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: 0.5994
-- Rewards/chosen: -0.1944
-- Rewards/rejected: -0.5072
-- Rewards/accuracies: 0.7188
-- Rewards/margins: 0.3128
-- Logps/rejected: -331.7664
-- Logps/chosen: -301.1218
-- Logits/rejected: -1.1664
-- Logits/chosen: -1.2640
+- Loss: 0.6003
+- Rewards/chosen: -0.1976
+- Rewards/rejected: -0.5074
+- Rewards/accuracies: 0.6992
+- Rewards/margins: 0.3098
+- Logps/rejected: -331.0023
+- Logps/chosen: -300.7068
+- Logits/rejected: -1.1613
+- Logits/chosen: -1.2587
 
 ## 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 |
 |:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.5895        | 0.9984 | 477  | 0.5994          | -0.1944        | -0.5072          | 0.7188             | 0.3128          | -331.7664      | -301.1218    | -1.1664         | -1.2640       |
+| 0.5913        | 0.9984 | 477  | 0.6003          | -0.1976        | -0.5074          | 0.6992             | 0.3098          | -331.0023      | -300.7068    | -1.1613         | -1.2587       |
 
 
 ### Framework versions

all_results.json

@@ -1,9 +1,9 @@
 {
     "epoch": 0.9984301412872841,
     "total_flos": 0.0,
-    "train_loss": 0.6282720775724208,
-    "train_runtime": 7595.833,
+    "train_loss": 0.6291670099494342,
+    "train_runtime": 7568.1764,
     "train_samples": 61134,
-    "train_samples_per_second": 8.048,
+    "train_samples_per_second": 8.078,
     "train_steps_per_second": 0.063
 }

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "rebuttal/qwen-3b-sft-full/checkpoint-700",
+  "_name_or_path": "rebuttal/qwen-3b-sft-full/checkpoint-900",
   "architectures": [
     "Qwen2ForCausalLM"
   ],

model-00001-of-00002.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:393d3aabbfa951ac734218b6768b3e1f5eab55006770f9cca5cb758aec624cb1
+oid sha256:142ceeabdc9efdd81585cb83ac993c7223d9e91a48f5fc0690e10efc925c624e
 size 4957560304

model-00002-of-00002.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:080f5d084ba63e89213be0f24b5bbf8f95d1c4a8a81d90813828e1f164fe25c2
+oid sha256:6a5473c0783737d1b1e4411294f6f605b24f0ea2a03beb17e2af556de0b43cc5
 size 1214366696

train_results.json

@@ -1,9 +1,9 @@
 {
     "epoch": 0.9984301412872841,
     "total_flos": 0.0,
-    "train_loss": 0.6282720775724208,
-    "train_runtime": 7595.833,
+    "train_loss": 0.6291670099494342,
+    "train_runtime": 7568.1764,
     "train_samples": 61134,
-    "train_samples_per_second": 8.048,
+    "train_samples_per_second": 8.078,
     "train_steps_per_second": 0.063
 }

