390625,
+      "loss": 1.8439,
+      "rewards/accuracies": 0.606249988079071,
+      "rewards/chosen": -13.912447929382324,
+      "rewards/margins": 2.488671064376831,
+      "rewards/rejected": -16.401119232177734,
       "step": 340
     },
     {
       "epoch": 0.3663003663003663,
-      "grad_norm": 33.60100508993655,
+      "grad_norm": 149.28700648360655,
       "learning_rate": 3.99669658015821e-07,
-      "logits/chosen": -0.23042722046375275,
-      "logits/rejected": 0.44881439208984375,
-      "logps/chosen": -416.906982421875,
-      "logps/rejected": -460.3355407714844,
-      "loss": 0.502,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -1.2567191123962402,
-      "rewards/margins": 0.7857658863067627,
-      "rewards/rejected": -2.042484998703003,
+      "logits/chosen": 0.006322336383163929,
+      "logits/rejected": 0.6332755088806152,
+      "logps/chosen": -1966.5765380859375,
+      "logps/rejected": -2201.843505859375,
+      "loss": 1.6671,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -16.753414154052734,
+      "rewards/margins": 2.7041499614715576,
+      "rewards/rejected": -19.457565307617188,
       "step": 350
     },
     {
       "epoch": 0.37676609105180536,
-      "grad_norm": 35.86790303415799,
+      "grad_norm": 237.65668361495474,
       "learning_rate": 3.92247625331392e-07,
-      "logits/chosen": -0.2522183060646057,
-      "logits/rejected": 0.6683695316314697,
-      "logps/chosen": -416.01641845703125,
-      "logps/rejected": -479.52294921875,
-      "loss": 0.4805,
-      "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -1.3013213872909546,
-      "rewards/margins": 0.8662745356559753,
-      "rewards/rejected": -2.1675961017608643,
+      "logits/chosen": -0.21500203013420105,
+      "logits/rejected": 0.6255682110786438,
+      "logps/chosen": -1989.7509765625,
+      "logps/rejected": -2207.83349609375,
+      "loss": 1.6927,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -17.038667678833008,
+      "rewards/margins": 2.4120330810546875,
+      "rewards/rejected": -19.450698852539062,
       "step": 360
     },
     {
       "epoch": 0.3872318158032444,
-      "grad_norm": 34.95835364160841,
+      "grad_norm": 152.55773033990448,
       "learning_rate": 3.846353490562664e-07,
-      "logits/chosen": 0.0829664096236229,
-      "logits/rejected": 0.9096466898918152,
-      "logps/chosen": -406.54913330078125,
-      "logps/rejected": -456.1133728027344,
-      "loss": 0.5136,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -1.4327296018600464,
-      "rewards/margins": 0.7719124555587769,
-      "rewards/rejected": -2.2046420574188232,
+      "logits/chosen": -0.39199286699295044,
+      "logits/rejected": -0.043508779257535934,
+      "logps/chosen": -1889.5286865234375,
+      "logps/rejected": -2139.589111328125,
+      "loss": 1.7098,
+      "rewards/accuracies": 0.59375,
+      "rewards/chosen": -16.262523651123047,
+      "rewards/margins": 2.7768733501434326,
+      "rewards/rejected": -19.039398193359375,
       "step": 370
     },
     {
       "epoch": 0.3976975405546834,
-      "grad_norm": 33.96402149473364,
+      "grad_norm": 239.86422108427834,
       "learning_rate": 3.768430099352445e-07,
-      "logits/chosen": -0.14455661177635193,
-      "logits/rejected": 0.8097721934318542,
-      "logps/chosen": -388.0363464355469,
-      "logps/rejected": -460.75048828125,
-      "loss": 0.5138,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -1.333893060684204,
-      "rewards/margins": 0.8102533221244812,
-      "rewards/rejected": -2.14414644241333,
+      "logits/chosen": -0.5338395833969116,
+      "logits/rejected": -0.10323655605316162,
+      "logps/chosen": -1830.7080078125,
+      "logps/rejected": -2104.773681640625,
+      "loss": 1.786,
+      "rewards/accuracies": 0.6187499761581421,
+      "rewards/chosen": -15.76060962677002,
+      "rewards/margins": 2.8237688541412354,
+      "rewards/rejected": -18.58437728881836,
       "step": 380
     },
     {
       "epoch": 0.40816326530612246,
-      "grad_norm": 32.427509955053814,
+      "grad_norm": 137.89263121746114,
       "learning_rate": 3.6888102953122304e-07,
-      "logits/chosen": 0.7358514070510864,
-      "logits/rejected": 1.5359573364257812,
-      "logps/chosen": -425.21014404296875,
-      "logps/rejected": -492.5250549316406,
-      "loss": 0.5231,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -1.4359853267669678,
-      "rewards/margins": 0.8055831789970398,
-      "rewards/rejected": -2.2415683269500732,
+      "logits/chosen": -0.3421451449394226,
+      "logits/rejected": 0.2877078056335449,
+      "logps/chosen": -1774.384765625,
+      "logps/rejected": -2007.7366943359375,
+      "loss": 1.9274,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -14.92773151397705,
+      "rewards/margins": 2.465951442718506,
+      "rewards/rejected": -17.393680572509766,
       "step": 390
     },
     {
       "epoch": 0.4186289900575615,
-      "grad_norm": 36.13318678339082,
+      "grad_norm": 164.86784545063486,
       "learning_rate": 3.607600562872785e-07,
-      "logits/chosen": 0.8064459562301636,
-      "logits/rejected": 1.430558681488037,
-      "logps/chosen": -414.16839599609375,
-      "logps/rejected": -482.3439025878906,
-      "loss": 0.5386,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -1.4417721033096313,
-      "rewards/margins": 0.7046888470649719,
-      "rewards/rejected": -2.146461009979248,
+      "logits/chosen": -0.7335325479507446,
+      "logits/rejected": -0.33919858932495117,
+      "logps/chosen": -1733.375244140625,
+      "logps/rejected": -1963.0279541015625,
+      "loss": 1.6642,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -14.63383960723877,
+      "rewards/margins": 2.319460391998291,
+      "rewards/rejected": -16.95330047607422,
       "step": 400
     },
     {
       "epoch": 0.4186289900575615,
-      "eval_logits/chosen": 0.024975202977657318,
-      "eval_logits/rejected": 0.9406751394271851,
-      "eval_logps/chosen": -412.3045654296875,
-      "eval_logps/rejected": -471.00274658203125,
-      "eval_loss": 0.510511040687561,
-      "eval_rewards/accuracies": 0.77182537317276,
-      "eval_rewards/chosen": -1.3033660650253296,
-      "eval_rewards/margins": 0.8045932054519653,
-      "eval_rewards/rejected": -2.107959032058716,
-      "eval_runtime": 177.0547,
-      "eval_samples_per_second": 11.296,
-      "eval_steps_per_second": 0.356,
+      "eval_logits/chosen": -0.7751028537750244,
+      "eval_logits/rejected": -0.08748837560415268,
+      "eval_logps/chosen": -1731.152587890625,
+      "eval_logps/rejected": -2045.1492919921875,
+      "eval_loss": 1.6421091556549072,
+      "eval_rewards/accuracies": 0.625,
+      "eval_rewards/chosen": -14.491846084594727,
+      "eval_rewards/margins": 3.3575782775878906,
+      "eval_rewards/rejected": -17.849422454833984,
+      "eval_runtime": 176.0651,
+      "eval_samples_per_second": 11.359,
+      "eval_steps_per_second": 0.358,
       "step": 400
     },
     {
       "epoch": 0.4290947148090005,
-      "grad_norm": 26.612881775481124,
+      "grad_norm": 128.91689311765836,
       "learning_rate": 3.5249095128531856e-07,
-      "logits/chosen": 0.2300519496202469,
-      "logits/rejected": 0.9924400448799133,
-      "logps/chosen": -398.78729248046875,
-      "logps/rejected": -436.8857421875,
-      "loss": 0.5301,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -1.270517110824585,
-      "rewards/margins": 0.6298418045043945,
-      "rewards/rejected": -1.9003589153289795,
+      "logits/chosen": -0.10633065551519394,
+      "logits/rejected": 0.350477933883667,
+      "logps/chosen": -1862.1099853515625,
+      "logps/rejected": -2067.15673828125,
+      "loss": 1.7556,
+      "rewards/accuracies": 0.59375,
+      "rewards/chosen": -15.903741836547852,
+      "rewards/margins": 2.2993245124816895,
+      "rewards/rejected": -18.203065872192383,
       "step": 410
     },
     {
       "epoch": 0.43956043956043955,
-      "grad_norm": 39.714482709669014,
+      "grad_norm": 187.2282869549343,
