Model save

README.md

@@ -16,15 +16,15 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model was trained from scratch on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.5040
-- Rewards/chosen: -0.1466
-- Rewards/rejected: -1.1658
-- Rewards/accuracies: 0.7812
-- Rewards/margins: 1.0192
-- Logps/rejected: -288.3346
-- Logps/chosen: -274.6251
-- Logits/rejected: -1.2466
-- Logits/chosen: -1.2542
+- Loss: 0.5188
+- Rewards/chosen: 0.1167
+- Rewards/rejected: -0.8264
+- Rewards/accuracies: 0.7578
+- Rewards/margins: 0.9431
+- Logps/rejected: -282.8627
+- Logps/chosen: -272.6900
+- Logits/rejected: -0.7841
+- Logits/chosen: -0.8114
 
 ## Model description
 
@@ -61,7 +61,7 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch  | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
 |:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.4801        | 0.9984 | 477  | 0.5040          | -0.1466        | -1.1658          | 0.7812             | 1.0192          | -288.3346      | -274.6251    | -1.2466         | -1.2542       |
+| 0.4978        | 0.9984 | 477  | 0.5188          | 0.1167         | -0.8264          | 0.7578             | 0.9431          | -282.8627      | -272.6900    | -0.7841         | -0.8114       |
 
 
 ### Framework versions

all_results.json

@@ -1,9 +1,9 @@
 {
     "epoch": 0.9984301412872841,
     "total_flos": 0.0,
-    "train_loss": 0.5369631534352493,
-    "train_runtime": 17159.3245,
+    "train_loss": 0.5465342798572916,
+    "train_runtime": 16447.8444,
     "train_samples": 61134,
-    "train_samples_per_second": 3.563,
-    "train_steps_per_second": 0.028
+    "train_samples_per_second": 3.717,
+    "train_steps_per_second": 0.029
 }

train_results.json

@@ -1,9 +1,9 @@
 {
     "epoch": 0.9984301412872841,
     "total_flos": 0.0,
-    "train_loss": 0.5369631534352493,
-    "train_runtime": 17159.3245,
+    "train_loss": 0.5465342798572916,
+    "train_runtime": 16447.8444,
     "train_samples": 61134,
-    "train_samples_per_second": 3.563,
-    "train_steps_per_second": 0.028
+    "train_samples_per_second": 3.717,
+    "train_steps_per_second": 0.029
 }

trainer_state.json

@@ -10,12 +10,12 @@
   "log_history": [
     {
       "epoch": 0.0020931449502878076,
-      "grad_norm": 32.67887508688321,
+      "grad_norm": 33.010625672344894,
       "learning_rate": 1.0416666666666666e-08,
-      "logits/chosen": -1.3137571811676025,
-      "logits/rejected": -1.3155428171157837,
-      "logps/chosen": -358.085205078125,
-      "logps/rejected": -360.6844482421875,
+      "logits/chosen": -0.9353300929069519,
+      "logits/rejected": -0.9031922817230225,
+      "logps/chosen": -358.0216064453125,
+      "logps/rejected": -357.3989562988281,
       "loss": 0.6931,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
@@ -25,733 +25,733 @@
     },
     {
       "epoch": 0.020931449502878074,
-      "grad_norm": 31.917048945929817,
+      "grad_norm": 32.45260821994328,
       "learning_rate": 1.0416666666666667e-07,
-      "logits/chosen": -1.2557947635650635,
-      "logits/rejected": -1.2471075057983398,
-      "logps/chosen": -299.4416198730469,
-      "logps/rejected": -277.6274108886719,
-      "loss": 0.6917,
-      "rewards/accuracies": 0.5347222089767456,
-      "rewards/chosen": 0.001689681550487876,
-      "rewards/margins": 0.013649268075823784,
-      "rewards/rejected": -0.011959584429860115,
+      "logits/chosen": -0.8636745810508728,
+      "logits/rejected": -0.8385691046714783,
+      "logps/chosen": -300.0736999511719,
+      "logps/rejected": -275.7908020019531,
+      "loss": 0.694,
+      "rewards/accuracies": 0.4236111044883728,
+      "rewards/chosen": 0.0011882925173267722,
+      "rewards/margins": -0.004837921820580959,
+      "rewards/rejected": 0.006026213988661766,
       "step": 10
     },
     {
       "epoch": 0.04186289900575615,
-      "grad_norm": 29.66358436990106,
+      "grad_norm": 30.077765563410388,
       "learning_rate": 2.0833333333333333e-07,
-      "logits/chosen": -1.219057321548462,
-      "logits/rejected": -1.2197545766830444,
-      "logps/chosen": -311.90618896484375,
-      "logps/rejected": -280.2146301269531,
-      "loss": 0.691,
-      "rewards/accuracies": 0.543749988079071,
-      "rewards/chosen": 0.01786460541188717,
-      "rewards/margins": 0.016024405136704445,
-      "rewards/rejected": 0.0018402017885819077,
+      "logits/chosen": -0.8328930139541626,
+      "logits/rejected": -0.8021430969238281,
+      "logps/chosen": -312.682861328125,
+      "logps/rejected": -277.70367431640625,
+      "loss": 0.6921,
+      "rewards/accuracies": 0.5062500238418579,
+      "rewards/chosen": -0.00014361817738972604,
+      "rewards/margins": -0.0021410987246781588,
+      "rewards/rejected": 0.0019974808674305677,
       "step": 20
     },
     {
       "epoch": 0.06279434850863422,
-      "grad_norm": 29.816640467315747,
+      "grad_norm": 30.524247060500045,
       "learning_rate": 3.1249999999999997e-07,
-      "logits/chosen": -1.2372534275054932,
-      "logits/rejected": -1.2391879558563232,
-      "logps/chosen": -289.08367919921875,
-      "logps/rejected": -249.3250274658203,
-      "loss": 0.6854,
+      "logits/chosen": -0.8413776159286499,
+      "logits/rejected": -0.8217487335205078,
+      "logps/chosen": -288.83856201171875,
+      "logps/rejected": -247.5811309814453,
+      "loss": 0.687,
       "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.06472653895616531,
-      "rewards/margins": 0.013585369102656841,
-      "rewards/rejected": 0.051141172647476196,
+      "rewards/chosen": 0.03682413697242737,
+      "rewards/margins": 0.01591811515390873,
+      "rewards/rejected": 0.02090602181851864,
       "step": 30
     },
     {
       "epoch": 0.0837257980115123,
-      "grad_norm": 28.843556314584543,
+      "grad_norm": 29.759026915242178,
       "learning_rate": 4.1666666666666667e-07,
-      "logits/chosen": -1.2231731414794922,
-      "logits/rejected": -1.2176843881607056,
-      "logps/chosen": -257.4178161621094,
-      "logps/rejected": -259.80511474609375,
