Model save

all_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 2.0,
-    "train_loss": 0.6161670185949492,
-    "train_runtime": 9955.6802,
+    "train_loss": 0.26489045179408527,
+    "train_runtime": 4520.7899,
     "train_samples": 47095,
-    "train_samples_per_second": 9.461,
-    "train_steps_per_second": 0.037
+    "train_samples_per_second": 20.835,
+    "train_steps_per_second": 0.081
 }

model-00001-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0378e033bd057ff4230c269dec32f7e6a2ae0600e579fc309e400f7996014301
+oid sha256:89021f925795319ec6cd79c57490ff100b4d209128279330c40d869a1257ea07
 size 4943162336

model-00002-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0e0f203f7c24487720e7e86483d08c9428bd96841fa9d4427eaa05442d00a663
+oid sha256:cc9bb6d1ff6e5e594487f490ec146b7666201b9e396d8c36acadcf28b8944e2e
 size 4999819336

model-00003-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2c5591b65d2ba8c3131411b497a8c0a8d0af91ce6f9c61707543a7ee405946ec
+oid sha256:d64a4a2eb8fd994c6c2e00218546b04c07ec10f9918e36094d7621932e29d265
 size 4540516344

runs/Jun20_09-58-48_n136-100-194/events.out.tfevents.1718848801.n136-100-194.155161.0

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ad66b0fb63b414c906409ee8053960f3cf88465d60719b79a58c1235bc60df21
-size 11813
+oid sha256:c162588195abc77ed21a2660c04be8b902d7ab73b7d3519958cc814c5462585c
+size 16295

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 2.0,
-    "train_loss": 0.6161670185949492,
-    "train_runtime": 9955.6802,
+    "train_loss": 0.26489045179408527,
+    "train_runtime": 4520.7899,
     "train_samples": 47095,
-    "train_samples_per_second": 9.461,
-    "train_steps_per_second": 0.037
+    "train_samples_per_second": 20.835,
+    "train_steps_per_second": 0.081
 }

trainer_state.json

@@ -10,13 +10,13 @@
   "log_history": [
     {
       "epoch": 0.01,
-      "grad_norm": 2062.9417756205603,
+      "grad_norm": 2733.5875966596914,
       "learning_rate": 2.702702702702703e-10,
       "logits/chosen": -1.3332719802856445,
       "logits/rejected": -1.246394395828247,
       "logps/chosen": -286.9539794921875,
       "logps/rejected": -263.3782958984375,
-      "loss": 0.7136,
+      "loss": 0.7283,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,552 +25,552 @@
     },
     {
       "epoch": 0.05,
-      "grad_norm": 2488.3980990852974,
+      "grad_norm": 3426.28374639058,
       "learning_rate": 2.702702702702703e-09,
-      "logits/chosen": -1.6142714023590088,
-      "logits/rejected": -1.3925563097000122,
-      "logps/chosen": -342.4814758300781,
-      "logps/rejected": -294.5446472167969,
-      "loss": 0.8226,
-      "rewards/accuracies": 0.4618055522441864,
-      "rewards/chosen": 0.079922616481781,
-      "rewards/margins": 0.09200635552406311,
-      "rewards/rejected": -0.012083739042282104,
+      "logits/chosen": -1.617490530014038,
+      "logits/rejected": -1.3964743614196777,
+      "logps/chosen": -342.53607177734375,
+      "logps/rejected": -294.5452575683594,
+      "loss": 0.9019,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": -0.0025859144516289234,
+      "rewards/margins": 0.014665775932371616,
+      "rewards/rejected": -0.017251690849661827,
       "step": 10
     },
     {
       "epoch": 0.11,
-      "grad_norm": 2085.30491295085,
+      "grad_norm": 3068.517544136213,
       "learning_rate": 5.405405405405406e-09,
-      "logits/chosen": -1.4863827228546143,
-      "logits/rejected": -1.3085709810256958,
-      "logps/chosen": -314.74273681640625,
-      "logps/rejected": -279.32977294921875,
-      "loss": 0.8217,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": 0.03496693819761276,
-      "rewards/margins": 0.07092654705047607,
-      "rewards/rejected": -0.03595960885286331,
+      "logits/chosen": -1.4905732870101929,
+      "logits/rejected": -1.3132953643798828,
+      "logps/chosen": -314.7499084472656,
+      "logps/rejected": -279.27752685546875,
+      "loss": 0.9225,
+      "rewards/accuracies": 0.4937500059604645,
+      "rewards/chosen": 0.03222837299108505,
+      "rewards/margins": -0.024351513013243675,
+      "rewards/rejected": 0.05657988786697388,
       "step": 20
     },
     {
       "epoch": 0.16,
-      "grad_norm": 2613.9787597915297,
+      "grad_norm": 3019.197796031677,
       "learning_rate": 8.108108108108109e-09,
-      "logits/chosen": -1.5464979410171509,
-      "logits/rejected": -1.3788726329803467,
-      "logps/chosen": -324.9065246582031,
-      "logps/rejected": -286.29925537109375,
-      "loss": 0.8318,