       "learning_rate": 3.4408477372034736e-07,
-      "logits/chosen": -0.2808998227119446,
-      "logits/rejected": 0.9214981198310852,
-      "logps/chosen": -379.78924560546875,
-      "logps/rejected": -428.7146911621094,
-      "loss": 0.5265,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -0.9761778116226196,
-      "rewards/margins": 0.7907910346984863,
-      "rewards/rejected": -1.7669687271118164,
+      "logits/chosen": -0.2209610939025879,
+      "logits/rejected": 0.7663095593452454,
+      "logps/chosen": -1825.959228515625,
+      "logps/rejected": -2182.580810546875,
+      "loss": 1.8542,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -15.437875747680664,
+      "rewards/margins": 3.867755174636841,
+      "rewards/rejected": -19.30562973022461,
       "step": 420
     },
     {
       "epoch": 0.4500261643118786,
-      "grad_norm": 36.7812247576203,
+      "grad_norm": 150.13979068919696,
       "learning_rate": 3.3555276610977276e-07,
-      "logits/chosen": 1.0532963275909424,
-      "logits/rejected": 1.682213544845581,
-      "logps/chosen": -372.61004638671875,
-      "logps/rejected": -455.31524658203125,
-      "loss": 0.507,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -1.18856680393219,
-      "rewards/margins": 0.8627974390983582,
-      "rewards/rejected": -2.0513641834259033,
+      "logits/chosen": -1.128701090812683,
+      "logits/rejected": -0.5558885335922241,
+      "logps/chosen": -1832.6103515625,
+      "logps/rejected": -2176.197265625,
+      "loss": 1.5079,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -15.788568496704102,
+      "rewards/margins": 3.47161602973938,
+      "rewards/rejected": -19.26018714904785,
       "step": 430
     },
     {
       "epoch": 0.4604918890633176,
-      "grad_norm": 37.84672118674808,
+      "grad_norm": 163.41066719667168,
       "learning_rate": 3.269063392575352e-07,
-      "logits/chosen": 2.015638828277588,
-      "logits/rejected": 2.8407273292541504,
-      "logps/chosen": -472.79931640625,
-      "logps/rejected": -504.94561767578125,
-      "loss": 0.5227,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -1.7762467861175537,
-      "rewards/margins": 0.7188802361488342,
-      "rewards/rejected": -2.4951272010803223,
+      "logits/chosen": -0.6949409246444702,
+      "logits/rejected": -0.05746125057339668,
+      "logps/chosen": -1597.5341796875,
+      "logps/rejected": -1821.0198974609375,
+      "loss": 1.4868,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -13.023595809936523,
+      "rewards/margins": 2.632272481918335,
+      "rewards/rejected": -15.655868530273438,
       "step": 440
     },
     {
       "epoch": 0.47095761381475665,
-      "grad_norm": 30.609934320862045,
+      "grad_norm": 133.46596474594617,
       "learning_rate": 3.1815705699316964e-07,
-      "logits/chosen": 1.3560642004013062,
-      "logits/rejected": 2.127135753631592,
-      "logps/chosen": -465.49188232421875,
-      "logps/rejected": -571.1328125,
-      "loss": 0.5099,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -1.8310394287109375,
-      "rewards/margins": 1.0971113443374634,
-      "rewards/rejected": -2.9281508922576904,
+      "logits/chosen": -0.4808398187160492,
+      "logits/rejected": 0.3264926075935364,
+      "logps/chosen": -1599.6370849609375,
+      "logps/rejected": -1936.6884765625,
+      "loss": 1.5413,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -13.172491073608398,
+      "rewards/margins": 3.4112179279327393,
+      "rewards/rejected": -16.583707809448242,
       "step": 450
     },
     {
       "epoch": 0.48142333856619574,
-      "grad_norm": 28.715258464621037,
+      "grad_norm": 155.84007478164062,
       "learning_rate": 3.0931662070620794e-07,
-      "logits/chosen": 0.8564838171005249,
-      "logits/rejected": 1.5636537075042725,
-      "logps/chosen": -452.06951904296875,
-      "logps/rejected": -517.620361328125,
-      "loss": 0.5258,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -1.9461126327514648,
-      "rewards/margins": 0.7798872590065002,
-      "rewards/rejected": -2.725999593734741,
+      "logits/chosen": -0.719369113445282,
+      "logits/rejected": -0.06152462959289551,
+      "logps/chosen": -1643.2447509765625,
+      "logps/rejected": -1872.9976806640625,
+      "loss": 1.7906,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -13.85786247253418,
+      "rewards/margins": 2.4219117164611816,
+      "rewards/rejected": -16.279773712158203,
       "step": 460
     },
     {
       "epoch": 0.49188906331763477,
-      "grad_norm": 43.25596740234111,
+      "grad_norm": 203.3322056694353,
       "learning_rate": 3.003968536966078e-07,
-      "logits/chosen": 0.4564128816127777,
-      "logits/rejected": 1.1015666723251343,
-      "logps/chosen": -456.0911560058594,
-      "logps/rejected": -495.827880859375,
-      "loss": 0.4802,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -1.7230240106582642,
-      "rewards/margins": 0.7358741164207458,
-      "rewards/rejected": -2.458897829055786,
+      "logits/chosen": -0.4609583020210266,
+      "logits/rejected": -0.09374441206455231,
+      "logps/chosen": -1654.1614990234375,
+      "logps/rejected": -1845.5618896484375,
+      "loss": 1.7718,
+      "rewards/accuracies": 0.59375,
+      "rewards/chosen": -13.703729629516602,
+      "rewards/margins": 2.2525086402893066,
+      "rewards/rejected": -15.956239700317383,
       "step": 470
     },
     {
       "epoch": 0.5023547880690737,
-      "grad_norm": 36.20202372184976,
+      "grad_norm": 156.4799546194198,
       "learning_rate": 2.9140968536213693e-07,
-      "logits/chosen": 0.7777734994888306,
-      "logits/rejected": 1.9267253875732422,
-      "logps/chosen": -460.71783447265625,
-      "logps/rejected": -554.2379150390625,
-      "loss": 0.5012,
-      "rewards/accuracies": 0.8125,
-      "rewards/chosen": -1.935050368309021,
-      "rewards/margins": 1.0943626165390015,
-      "rewards/rejected": -3.0294129848480225,
+      "logits/chosen": -0.2353781908750534,
+      "logits/rejected": 0.5946909785270691,
+      "logps/chosen": -1859.3265380859375,
+      "logps/rejected": -2325.88134765625,
+      "loss": 1.3829,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -15.921140670776367,
+      "rewards/margins": 4.824706077575684,
+      "rewards/rejected": -20.745845794677734,
       "step": 480
     },
     {
       "epoch": 0.5128205128205128,
-      "grad_norm": 32.834417417864124,
+      "grad_norm": 160.19325879757844,
       "learning_rate": 2.823671352438608e-07,
-      "logits/chosen": 0.05315951257944107,
-      "logits/rejected": 2.070735454559326,
-      "logps/chosen": -481.5361328125,
-      "logps/rejected": -537.8575439453125,
-      "loss": 0.4958,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -1.6960468292236328,
-      "rewards/margins": 1.1183526515960693,
-      "rewards/rejected": -2.814399242401123,
+      "logits/chosen": -0.9654836654663086,
+      "logits/rejected": -0.002035105135291815,
+      "logps/chosen": -1637.873291015625,
+      "logps/rejected": -2143.010986328125,
+      "loss": 1.6206,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -13.259417533874512,
+      "rewards/margins": 5.606515407562256,
+      "rewards/rejected": -18.86593246459961,
       "step": 490
     },
     {
       "epoch": 0.5232862375719518,
-      "grad_norm": 37.03566337015388,
+      "grad_norm": 221.83952267135834,
       "learning_rate": 2.73281296951072e-07,
-      "logits/chosen": 0.2395954132080078,
-      "logits/rejected": 1.218072772026062,
-      "logps/chosen": -431.42626953125,
-      "logps/rejected": -487.06756591796875,
-      "loss": 0.4996,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -1.7278209924697876,
-      "rewards/margins": 0.868577778339386,
-      "rewards/rejected": -2.5963988304138184,
+      "logits/chosen": -0.6597784161567688,
+      "logits/rejected": -0.14649493992328644,
+      "logps/chosen": -1530.5738525390625,
+      "logps/rejected": -1781.8070068359375,
+      "loss": 1.6328,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -12.7192964553833,