-      "loss": 0.6642,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.20605416595935822,
-      "rewards/margins": 0.0631551519036293,
-      "rewards/rejected": 0.14289899170398712,
+      "logits/chosen": -0.8259631395339966,
+      "logits/rejected": -0.7971775531768799,
+      "logps/chosen": -257.2466735839844,
+      "logps/rejected": -258.15093994140625,
+      "loss": 0.6662,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.08521110564470291,
+      "rewards/margins": 0.06521101295948029,
+      "rewards/rejected": 0.02000008150935173,
       "step": 40
     },
     {
       "epoch": 0.10465724751439037,
-      "grad_norm": 27.177864701762527,
+      "grad_norm": 28.40954484218416,
       "learning_rate": 4.999731868769026e-07,
-      "logits/chosen": -1.2115309238433838,
-      "logits/rejected": -1.2084529399871826,
-      "logps/chosen": -284.76275634765625,
-      "logps/rejected": -270.4881591796875,
-      "loss": 0.6374,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": 0.3667793869972229,
-      "rewards/margins": 0.15095432102680206,
-      "rewards/rejected": 0.21582508087158203,
+      "logits/chosen": -0.8126314282417297,
+      "logits/rejected": -0.7864682078361511,
+      "logps/chosen": -286.5724182128906,
+      "logps/rejected": -271.7125244140625,
+      "loss": 0.636,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.18741461634635925,
+      "rewards/margins": 0.14045223593711853,
+      "rewards/rejected": 0.04696240276098251,
       "step": 50
     },
     {
       "epoch": 0.12558869701726844,
-      "grad_norm": 26.377177341299028,
+      "grad_norm": 26.575331400959886,
       "learning_rate": 4.990353313429303e-07,
-      "logits/chosen": -1.2306678295135498,
-      "logits/rejected": -1.2262895107269287,
-      "logps/chosen": -250.672119140625,
-      "logps/rejected": -242.7108917236328,
-      "loss": 0.6075,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": 0.5981600284576416,
-      "rewards/margins": 0.28279852867126465,
-      "rewards/rejected": 0.31536155939102173,
+      "logits/chosen": -0.8289661407470703,
+      "logits/rejected": -0.8004046678543091,
+      "logps/chosen": -252.25851440429688,
+      "logps/rejected": -243.3365478515625,
+      "loss": 0.6094,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": 0.3671998381614685,
+      "rewards/margins": 0.27907320857048035,
+      "rewards/rejected": 0.08812658488750458,
       "step": 60
     },
     {
       "epoch": 0.14652014652014653,
-      "grad_norm": 26.946200636877062,
+      "grad_norm": 29.71245855134311,
       "learning_rate": 4.967625656594781e-07,
-      "logits/chosen": -1.2291628122329712,
-      "logits/rejected": -1.2251144647598267,
-      "logps/chosen": -291.495849609375,
-      "logps/rejected": -280.19140625,
-      "loss": 0.5843,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.5898247957229614,
-      "rewards/margins": 0.3466072976589203,
-      "rewards/rejected": 0.24321749806404114,
+      "logits/chosen": -0.8133388757705688,
+      "logits/rejected": -0.7937413454055786,
+      "logps/chosen": -292.16748046875,
+      "logps/rejected": -281.007080078125,
+      "loss": 0.59,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.4147632122039795,
+      "rewards/margins": 0.36940422654151917,
+      "rewards/rejected": 0.04535900428891182,
       "step": 70
     },
     {
       "epoch": 0.1674515960230246,
-      "grad_norm": 30.93097491171412,
+      "grad_norm": 30.87247601783358,
       "learning_rate": 4.93167072587771e-07,
-      "logits/chosen": -1.2335067987442017,
-      "logits/rejected": -1.2280460596084595,
-      "logps/chosen": -322.7510681152344,
-      "logps/rejected": -254.08969116210938,
-      "loss": 0.5751,
-      "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": 0.5560204386711121,
-      "rewards/margins": 0.521833598613739,
-      "rewards/rejected": 0.03418678790330887,
+      "logits/chosen": -0.8584713935852051,
+      "logits/rejected": -0.793620765209198,
+      "logps/chosen": -323.61346435546875,
+      "logps/rejected": -252.2790069580078,
+      "loss": 0.5816,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": 0.4810527265071869,
+      "rewards/margins": 0.5530245900154114,
+      "rewards/rejected": -0.0719718337059021,
       "step": 80
     },
     {
       "epoch": 0.18838304552590268,
-      "grad_norm": 24.87576635994796,
+      "grad_norm": 28.471314182224006,
       "learning_rate": 4.882681251368548e-07,
-      "logits/chosen": -1.2730495929718018,
-      "logits/rejected": -1.2472363710403442,
-      "logps/chosen": -251.80154418945312,
-      "logps/rejected": -260.86273193359375,
-      "loss": 0.5494,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": 0.3186618685722351,
-      "rewards/margins": 0.5627286434173584,
-      "rewards/rejected": -0.2440667599439621,
+      "logits/chosen": -0.8546805381774902,
+      "logits/rejected": -0.81480872631073,
+      "logps/chosen": -250.3916015625,
+      "logps/rejected": -258.33978271484375,
+      "loss": 0.5539,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.38906946778297424,
+      "rewards/margins": 0.5449206233024597,
+      "rewards/rejected": -0.15585120022296906,
       "step": 90
     },
     {
       "epoch": 0.20931449502878074,
-      "grad_norm": 22.700829979602176,
+      "grad_norm": 24.327372745149354,
       "learning_rate": 4.820919832540181e-07,
-      "logits/chosen": -1.2834389209747314,
-      "logits/rejected": -1.2922523021697998,
-      "logps/chosen": -297.82958984375,
-      "logps/rejected": -286.1127014160156,
-      "loss": 0.5503,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": 0.20889770984649658,
-      "rewards/margins": 0.736894965171814,
-      "rewards/rejected": -0.5279972553253174,
+      "logits/chosen": -0.8973363041877747,
+      "logits/rejected": -0.8621295094490051,
+      "logps/chosen": -294.51544189453125,
+      "logps/rejected": -281.0751647949219,
+      "loss": 0.5545,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": 0.4683733582496643,