-      "rewards/accuracies": 0.515625,
-      "rewards/chosen": -0.0007322698947973549,
-      "rewards/margins": 0.02973010204732418,
-      "rewards/rejected": -0.030462373048067093,
+      "logits/chosen": -1.5479624271392822,
+      "logits/rejected": -1.3802028894424438,
+      "logps/chosen": -324.89044189453125,
+      "logps/rejected": -286.2395324707031,
+      "loss": 0.9562,
+      "rewards/accuracies": 0.5218750238418579,
+      "rewards/chosen": 0.031223665922880173,
+      "rewards/margins": -0.0476018562912941,
+      "rewards/rejected": 0.07882551848888397,
       "step": 30
     },
     {
       "epoch": 0.22,
-      "grad_norm": 2309.6989479898994,
+      "grad_norm": 2861.837005400921,
       "learning_rate": 9.997973265157192e-09,
-      "logits/chosen": -1.5338213443756104,
-      "logits/rejected": -1.356065034866333,
-      "logps/chosen": -325.39349365234375,
-      "logps/rejected": -285.630859375,
-      "loss": 0.8544,
-      "rewards/accuracies": 0.5093749761581421,
-      "rewards/chosen": -0.00019043684005737305,
-      "rewards/margins": -0.028223956003785133,
-      "rewards/rejected": 0.02803351916372776,
+      "logits/chosen": -1.5354044437408447,
+      "logits/rejected": -1.3576419353485107,
+      "logps/chosen": -325.43408203125,
+      "logps/rejected": -285.6204528808594,
+      "loss": 0.9309,
+      "rewards/accuracies": 0.484375,
+      "rewards/chosen": -0.08140890300273895,
+      "rewards/margins": -0.13951030373573303,
+      "rewards/rejected": 0.05810140445828438,
       "step": 40
     },
     {
       "epoch": 0.27,
-      "grad_norm": 2372.8781916000794,
+      "grad_norm": 3285.7135249709568,
       "learning_rate": 9.961988113473708e-09,
-      "logits/chosen": -1.540814757347107,
-      "logits/rejected": -1.3939155340194702,
-      "logps/chosen": -337.01385498046875,
-      "logps/rejected": -297.3047790527344,
-      "loss": 0.7925,
-      "rewards/accuracies": 0.5062500238418579,
-      "rewards/chosen": 0.010568022727966309,
-      "rewards/margins": 0.0009421706199645996,
-      "rewards/rejected": 0.009625854901969433,
+      "logits/chosen": -1.534355640411377,
+      "logits/rejected": -1.3875898122787476,
+      "logps/chosen": -337.02044677734375,
+      "logps/rejected": -297.35101318359375,
+      "loss": 0.88,
+      "rewards/accuracies": 0.546875,
+      "rewards/chosen": 0.0009558796882629395,
+      "rewards/margins": 0.08057532459497452,
+      "rewards/rejected": -0.07961944490671158,
       "step": 50
     },
     {
       "epoch": 0.33,
-      "grad_norm": 1906.9193219897543,
+      "grad_norm": 2547.1665545835035,
       "learning_rate": 9.881337335184878e-09,
-      "logits/chosen": -1.5821880102157593,
-      "logits/rejected": -1.433316707611084,
-      "logps/chosen": -319.8349609375,
-      "logps/rejected": -285.03131103515625,
-      "loss": 0.7444,
-      "rewards/accuracies": 0.59375,
-      "rewards/chosen": 0.011926290579140186,
-      "rewards/margins": 0.23517760634422302,
-      "rewards/rejected": -0.22325129806995392,
+      "logits/chosen": -1.5822935104370117,
+      "logits/rejected": -1.4333903789520264,
+      "logps/chosen": -319.79644775390625,
+      "logps/rejected": -285.0381164550781,
+      "loss": 0.8105,
+      "rewards/accuracies": 0.6156250238418579,
+      "rewards/chosen": 0.09285839647054672,
+      "rewards/margins": 0.40408092737197876,
+      "rewards/rejected": -0.31122252345085144,
       "step": 60
     },
     {
       "epoch": 0.38,
-      "grad_norm": 2229.621479388874,
+      "grad_norm": 3178.8313195657583,
       "learning_rate": 9.756746912994832e-09,
-      "logits/chosen": -1.5089519023895264,
-      "logits/rejected": -1.3478004932403564,
-      "logps/chosen": -312.11767578125,
-      "logps/rejected": -275.03704833984375,
-      "loss": 0.7381,
-      "rewards/accuracies": 0.5531250238418579,
-      "rewards/chosen": -0.015234187245368958,
-      "rewards/margins": 0.07565010339021683,
-      "rewards/rejected": -0.09088429063558578,
+      "logits/chosen": -1.5119212865829468,
+      "logits/rejected": -1.350838541984558,
+      "logps/chosen": -312.1349182128906,
+      "logps/rejected": -275.08660888671875,
+      "loss": 0.7993,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": -0.05482473224401474,
+      "rewards/margins": 0.165395587682724,
+      "rewards/rejected": -0.22022032737731934,
       "step": 70
     },
     {
       "epoch": 0.43,
-      "grad_norm": 1970.0426820414286,
+      "grad_norm": 2596.199281277795,
       "learning_rate": 9.589338354885628e-09,