+      "rewards/margins": 2.8244967460632324,
+      "rewards/rejected": -15.543792724609375,
       "step": 500
     },
     {
       "epoch": 0.5232862375719518,
-      "eval_logits/chosen": 0.005896765738725662,
-      "eval_logits/rejected": 1.1533339023590088,
-      "eval_logps/chosen": -460.39996337890625,
-      "eval_logps/rejected": -534.9353637695312,
-      "eval_loss": 0.49528467655181885,
-      "eval_rewards/accuracies": 0.7678571343421936,
-      "eval_rewards/chosen": -1.7843199968338013,
-      "eval_rewards/margins": 0.9629656076431274,
-      "eval_rewards/rejected": -2.7472856044769287,
-      "eval_runtime": 175.591,
-      "eval_samples_per_second": 11.39,
-      "eval_steps_per_second": 0.359,
+      "eval_logits/chosen": -0.6590258479118347,
+      "eval_logits/rejected": -0.091790109872818,
+      "eval_logps/chosen": -1589.3370361328125,
+      "eval_logps/rejected": -1890.562255859375,
+      "eval_loss": 1.5119922161102295,
+      "eval_rewards/accuracies": 0.6388888955116272,
+      "eval_rewards/chosen": -13.073691368103027,
+      "eval_rewards/margins": 3.229863166809082,
+      "eval_rewards/rejected": -16.303556442260742,
+      "eval_runtime": 177.8158,
+      "eval_samples_per_second": 11.248,
+      "eval_steps_per_second": 0.354,
       "step": 500
     },
     {
       "epoch": 0.533751962323391,
-      "grad_norm": 31.83813721910833,
+      "grad_norm": 187.4336485549293,
       "learning_rate": 2.641643219871597e-07,
-      "logits/chosen": -0.09106893837451935,
-      "logits/rejected": 1.0047601461410522,
-      "logps/chosen": -448.24365234375,
-      "logps/rejected": -547.2485961914062,
-      "loss": 0.4759,
-      "rewards/accuracies": 0.824999988079071,
-      "rewards/chosen": -1.6430542469024658,
-      "rewards/margins": 1.1913459300994873,
-      "rewards/rejected": -2.834400177001953,
+      "logits/chosen": -0.5598984360694885,
+      "logits/rejected": -0.2727218270301819,
+      "logps/chosen": -1694.568359375,
+      "logps/rejected": -2086.98193359375,
+      "loss": 1.4069,
+      "rewards/accuracies": 0.606249988079071,
+      "rewards/chosen": -14.106300354003906,
+      "rewards/margins": 4.125433921813965,
+      "rewards/rejected": -18.231733322143555,
       "step": 510
     },
     {
       "epoch": 0.54421768707483,
-      "grad_norm": 31.094283760600373,
+      "grad_norm": 180.24950333654212,
       "learning_rate": 2.550284034980507e-07,
-      "logits/chosen": 0.5583597421646118,
-      "logits/rejected": 1.752502202987671,
-      "logps/chosen": -487.286376953125,
-      "logps/rejected": -552.1371459960938,
-      "loss": 0.5187,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -2.1521472930908203,
-      "rewards/margins": 0.8697667121887207,
-      "rewards/rejected": -3.021914005279541,
+      "logits/chosen": -0.652435302734375,
+      "logits/rejected": -0.25857192277908325,
+      "logps/chosen": -1941.6849365234375,
+      "logps/rejected": -2231.46337890625,
+      "loss": 1.6022,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -16.69613265991211,
+      "rewards/margins": 3.1190426349639893,
+      "rewards/rejected": -19.815174102783203,
       "step": 520
     },
     {
       "epoch": 0.554683411826269,
-      "grad_norm": 46.54511021652677,
+      "grad_norm": 147.71519410172087,
       "learning_rate": 2.4588575996495794e-07,
-      "logits/chosen": -0.278386652469635,
-      "logits/rejected": 0.8032172918319702,
-      "logps/chosen": -446.83624267578125,
-      "logps/rejected": -526.9205322265625,
-      "loss": 0.4996,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -1.7808927297592163,
-      "rewards/margins": 0.9380749464035034,
-      "rewards/rejected": -2.7189676761627197,
+      "logits/chosen": -0.6198351979255676,
+      "logits/rejected": -0.19036616384983063,
+      "logps/chosen": -1601.6470947265625,
+      "logps/rejected": -1820.4556884765625,
+      "loss": 1.5136,
+      "rewards/accuracies": 0.5062500238418579,
+      "rewards/chosen": -13.329002380371094,
+      "rewards/margins": 2.3253164291381836,
+      "rewards/rejected": -15.654316902160645,
       "step": 530
     },
     {
       "epoch": 0.565149136577708,
-      "grad_norm": 32.98937848769124,
+      "grad_norm": 146.6770433780799,
       "learning_rate": 2.367486188632446e-07,
-      "logits/chosen": -0.7310466170310974,
-      "logits/rejected": 0.3242616355419159,
-      "logps/chosen": -452.5191345214844,
-      "logps/rejected": -511.3818359375,
-      "loss": 0.513,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -1.6216903924942017,
-      "rewards/margins": 0.8862103223800659,
-      "rewards/rejected": -2.5079007148742676,
+      "logits/chosen": -0.7303057909011841,
+      "logits/rejected": 0.15564236044883728,
+      "logps/chosen": -1670.916015625,
+      "logps/rejected": -2011.5406494140625,
+      "loss": 1.5458,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -13.805659294128418,
+      "rewards/margins": 3.703829288482666,
+      "rewards/rejected": -17.509489059448242,
       "step": 540
     },
     {
       "epoch": 0.5756148613291471,
-      "grad_norm": 28.08745252820734,
+      "grad_norm": 206.94359776232758,
       "learning_rate": 2.276292003092593e-07,
-      "logits/chosen": -0.819686770439148,
-      "logits/rejected": 0.32188865542411804,
-      "logps/chosen": -443.1897888183594,
-      "logps/rejected": -494.9161071777344,
-      "loss": 0.526,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -1.719175100326538,
-      "rewards/margins": 0.9371669888496399,
-      "rewards/rejected": -2.656342029571533,
+      "logits/chosen": -0.22513580322265625,
+      "logits/rejected": 0.4895138740539551,
+      "logps/chosen": -1914.7532958984375,
+      "logps/rejected": -2300.30322265625,
+      "loss": 1.6801,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -16.434810638427734,
+      "rewards/margins": 4.275403022766113,
+      "rewards/rejected": -20.71021270751953,
       "step": 550
     },
     {
       "epoch": 0.5860805860805861,
-      "grad_norm": 34.81864414663528,
+      "grad_norm": 175.41735239090949,
       "learning_rate": 2.185397007170141e-07,
-      "logits/chosen": -0.32582345604896545,
-      "logits/rejected": 1.0776915550231934,
-      "logps/chosen": -462.23406982421875,
-      "logps/rejected": -525.7540893554688,
-      "loss": 0.5104,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -1.7941694259643555,
-      "rewards/margins": 0.9143456220626831,
-      "rewards/rejected": -2.708515167236328,
+      "logits/chosen": -0.1453290730714798,
+      "logits/rejected": 0.3121495842933655,
+      "logps/chosen": -1876.300537109375,
+      "logps/rejected": -2229.38134765625,
+      "loss": 1.3878,
+      "rewards/accuracies": 0.6187499761581421,
+      "rewards/chosen": -15.934832572937012,
+      "rewards/margins": 3.8099570274353027,
+      "rewards/rejected": -19.744789123535156,
       "step": 560
     },
     {
       "epoch": 0.5965463108320251,
-      "grad_norm": 32.06431987316443,
+      "grad_norm": 142.79294258337345,
       "learning_rate": 2.094922764865619e-07,
-      "logits/chosen": -0.5136072635650635,
-      "logits/rejected": 0.8532153367996216,
-      "logps/chosen": -462.86260986328125,
-      "logps/rejected": -532.7152099609375,
-      "loss": 0.4848,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -1.8336808681488037,
-      "rewards/margins": 0.8313940167427063,
-      "rewards/rejected": -2.6650748252868652,
+      "logits/chosen": -0.276650995016098,
+      "logits/rejected": 0.13945253193378448,
+      "logps/chosen": -1827.0634765625,
+      "logps/rejected": -2034.280517578125,
+      "loss": 1.4902,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -15.475687980651855,
+      "rewards/margins": 2.2050392627716064,
+      "rewards/rejected": -17.680728912353516,
       "step": 570
     },
     {
       "epoch": 0.6070120355834642,
-      "grad_norm": 38.10097117955034,
+      "grad_norm": 245.80968468908674,
       "learning_rate": 2.0049902774588797e-07,