trainer_state.json

@@ -10,12 +10,12 @@
   "log_history": [
     {
       "epoch": 0.0020931449502878076,
-      "grad_norm": 2.2012234271930335,
+      "grad_norm": 2.209093805007533,
       "learning_rate": 1.0416666666666666e-08,
-      "logits/chosen": -1.6323597431182861,
-      "logits/rejected": -1.3586889505386353,
-      "logps/chosen": -369.83544921875,
-      "logps/rejected": -371.8328857421875,
+      "logits/chosen": -1.6361453533172607,
+      "logits/rejected": -1.3620976209640503,
+      "logps/chosen": -368.01727294921875,
+      "logps/rejected": -370.546875,
       "loss": 0.6931,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
@@ -25,722 +25,722 @@
     },
     {
       "epoch": 0.020931449502878074,
-      "grad_norm": 2.1677556807963656,
+      "grad_norm": 2.1732615894977494,
       "learning_rate": 1.0416666666666667e-07,
-      "logits/chosen": -1.243551254272461,
-      "logits/rejected": -1.2145392894744873,
-      "logps/chosen": -314.3976745605469,
-      "logps/rejected": -287.333740234375,
-      "loss": 0.6931,
-      "rewards/accuracies": 0.4444444477558136,
-      "rewards/chosen": -0.00020455155754461884,
-      "rewards/margins": 9.76992305368185e-05,
-      "rewards/rejected": -0.000302250700769946,
+      "logits/chosen": -1.2434254884719849,
+      "logits/rejected": -1.2148737907409668,
+      "logps/chosen": -313.6167907714844,
+      "logps/rejected": -286.7457580566406,
+      "loss": 0.6934,
+      "rewards/accuracies": 0.4861111044883728,
+      "rewards/chosen": -0.0004693272348958999,
+      "rewards/margins": 0.00012820077245123684,
+      "rewards/rejected": -0.0005975279491394758,
       "step": 10
     },
     {
       "epoch": 0.04186289900575615,
-      "grad_norm": 2.1076716968602103,
+      "grad_norm": 2.1294544043702355,
       "learning_rate": 2.0833333333333333e-07,
-      "logits/chosen": -1.3115403652191162,
-      "logits/rejected": -1.1970912218093872,
-      "logps/chosen": -321.0391845703125,
-      "logps/rejected": -284.40985107421875,
-      "loss": 0.6929,
-      "rewards/accuracies": 0.5062500238418579,
-      "rewards/chosen": 0.0009345440194010735,
-      "rewards/margins": 0.0011811652220785618,
-      "rewards/rejected": -0.0002466212317813188,
+      "logits/chosen": -1.3135641813278198,
+      "logits/rejected": -1.1997337341308594,
+      "logps/chosen": -320.4668884277344,
+      "logps/rejected": -283.766357421875,
+      "loss": 0.6932,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.0006623806548304856,
+      "rewards/margins": 0.000770134967751801,
+      "rewards/rejected": -0.00010775432747323066,
       "step": 20
     },
     {
       "epoch": 0.06279434850863422,
-      "grad_norm": 2.125315368033945,
+      "grad_norm": 2.1326205281225716,
       "learning_rate": 3.1249999999999997e-07,
-      "logits/chosen": -1.331930160522461,
-      "logits/rejected": -1.2224106788635254,
-      "logps/chosen": -295.735107421875,
-      "logps/rejected": -252.8268280029297,
-      "loss": 0.6928,
-      "rewards/accuracies": 0.543749988079071,
-      "rewards/chosen": 0.0020953179337084293,
-      "rewards/margins": 0.0013438600581139326,
-      "rewards/rejected": 0.0007514576427638531,
+      "logits/chosen": -1.3327853679656982,
+      "logits/rejected": -1.2241847515106201,
+      "logps/chosen": -295.09368896484375,
+      "logps/rejected": -251.9848175048828,
+      "loss": 0.6929,
+      "rewards/accuracies": 0.53125,
+      "rewards/chosen": 0.001088696182705462,
+      "rewards/margins": 3.837456461042166e-06,
+      "rewards/rejected": 0.0010848587844520807,
       "step": 30
     },
     {
       "epoch": 0.0837257980115123,
-      "grad_norm": 2.0068651959377166,
+      "grad_norm": 2.0204925028633043,
       "learning_rate": 4.1666666666666667e-07,
-      "logits/chosen": -1.262263298034668,
-      "logits/rejected": -1.154270052909851,
-      "logps/chosen": -271.7317810058594,
-      "logps/rejected": -267.47869873046875,
-      "loss": 0.6921,
-      "rewards/accuracies": 0.6187499761581421,
-      "rewards/chosen": 0.006908699870109558,
-      "rewards/margins": 0.002343912376090884,
-      "rewards/rejected": 0.00456478726118803,