-      "logits/chosen": -1.591552734375,
-      "logits/rejected": -1.4374128580093384,
-      "logps/chosen": -323.3088684082031,
-      "logps/rejected": -288.12445068359375,
-      "loss": 0.7257,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.1117367148399353,
-      "rewards/margins": 0.34563174843788147,
-      "rewards/rejected": -0.23389501869678497,
+      "logits/chosen": -1.5992329120635986,
+      "logits/rejected": -1.4463211297988892,
+      "logps/chosen": -323.2821960449219,
+      "logps/rejected": -288.0993347167969,
+      "loss": 0.7772,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.20231203734874725,
+      "rewards/margins": 0.4638887345790863,
+      "rewards/rejected": -0.26157671213150024,
       "step": 80
     },
     {
       "epoch": 0.49,
-      "grad_norm": 1647.476042777907,
+      "grad_norm": 2404.162696635418,
       "learning_rate": 9.380618598797472e-09,
-      "logits/chosen": -1.6083869934082031,
-      "logits/rejected": -1.4117141962051392,
-      "logps/chosen": -319.9634094238281,
-      "logps/rejected": -281.79248046875,
-      "loss": 0.6768,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.1753208488225937,
-      "rewards/margins": 0.44467267394065857,
-      "rewards/rejected": -0.2693518102169037,
+      "logits/chosen": -1.6108148097991943,
+      "logits/rejected": -1.4147026538848877,
+      "logps/chosen": -319.95526123046875,
+      "logps/rejected": -281.7666015625,
+      "loss": 0.7649,
+      "rewards/accuracies": 0.640625,
+      "rewards/chosen": 0.2500740885734558,
+      "rewards/margins": 0.5574880838394165,
+      "rewards/rejected": -0.3074139356613159,
       "step": 90
     },
     {
       "epoch": 0.54,
-      "grad_norm": 1779.591190181612,
+      "grad_norm": 2500.271614922207,
       "learning_rate": 9.132466447838596e-09,
-      "logits/chosen": -1.5439790487289429,
-      "logits/rejected": -1.368858814239502,
-      "logps/chosen": -321.8800964355469,
-      "logps/rejected": -282.66168212890625,
-      "loss": 0.6482,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": 0.34998807311058044,
-      "rewards/margins": 0.6073418855667114,
-      "rewards/rejected": -0.25735384225845337,
+      "logits/chosen": -1.542976975440979,
+      "logits/rejected": -1.3676128387451172,
+      "logps/chosen": -321.9007263183594,
+      "logps/rejected": -282.65899658203125,
+      "loss": 0.7305,
+      "rewards/accuracies": 0.659375011920929,
+      "rewards/chosen": 0.4254188537597656,
+      "rewards/margins": 0.763160228729248,
+      "rewards/rejected": -0.3377414047718048,
       "step": 100
     },
     {
       "epoch": 0.6,
-      "grad_norm": 1799.5128068859713,
+      "grad_norm": 2525.572173445101,
       "learning_rate": 8.847115658129039e-09,
-      "logits/chosen": -1.5068881511688232,
-      "logits/rejected": -1.3783992528915405,
-      "logps/chosen": -318.10797119140625,
-      "logps/rejected": -287.1791076660156,
-      "loss": 0.6577,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.35399100184440613,
-      "rewards/margins": 0.5296486616134644,
-      "rewards/rejected": -0.17565762996673584,
+      "logits/chosen": -1.512939453125,
+      "logits/rejected": -1.3849382400512695,
+      "logps/chosen": -318.14813232421875,
+      "logps/rejected": -287.1947937011719,
+      "loss": 0.7164,
+      "rewards/accuracies": 0.628125011920929,
+      "rewards/chosen": 0.3916184604167938,
+      "rewards/margins": 0.657262921333313,
+      "rewards/rejected": -0.265644371509552,
       "step": 110
     },
     {
       "epoch": 0.65,
-      "grad_norm": 1676.764876114058,
+      "grad_norm": 2498.220455730566,
       "learning_rate": 8.527134831514116e-09,
-      "logits/chosen": -1.5781362056732178,
-      "logits/rejected": -1.4229751825332642,
-      "logps/chosen": -331.3733825683594,
-      "logps/rejected": -297.85699462890625,
-      "loss": 0.6575,
-      "rewards/accuracies": 0.609375,
-      "rewards/chosen": 0.3793606460094452,
-      "rewards/margins": 0.4118588864803314,
-      "rewards/rejected": -0.03249818831682205,
+      "logits/chosen": -1.5739517211914062,
+      "logits/rejected": -1.41860032081604,
+      "logps/chosen": -331.3175354003906,
+      "logps/rejected": -297.8718566894531,
+      "loss": 0.7018,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.6175383925437927,
+      "rewards/margins": 0.6906081438064575,
+      "rewards/rejected": -0.07306969165802002,
       "step": 120
     },
     {
       "epoch": 0.71,
-      "grad_norm": 1566.6901996912077,