-      "logits/chosen": -0.3451794385910034,
-      "logits/rejected": 0.9854526519775391,
-      "logps/chosen": -424.631103515625,
-      "logps/rejected": -478.0337829589844,
-      "loss": 0.5039,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -1.7129182815551758,
-      "rewards/margins": 0.9391329884529114,
-      "rewards/rejected": -2.6520514488220215,
+      "logits/chosen": -0.011815989390015602,
+      "logits/rejected": 0.42436084151268005,
+      "logps/chosen": -1794.5543212890625,
+      "logps/rejected": -2061.93310546875,
+      "loss": 1.4461,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -15.412150382995605,
+      "rewards/margins": 3.078895330429077,
+      "rewards/rejected": -18.491044998168945,
       "step": 580
     },
     {
       "epoch": 0.6174777603349032,
-      "grad_norm": 31.542114777873778,
+      "grad_norm": 175.38280547329734,
       "learning_rate": 1.9157198216806238e-07,
-      "logits/chosen": -0.5171893239021301,
-      "logits/rejected": 0.9677211046218872,
-      "logps/chosen": -454.8963928222656,
-      "logps/rejected": -536.4180908203125,
-      "loss": 0.4838,
+      "logits/chosen": -0.3044319152832031,
+      "logits/rejected": 0.3406422734260559,
+      "logps/chosen": -1649.8509521484375,
+      "logps/rejected": -2006.366455078125,
+      "loss": 1.5446,
       "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -1.7694385051727295,
-      "rewards/margins": 0.9072266817092896,
-      "rewards/rejected": -2.6766653060913086,
+      "rewards/chosen": -13.71898365020752,
+      "rewards/margins": 3.657163143157959,
+      "rewards/rejected": -17.376148223876953,
       "step": 590
     },
     {
       "epoch": 0.6279434850863422,
-      "grad_norm": 29.346076446463687,
+      "grad_norm": 203.04339818262545,
       "learning_rate": 1.8272307888529274e-07,
-      "logits/chosen": -0.03304073214530945,
-      "logits/rejected": 1.3204997777938843,
-      "logps/chosen": -449.006103515625,
-      "logps/rejected": -523.1787719726562,
-      "loss": 0.4664,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -1.6648248434066772,
-      "rewards/margins": 0.9661164283752441,
-      "rewards/rejected": -2.630941390991211,
+      "logits/chosen": 0.16477735340595245,
+      "logits/rejected": 0.6171606183052063,
+      "logps/chosen": -1870.41015625,
+      "logps/rejected": -2165.638427734375,
+      "loss": 1.6032,
+      "rewards/accuracies": 0.5687500238418579,
+      "rewards/chosen": -15.878863334655762,
+      "rewards/margins": 3.1766743659973145,
+      "rewards/rejected": -19.055538177490234,
       "step": 600
     },
     {
       "epoch": 0.6279434850863422,
-      "eval_logits/chosen": -0.26891058683395386,
-      "eval_logits/rejected": 1.055680513381958,
-      "eval_logps/chosen": -450.7947692871094,
-      "eval_logps/rejected": -530.9400634765625,
-      "eval_loss": 0.49440109729766846,
-      "eval_rewards/accuracies": 0.7698412537574768,
-      "eval_rewards/chosen": -1.6882680654525757,
-      "eval_rewards/margins": 1.01906418800354,
-      "eval_rewards/rejected": -2.7073323726654053,
-      "eval_runtime": 176.7399,
-      "eval_samples_per_second": 11.316,
-      "eval_steps_per_second": 0.356,
+      "eval_logits/chosen": 0.01903720200061798,
+      "eval_logits/rejected": 0.6402472853660583,
+      "eval_logps/chosen": -2015.7071533203125,
+      "eval_logps/rejected": -2402.58447265625,
+      "eval_loss": 1.4751698970794678,
+      "eval_rewards/accuracies": 0.6230158805847168,
+      "eval_rewards/chosen": -17.33738899230957,
+      "eval_rewards/margins": 4.086385250091553,
+      "eval_rewards/rejected": -21.42377471923828,
+      "eval_runtime": 176.4506,
+      "eval_samples_per_second": 11.335,
+      "eval_steps_per_second": 0.357,
       "step": 600
     },
     {
       "epoch": 0.6384092098377813,
-      "grad_norm": 37.553002444239766,
+      "grad_norm": 184.64896406440843,
       "learning_rate": 1.7396415252139288e-07,
-      "logits/chosen": -0.38198137283325195,
-      "logits/rejected": 1.3065571784973145,
-      "logps/chosen": -473.23046875,
-      "logps/rejected": -518.6094360351562,
-      "loss": 0.4964,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.6766010522842407,
-      "rewards/margins": 1.062415361404419,
-      "rewards/rejected": -2.73901629447937,
+      "logits/chosen": 0.0034784465096890926,
+      "logits/rejected": 0.6044633388519287,
+      "logps/chosen": -2050.113037109375,
+      "logps/rejected": -2622.564453125,
+      "loss": 1.5229,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -17.44542694091797,
+      "rewards/margins": 6.333140850067139,
+      "rewards/rejected": -23.778566360473633,
       "step": 610
     },
     {
       "epoch": 0.6488749345892203,
-      "grad_norm": 42.193220673193395,
+      "grad_norm": 150.92780625161797,
       "learning_rate": 1.6530691736402316e-07,
-      "logits/chosen": -0.21174755692481995,
-      "logits/rejected": 0.9794837236404419,
-      "logps/chosen": -429.8130798339844,
-      "logps/rejected": -492.4026794433594,
-      "loss": 0.4665,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -1.5878998041152954,
-      "rewards/margins": 0.9720605611801147,
-      "rewards/rejected": -2.5599606037139893,
+      "logits/chosen": -0.05873150750994682,
+      "logits/rejected": 0.2572210133075714,
+      "logps/chosen": -1822.690185546875,
+      "logps/rejected": -2140.002685546875,
+      "loss": 1.3047,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -15.516670227050781,
+      "rewards/margins": 3.519291400909424,
+      "rewards/rejected": -19.035961151123047,
       "step": 620
     },
     {
       "epoch": 0.6593406593406593,
-      "grad_norm": 38.97702529967974,
+      "grad_norm": 158.62413320623054,
       "learning_rate": 1.5676295169786864e-07,
-      "logits/chosen": -0.6409791707992554,
-      "logits/rejected": 0.6868582367897034,
-      "logps/chosen": -423.3990783691406,
-      "logps/rejected": -487.27117919921875,
-      "loss": 0.4718,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -1.4871389865875244,
-      "rewards/margins": 1.025777816772461,
-      "rewards/rejected": -2.5129168033599854,
+      "logits/chosen": -0.5535549521446228,
+      "logits/rejected": -0.16974008083343506,
+      "logps/chosen": -1799.411376953125,
+      "logps/rejected": -2184.095458984375,
+      "loss": 1.4004,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -15.247261047363281,
+      "rewards/margins": 4.233900547027588,
+      "rewards/rejected": -19.481159210205078,
       "step": 630
     },
     {
       "epoch": 0.6698063840920984,
-      "grad_norm": 47.68708831419142,
+      "grad_norm": 174.63990723873954,
       "learning_rate": 1.483436823197092e-07,
-      "logits/chosen": -0.141208216547966,
-      "logits/rejected": 1.2582366466522217,
-      "logps/chosen": -447.75701904296875,
-      "logps/rejected": -507.4087829589844,
-      "loss": 0.4792,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.8073606491088867,
-      "rewards/margins": 0.9799903631210327,
-      "rewards/rejected": -2.787351131439209,
+      "logits/chosen": -0.49727511405944824,
+      "logits/rejected": -0.09024439752101898,
+      "logps/chosen": -1910.181396484375,
+      "logps/rejected": -2272.175537109375,
+      "loss": 1.2582,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -16.431602478027344,
+      "rewards/margins": 4.003415107727051,
+      "rewards/rejected": -20.43501853942871,
       "step": 640
     },
     {
       "epoch": 0.6802721088435374,
-      "grad_norm": 58.635870372435406,
+      "grad_norm": 212.30897956956616,
       "learning_rate": 1.4006036925609243e-07,
-      "logits/chosen": -0.6169118881225586,
-      "logits/rejected": 0.4020913243293762,
-      "logps/chosen": -419.74786376953125,
-      "logps/rejected": -509.29840087890625,
-      "loss": 0.4833,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -1.5206490755081177,
-      "rewards/margins": 0.8552547693252563,
-      "rewards/rejected": -2.375903606414795,
+      "logits/chosen": -0.5441917777061462,