+      "logits/chosen": -1.2614781856536865,
+      "logits/rejected": -1.1543315649032593,
+      "logps/chosen": -270.7391662597656,
+      "logps/rejected": -266.8966369628906,
+      "loss": 0.692,
+      "rewards/accuracies": 0.581250011920929,
+      "rewards/chosen": 0.007228652946650982,
+      "rewards/margins": 0.0024825455620884895,
+      "rewards/rejected": 0.004746107384562492,
       "step": 40
     },
     {
       "epoch": 0.10465724751439037,
-      "grad_norm": 2.1892570691833337,
+      "grad_norm": 2.212404949141082,
       "learning_rate": 4.999731868769026e-07,
-      "logits/chosen": -1.2883636951446533,
-      "logits/rejected": -1.17487633228302,
-      "logps/chosen": -300.61627197265625,
-      "logps/rejected": -280.78704833984375,
-      "loss": 0.6906,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.01698208972811699,
-      "rewards/margins": 0.006509358994662762,
-      "rewards/rejected": 0.010472728870809078,
+      "logits/chosen": -1.289907455444336,
+      "logits/rejected": -1.176802158355713,
+      "logps/chosen": -299.7214050292969,
+      "logps/rejected": -279.8206787109375,
+      "loss": 0.6903,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": 0.01695079728960991,
+      "rewards/margins": 0.005069266073405743,
+      "rewards/rejected": 0.011881532147526741,
       "step": 50
     },
     {
       "epoch": 0.12558869701726844,
-      "grad_norm": 2.290084176301081,
+      "grad_norm": 2.3157132849696085,
       "learning_rate": 4.990353313429303e-07,
-      "logits/chosen": -1.3334367275238037,
-      "logits/rejected": -1.2180477380752563,
-      "logps/chosen": -260.46710205078125,
-      "logps/rejected": -252.9662628173828,
-      "loss": 0.6883,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": 0.029394078999757767,
-      "rewards/margins": 0.010875710286200047,
-      "rewards/rejected": 0.018518367782235146,
+      "logits/chosen": -1.332875370979309,
+      "logits/rejected": -1.217449426651001,
+      "logps/chosen": -259.3828125,
+      "logps/rejected": -251.9915008544922,
+      "loss": 0.688,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.03174789622426033,
+      "rewards/margins": 0.011656029149889946,
+      "rewards/rejected": 0.020091867074370384,
       "step": 60
     },
     {
       "epoch": 0.14652014652014653,
-      "grad_norm": 2.0758544279023408,
+      "grad_norm": 2.1379350799498793,
       "learning_rate": 4.967625656594781e-07,
-      "logits/chosen": -1.2558209896087646,
-      "logits/rejected": -1.2046929597854614,
-      "logps/chosen": -305.46661376953125,
-      "logps/rejected": -288.0334167480469,
+      "logits/chosen": -1.2581872940063477,
+      "logits/rejected": -1.2066874504089355,
+      "logps/chosen": -304.32537841796875,
+      "logps/rejected": -287.1934814453125,
       "loss": 0.6849,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": 0.036026474088430405,
-      "rewards/margins": 0.014729541726410389,
-      "rewards/rejected": 0.02129693701863289,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.03804336115717888,
+      "rewards/margins": 0.015704389661550522,
+      "rewards/rejected": 0.022338975220918655,
       "step": 70
     },
     {
       "epoch": 0.1674515960230246,
-      "grad_norm": 2.207772795120628,
+      "grad_norm": 2.2394142872730223,
       "learning_rate": 4.93167072587771e-07,
-      "logits/chosen": -1.3690812587738037,
-      "logits/rejected": -1.1861250400543213,
-      "logps/chosen": -338.672607421875,
-      "logps/rejected": -258.03009033203125,
-      "loss": 0.6804,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.04112464562058449,
-      "rewards/margins": 0.030477171763777733,
-      "rewards/rejected": 0.01064747292548418,
+      "logits/chosen": -1.369598627090454,
+      "logits/rejected": -1.1864904165267944,
+      "logps/chosen": -337.746337890625,
+      "logps/rejected": -257.49017333984375,
+      "loss": 0.68,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": 0.042612869292497635,
+      "rewards/margins": 0.0334814190864563,
+      "rewards/rejected": 0.009131452068686485,
       "step": 80
     },
     {
       "epoch": 0.18838304552590268,
-      "grad_norm": 2.2757195422832237,