+      "grad_norm": 2170.854176631694,
       "learning_rate": 8.175404294144481e-09,
-      "logits/chosen": -1.6145737171173096,
-      "logits/rejected": -1.4269483089447021,
-      "logps/chosen": -317.0880432128906,
-      "logps/rejected": -271.5414123535156,
-      "loss": 0.6044,
-      "rewards/accuracies": 0.671875,
-      "rewards/chosen": 0.6310849189758301,
-      "rewards/margins": 0.7299145460128784,
-      "rewards/rejected": -0.09882961958646774,
+      "logits/chosen": -1.616276502609253,
+      "logits/rejected": -1.429518699645996,
+      "logps/chosen": -317.1609802246094,
+      "logps/rejected": -271.5557861328125,
+      "loss": 0.6719,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.6955646872520447,
+      "rewards/margins": 0.8560658693313599,
+      "rewards/rejected": -0.160501167178154,
       "step": 130
     },
     {
       "epoch": 0.76,
-      "grad_norm": 1706.595775593044,
+      "grad_norm": 2147.543587634845,
       "learning_rate": 7.79509016905158e-09,
-      "logits/chosen": -1.5648548603057861,
-      "logits/rejected": -1.4158308506011963,
-      "logps/chosen": -331.06622314453125,
-      "logps/rejected": -294.2123718261719,
-      "loss": 0.6171,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.7887445092201233,
-      "rewards/margins": 0.765161395072937,
-      "rewards/rejected": 0.023583168163895607,
+      "logits/chosen": -1.5726101398468018,
+      "logits/rejected": -1.4245671033859253,
+      "logps/chosen": -331.12109375,
+      "logps/rejected": -294.2488098144531,
+      "loss": 0.6686,
+      "rewards/accuracies": 0.6968749761581421,
+      "rewards/chosen": 0.9418653249740601,
+      "rewards/margins": 0.9832620620727539,
+      "rewards/rejected": -0.04139674827456474,
       "step": 140
     },
     {
       "epoch": 0.82,
-      "grad_norm": 1648.2049279025357,
+      "grad_norm": 2298.0130679506387,
       "learning_rate": 7.389615876105773e-09,
-      "logits/chosen": -1.5560743808746338,
-      "logits/rejected": -1.4283266067504883,
-      "logps/chosen": -314.5069274902344,
-      "logps/rejected": -291.7706298828125,
-      "loss": 0.6127,
-      "rewards/accuracies": 0.6656249761581421,
-      "rewards/chosen": 0.8379846811294556,
-      "rewards/margins": 0.7371869087219238,
-      "rewards/rejected": 0.10079775750637054,
+      "logits/chosen": -1.5536715984344482,
+      "logits/rejected": -1.4254592657089233,
+      "logps/chosen": -314.55267333984375,
+      "logps/rejected": -291.81536865234375,
+      "loss": 0.6793,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 1.0258944034576416,
+      "rewards/margins": 0.981005072593689,
+      "rewards/rejected": 0.04488936811685562,
       "step": 150
     },
     {
       "epoch": 0.87,
-      "grad_norm": 1635.8235385722824,
+      "grad_norm": 2220.286727308493,
       "learning_rate": 6.962631315901861e-09,
-      "logits/chosen": -1.5186518430709839,
-      "logits/rejected": -1.4028724431991577,
-      "logps/chosen": -317.958251953125,
-      "logps/rejected": -291.0096435546875,
-      "loss": 0.6088,
-      "rewards/accuracies": 0.653124988079071,
-      "rewards/chosen": 0.8378221392631531,
-      "rewards/margins": 0.6740074753761292,
-      "rewards/rejected": 0.16381461918354034,
+      "logits/chosen": -1.5181314945220947,
+      "logits/rejected": -1.4019381999969482,
+      "logps/chosen": -318.02752685546875,
+      "logps/rejected": -291.03936767578125,
+      "loss": 0.6742,
+      "rewards/accuracies": 0.628125011920929,
+      "rewards/chosen": 0.9784830808639526,
+      "rewards/margins": 0.819505512714386,
+      "rewards/rejected": 0.15897764265537262,
       "step": 160
     },
     {
       "epoch": 0.92,
-      "grad_norm": 1575.6130834814026,
+      "grad_norm": 2125.4983562302123,
       "learning_rate": 6.517980014965139e-09,
-      "logits/chosen": -1.6025912761688232,
-      "logits/rejected": -1.4152277708053589,
-      "logps/chosen": -331.40386962890625,
-      "logps/rejected": -289.4659729003906,
-      "loss": 0.5997,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": 0.8780991435050964,
-      "rewards/margins": 0.8349622488021851,
-      "rewards/rejected": 0.04313689470291138,
+      "logits/chosen": -1.5958881378173828,
+      "logits/rejected": -1.4071909189224243,
+      "logps/chosen": -331.4378356933594,
+      "logps/rejected": -289.5236511230469,
+      "loss": 0.6456,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": 1.1029136180877686,
+      "rewards/margins": 1.160766363143921,
+      "rewards/rejected": -0.05785265564918518,