+      "logits/rejected": -0.3759006857872009,
+      "logps/chosen": -1762.1038818359375,
+      "logps/rejected": -1993.853515625,
+      "loss": 1.3183,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -14.94421100616455,
+      "rewards/margins": 2.2772457599639893,
+      "rewards/rejected": -17.221454620361328,
       "step": 650
     },
     {
       "epoch": 0.6907378335949764,
-      "grad_norm": 38.14798973487795,
+      "grad_norm": 122.40725726992933,
       "learning_rate": 1.319240907040458e-07,
-      "logits/chosen": -0.6595967411994934,
-      "logits/rejected": 0.701393723487854,
-      "logps/chosen": -430.74365234375,
-      "logps/rejected": -514.9932250976562,
-      "loss": 0.4825,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -1.485804557800293,
-      "rewards/margins": 1.0895816087722778,
-      "rewards/rejected": -2.5753860473632812,
+      "logits/chosen": -0.578727126121521,
+      "logits/rejected": -0.15290720760822296,
+      "logps/chosen": -1786.3648681640625,
+      "logps/rejected": -2103.92919921875,
+      "loss": 1.5482,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -15.042015075683594,
+      "rewards/margins": 3.422727584838867,
+      "rewards/rejected": -18.46474266052246,
       "step": 660
     },
     {
       "epoch": 0.7012035583464155,
-      "grad_norm": 45.96703369486118,
+      "grad_norm": 273.40146184819037,
       "learning_rate": 1.239457282149695e-07,
-      "logits/chosen": -0.6679301261901855,
-      "logits/rejected": 0.42252644896507263,
-      "logps/chosen": -438.337158203125,
-      "logps/rejected": -540.0033569335938,
-      "loss": 0.472,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -1.4978899955749512,
-      "rewards/margins": 1.0711971521377563,
-      "rewards/rejected": -2.569087028503418,
+      "logits/chosen": -0.6542818546295166,
+      "logits/rejected": -0.6405806541442871,
+      "logps/chosen": -1718.8697509765625,
+      "logps/rejected": -2025.167236328125,
+      "loss": 1.1528,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -14.303213119506836,
+      "rewards/margins": 3.1175124645233154,
+      "rewards/rejected": -17.420726776123047,
       "step": 670
     },
     {
       "epoch": 0.7116692830978545,
-      "grad_norm": 36.89505960468928,
+      "grad_norm": 246.28508875936285,
       "learning_rate": 1.1613595214152711e-07,
-      "logits/chosen": -0.2877568006515503,
-      "logits/rejected": 0.7360206246376038,
-      "logps/chosen": -468.3255920410156,
-      "logps/rejected": -552.3143310546875,
-      "loss": 0.5059,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.5666911602020264,
-      "rewards/margins": 1.0774924755096436,
-      "rewards/rejected": -2.64418363571167,
+      "logits/chosen": -0.6755629777908325,
+      "logits/rejected": -0.26193898916244507,
+      "logps/chosen": -1862.4000244140625,
+      "logps/rejected": -2191.969482421875,
+      "loss": 1.3671,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -15.507433891296387,
+      "rewards/margins": 3.5333023071289062,
+      "rewards/rejected": -19.04073715209961,
       "step": 680
     },
     {
       "epoch": 0.7221350078492935,
-      "grad_norm": 26.931344238941527,
+      "grad_norm": 216.14843384209277,
       "learning_rate": 1.0850520736699362e-07,
-      "logits/chosen": -0.49702662229537964,
-      "logits/rejected": 0.9417774081230164,
-      "logps/chosen": -430.6412658691406,
-      "logps/rejected": -496.180419921875,
-      "loss": 0.4527,
-      "rewards/accuracies": 0.8062499761581421,
-      "rewards/chosen": -1.5998663902282715,
-      "rewards/margins": 1.0349512100219727,
-      "rewards/rejected": -2.634817361831665,
+      "logits/chosen": -0.6002136468887329,
+      "logits/rejected": 0.03606845811009407,
+      "logps/chosen": -1838.6982421875,
+      "logps/rejected": -2214.07470703125,
+      "loss": 1.3895,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -15.680435180664062,
+      "rewards/margins": 4.133326053619385,
+      "rewards/rejected": -19.813762664794922,
       "step": 690
     },
     {
       "epoch": 0.7326007326007326,
-      "grad_norm": 34.042068496483864,
+      "grad_norm": 162.01079027631573,
       "learning_rate": 1.0106369933615042e-07,
-      "logits/chosen": -0.22161564230918884,
-      "logits/rejected": 0.7004293203353882,
-      "logps/chosen": -416.830322265625,
-      "logps/rejected": -488.65106201171875,
-      "loss": 0.4716,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -1.6273488998413086,
-      "rewards/margins": 0.8736777305603027,
-      "rewards/rejected": -2.5010266304016113,
+      "logits/chosen": -0.7846351861953735,
+      "logits/rejected": -0.5915166735649109,
+      "logps/chosen": -1752.784423828125,
+      "logps/rejected": -2021.7802734375,
+      "loss": 1.5039,
+      "rewards/accuracies": 0.581250011920929,
+      "rewards/chosen": -14.986889839172363,
+      "rewards/margins": 2.845428943634033,
+      "rewards/rejected": -17.832317352294922,
       "step": 700
     },
     {
       "epoch": 0.7326007326007326,
-      "eval_logits/chosen": -0.3817674517631531,
-      "eval_logits/rejected": 0.7091061472892761,
-      "eval_logps/chosen": -438.7652893066406,
-      "eval_logps/rejected": -520.2406005859375,
-      "eval_loss": 0.49287474155426025,
-      "eval_rewards/accuracies": 0.7777777910232544,
-      "eval_rewards/chosen": -1.5679733753204346,
-      "eval_rewards/margins": 1.032364010810852,
-      "eval_rewards/rejected": -2.600337505340576,
-      "eval_runtime": 175.3398,
-      "eval_samples_per_second": 11.406,
-      "eval_steps_per_second": 0.359,
+      "eval_logits/chosen": -0.8898468017578125,
+      "eval_logits/rejected": -0.4967605769634247,
+      "eval_logps/chosen": -1694.96240234375,
+      "eval_logps/rejected": -2016.4490966796875,
+      "eval_loss": 1.3852962255477905,
+      "eval_rewards/accuracies": 0.6527777910232544,
+      "eval_rewards/chosen": -14.129942893981934,
+      "eval_rewards/margins": 3.432478666305542,
+      "eval_rewards/rejected": -17.562421798706055,
+      "eval_runtime": 176.0679,
+      "eval_samples_per_second": 11.359,
+      "eval_steps_per_second": 0.358,
       "step": 700
     },
     {
       "epoch": 0.7430664573521716,
-      "grad_norm": 34.12335024104434,
+      "grad_norm": 177.45761000957364,
       "learning_rate": 9.382138040640714e-08,
-      "logits/chosen": -0.200923353433609,
-      "logits/rejected": 1.0540679693222046,
-      "logps/chosen": -429.7672424316406,
-      "logps/rejected": -506.0777282714844,
-      "loss": 0.5254,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -1.7129242420196533,
-      "rewards/margins": 0.9184734225273132,
-      "rewards/rejected": -2.6313977241516113,
+      "logits/chosen": -1.012629747390747,
+      "logits/rejected": -0.6268833875656128,
+      "logps/chosen": -1776.499755859375,
+      "logps/rejected": -2017.539794921875,
+      "loss": 1.5264,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -15.180249214172363,
+      "rewards/margins": 2.565770387649536,
+      "rewards/rejected": -17.74601936340332,
       "step": 710
     },
     {
       "epoch": 0.7535321821036107,
-      "grad_norm": 30.02495734311258,
+      "grad_norm": 140.94359920373847,
       "learning_rate": 8.678793653740632e-08,
-      "logits/chosen": -0.33009666204452515,
-      "logits/rejected": 0.9047529101371765,
-      "logps/chosen": -432.42010498046875,
-      "logps/rejected": -512.3065185546875,
-      "loss": 0.4648,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -1.4868285655975342,
-      "rewards/margins": 1.1614835262298584,
-      "rewards/rejected": -2.6483120918273926,
+      "logits/chosen": -0.9271895289421082,
+      "logits/rejected": -0.47789469361305237,
+      "logps/chosen": -1664.4437255859375,
+      "logps/rejected": -1977.908447265625,
+      "loss": 1.3295,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -13.8070650100708,
+      "rewards/margins": 3.4972636699676514,
+      "rewards/rejected": -17.3043270111084,
       "step": 720
     },
     {