+      "grad_norm": 2.268069081593833,
       "learning_rate": 4.882681251368548e-07,
-      "logits/chosen": -1.2589188814163208,
-      "logits/rejected": -1.1841020584106445,
-      "logps/chosen": -261.81866455078125,
-      "logps/rejected": -260.72711181640625,
-      "loss": 0.6758,
+      "logits/chosen": -1.25824773311615,
+      "logits/rejected": -1.1849478483200073,
+      "logps/chosen": -261.4163513183594,
+      "logps/rejected": -260.33209228515625,
+      "loss": 0.6755,
       "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.014796595089137554,
-      "rewards/margins": 0.031740702688694,
-      "rewards/rejected": -0.016944104805588722,
+      "rewards/chosen": 0.012937061488628387,
+      "rewards/margins": 0.0325547531247139,
+      "rewards/rejected": -0.01961769163608551,
       "step": 90
     },
     {
       "epoch": 0.20931449502878074,
-      "grad_norm": 2.4570891393589025,
+      "grad_norm": 2.5344949458322312,
       "learning_rate": 4.820919832540181e-07,
-      "logits/chosen": -1.3160041570663452,
-      "logits/rejected": -1.15372633934021,
-      "logps/chosen": -314.7441711425781,
-      "logps/rejected": -296.74298095703125,
-      "loss": 0.6701,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.0007129657897166908,
-      "rewards/margins": 0.06460981070995331,
-      "rewards/rejected": -0.06389684975147247,
+      "logits/chosen": -1.3172709941864014,
+      "logits/rejected": -1.1559362411499023,
+      "logps/chosen": -314.45269775390625,
+      "logps/rejected": -296.49200439453125,
+      "loss": 0.6703,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.006442523095756769,
+      "rewards/margins": 0.06349583715200424,
+      "rewards/rejected": -0.06993836164474487,
       "step": 100
     },
     {
       "epoch": 0.2302459445316588,
-      "grad_norm": 2.8235839193346903,
+      "grad_norm": 2.765489028108744,
       "learning_rate": 4.7467175306295647e-07,
-      "logits/chosen": -1.3190224170684814,
-      "logits/rejected": -1.21718430519104,
-      "logps/chosen": -300.4778747558594,
-      "logps/rejected": -288.29754638671875,
-      "loss": 0.6695,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -0.019010253250598907,
-      "rewards/margins": 0.06232859566807747,
-      "rewards/rejected": -0.08133885264396667,
+      "logits/chosen": -1.3196508884429932,
+      "logits/rejected": -1.2175612449645996,
+      "logps/chosen": -299.80841064453125,
+      "logps/rejected": -288.1402282714844,
+      "loss": 0.6694,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": -0.020120445638895035,
+      "rewards/margins": 0.0659942552447319,
+      "rewards/rejected": -0.08611470460891724,
       "step": 110
     },
     {
       "epoch": 0.25117739403453687,
-      "grad_norm": 2.6083124454694575,
+      "grad_norm": 2.629057180782522,
       "learning_rate": 4.6604720940421207e-07,
-      "logits/chosen": -1.3359606266021729,
-      "logits/rejected": -1.3284838199615479,
-      "logps/chosen": -275.46221923828125,
-      "logps/rejected": -280.50030517578125,
-      "loss": 0.6566,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -0.04199109226465225,
-      "rewards/margins": 0.07056319713592529,
-      "rewards/rejected": -0.11255427449941635,
+      "logits/chosen": -1.334337592124939,
+      "logits/rejected": -1.3264808654785156,
+      "logps/chosen": -275.4209899902344,
+      "logps/rejected": -280.6207275390625,
+      "loss": 0.6562,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -0.04990966618061066,
+      "rewards/margins": 0.0697593092918396,
+      "rewards/rejected": -0.11966897547245026,
       "step": 120
     },
     {
       "epoch": 0.272108843537415,
-      "grad_norm": 2.837520507016576,
+      "grad_norm": 2.927320582140714,
       "learning_rate": 4.5626458262912735e-07,
-      "logits/chosen": -1.3216521739959717,
-      "logits/rejected": -1.2274608612060547,
-      "logps/chosen": -299.32415771484375,
-      "logps/rejected": -295.39166259765625,
-      "loss": 0.6489,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -0.09194846451282501,
-      "rewards/margins": 0.0814133882522583,
-      "rewards/rejected": -0.17336183786392212,