       "step": 170
     },
     {
       "epoch": 0.98,
-      "grad_norm": 1546.3751249922345,
+      "grad_norm": 2116.9259500110184,
       "learning_rate": 6.059664528022266e-09,
-      "logits/chosen": -1.5942988395690918,
-      "logits/rejected": -1.44364333152771,
-      "logps/chosen": -315.07196044921875,
-      "logps/rejected": -276.7376708984375,
-      "loss": 0.5773,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": 0.8913241624832153,
-      "rewards/margins": 0.9472495317459106,
-      "rewards/rejected": -0.05592530965805054,
+      "logits/chosen": -1.5962104797363281,
+      "logits/rejected": -1.445967197418213,
+      "logps/chosen": -315.10467529296875,
+      "logps/rejected": -276.73443603515625,
+      "loss": 0.6191,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": 1.123002052307129,
+      "rewards/margins": 1.1910655498504639,
+      "rewards/rejected": -0.0680634081363678,
       "step": 180
     },
     {
       "epoch": 1.03,
-      "grad_norm": 1681.3148479750444,
+      "grad_norm": 2067.8979230397717,
       "learning_rate": 5.591810408770492e-09,
-      "logits/chosen": -1.5504480600357056,
-      "logits/rejected": -1.3759148120880127,
-      "logps/chosen": -315.5844421386719,
-      "logps/rejected": -278.6695861816406,
-      "loss": 0.5632,
-      "rewards/accuracies": 0.7093750238418579,
-      "rewards/chosen": 0.8848656415939331,
-      "rewards/margins": 0.8844806551933289,
-      "rewards/rejected": 0.00038505197153426707,
+      "logits/chosen": -1.55275559425354,
+      "logits/rejected": -1.3787180185317993,
+      "logps/chosen": -315.572509765625,
+      "logps/rejected": -278.71087646484375,
+      "loss": 0.6052,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 1.2037546634674072,
+      "rewards/margins": 1.2858160734176636,
+      "rewards/rejected": -0.0820615291595459,
       "step": 190
     },
     {
       "epoch": 1.09,
-      "grad_norm": 1651.7882136807318,
+      "grad_norm": 2203.6189854484824,
       "learning_rate": 5.118629073464423e-09,
-      "logits/chosen": -1.571003794670105,
-      "logits/rejected": -1.3608561754226685,
-      "logps/chosen": -325.93023681640625,
-      "logps/rejected": -282.7080993652344,
-      "loss": 0.5605,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": 1.0313498973846436,
-      "rewards/margins": 0.9450349807739258,
-      "rewards/rejected": 0.08631500601768494,
+      "logits/chosen": -1.5673738718032837,
+      "logits/rejected": -1.3565856218338013,
+      "logps/chosen": -325.91680908203125,
+      "logps/rejected": -282.65869140625,
+      "loss": 0.6024,
+      "rewards/accuracies": 0.746874988079071,
+      "rewards/chosen": 1.4020355939865112,
+      "rewards/margins": 1.1880966424942017,
+      "rewards/rejected": 0.21393892168998718,
       "step": 200
     },
     {
       "epoch": 1.14,
-      "grad_norm": 1538.4386313699126,
+      "grad_norm": 2263.504332795979,
       "learning_rate": 4.644379891605983e-09,
-      "logits/chosen": -1.608812689781189,
-      "logits/rejected": -1.4315342903137207,
-      "logps/chosen": -324.66522216796875,
-      "logps/rejected": -291.33428955078125,
-      "loss": 0.5478,
-      "rewards/accuracies": 0.7281249761581421,
-      "rewards/chosen": 1.0752595663070679,
-      "rewards/margins": 1.0428497791290283,
-      "rewards/rejected": 0.03240995481610298,
+      "logits/chosen": -1.611310601234436,
+      "logits/rejected": -1.4343440532684326,
+      "logps/chosen": -324.752197265625,
+      "logps/rejected": -291.36102294921875,
+      "loss": 0.5985,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 1.2598803043365479,
+      "rewards/margins": 1.2701908349990845,
+      "rewards/rejected": -0.010310685262084007,
       "step": 210
     },
     {
       "epoch": 1.2,
-      "grad_norm": 1737.3887570467818,
+      "grad_norm": 2324.309417872748,
       "learning_rate": 4.173331844980362e-09,
-      "logits/chosen": -1.5384166240692139,
-      "logits/rejected": -1.4137290716171265,
-      "logps/chosen": -323.9536437988281,
-      "logps/rejected": -293.42535400390625,
-      "loss": 0.563,
-      "rewards/accuracies": 0.6968749761581421,
-      "rewards/chosen": 0.9658479690551758,
-      "rewards/margins": 0.9138795137405396,
-      "rewards/rejected": 0.051968496292829514,
+      "logits/chosen": -1.5291264057159424,
+      "logits/rejected": -1.4033840894699097,
+      "logps/chosen": -323.9982604980469,
+      "logps/rejected": -293.4136047363281,
+      "loss": 0.5948,
+      "rewards/accuracies": 0.7281249761581421,