       "epoch": 0.7639979068550498,
-      "grad_norm": 38.12612104973478,
+      "grad_norm": 190.75937551525504,
       "learning_rate": 7.997277433690983e-08,
-      "logits/chosen": -0.24449904263019562,
-      "logits/rejected": 0.697944164276123,
-      "logps/chosen": -442.3524475097656,
-      "logps/rejected": -493.40228271484375,
-      "loss": 0.4935,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.7699024677276611,
-      "rewards/margins": 0.7942432761192322,
-      "rewards/rejected": -2.564145565032959,
+      "logits/chosen": -0.8303499221801758,
+      "logits/rejected": -0.2948521077632904,
+      "logps/chosen": -1813.2340087890625,
+      "logps/rejected": -2049.240234375,
+      "loss": 1.4631,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": -15.478715896606445,
+      "rewards/margins": 2.6438088417053223,
+      "rewards/rejected": -18.12252426147461,
       "step": 730
     },
     {
       "epoch": 0.7744636316064888,
-      "grad_norm": 34.63386515718176,
+      "grad_norm": 164.74206538760382,
       "learning_rate": 7.338500848029602e-08,
-      "logits/chosen": -0.09529142081737518,
-      "logits/rejected": 0.7665776610374451,
-      "logps/chosen": -415.37127685546875,
-      "logps/rejected": -480.74981689453125,
-      "loss": 0.5094,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -1.6250483989715576,
-      "rewards/margins": 0.8600455522537231,
-      "rewards/rejected": -2.485093832015991,
+      "logits/chosen": -0.6835179924964905,
+      "logits/rejected": -0.42263850569725037,
+      "logps/chosen": -1808.6490478515625,
+      "logps/rejected": -2096.81396484375,
+      "loss": 1.2242,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -15.557826042175293,
+      "rewards/margins": 3.0879101753234863,
+      "rewards/rejected": -18.645736694335938,
       "step": 740
     },
     {
       "epoch": 0.7849293563579278,
-      "grad_norm": 38.170157629518556,
+      "grad_norm": 135.0757551116068,
       "learning_rate": 6.70334495204884e-08,
-      "logits/chosen": -0.16666540503501892,
-      "logits/rejected": 0.6466744542121887,
-      "logps/chosen": -414.00482177734375,
-      "logps/rejected": -509.15478515625,
-      "loss": 0.4725,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -1.6713943481445312,
-      "rewards/margins": 0.9836255311965942,
-      "rewards/rejected": -2.655019760131836,
+      "logits/chosen": -0.5583680868148804,
+      "logits/rejected": -0.36530551314353943,
+      "logps/chosen": -1854.912353515625,
+      "logps/rejected": -2177.262451171875,
+      "loss": 1.3344,
+      "rewards/accuracies": 0.5562499761581421,
+      "rewards/chosen": -16.08046531677246,
+      "rewards/margins": 3.2556281089782715,
+      "rewards/rejected": -19.33609390258789,
       "step": 750
     },
     {
       "epoch": 0.7953950811093669,
-      "grad_norm": 32.154322102412,
+      "grad_norm": 158.01784405358154,
       "learning_rate": 6.092659210462231e-08,
-      "logits/chosen": -0.10375770181417465,
-      "logits/rejected": 0.7752261757850647,
-      "logps/chosen": -410.65399169921875,
-      "logps/rejected": -507.827880859375,
-      "loss": 0.5116,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -1.6930710077285767,
-      "rewards/margins": 0.9550973773002625,
-      "rewards/rejected": -2.6481685638427734,
+      "logits/chosen": -0.653573215007782,
+      "logits/rejected": -0.4876467287540436,
+      "logps/chosen": -1903.880615234375,
+      "logps/rejected": -2182.48291015625,
+      "loss": 1.4038,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -16.625337600708008,
+      "rewards/margins": 2.7693800926208496,
+      "rewards/rejected": -19.394718170166016,
       "step": 760
     },
     {
       "epoch": 0.8058608058608059,
-      "grad_norm": 29.7627819143098,
+      "grad_norm": 169.97964049682443,
       "learning_rate": 5.507260361320737e-08,
-      "logits/chosen": -0.4120585322380066,
-      "logits/rejected": 0.24446937441825867,
-      "logps/chosen": -466.15069580078125,
-      "logps/rejected": -568.6295776367188,
-      "loss": 0.4641,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -1.6279999017715454,
-      "rewards/margins": 0.8385451436042786,
-      "rewards/rejected": -2.4665448665618896,
+      "logits/chosen": -1.0366981029510498,
+      "logits/rejected": -0.9037246704101562,
+      "logps/chosen": -1879.755126953125,
+      "logps/rejected": -2001.697265625,
+      "loss": 1.286,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -15.764042854309082,
+      "rewards/margins": 1.033178687095642,
+      "rewards/rejected": -16.797222137451172,
       "step": 770
     },
     {
       "epoch": 0.8163265306122449,
-      "grad_norm": 40.436476118941954,
+      "grad_norm": 162.02338031146334,
       "learning_rate": 4.947931323697982e-08,
-      "logits/chosen": -0.4582279324531555,
-      "logits/rejected": 0.3044646382331848,
-      "logps/chosen": -473.2518005371094,
-      "logps/rejected": -511.49334716796875,
-      "loss": 0.5022,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -1.542907953262329,
-      "rewards/margins": 0.8963887095451355,
-      "rewards/rejected": -2.4392964839935303,
+      "logits/chosen": -1.0304605960845947,
+      "logits/rejected": -0.9400796890258789,
+      "logps/chosen": -1669.2073974609375,
+      "logps/rejected": -2004.525390625,
+      "loss": 1.5927,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -13.502462387084961,
+      "rewards/margins": 3.8671538829803467,
+      "rewards/rejected": -17.369617462158203,
       "step": 780
     },
     {
       "epoch": 0.826792255363684,
-      "grad_norm": 38.14021936834171,
+      "grad_norm": 140.7368428333841,
       "learning_rate": 4.415420150605398e-08,
-      "logits/chosen": -0.21675653755664825,
-      "logits/rejected": 1.2191559076309204,
-      "logps/chosen": -446.891357421875,
-      "logps/rejected": -514.4319458007812,
-      "loss": 0.5074,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -1.6831605434417725,
-      "rewards/margins": 1.0085071325302124,
-      "rewards/rejected": -2.6916680335998535,
+      "logits/chosen": -1.0811887979507446,
+      "logits/rejected": -0.5253428220748901,
+      "logps/chosen": -1726.182373046875,
+      "logps/rejected": -2063.27099609375,
+      "loss": 1.4648,
+      "rewards/accuracies": 0.581250011920929,
+      "rewards/chosen": -14.476069450378418,
+      "rewards/margins": 3.7039875984191895,
+      "rewards/rejected": -18.180057525634766,
       "step": 790
     },
     {
       "epoch": 0.837257980115123,
-      "grad_norm": 33.872578527467745,
+      "grad_norm": 202.82775780509928,
       "learning_rate": 3.9104390285376374e-08,
-      "logits/chosen": 0.09026177227497101,
-      "logits/rejected": 0.9493114352226257,
-      "logps/chosen": -439.46429443359375,
-      "logps/rejected": -529.3217163085938,
-      "loss": 0.4741,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -1.7335838079452515,
-      "rewards/margins": 0.9264078140258789,
-      "rewards/rejected": -2.65999174118042,
+      "logits/chosen": -0.835501492023468,
+      "logits/rejected": -0.5900505781173706,
+      "logps/chosen": -1749.853759765625,
+      "logps/rejected": -1951.329345703125,
+      "loss": 1.3527,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -14.83747673034668,
+      "rewards/margins": 2.0425891876220703,
+      "rewards/rejected": -16.88006591796875,
       "step": 800
     },
     {
       "epoch": 0.837257980115123,
-      "eval_logits/chosen": -0.3940890431404114,
-      "eval_logits/rejected": 0.6599191427230835,
-      "eval_logps/chosen": -437.01190185546875,
-      "eval_logps/rejected": -517.6414794921875,
-      "eval_loss": 0.49065762758255005,
-      "eval_rewards/accuracies": 0.7777777910232544,
-      "eval_rewards/chosen": -1.5504390001296997,
-      "eval_rewards/margins": 1.02390718460083,
-      "eval_rewards/rejected": -2.5743465423583984,
-      "eval_runtime": 176.8304,
-      "eval_samples_per_second": 11.31,
-      "eval_steps_per_second": 0.356,
+      "eval_logits/chosen": -1.0374784469604492,