+      "logits/chosen": -1.3210299015045166,
+      "logits/rejected": -1.227181315422058,
+      "logps/chosen": -299.58709716796875,
+      "logps/rejected": -295.92364501953125,
+      "loss": 0.6486,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -0.10010261833667755,
+      "rewards/margins": 0.08470647037029266,
+      "rewards/rejected": -0.1848091185092926,
       "step": 130
     },
     {
       "epoch": 0.29304029304029305,
-      "grad_norm": 3.6850978689525142,
+      "grad_norm": 4.653751503139143,
       "learning_rate": 4.453763107901675e-07,
-      "logits/chosen": -1.3599679470062256,
-      "logits/rejected": -1.2200071811676025,
-      "logps/chosen": -341.9486999511719,
-      "logps/rejected": -295.4434509277344,
-      "loss": 0.6457,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -0.06680432707071304,
-      "rewards/margins": 0.13532930612564087,
-      "rewards/rejected": -0.20213361084461212,
+      "logits/chosen": -1.3587443828582764,
+      "logits/rejected": -1.218764066696167,
+      "logps/chosen": -340.66259765625,
+      "logps/rejected": -294.0758361816406,
+      "loss": 0.6454,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.062156885862350464,
+      "rewards/margins": 0.13383856415748596,
+      "rewards/rejected": -0.19599545001983643,
       "step": 140
     },
     {
       "epoch": 0.3139717425431711,
-      "grad_norm": 3.4626581883122793,
+      "grad_norm": 3.782599320828944,
       "learning_rate": 4.3344075855595097e-07,
-      "logits/chosen": -1.365296721458435,
-      "logits/rejected": -1.2838973999023438,
-      "logps/chosen": -303.7018127441406,
-      "logps/rejected": -280.51727294921875,
-      "loss": 0.6437,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": -0.11250151693820953,
-      "rewards/margins": 0.13880370557308197,
-      "rewards/rejected": -0.2513051927089691,
+      "logits/chosen": -1.3679354190826416,
+      "logits/rejected": -1.2858151197433472,
+      "logps/chosen": -304.27508544921875,
+      "logps/rejected": -281.0719909667969,
+      "loss": 0.6434,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": -0.12646649777889252,
+      "rewards/margins": 0.13605287671089172,
+      "rewards/rejected": -0.26251935958862305,
       "step": 150
     },
     {
       "epoch": 0.3349031920460492,
-      "grad_norm": 6.554231470712301,
+      "grad_norm": 6.066172015511962,
       "learning_rate": 4.2052190435769554e-07,
-      "logits/chosen": -1.400604486465454,
-      "logits/rejected": -1.2205251455307007,
-      "logps/chosen": -295.25970458984375,
-      "logps/rejected": -286.7093811035156,
-      "loss": 0.6303,
+      "logits/chosen": -1.4009732007980347,
+      "logits/rejected": -1.2202575206756592,
+      "logps/chosen": -293.8358459472656,
+      "logps/rejected": -285.55279541015625,
+      "loss": 0.6307,
       "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -0.09192417562007904,
-      "rewards/margins": 0.18175065517425537,
-      "rewards/rejected": -0.2736748456954956,
+      "rewards/chosen": -0.08570082485675812,
+      "rewards/margins": 0.18192997574806213,
+      "rewards/rejected": -0.26763075590133667,
       "step": 160
     },
     {
       "epoch": 0.35583464154892724,
-      "grad_norm": 5.255676474646723,
+      "grad_norm": 4.044043694362463,
       "learning_rate": 4.0668899744407567e-07,
-      "logits/chosen": -1.3328770399093628,
-      "logits/rejected": -1.241591453552246,
-      "logps/chosen": -288.12481689453125,
-      "logps/rejected": -274.1601257324219,
-      "loss": 0.6337,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.1892765462398529,
-      "rewards/margins": 0.14230875670909882,
-      "rewards/rejected": -0.3315853178501129,
+      "logits/chosen": -1.334903359413147,
+      "logits/rejected": -1.2431747913360596,
+      "logps/chosen": -287.99957275390625,
+      "logps/rejected": -273.060546875,
+      "loss": 0.6347,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.19482368230819702,
+      "rewards/margins": 0.13516560196876526,
+      "rewards/rejected": -0.3299892544746399,
       "step": 170
     },
     {
       "epoch": 0.37676609105180536,
-      "grad_norm": 4.9849427125913515,