+      "rewards/chosen": 1.1985571384429932,
+      "rewards/margins": 1.105764627456665,
+      "rewards/rejected": 0.09279236942529678,
       "step": 220
     },
     {
       "epoch": 1.25,
-      "grad_norm": 1605.3661746462226,
+      "grad_norm": 2275.313721629071,
       "learning_rate": 3.7097251001664824e-09,
-      "logits/chosen": -1.537548542022705,
-      "logits/rejected": -1.3787362575531006,
-      "logps/chosen": -323.85125732421875,
-      "logps/rejected": -286.95379638671875,
-      "loss": 0.526,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": 1.146087408065796,
-      "rewards/margins": 1.0939618349075317,
-      "rewards/rejected": 0.0521254763007164,
+      "logits/chosen": -1.5342741012573242,
+      "logits/rejected": -1.3754017353057861,
+      "logps/chosen": -323.9897766113281,
+      "logps/rejected": -287.0173645019531,
+      "loss": 0.577,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 1.2511075735092163,
+      "rewards/margins": 1.3087403774261475,
+      "rewards/rejected": -0.05763290077447891,
       "step": 230
     },
     {
       "epoch": 1.3,
-      "grad_norm": 1689.839854162397,
+      "grad_norm": 2261.889907648677,
       "learning_rate": 3.2577328404292057e-09,
-      "logits/chosen": -1.5391089916229248,
-      "logits/rejected": -1.4084638357162476,
-      "logps/chosen": -312.51373291015625,
-      "logps/rejected": -285.9711608886719,
-      "loss": 0.5418,
-      "rewards/accuracies": 0.7093750238418579,
-      "rewards/chosen": 1.0901774168014526,
-      "rewards/margins": 1.016390085220337,
-      "rewards/rejected": 0.07378745824098587,
+      "logits/chosen": -1.5497777462005615,
+      "logits/rejected": -1.4208284616470337,
+      "logps/chosen": -312.53802490234375,
+      "logps/rejected": -285.97076416015625,
+      "loss": 0.5741,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 1.4049217700958252,
+      "rewards/margins": 1.3057136535644531,
+      "rewards/rejected": 0.09920807182788849,
       "step": 240
     },
     {
       "epoch": 1.36,
-      "grad_norm": 1710.94558540331,
+      "grad_norm": 2520.220972155196,
       "learning_rate": 2.821423700565763e-09,
-      "logits/chosen": -1.5968081951141357,
-      "logits/rejected": -1.4188272953033447,
-      "logps/chosen": -350.68487548828125,
-      "logps/rejected": -306.6036071777344,
-      "loss": 0.532,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": 1.2678377628326416,
-      "rewards/margins": 1.2405023574829102,
-      "rewards/rejected": 0.027335500344634056,
+      "logits/chosen": -1.5996572971343994,
+      "logits/rejected": -1.4216984510421753,
+      "logps/chosen": -350.76129150390625,
+      "logps/rejected": -306.58831787109375,
+      "loss": 0.5681,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 1.537647008895874,
+      "rewards/margins": 1.4706511497497559,
+      "rewards/rejected": 0.06699595600366592,
       "step": 250
     },
     {
       "epoch": 1.41,
-      "grad_norm": 1638.2367115980887,
+      "grad_norm": 2185.2135747623915,
       "learning_rate": 2.4047251428513483e-09,
-      "logits/chosen": -1.6129051446914673,
-      "logits/rejected": -1.4581451416015625,
-      "logps/chosen": -325.2450256347656,
-      "logps/rejected": -291.1476745605469,
-      "loss": 0.5289,
-      "rewards/accuracies": 0.734375,
-      "rewards/chosen": 1.2301806211471558,
-      "rewards/margins": 1.2308820486068726,
-      "rewards/rejected": -0.0007013082504272461,
+      "logits/chosen": -1.61586594581604,
+      "logits/rejected": -1.4618706703186035,
+      "logps/chosen": -325.3050537109375,
+      "logps/rejected": -291.10345458984375,
+      "loss": 0.5977,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": 1.5201152563095093,
+      "rewards/margins": 1.4326369762420654,
+      "rewards/rejected": 0.08747831732034683,
       "step": 260
     },
     {
       "epoch": 1.47,
-      "grad_norm": 1199.4883951774482,
+      "grad_norm": 1735.3140228188304,
       "learning_rate": 2.011388103757442e-09,
-      "logits/chosen": -1.5265954732894897,
-      "logits/rejected": -1.3828239440917969,
-      "logps/chosen": -316.2944641113281,
-      "logps/rejected": -285.7884826660156,
-      "loss": 0.5191,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": 1.3710923194885254,
-      "rewards/margins": 1.2594387531280518,
-      "rewards/rejected": 0.11165344715118408,
+      "logits/chosen": -1.5243465900421143,
+      "logits/rejected": -1.3802506923675537,
+      "logps/chosen": -316.4330139160156,