+      "eval_logits/rejected": -0.6750361919403076,
+      "eval_logps/chosen": -1672.130615234375,
+      "eval_logps/rejected": -1986.035888671875,
+      "eval_loss": 1.366306185722351,
+      "eval_rewards/accuracies": 0.6448412537574768,
+      "eval_rewards/chosen": -13.901623725891113,
+      "eval_rewards/margins": 3.3566668033599854,
+      "eval_rewards/rejected": -17.25829315185547,
+      "eval_runtime": 176.0547,
+      "eval_samples_per_second": 11.36,
+      "eval_steps_per_second": 0.358,
       "step": 800
     },
     {
       "epoch": 0.847723704866562,
-      "grad_norm": 32.1124717145771,
+      "grad_norm": 218.80895490878117,
       "learning_rate": 3.433663324986208e-08,
-      "logits/chosen": -0.3709171712398529,
-      "logits/rejected": 0.708153486251831,
-      "logps/chosen": -451.8567810058594,
-      "logps/rejected": -501.13153076171875,
-      "loss": 0.4866,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.606183409690857,
-      "rewards/margins": 0.9406149983406067,
-      "rewards/rejected": -2.5467987060546875,
+      "logits/chosen": -1.2597501277923584,
+      "logits/rejected": -0.7243804931640625,
+      "logps/chosen": -1665.3489990234375,
+      "logps/rejected": -2045.541259765625,
+      "loss": 1.4186,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -13.741106986999512,
+      "rewards/margins": 4.249786853790283,
+      "rewards/rejected": -17.990894317626953,
       "step": 810
     },
     {
       "epoch": 0.858189429618001,
-      "grad_norm": 36.42213321353283,
+      "grad_norm": 175.67069527310957,
       "learning_rate": 2.9857306851953897e-08,
-      "logits/chosen": -0.5197598934173584,
-      "logits/rejected": 0.44594240188598633,
-      "logps/chosen": -462.8509216308594,
-      "logps/rejected": -537.5787963867188,
-      "loss": 0.498,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -1.5807650089263916,
-      "rewards/margins": 0.9640032052993774,
-      "rewards/rejected": -2.5447685718536377,
+      "logits/chosen": -1.075448751449585,
+      "logits/rejected": -0.8459098935127258,
+      "logps/chosen": -1705.802734375,
+      "logps/rejected": -1971.207275390625,
+      "loss": 1.1819,
+      "rewards/accuracies": 0.606249988079071,
+      "rewards/chosen": -14.010282516479492,
+      "rewards/margins": 2.8707687854766846,
+      "rewards/rejected": -16.881052017211914,
       "step": 820
     },
     {
       "epoch": 0.8686551543694401,
-      "grad_norm": 33.241447274140704,
+      "grad_norm": 150.14969837730865,
       "learning_rate": 2.567240179368185e-08,
-      "logits/chosen": -0.35766562819480896,
-      "logits/rejected": 0.4714294373989105,
-      "logps/chosen": -418.21746826171875,
-      "logps/rejected": -512.0104370117188,
-      "loss": 0.5054,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -1.600725531578064,
-      "rewards/margins": 1.0214117765426636,
-      "rewards/rejected": -2.6221375465393066,
+      "logits/chosen": -0.8211779594421387,
+      "logits/rejected": -0.672277569770813,
+      "logps/chosen": -1724.1959228515625,
+      "logps/rejected": -1975.289306640625,
+      "loss": 1.3771,
+      "rewards/accuracies": 0.6187499761581421,
+      "rewards/chosen": -14.66050910949707,
+      "rewards/margins": 2.5944151878356934,
+      "rewards/rejected": -17.25492286682129,
       "step": 830
     },
     {
       "epoch": 0.8791208791208791,
-      "grad_norm": 30.190104504480274,
+      "grad_norm": 143.51050018041488,
       "learning_rate": 2.1787515014630357e-08,
-      "logits/chosen": -0.35540395975112915,
-      "logits/rejected": 0.8115441203117371,
-      "logps/chosen": -448.0243225097656,
-      "logps/rejected": -531.029541015625,
-      "loss": 0.4616,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -1.6575037240982056,
-      "rewards/margins": 0.9920711517333984,
-      "rewards/rejected": -2.6495752334594727,
+      "logits/chosen": -0.9592329263687134,
+      "logits/rejected": -0.6304475665092468,
+      "logps/chosen": -1664.050048828125,
+      "logps/rejected": -2090.85107421875,
+      "loss": 1.1841,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -13.817761421203613,
+      "rewards/margins": 4.430028915405273,
+      "rewards/rejected": -18.247791290283203,
       "step": 840
     },
     {
       "epoch": 0.8895866038723181,
-      "grad_norm": 41.82041936139999,
+      "grad_norm": 160.09590738302992,
       "learning_rate": 1.820784220652766e-08,
-      "logits/chosen": -0.12738969922065735,
-      "logits/rejected": 1.10179603099823,
-      "logps/chosen": -434.4571228027344,
-      "logps/rejected": -497.9303283691406,
-      "loss": 0.455,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -1.5362211465835571,
-      "rewards/margins": 1.1085121631622314,
-      "rewards/rejected": -2.64473295211792,
+      "logits/chosen": -0.8976573944091797,
+      "logits/rejected": -0.619744598865509,
+      "logps/chosen": -1732.6185302734375,
+      "logps/rejected": -2009.6126708984375,
+      "loss": 1.3946,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -14.51783561706543,
+      "rewards/margins": 3.2437214851379395,
+      "rewards/rejected": -17.761554718017578,
       "step": 850
     },
     {
       "epoch": 0.9000523286237572,
-      "grad_norm": 34.694500021885084,
+      "grad_norm": 140.45079725700174,
       "learning_rate": 1.4938170864468636e-08,
-      "logits/chosen": -0.25795820355415344,
-      "logits/rejected": 0.9191703796386719,
-      "logps/chosen": -451.52752685546875,
-      "logps/rejected": -523.0709838867188,
-      "loss": 0.4787,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -1.6323562860488892,
-      "rewards/margins": 0.984820544719696,
-      "rewards/rejected": -2.6171765327453613,
+      "logits/chosen": -1.2183126211166382,
+      "logits/rejected": -0.7451462149620056,
+      "logps/chosen": -1663.8861083984375,
+      "logps/rejected": -2030.5501708984375,
+      "loss": 1.403,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -13.755941390991211,
+      "rewards/margins": 3.9360270500183105,
+      "rewards/rejected": -17.691970825195312,
       "step": 860
     },
     {
       "epoch": 0.9105180533751962,
-      "grad_norm": 34.514941887456985,
+      "grad_norm": 177.87764974909854,
       "learning_rate": 1.1982873884064465e-08,
-      "logits/chosen": -0.48437461256980896,
-      "logits/rejected": 0.7289548516273499,
-      "logps/chosen": -428.81390380859375,
-      "logps/rejected": -501.2752990722656,
-      "loss": 0.489,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.5704351663589478,
-      "rewards/margins": 0.9770514369010925,
-      "rewards/rejected": -2.5474865436553955,
+      "logits/chosen": -1.142114281654358,
+      "logits/rejected": -0.8570957183837891,
+      "logps/chosen": -1702.1165771484375,
+      "logps/rejected": -2053.07568359375,
+      "loss": 1.364,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -14.30346393585205,
+      "rewards/margins": 3.762028932571411,
+      "rewards/rejected": -18.06549072265625,
       "step": 870
     },
     {
       "epoch": 0.9209837781266352,
-      "grad_norm": 33.54810196280602,
+      "grad_norm": 138.3301348415624,
       "learning_rate": 9.345903713082304e-09,
-      "logits/chosen": -0.3731919825077057,
-      "logits/rejected": 0.4349101185798645,
-      "logps/chosen": -459.6202697753906,
-      "logps/rejected": -546.8380126953125,
-      "loss": 0.4823,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -1.5548641681671143,
-      "rewards/margins": 0.9671527743339539,
-      "rewards/rejected": -2.522017002105713,
+      "logits/chosen": -1.0760080814361572,
+      "logits/rejected": -0.866096019744873,
+      "logps/chosen": -1735.3382568359375,
+      "logps/rejected": -2023.660888671875,
+      "loss": 1.355,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -14.312044143676758,
+      "rewards/margins": 2.978205680847168,
+      "rewards/rejected": -17.290246963500977,
       "step": 880
     },
     {
       "epoch": 0.9314495028780743,
-      "grad_norm": 32.531647089433314,