+      "rewards/margins": 0.7225647568702698,
+      "rewards/rejected": -0.2541913688182831,
       "step": 100
     },
     {
       "epoch": 0.2302459445316588,
-      "grad_norm": 25.7861610694365,
+      "grad_norm": 28.090520904051154,
       "learning_rate": 4.7467175306295647e-07,
-      "logits/chosen": -1.252528429031372,
-      "logits/rejected": -1.2519152164459229,
-      "logps/chosen": -287.58612060546875,
-      "logps/rejected": -277.14947509765625,
-      "loss": 0.5699,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": 0.10858204215765,
-      "rewards/margins": 0.6252473592758179,
-      "rewards/rejected": -0.5166653394699097,
+      "logits/chosen": -0.8554052114486694,
+      "logits/rejected": -0.8270560503005981,
+      "logps/chosen": -285.19793701171875,
+      "logps/rejected": -274.04998779296875,
+      "loss": 0.5708,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": 0.32352814078330994,
+      "rewards/margins": 0.6342454552650452,
+      "rewards/rejected": -0.31071722507476807,
       "step": 110
     },
     {
       "epoch": 0.25117739403453687,
-      "grad_norm": 25.229488240618327,
+      "grad_norm": 27.328358131262704,
       "learning_rate": 4.6604720940421207e-07,
-      "logits/chosen": -1.1948816776275635,
-      "logits/rejected": -1.2172932624816895,
-      "logps/chosen": -266.800537109375,
-      "logps/rejected": -268.51800537109375,
-      "loss": 0.5436,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.12671348452568054,
-      "rewards/margins": 0.6409411430358887,
-      "rewards/rejected": -0.5142276287078857,
+      "logits/chosen": -0.7863410711288452,
+      "logits/rejected": -0.7909287810325623,
+      "logps/chosen": -265.45599365234375,
+      "logps/rejected": -265.26190185546875,
+      "loss": 0.5456,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.25525203347206116,
+      "rewards/margins": 0.6262764930725098,
+      "rewards/rejected": -0.3710244596004486,
       "step": 120
     },
     {
       "epoch": 0.272108843537415,
-      "grad_norm": 24.47701447808746,
+      "grad_norm": 28.030552921340494,
       "learning_rate": 4.5626458262912735e-07,
-      "logits/chosen": -1.230271339416504,
-      "logits/rejected": -1.2303745746612549,
-      "logps/chosen": -278.7350158691406,
-      "logps/rejected": -277.00469970703125,
-      "loss": 0.525,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.17859216034412384,
-      "rewards/margins": 0.7014043927192688,
-      "rewards/rejected": -0.5228122472763062,
+      "logits/chosen": -0.8188163638114929,
+      "logits/rejected": -0.7965669631958008,
+      "logps/chosen": -276.0892333984375,
+      "logps/rejected": -273.39666748046875,
+      "loss": 0.5299,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": 0.33988282084465027,
+      "rewards/margins": 0.655505895614624,
+      "rewards/rejected": -0.31562304496765137,
       "step": 130
     },
     {
       "epoch": 0.29304029304029305,
-      "grad_norm": 31.64152947602703,
+      "grad_norm": 42.83275029235473,
       "learning_rate": 4.453763107901675e-07,
-      "logits/chosen": -1.2872731685638428,
-      "logits/rejected": -1.2797777652740479,
-      "logps/chosen": -316.21807861328125,
-      "logps/rejected": -273.5144958496094,
-      "loss": 0.531,
+      "logits/chosen": -0.8900020718574524,
+      "logits/rejected": -0.8431411981582642,
+      "logps/chosen": -315.87835693359375,
+      "logps/rejected": -269.9021911621094,
+      "loss": 0.5391,
       "rewards/accuracies": 0.78125,
-      "rewards/chosen": 0.3403511643409729,
-      "rewards/margins": 0.9147707223892212,
-      "rewards/rejected": -0.5744196176528931,
+      "rewards/chosen": 0.45897454023361206,
+      "rewards/margins": 0.8581539988517761,
+      "rewards/rejected": -0.39917951822280884,
       "step": 140
     },
     {
       "epoch": 0.3139717425431711,
-      "grad_norm": 28.473493709163563,
+      "grad_norm": 30.67114502558656,
       "learning_rate": 4.3344075855595097e-07,
-      "logits/chosen": -1.2500557899475098,
-      "logits/rejected": -1.2551041841506958,
-      "logps/chosen": -279.05126953125,
-      "logps/rejected": -254.45742797851562,
-      "loss": 0.5396,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.07857184112071991,
-      "rewards/margins": 0.7842037677764893,
-      "rewards/rejected": -0.7056319117546082,
+      "logits/chosen": -0.8563919067382812,
+      "logits/rejected": -0.8316940069198608,
+      "logps/chosen": -276.3800354003906,
+      "logps/rejected": -249.8993682861328,
+      "loss": 0.5427,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": 0.2782554030418396,
+      "rewards/margins": 0.7451222538948059,
+      "rewards/rejected": -0.4668668210506439,
       "step": 150
     },
     {
       "epoch": 0.3349031920460492,
-      "grad_norm": 24.237082918392105,
+      "grad_norm": 25.734598107143015,
       "learning_rate": 4.2052190435769554e-07,
-      "logits/chosen": -1.2701295614242554,
-      "logits/rejected": -1.2917912006378174,
-      "logps/chosen": -276.0731201171875,
-      "logps/rejected": -260.68072509765625,
-      "loss": 0.5335,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.2085450440645218,
-      "rewards/margins": 0.8893882036209106,
-      "rewards/rejected": -0.6808432340621948,
+      "logits/chosen": -0.8762856721878052,
+      "logits/rejected": -0.8501691818237305,
+      "logps/chosen": -275.22332763671875,
+      "logps/rejected": -256.7464294433594,
+      "loss": 0.5468,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": 0.3591170907020569,
+      "rewards/margins": 0.861784815788269,
+      "rewards/rejected": -0.5026678442955017,
       "step": 160
     },
     {
       "epoch": 0.35583464154892724,
-      "grad_norm": 25.58214081499019,
+      "grad_norm": 26.617553811529366,
       "learning_rate": 4.0668899744407567e-07,
-      "logits/chosen": -1.2040237188339233,
-      "logits/rejected": -1.2073538303375244,
-      "logps/chosen": -258.0507507324219,