+      "logps/rejected": -285.81353759765625,
+      "loss": 0.5429,
+      "rewards/accuracies": 0.7406250238418579,
+      "rewards/chosen": 1.550986647605896,
+      "rewards/margins": 1.45218026638031,
+      "rewards/rejected": 0.09880634397268295,
       "step": 270
     },
     {
       "epoch": 1.52,
-      "grad_norm": 1472.2115597857592,
+      "grad_norm": 2121.5830866823144,
       "learning_rate": 1.644953229677474e-09,
-      "logits/chosen": -1.600651502609253,
-      "logits/rejected": -1.4179413318634033,
-      "logps/chosen": -326.00335693359375,
-      "logps/rejected": -284.74188232421875,
-      "loss": 0.5459,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 1.3610546588897705,
-      "rewards/margins": 1.2091944217681885,
-      "rewards/rejected": 0.1518600881099701,
+      "logits/chosen": -1.6015859842300415,
+      "logits/rejected": -1.4193016290664673,
+      "logps/chosen": -326.1202087402344,
+      "logps/rejected": -284.7384033203125,
+      "loss": 0.5887,
+      "rewards/accuracies": 0.746874988079071,
+      "rewards/chosen": 1.580999732017517,
+      "rewards/margins": 1.3716144561767578,
+      "rewards/rejected": 0.20938535034656525,
       "step": 280
     },
     {
       "epoch": 1.58,
-      "grad_norm": 1566.9737970600454,
+      "grad_norm": 2103.020456118981,
       "learning_rate": 1.308719005590957e-09,
-      "logits/chosen": -1.5032551288604736,
-      "logits/rejected": -1.3876453638076782,
-      "logps/chosen": -318.40948486328125,
-      "logps/rejected": -282.49554443359375,
-      "loss": 0.5407,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": 1.2658413648605347,
-      "rewards/margins": 1.187675952911377,
-      "rewards/rejected": 0.07816555351018906,
+      "logits/chosen": -1.509340524673462,
+      "logits/rejected": -1.3944005966186523,
+      "logps/chosen": -318.451416015625,
+      "logps/rejected": -282.4563293457031,
+      "loss": 0.5721,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": 1.603921890258789,
+      "rewards/margins": 1.4211392402648926,
+      "rewards/rejected": 0.182782843708992,
       "step": 290
     },
     {
       "epoch": 1.63,
-      "grad_norm": 1348.7257224769698,
+      "grad_norm": 2010.5734100649909,
       "learning_rate": 1.005712063557776e-09,
-      "logits/chosen": -1.6333671808242798,
-      "logits/rejected": -1.455556869506836,
-      "logps/chosen": -324.13885498046875,
-      "logps/rejected": -290.60186767578125,
-      "loss": 0.5346,
+      "logits/chosen": -1.6272541284561157,
+      "logits/rejected": -1.4480578899383545,
+      "logps/chosen": -324.20068359375,
+      "logps/rejected": -290.54803466796875,
+      "loss": 0.5898,
       "rewards/accuracies": 0.6968749761581421,
-      "rewards/chosen": 1.1175706386566162,
-      "rewards/margins": 1.0337438583374023,
-      "rewards/rejected": 0.08382664620876312,
+      "rewards/chosen": 1.3664501905441284,
+      "rewards/margins": 1.1471518278121948,
+      "rewards/rejected": 0.21929831802845,
       "step": 300
     },
     {
       "epoch": 1.68,
-      "grad_norm": 1356.5441208888985,
+      "grad_norm": 1862.455491029429,
       "learning_rate": 7.386599383124321e-10,
-      "logits/chosen": -1.565224051475525,
-      "logits/rejected": -1.3825923204421997,
-      "logps/chosen": -321.80316162109375,
-      "logps/rejected": -285.7908630371094,
-      "loss": 0.5304,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 1.2159234285354614,
-      "rewards/margins": 1.1465200185775757,
-      "rewards/rejected": 0.06940338760614395,
+      "logits/chosen": -1.563561201095581,
+      "logits/rejected": -1.3803369998931885,
+      "logps/chosen": -321.889404296875,
+      "logps/rejected": -285.8083190917969,
+      "loss": 0.5879,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 1.4487113952636719,
+      "rewards/margins": 1.3910987377166748,
+      "rewards/rejected": 0.057612527161836624,
       "step": 310
     },
     {
       "epoch": 1.74,
-      "grad_norm": 1445.3559110776998,
+      "grad_norm": 1875.0391267528262,
       "learning_rate": 5.099665152003929e-10,
-      "logits/chosen": -1.5921494960784912,
-      "logits/rejected": -1.3807857036590576,
-      "logps/chosen": -333.7308654785156,
-      "logps/rejected": -289.9362487792969,
-      "loss": 0.5241,
-      "rewards/accuracies": 0.7718750238418579,
-      "rewards/chosen": 1.3256893157958984,
-      "rewards/margins": 1.292041540145874,
-      "rewards/rejected": 0.03364778310060501,
+      "logits/chosen": -1.5980346202850342,