+      "grad_norm": 179.16273994251034,
       "learning_rate": 7.030787065396865e-09,
-      "logits/chosen": -0.3866179585456848,
-      "logits/rejected": 0.5604432225227356,
-      "logps/chosen": -444.09967041015625,
-      "logps/rejected": -537.5520629882812,
-      "loss": 0.4825,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -1.5642650127410889,
-      "rewards/margins": 1.0092883110046387,
-      "rewards/rejected": -2.5735533237457275,
+      "logits/chosen": -1.0234577655792236,
+      "logits/rejected": -0.9720734357833862,
+      "logps/chosen": -1736.5269775390625,
+      "logps/rejected": -2083.37939453125,
+      "loss": 1.4332,
+      "rewards/accuracies": 0.581250011920929,
+      "rewards/chosen": -14.488537788391113,
+      "rewards/margins": 3.543290376663208,
+      "rewards/rejected": -18.031827926635742,
       "step": 890
     },
     {
       "epoch": 0.9419152276295133,
-      "grad_norm": 41.37080153124819,
+      "grad_norm": 163.4835379161221,
       "learning_rate": 5.04062020432286e-09,
-      "logits/chosen": -0.411233514547348,
-      "logits/rejected": 0.5156325101852417,
-      "logps/chosen": -434.09906005859375,
-      "logps/rejected": -530.5025024414062,
-      "loss": 0.4741,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -1.5370757579803467,
-      "rewards/margins": 0.953914999961853,
-      "rewards/rejected": -2.49099063873291,
+      "logits/chosen": -0.8189510107040405,
+      "logits/rejected": -0.8584410548210144,
+      "logps/chosen": -1706.8818359375,
+      "logps/rejected": -1968.8441162109375,
+      "loss": 1.5137,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -14.264904975891113,
+      "rewards/margins": 2.609503984451294,
+      "rewards/rejected": -16.874408721923828,
       "step": 900
     },
     {
       "epoch": 0.9419152276295133,
-      "eval_logits/chosen": -0.3568094074726105,
-      "eval_logits/rejected": 0.7152466773986816,
-      "eval_logps/chosen": -440.10498046875,
-      "eval_logps/rejected": -523.0108032226562,
-      "eval_loss": 0.49055662751197815,
-      "eval_rewards/accuracies": 0.7698412537574768,
-      "eval_rewards/chosen": -1.5813697576522827,
-      "eval_rewards/margins": 1.0466694831848145,
-      "eval_rewards/rejected": -2.6280391216278076,
-      "eval_runtime": 175.5396,
-      "eval_samples_per_second": 11.393,
-      "eval_steps_per_second": 0.359,
+      "eval_logits/chosen": -1.001752495765686,
+      "eval_logits/rejected": -0.673967182636261,
+      "eval_logps/chosen": -1735.9151611328125,
+      "eval_logps/rejected": -2073.3388671875,
+      "eval_loss": 1.3373700380325317,
+      "eval_rewards/accuracies": 0.6408730149269104,
+      "eval_rewards/chosen": -14.539473533630371,
+      "eval_rewards/margins": 3.5918467044830322,
+      "eval_rewards/rejected": -18.13132095336914,
+      "eval_runtime": 176.3334,
+      "eval_samples_per_second": 11.342,
+      "eval_steps_per_second": 0.357,
       "step": 900
     },
     {
       "epoch": 0.9523809523809523,
-      "grad_norm": 47.043240406771055,
+      "grad_norm": 190.32378571700949,
       "learning_rate": 3.3780648016376866e-09,
-      "logits/chosen": -0.13907718658447266,
-      "logits/rejected": 0.8174427151679993,
-      "logps/chosen": -380.21087646484375,
-      "logps/rejected": -457.331298828125,
-      "loss": 0.4873,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -1.5678941011428833,
-      "rewards/margins": 0.8766323924064636,
-      "rewards/rejected": -2.444526433944702,
+      "logits/chosen": -0.9321626424789429,
+      "logits/rejected": -0.5902298092842102,
+      "logps/chosen": -1696.779296875,
+      "logps/rejected": -1922.1607666015625,
+      "loss": 1.4578,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": -14.7335786819458,
+      "rewards/margins": 2.3592441082000732,
+      "rewards/rejected": -17.092823028564453,
       "step": 910
     },
     {
       "epoch": 0.9628466771323915,
-      "grad_norm": 41.04154104780731,
+      "grad_norm": 183.98567167006505,
       "learning_rate": 2.0453443778310766e-09,
-      "logits/chosen": -0.3518763482570648,
-      "logits/rejected": 0.7049046754837036,
-      "logps/chosen": -444.53271484375,
-      "logps/rejected": -510.38812255859375,
-      "loss": 0.4989,
-      "rewards/accuracies": 0.768750011920929,
-      "rewards/chosen": -1.5789575576782227,
-      "rewards/margins": 1.008893370628357,
-      "rewards/rejected": -2.587851047515869,
+      "logits/chosen": -1.0600922107696533,
+      "logits/rejected": -0.7931039929389954,
+      "logps/chosen": -1763.392822265625,
+      "logps/rejected": -2107.805419921875,
+      "loss": 1.3202,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -14.767558097839355,
+      "rewards/margins": 3.794466495513916,
+      "rewards/rejected": -18.562023162841797,
       "step": 920
     },
     {
       "epoch": 0.9733124018838305,
-      "grad_norm": 31.21202125429197,
+      "grad_norm": 181.56437725274117,
       "learning_rate": 1.0442413283435758e-09,
-      "logits/chosen": -0.5221754312515259,
-      "logits/rejected": 1.012073040008545,
-      "logps/chosen": -460.8104553222656,
-      "logps/rejected": -517.5181884765625,
-      "loss": 0.4808,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -1.4597631692886353,
-      "rewards/margins": 1.07325279712677,
-      "rewards/rejected": -2.5330162048339844,
+      "logits/chosen": -1.1890182495117188,
+      "logits/rejected": -0.5295430421829224,
+      "logps/chosen": -1729.0921630859375,
+      "logps/rejected": -1985.2783203125,
+      "loss": 1.5669,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -14.14258098602295,
+      "rewards/margins": 3.068037748336792,
+      "rewards/rejected": -17.210617065429688,
       "step": 930
     },
     {
       "epoch": 0.9837781266352695,
-      "grad_norm": 31.068531683647514,
+      "grad_norm": 173.28786175289625,
       "learning_rate": 3.760945397705828e-10,
-      "logits/chosen": -0.38587886095046997,
-      "logits/rejected": 0.7057468891143799,
-      "logps/chosen": -410.48834228515625,
-      "logps/rejected": -507.0341796875,
-      "loss": 0.4773,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -1.4730709791183472,
-      "rewards/margins": 1.0000593662261963,
-      "rewards/rejected": -2.473130464553833,
+      "logits/chosen": -0.856045126914978,
+      "logits/rejected": -0.7398639917373657,
+      "logps/chosen": -1713.3883056640625,
+      "logps/rejected": -2039.740966796875,
+      "loss": 1.266,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -14.502069473266602,
+      "rewards/margins": 3.2981293201446533,
+      "rewards/rejected": -17.800199508666992,
       "step": 940
     },
     {
       "epoch": 0.9942438513867086,
-      "grad_norm": 32.463686030627095,
+      "grad_norm": 188.65879146663107,
       "learning_rate": 4.17975992204056e-11,
-      "logits/chosen": -0.2789713740348816,
-      "logits/rejected": 0.4086712896823883,
-      "logps/chosen": -464.32086181640625,
-      "logps/rejected": -530.4679565429688,
-      "loss": 0.4828,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -1.6147760152816772,
-      "rewards/margins": 0.8401002883911133,
-      "rewards/rejected": -2.45487642288208,
+      "logits/chosen": -1.168084740638733,
+      "logits/rejected": -0.8855546116828918,
+      "logps/chosen": -1736.102783203125,
+      "logps/rejected": -1955.3255615234375,
+      "loss": 1.4604,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": -14.33259391784668,
+      "rewards/margins": 2.370856761932373,
+      "rewards/rejected": -16.70345115661621,
       "step": 950
     },
     {
       "epoch": 0.9994767137624281,
       "step": 955,
       "total_flos": 0.0,
-      "train_loss": 0.5247097397349891,
-      "train_runtime": 18219.8315,
-      "train_samples_per_second": 3.355,
-      "train_steps_per_second": 0.052
+      "train_loss": 2.1165736393154604,
+      "train_runtime": 18133.1885,
+      "train_samples_per_second": 3.371,
+      "train_steps_per_second": 0.053
     }
   ],
   "logging_steps": 10,