-      "logps/rejected": -239.2794647216797,
-      "loss": 0.5381,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": 0.22140569984912872,
-      "rewards/margins": 0.6377298831939697,
-      "rewards/rejected": -0.41632407903671265,
+      "logits/chosen": -0.7889877557754517,
+      "logits/rejected": -0.7687152028083801,
+      "logps/chosen": -257.8996276855469,
+      "logps/rejected": -236.44815063476562,
+      "loss": 0.5372,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.3476775288581848,
+      "rewards/margins": 0.6377132534980774,
+      "rewards/rejected": -0.29003575444221497,
       "step": 170
     },
     {
       "epoch": 0.37676609105180536,
-      "grad_norm": 24.563960047050507,
+      "grad_norm": 24.425125498336687,
       "learning_rate": 3.920161866827889e-07,
-      "logits/chosen": -1.2623085975646973,
-      "logits/rejected": -1.2519021034240723,
-      "logps/chosen": -269.4590759277344,
-      "logps/rejected": -253.0479736328125,
-      "loss": 0.5109,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": 0.21817712485790253,
-      "rewards/margins": 0.7618513107299805,
-      "rewards/rejected": -0.5436741709709167,
+      "logits/chosen": -0.8434957265853882,
+      "logits/rejected": -0.8080987930297852,
+      "logps/chosen": -266.18658447265625,
+      "logps/rejected": -249.037353515625,
+      "loss": 0.5124,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.40633445978164673,
+      "rewards/margins": 0.7494776844978333,
+      "rewards/rejected": -0.3431432247161865,
       "step": 180
     },
     {
       "epoch": 0.3976975405546834,
-      "grad_norm": 27.2347337384506,
+      "grad_norm": 30.662514514517564,
       "learning_rate": 3.765821230985757e-07,
-      "logits/chosen": -1.2562354803085327,
-      "logits/rejected": -1.2483447790145874,
-      "logps/chosen": -263.9108581542969,
-      "logps/rejected": -266.5344543457031,
-      "loss": 0.5268,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.125687837600708,
-      "rewards/margins": 0.643466591835022,
-      "rewards/rejected": -0.517778754234314,
+      "logits/chosen": -0.8309434652328491,
+      "logits/rejected": -0.8197731971740723,
+      "logps/chosen": -262.6062927246094,
+      "logps/rejected": -264.48297119140625,
+      "loss": 0.5371,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.33688968420028687,
+      "rewards/margins": 0.6359924077987671,
+      "rewards/rejected": -0.2991027534008026,
       "step": 190
     },
     {
       "epoch": 0.4186289900575615,
-      "grad_norm": 119.61156679155792,
+      "grad_norm": 72.1247270129297,
       "learning_rate": 3.604695382782159e-07,
-      "logits/chosen": -1.2198866605758667,
-      "logits/rejected": -1.235428810119629,
-      "logps/chosen": -251.72509765625,
-      "logps/rejected": -276.9458923339844,
-      "loss": 0.5468,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.04041861370205879,
-      "rewards/margins": 0.46673449873924255,
-      "rewards/rejected": -0.5071530938148499,
+      "logits/chosen": -0.7932894229888916,
+      "logits/rejected": -0.8034732937812805,
+      "logps/chosen": -247.0430450439453,
+      "logps/rejected": -273.28717041015625,
+      "loss": 0.5509,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.2075834572315216,
+      "rewards/margins": 0.5311359167098999,
+      "rewards/rejected": -0.3235524296760559,
       "step": 200
     },
     {
       "epoch": 0.43956043956043955,
-      "grad_norm": 28.967310114309072,
+      "grad_norm": 30.791266268313425,
       "learning_rate": 3.4376480090239047e-07,
-      "logits/chosen": -1.2494560480117798,
-      "logits/rejected": -1.2435802221298218,
-      "logps/chosen": -309.4058837890625,
-      "logps/rejected": -264.42999267578125,
-      "loss": 0.5412,
+      "logits/chosen": -0.8500334620475769,
+      "logits/rejected": -0.7941908240318298,
+      "logps/chosen": -307.06243896484375,
+      "logps/rejected": -260.5225830078125,
+      "loss": 0.5554,
       "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.18436923623085022,
-      "rewards/margins": 0.7689916491508484,
-      "rewards/rejected": -0.5846225023269653,
+      "rewards/chosen": 0.3245794475078583,
+      "rewards/margins": 0.7126837968826294,
+      "rewards/rejected": -0.38810431957244873,
       "step": 210
     },
     {
       "epoch": 0.4604918890633176,
-      "grad_norm": 26.092643646706687,
+      "grad_norm": 28.4806594774881,
       "learning_rate": 3.265574537815398e-07,
-      "logits/chosen": -1.201503038406372,
-      "logits/rejected": -1.2363094091415405,
-      "logps/chosen": -232.6184844970703,
-      "logps/rejected": -249.8125,
-      "loss": 0.5175,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.033870868384838104,
-      "rewards/margins": 0.8461370468139648,
-      "rewards/rejected": -0.8122660517692566,
+      "logits/chosen": -0.7605562210083008,
+      "logits/rejected": -0.7752619385719299,
+      "logps/chosen": -230.6273956298828,
+      "logps/rejected": -246.4652862548828,
+      "loss": 0.5332,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.20168304443359375,
+      "rewards/margins": 0.7937971949577332,
+      "rewards/rejected": -0.5921142101287842,
       "step": 220
     },
     {
       "epoch": 0.48142333856619574,
-      "grad_norm": 28.1867683230448,
+      "grad_norm": 30.177561277957544,
       "learning_rate": 3.0893973387735683e-07,
-      "logits/chosen": -1.2477219104766846,
-      "logits/rejected": -1.2469167709350586,
-      "logps/chosen": -257.1601867675781,
-      "logps/rejected": -246.19308471679688,
-      "loss": 0.5214,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": 0.1131761223077774,
-      "rewards/margins": 1.0220317840576172,
-      "rewards/rejected": -0.9088556170463562,
+      "logits/chosen": -0.8394104838371277,
+      "logits/rejected": -0.8031412363052368,
+      "logps/chosen": -254.2725372314453,
+      "logps/rejected": -240.7719268798828,
+      "loss": 0.5367,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.29248276352882385,