+      "logits/rejected": -1.3878109455108643,
+      "logps/chosen": -333.843994140625,
+      "logps/rejected": -289.8874816894531,
+      "loss": 0.5634,
+      "rewards/accuracies": 0.7593749761581421,
+      "rewards/chosen": 1.5412708520889282,
+      "rewards/margins": 1.3988701105117798,
+      "rewards/rejected": 0.14240065217018127,
       "step": 320
     },
     {
       "epoch": 1.79,
-      "grad_norm": 1681.5042999261696,
+      "grad_norm": 2232.545936783362,
       "learning_rate": 3.216903914633745e-10,
-      "logits/chosen": -1.5627129077911377,
-      "logits/rejected": -1.4408833980560303,
-      "logps/chosen": -325.2505187988281,
-      "logps/rejected": -296.106201171875,
-      "loss": 0.5429,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": 1.165374517440796,
-      "rewards/margins": 1.0651426315307617,
-      "rewards/rejected": 0.1002318263053894,
+      "logits/chosen": -1.5563807487487793,
+      "logits/rejected": -1.4342092275619507,
+      "logps/chosen": -325.34674072265625,
+      "logps/rejected": -296.1554870605469,
+      "loss": 0.5762,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": 1.3614461421966553,
+      "rewards/margins": 1.3264009952545166,
+      "rewards/rejected": 0.03504505008459091,
       "step": 330
     },
     {
       "epoch": 1.85,
-      "grad_norm": 1536.75287567762,
+      "grad_norm": 1895.2431252651277,
       "learning_rate": 1.7552634565570324e-10,
-      "logits/chosen": -1.5574743747711182,
-      "logits/rejected": -1.3901411294937134,
-      "logps/chosen": -329.89141845703125,
-      "logps/rejected": -292.8751525878906,
-      "loss": 0.5342,
-      "rewards/accuracies": 0.753125011920929,
-      "rewards/chosen": 1.4129165410995483,
-      "rewards/margins": 1.3112914562225342,
-      "rewards/rejected": 0.10162514448165894,
+      "logits/chosen": -1.5503065586090088,
+      "logits/rejected": -1.381874680519104,
+      "logps/chosen": -329.953857421875,
+      "logps/rejected": -292.89447021484375,
+      "loss": 0.5681,
+      "rewards/accuracies": 0.7718750238418579,
+      "rewards/chosen": 1.7590471506118774,
+      "rewards/margins": 1.6621748208999634,
+      "rewards/rejected": 0.09687252342700958,
       "step": 340
     },
     {
       "epoch": 1.9,
-      "grad_norm": 1492.8399510840338,
+      "grad_norm": 2070.257116146559,
       "learning_rate": 7.279008199590543e-11,
-      "logits/chosen": -1.5503973960876465,
-      "logits/rejected": -1.3889100551605225,
-      "logps/chosen": -326.42120361328125,
-      "logps/rejected": -291.9585266113281,
-      "loss": 0.5261,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 1.3398044109344482,
-      "rewards/margins": 1.2421011924743652,
-      "rewards/rejected": 0.09770330041646957,
+      "logits/chosen": -1.541084885597229,
+      "logits/rejected": -1.378144383430481,
+      "logps/chosen": -326.51507568359375,
+      "logps/rejected": -291.95123291015625,
+      "loss": 0.5797,
+      "rewards/accuracies": 0.7281249761581421,
+      "rewards/chosen": 1.5987285375595093,
+      "rewards/margins": 1.4538962841033936,
+      "rewards/rejected": 0.1448323279619217,
       "step": 350
     },
     {
       "epoch": 1.96,
-      "grad_norm": 1452.281513333118,
+      "grad_norm": 2166.9724743285583,
       "learning_rate": 1.4406386978128017e-11,
-      "logits/chosen": -1.6207876205444336,
-      "logits/rejected": -1.424393653869629,
-      "logps/chosen": -331.06390380859375,
-      "logps/rejected": -291.6929626464844,
-      "loss": 0.5043,
-      "rewards/accuracies": 0.7906249761581421,
-      "rewards/chosen": 1.518845558166504,
-      "rewards/margins": 1.381410837173462,
-      "rewards/rejected": 0.13743488490581512,
+      "logits/chosen": -1.6209802627563477,
+      "logits/rejected": -1.42485511302948,
+      "logps/chosen": -331.2142639160156,
+      "logps/rejected": -291.69842529296875,
+      "loss": 0.5582,
+      "rewards/accuracies": 0.778124988079071,
+      "rewards/chosen": 1.724250078201294,
+      "rewards/margins": 1.5519315004348755,
+      "rewards/rejected": 0.1723184883594513,
       "step": 360
     },
     {
       "epoch": 2.0,
       "step": 368,
       "total_flos": 0.0,
-      "train_loss": 0.6161670185949492,
-      "train_runtime": 9955.6802,
-      "train_samples_per_second": 9.461,
-      "train_steps_per_second": 0.037
+      "train_loss": 0.26489045179408527,
+      "train_runtime": 4520.7899,
+      "train_samples_per_second": 20.835,
+      "train_steps_per_second": 0.081
     }
   ],
   "logging_steps": 10,