+      "rewards/margins": 0.8323594927787781,
+      "rewards/rejected": -0.5398766994476318,
       "step": 230
     },
     {
       "epoch": 0.5023547880690737,
-      "grad_norm": 25.40371570243187,
+      "grad_norm": 27.13662842488361,
       "learning_rate": 2.910060778827554e-07,
-      "logits/chosen": -1.249792456626892,
-      "logits/rejected": -1.225212812423706,
-      "logps/chosen": -284.3082580566406,
-      "logps/rejected": -275.9374694824219,
-      "loss": 0.4936,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -0.03040931187570095,
-      "rewards/margins": 0.9342262148857117,
-      "rewards/rejected": -0.9646355509757996,
+      "logits/chosen": -0.8476204872131348,
+      "logits/rejected": -0.7989686131477356,
+      "logps/chosen": -281.7205505371094,
+      "logps/rejected": -270.6007385253906,
+      "loss": 0.506,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.29095786809921265,
+      "rewards/margins": 0.8068556785583496,
+      "rewards/rejected": -0.5158977508544922,
       "step": 240
     },
     {
       "epoch": 0.5232862375719518,
-      "grad_norm": 27.87875323309372,
+      "grad_norm": 28.003194585077747,
       "learning_rate": 2.7285261601056697e-07,
-      "logits/chosen": -1.2633453607559204,
-      "logits/rejected": -1.2385962009429932,
-      "logps/chosen": -281.73394775390625,
-      "logps/rejected": -253.89901733398438,
-      "loss": 0.511,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -0.20308038592338562,
-      "rewards/margins": 1.0585527420043945,
-      "rewards/rejected": -1.261633276939392,
+      "logits/chosen": -0.8499590158462524,
+      "logits/rejected": -0.7796657085418701,
+      "logps/chosen": -277.3616638183594,
+      "logps/rejected": -247.3683624267578,
+      "loss": 0.5216,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": 0.2743334174156189,
+      "rewards/margins": 0.980247974395752,
+      "rewards/rejected": -0.7059146165847778,
       "step": 250
     },
     {
       "epoch": 0.54421768707483,
-      "grad_norm": 25.946250812747735,
+      "grad_norm": 28.400347125832315,
       "learning_rate": 2.5457665670441937e-07,
-      "logits/chosen": -1.3085938692092896,
-      "logits/rejected": -1.3239829540252686,
-      "logps/chosen": -263.26397705078125,
-      "logps/rejected": -255.79824829101562,
-      "loss": 0.5073,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.36238956451416016,
-      "rewards/margins": 0.8067954182624817,
-      "rewards/rejected": -1.1691850423812866,
+      "logits/chosen": -0.88239985704422,
+      "logits/rejected": -0.879532516002655,
+      "logps/chosen": -256.6146545410156,
+      "logps/rejected": -248.2431640625,
+      "loss": 0.5235,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": 0.15996988117694855,
+      "rewards/margins": 0.6795960664749146,
+      "rewards/rejected": -0.5196262001991272,
       "step": 260
     },
     {
       "epoch": 0.565149136577708,
-      "grad_norm": 25.43204987268852,
+      "grad_norm": 24.336672887905703,
       "learning_rate": 2.3627616503391812e-07,
-      "logits/chosen": -1.259594202041626,
-      "logits/rejected": -1.2623211145401,
-      "logps/chosen": -301.30059814453125,
-      "logps/rejected": -282.37060546875,
-      "loss": 0.5244,
+      "logits/chosen": -0.8440641164779663,
+      "logits/rejected": -0.8187896013259888,
+      "logps/chosen": -296.5509948730469,
+      "logps/rejected": -274.3413391113281,
+      "loss": 0.5245,
       "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -0.22281932830810547,
-      "rewards/margins": 1.033884882926941,
-      "rewards/rejected": -1.2567042112350464,
+      "rewards/chosen": 0.28971901535987854,
+      "rewards/margins": 0.9396535754203796,
+      "rewards/rejected": -0.6499345898628235,
       "step": 270
     },
     {
       "epoch": 0.5860805860805861,
-      "grad_norm": 32.334489920343785,
+      "grad_norm": 33.18965003358362,
       "learning_rate": 2.1804923757009882e-07,
-      "logits/chosen": -1.2075401544570923,
-      "logits/rejected": -1.2296500205993652,
-      "logps/chosen": -259.00225830078125,
-      "logps/rejected": -242.996826171875,
-      "loss": 0.5159,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": -0.18557050824165344,
-      "rewards/margins": 0.8612161874771118,
-      "rewards/rejected": -1.0467866659164429,
+      "logits/chosen": -0.7617496252059937,
+      "logits/rejected": -0.753891110420227,
+      "logps/chosen": -254.390625,
+      "logps/rejected": -235.6627960205078,
+      "loss": 0.5353,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.18729178607463837,
+      "rewards/margins": 0.7195244431495667,
+      "rewards/rejected": -0.5322326421737671,
       "step": 280
     },
     {
       "epoch": 0.6070120355834642,
-      "grad_norm": 26.55876737503656,
+      "grad_norm": 28.187492485886555,
       "learning_rate": 1.9999357655598891e-07,
-      "logits/chosen": -1.2683426141738892,
-      "logits/rejected": -1.2434879541397095,
-      "logps/chosen": -250.16262817382812,
-      "logps/rejected": -262.10491943359375,
-      "loss": 0.515,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -0.2073489874601364,
-      "rewards/margins": 0.8695666193962097,
-      "rewards/rejected": -1.0769156217575073,
+      "logits/chosen": -0.8251991271972656,
+      "logits/rejected": -0.7897046208381653,
+      "logps/chosen": -245.2775115966797,
+      "logps/rejected": -255.91702270507812,
+      "loss": 0.5273,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": 0.156294047832489,
+      "rewards/margins": 0.7444198131561279,
+      "rewards/rejected": -0.5881257653236389,
       "step": 290
     },
     {
       "epoch": 0.6279434850863422,
-      "grad_norm": 25.006333507207025,
+      "grad_norm": 27.921024679648337,
       "learning_rate": 1.8220596619089573e-07,
-      "logits/chosen": -1.234788179397583,
-      "logits/rejected": -1.2198808193206787,
-      "logps/chosen": -328.10076904296875,
-      "logps/rejected": -289.6324157714844,
-      "loss": 0.4989,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -0.14694449305534363,
-      "rewards/margins": 0.8379470705986023,
-      "rewards/rejected": -0.9848915338516235,
+      "logits/chosen": -0.8231035470962524,
+      "logits/rejected": -0.7743059396743774,
+      "logps/chosen": -325.2027282714844,
+      "logps/rejected": -283.76141357421875,
+      "loss": 0.5098,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": 0.2244129180908203,
+      "rewards/margins": 0.7570952773094177,
+      "rewards/rejected": -0.5326822996139526,
       "step": 300
     },
     {
       "epoch": 0.6488749345892203,
-      "grad_norm": 28.054496962517977,
+      "grad_norm": 25.528912079564087,
       "learning_rate": 1.647817538357072e-07,
-      "logits/chosen": -1.2231569290161133,
-      "logits/rejected": -1.2060788869857788,
-      "logps/chosen": -307.0669860839844,
-      "logps/rejected": -263.49468994140625,
-      "loss": 0.5027,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -0.13182426989078522,
-      "rewards/margins": 0.9834961891174316,
-      "rewards/rejected": -1.1153204441070557,
+      "logits/chosen": -0.8117376565933228,
+      "logits/rejected": -0.7634798884391785,
+      "logps/chosen": -304.58367919921875,
+      "logps/rejected": -258.6290588378906,
+      "loss": 0.52,
+      "rewards/accuracies": 0.768750011920929,
+      "rewards/chosen": 0.1614515334367752,
+      "rewards/margins": 0.9296430349349976,
+      "rewards/rejected": -0.7681915163993835,
       "step": 310
     },
     {
       "epoch": 0.6698063840920984,
-      "grad_norm": 29.07624093072449,
+      "grad_norm": 28.932687234297475,
       "learning_rate": 1.478143389201113e-07,
-      "logits/chosen": -1.2904080152511597,
-      "logits/rejected": -1.2708590030670166,
-      "logps/chosen": -258.17510986328125,
-      "logps/rejected": -237.592041015625,
-      "loss": 0.483,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -0.0982074961066246,
-      "rewards/margins": 0.9570711851119995,
-      "rewards/rejected": -1.0552786588668823,
+      "logits/chosen": -0.8632524609565735,
+      "logits/rejected": -0.8123928308486938,
+      "logps/chosen": -253.69540405273438,
+      "logps/rejected": -231.25924682617188,
+      "loss": 0.4989,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": 0.1885661631822586,
+      "rewards/margins": 0.8719049692153931,
+      "rewards/rejected": -0.6833387613296509,
       "step": 320
     },
     {
       "epoch": 0.6907378335949764,
-      "grad_norm": 26.27198987673298,
+      "grad_norm": 26.099734467633418,
       "learning_rate": 1.3139467229135998e-07,
-      "logits/chosen": -1.2492620944976807,
-      "logits/rejected": -1.249738335609436,
-      "logps/chosen": -276.73968505859375,
-      "logps/rejected": -278.4327392578125,
-      "loss": 0.4973,
-      "rewards/accuracies": 0.800000011920929,
-      "rewards/chosen": -0.047559954226017,
-      "rewards/margins": 0.8613998293876648,
-      "rewards/rejected": -0.9089597463607788,
+      "logits/chosen": -0.8447322845458984,
+      "logits/rejected": -0.8383687734603882,
+      "logps/chosen": -272.8449401855469,
+      "logps/rejected": -272.4441833496094,
+      "loss": 0.515,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.23482713103294373,
+      "rewards/margins": 0.7712266445159912,
+      "rewards/rejected": -0.5363994836807251,
       "step": 330
     },
     {
       "epoch": 0.7116692830978545,
-      "grad_norm": 30.87966852792751,
+      "grad_norm": 33.44802483538591,
       "learning_rate": 1.1561076868822755e-07,
-      "logits/chosen": -1.1826813220977783,
-      "logits/rejected": -1.193650484085083,
-      "logps/chosen": -297.5704345703125,
-      "logps/rejected": -278.57781982421875,
-      "loss": 0.5121,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -0.20869632065296173,
-      "rewards/margins": 0.8355530500411987,
-      "rewards/rejected": -1.044249415397644,
+      "logits/chosen": -0.7767515778541565,
+      "logits/rejected": -0.7614132165908813,
+      "logps/chosen": -295.1501159667969,
+      "logps/rejected": -273.54913330078125,
+      "loss": 0.5175,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.06397654116153717,
+      "rewards/margins": 0.7476279139518738,
+      "rewards/rejected": -0.683651328086853,
       "step": 340
     },
     {
       "epoch": 0.7326007326007326,
-      "grad_norm": 29.877476905351152,
+      "grad_norm": 30.17255483577115,
       "learning_rate": 1.0054723495346482e-07,
-      "logits/chosen": -1.2712528705596924,
-      "logits/rejected": -1.2707802057266235,
-      "logps/chosen": -263.9284362792969,
-      "logps/rejected": -245.35061645507812,
-      "loss": 0.4838,
-      "rewards/accuracies": 0.8187500238418579,
-      "rewards/chosen": -0.14454463124275208,
-      "rewards/margins": 1.062632441520691,
-      "rewards/rejected": -1.2071770429611206,
+      "logits/chosen": -0.8592581748962402,
+      "logits/rejected": -0.8231215476989746,
+      "logps/chosen": -259.3691101074219,
+      "logps/rejected": -239.63229370117188,
+      "loss": 0.4965,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.17322279512882233,
+      "rewards/margins": 0.9676720499992371,
+      "rewards/rejected": -0.7944492101669312,
       "step": 350
     },
     {
       "epoch": 0.7535321821036107,
-      "grad_norm": 25.334588017324133,
+      "grad_norm": 26.224162546986935,
       "learning_rate": 8.628481651367875e-08,
-      "logits/chosen": -1.2483649253845215,
-      "logits/rejected": -1.2495536804199219,
-      "logps/chosen": -314.11920166015625,
-      "logps/rejected": -283.3104248046875,
-      "loss": 0.5272,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -0.16097046434879303,
-      "rewards/margins": 0.9746401906013489,
-      "rewards/rejected": -1.1356106996536255,
+      "logits/chosen": -0.8377500772476196,
+      "logits/rejected": -0.7995637655258179,
+      "logps/chosen": -312.7345275878906,
+      "logps/rejected": -277.9017333984375,
+      "loss": 0.5325,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": 0.13472691178321838,
+      "rewards/margins": 0.9044864773750305,
+      "rewards/rejected": -0.7697595953941345,
       "step": 360
     },
     {
       "epoch": 0.7744636316064888,
-      "grad_norm": 28.708071038705555,
+      "grad_norm": 29.675575495805287,
       "learning_rate": 7.289996455765748e-08,
-      "logits/chosen": -1.296809434890747,
-      "logits/rejected": -1.2918803691864014,
-      "logps/chosen": -264.4002990722656,
-      "logps/rejected": -242.81118774414062,
-      "loss": 0.5196,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.11513688415288925,
-      "rewards/margins": 0.881182849407196,
-      "rewards/rejected": -0.9963196516036987,
+      "logits/chosen": -0.878030002117157,
+      "logits/rejected": -0.8434737920761108,
+      "logps/chosen": -261.632080078125,
+      "logps/rejected": -238.229248046875,
+      "loss": 0.5339,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": 0.07875274866819382,
+      "rewards/margins": 0.7951083183288574,
+      "rewards/rejected": -0.7163556218147278,
       "step": 370
     },
     {
       "epoch": 0.7953950811093669,
-      "grad_norm": 23.458380321356813,
+      "grad_norm": 24.60988736745974,
       "learning_rate": 6.046442623320145e-08,
-      "logits/chosen": -1.2197574377059937,
-      "logits/rejected": -1.2310526371002197,
-      "logps/chosen": -249.80551147460938,
-      "logps/rejected": -277.2812194824219,
-      "loss": 0.5089,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -0.4139748513698578,
-      "rewards/margins": 1.0405232906341553,
-      "rewards/rejected": -1.454498291015625,
+      "logits/chosen": -0.7838923335075378,
+      "logits/rejected": -0.7837015986442566,
+      "logps/chosen": -245.961181640625,
+      "logps/rejected": -272.2265930175781,
+      "loss": 0.5217,
+      "rewards/accuracies": 0.7875000238418579,
+      "rewards/chosen": -0.14545968174934387,
+      "rewards/margins": 0.9610022306442261,
+      "rewards/rejected": -1.106461763381958,
       "step": 380
     },
     {
       "epoch": 0.8163265306122449,
-      "grad_norm": 36.634474237425266,
+      "grad_norm": 34.79197368528132,
       "learning_rate": 4.904486005914027e-08,
-      "logits/chosen": -1.2632532119750977,
-      "logits/rejected": -1.229184865951538,
-      "logps/chosen": -337.5649719238281,
-      "logps/rejected": -315.7585754394531,
-      "loss": 0.4854,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -0.16942563652992249,
-      "rewards/margins": 0.9667377471923828,
-      "rewards/rejected": -1.1361634731292725,
+      "logits/chosen": -0.8651834726333618,
+      "logits/rejected": -0.804705023765564,
+      "logps/chosen": -333.53594970703125,
+      "logps/rejected": -309.1046447753906,
+      "loss": 0.5025,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": 0.1259995996952057,
+      "rewards/margins": 0.8870970010757446,
+      "rewards/rejected": -0.7610973715782166,
       "step": 390
     },
     {
       "epoch": 0.837257980115123,
-      "grad_norm": 27.997965293261185,
+      "grad_norm": 29.60112890968661,
       "learning_rate": 3.8702478614051345e-08,
-      "logits/chosen": -1.2391654253005981,
-      "logits/rejected": -1.2487369775772095,
-      "logps/chosen": -265.0726623535156,
-      "logps/rejected": -252.5985870361328,
-      "loss": 0.5018,
-      "rewards/accuracies": 0.7875000238418579,
-      "rewards/chosen": -0.21206898987293243,
-      "rewards/margins": 0.9942883253097534,
-      "rewards/rejected": -1.2063573598861694,
+      "logits/chosen": -0.8112460970878601,
+      "logits/rejected": -0.7928680181503296,
+      "logps/chosen": -261.1005859375,
+      "logps/rejected": -247.94161987304688,
+      "loss": 0.509,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": 0.09377719461917877,
+      "rewards/margins": 0.9371738433837891,
+      "rewards/rejected": -0.8433966636657715,
       "step": 400
     },
     {
       "epoch": 0.858189429618001,
-      "grad_norm": 30.320451161843337,
+      "grad_norm": 31.79436144433176,
       "learning_rate": 2.9492720416985e-08,
-      "logits/chosen": -1.2879774570465088,
-      "logits/rejected": -1.2872713804244995,
-      "logps/chosen": -310.5592956542969,
-      "logps/rejected": -277.61102294921875,
-      "loss": 0.51,
-      "rewards/accuracies": 0.793749988079071,
-      "rewards/chosen": -0.20286865532398224,
-      "rewards/margins": 1.0352991819381714,
-      "rewards/rejected": -1.2381680011749268,
+      "logits/chosen": -0.9002103805541992,
+      "logits/rejected": -0.8619481921195984,
+      "logps/chosen": -307.99755859375,
+      "logps/rejected": -271.2117919921875,
+      "loss": 0.5347,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": 0.08911776542663574,
+      "rewards/margins": 0.9049032926559448,
+      "rewards/rejected": -0.8157855272293091,
       "step": 410
     },
     {
       "epoch": 0.8791208791208791,
-      "grad_norm": 27.203766432310225,
+      "grad_norm": 26.61951731882698,
       "learning_rate": 2.1464952759020856e-08,
-      "logits/chosen": -1.154717206954956,
-      "logits/rejected": -1.1917306184768677,
-      "logps/chosen": -267.16241455078125,
-      "logps/rejected": -283.44769287109375,
-      "loss": 0.4977,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -0.34657976031303406,
-      "rewards/margins": 1.0054247379302979,
-      "rewards/rejected": -1.3520046472549438,
+      "logits/chosen": -0.723199725151062,
+      "logits/rejected": -0.7414354681968689,
+      "logps/chosen": -262.56146240234375,
+      "logps/rejected": -279.45758056640625,
+      "loss": 0.5108,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.04666225239634514,
+      "rewards/margins": 0.907934844493866,
+      "rewards/rejected": -0.9545971155166626,
       "step": 420
     },
     {
       "epoch": 0.9000523286237572,
-      "grad_norm": 24.530741171541777,
+      "grad_norm": 25.933912807613094,
       "learning_rate": 1.4662207078575684e-08,
-      "logits/chosen": -1.2220633029937744,
-      "logits/rejected": -1.1990492343902588,
-      "logps/chosen": -294.7621765136719,
-      "logps/rejected": -281.249755859375,
-      "loss": 0.4783,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": -0.23210814595222473,
-      "rewards/margins": 0.9327493906021118,
-      "rewards/rejected": -1.1648573875427246,
+      "logits/chosen": -0.8103313446044922,
+      "logits/rejected": -0.7611863613128662,
+      "logps/chosen": -292.8187561035156,
+      "logps/rejected": -275.9464111328125,
+      "loss": 0.4973,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.03782375156879425,
+      "rewards/margins": 0.8563901782035828,
+      "rewards/rejected": -0.8185665011405945,
       "step": 430
     },
     {
       "epoch": 0.9209837781266352,
-      "grad_norm": 27.913797281437322,
+      "grad_norm": 27.040363752890194,
       "learning_rate": 9.12094829893642e-09,
-      "logits/chosen": -1.270138144493103,
-      "logits/rejected": -1.2575105428695679,
-      "logps/chosen": -267.31134033203125,
-      "logps/rejected": -235.6915283203125,
-      "loss": 0.5077,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -0.16507378220558167,
-      "rewards/margins": 0.88013756275177,
-      "rewards/rejected": -1.0452113151550293,
+      "logits/chosen": -0.867548942565918,
+      "logits/rejected": -0.8313533663749695,
+      "logps/chosen": -264.0154724121094,
+      "logps/rejected": -230.5113067626953,
+      "loss": 0.5192,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": 0.09564376622438431,
+      "rewards/margins": 0.8098851442337036,
+      "rewards/rejected": -0.7142413258552551,
       "step": 440
     },
     {
       "epoch": 0.9419152276295133,
-      "grad_norm": 24.734254880471383,
+      "grad_norm": 25.39623084095404,
       "learning_rate": 4.8708793644441086e-09,
-      "logits/chosen": -1.1567983627319336,
-      "logits/rejected": -1.17081618309021,
-      "logps/chosen": -274.9759216308594,
-      "logps/rejected": -272.7271423339844,
-      "loss": 0.4971,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.3414493501186371,
-      "rewards/margins": 0.9493969082832336,
-      "rewards/rejected": -1.2908462285995483,
+      "logits/chosen": -0.7358702421188354,
+      "logits/rejected": -0.7247830033302307,
+      "logps/chosen": -272.40289306640625,
+      "logps/rejected": -267.282470703125,
+      "loss": 0.5173,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.09212496876716614,
+      "rewards/margins": 0.8651455640792847,
+      "rewards/rejected": -0.957270622253418,
       "step": 450
     },
     {
       "epoch": 0.9628466771323915,
-      "grad_norm": 28.240480553266362,
+      "grad_norm": 30.67806976004459,
       "learning_rate": 1.9347820230782295e-09,
-      "logits/chosen": -1.2195831537246704,
-      "logits/rejected": -1.2097227573394775,
-      "logps/chosen": -275.17266845703125,
-      "logps/rejected": -242.7679901123047,
-      "loss": 0.5119,
+      "logits/chosen": -0.81078040599823,
+      "logits/rejected": -0.7622171640396118,
+      "logps/chosen": -271.894775390625,
+      "logps/rejected": -236.7905731201172,
+      "loss": 0.523,
       "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -0.174423485994339,
-      "rewards/margins": 1.005141258239746,
-      "rewards/rejected": -1.1795645952224731,
+      "rewards/chosen": 0.05093681812286377,
+      "rewards/margins": 0.9445697069168091,
+      "rewards/rejected": -0.8936327695846558,
       "step": 460
     },
     {
       "epoch": 0.9837781266352695,
-      "grad_norm": 28.592886469999605,
+      "grad_norm": 30.19189408633993,
       "learning_rate": 3.2839470889836627e-10,
-      "logits/chosen": -1.2384300231933594,
-      "logits/rejected": -1.2269245386123657,
-      "logps/chosen": -297.8599548339844,
-      "logps/rejected": -291.4131774902344,
-      "loss": 0.4801,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.14566555619239807,
-      "rewards/margins": 0.9302905797958374,
-      "rewards/rejected": -1.075956106185913,
+      "logits/chosen": -0.8224647641181946,
+      "logits/rejected": -0.7915753126144409,
+      "logps/chosen": -294.343994140625,
+      "logps/rejected": -284.93365478515625,
+      "loss": 0.4978,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": 0.1548490971326828,
+      "rewards/margins": 0.8648107647895813,
+      "rewards/rejected": -0.7099617719650269,
       "step": 470
     },
     {
       "epoch": 0.9984301412872841,
-      "eval_logits/chosen": -1.2541908025741577,
-      "eval_logits/rejected": -1.2466377019882202,
-      "eval_logps/chosen": -274.62506103515625,
-      "eval_logps/rejected": -288.3345947265625,
-      "eval_loss": 0.5040282011032104,
-      "eval_rewards/accuracies": 0.78125,
-      "eval_rewards/chosen": -0.1466209888458252,
-      "eval_rewards/margins": 1.0191689729690552,
-      "eval_rewards/rejected": -1.1657898426055908,
-      "eval_runtime": 226.2735,
-      "eval_samples_per_second": 8.839,
-      "eval_steps_per_second": 0.141,
+      "eval_logits/chosen": -0.8113551735877991,
+      "eval_logits/rejected": -0.784081220626831,
+      "eval_logps/chosen": -272.69000244140625,
+      "eval_logps/rejected": -282.8626708984375,
+      "eval_loss": 0.5187761783599854,
+      "eval_rewards/accuracies": 0.7578125,
+      "eval_rewards/chosen": 0.1167038306593895,
+      "eval_rewards/margins": 0.9430557489395142,
+      "eval_rewards/rejected": -0.8263520002365112,
+      "eval_runtime": 252.9492,
+      "eval_samples_per_second": 7.907,
+      "eval_steps_per_second": 0.127,
       "step": 477
     },
     {
       "epoch": 0.9984301412872841,
       "step": 477,
       "total_flos": 0.0,
-      "train_loss": 0.5369631534352493,
-      "train_runtime": 17159.3245,
-      "train_samples_per_second": 3.563,
-      "train_steps_per_second": 0.028
+      "train_loss": 0.5465342798572916,
+      "train_runtime": 16447.8444,
+      "train_samples_per_second": 3.717,
+      "train_steps_per_second": 0.029
     }
   ],
   "logging_steps": 10,