+      "grad_norm": 4.261893089349372,
       "learning_rate": 3.920161866827889e-07,
-      "logits/chosen": -1.283872365951538,
-      "logits/rejected": -1.2232367992401123,
-      "logps/chosen": -295.1382141113281,
-      "logps/rejected": -284.7275390625,
-      "loss": 0.6246,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.15464556217193604,
-      "rewards/margins": 0.14419159293174744,
-      "rewards/rejected": -0.29883715510368347,
+      "logits/chosen": -1.2868341207504272,
+      "logits/rejected": -1.2260396480560303,
+      "logps/chosen": -292.34771728515625,
+      "logps/rejected": -282.02740478515625,
+      "loss": 0.6252,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -0.1360318958759308,
+      "rewards/margins": 0.14469388127326965,
+      "rewards/rejected": -0.28072577714920044,
       "step": 180
     },
     {
       "epoch": 0.3976975405546834,
-      "grad_norm": 3.7547474704915236,
+      "grad_norm": 3.8979942896091306,
       "learning_rate": 3.765821230985757e-07,
-      "logits/chosen": -1.2838518619537354,
-      "logits/rejected": -1.2677780389785767,
-      "logps/chosen": -282.908447265625,
-      "logps/rejected": -292.8689880371094,
-      "loss": 0.6244,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -0.08928468078374863,
-      "rewards/margins": 0.1427617073059082,
-      "rewards/rejected": -0.23204639554023743,
+      "logits/chosen": -1.287151575088501,
+      "logits/rejected": -1.2711493968963623,
+      "logps/chosen": -284.5529479980469,
+      "logps/rejected": -294.7345275878906,
+      "loss": 0.6252,
+      "rewards/accuracies": 0.6187499761581421,
+      "rewards/chosen": -0.11440813541412354,
+      "rewards/margins": 0.14361563324928284,
+      "rewards/rejected": -0.25802376866340637,
       "step": 190
     },
     {
       "epoch": 0.4186289900575615,
-      "grad_norm": 5.896948199734476,
+      "grad_norm": 5.501329973423857,
       "learning_rate": 3.604695382782159e-07,
-      "logits/chosen": -1.2904465198516846,
-      "logits/rejected": -1.2478965520858765,
-      "logps/chosen": -276.3597717285156,
-      "logps/rejected": -307.5818786621094,
+      "logits/chosen": -1.2931939363479614,
+      "logits/rejected": -1.2511314153671265,
+      "logps/chosen": -277.0694885253906,
+      "logps/rejected": -307.970458984375,
       "loss": 0.6293,
-      "rewards/accuracies": 0.6499999761581421,
-      "rewards/chosen": -0.15242597460746765,
-      "rewards/margins": 0.14724506437778473,
-      "rewards/rejected": -0.2996710240840912,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": -0.1651896983385086,
+      "rewards/margins": 0.14419811964035034,
+      "rewards/rejected": -0.30938783288002014,
       "step": 200
     },
     {
       "epoch": 0.43956043956043955,
-      "grad_norm": 4.451047420080709,
+      "grad_norm": 4.732203165730444,
       "learning_rate": 3.4376480090239047e-07,
-      "logits/chosen": -1.3529211282730103,
-      "logits/rejected": -1.1762115955352783,
-      "logps/chosen": -339.2208251953125,
-      "logps/rejected": -304.05657958984375,
-      "loss": 0.6262,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.15597447752952576,
-      "rewards/margins": 0.23740153014659882,
-      "rewards/rejected": -0.39337602257728577,
+      "logits/chosen": -1.3534619808197021,
+      "logits/rejected": -1.1777923107147217,
+      "logps/chosen": -341.540771484375,
+      "logps/rejected": -305.64324951171875,
+      "loss": 0.628,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.1867484748363495,
+      "rewards/margins": 0.22952301800251007,
+      "rewards/rejected": -0.41627150774002075,
       "step": 210
     },
     {
       "epoch": 0.4604918890633176,
-      "grad_norm": 8.848882446398921,
+      "grad_norm": 7.387774124180662,
       "learning_rate": 3.265574537815398e-07,
-      "logits/chosen": -1.303682565689087,
-      "logits/rejected": -1.2175633907318115,
-      "logps/chosen": -264.3138122558594,
-      "logps/rejected": -287.0865478515625,
-      "loss": 0.6218,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -0.20161668956279755,
-      "rewards/margins": 0.18881258368492126,
-      "rewards/rejected": -0.3904293179512024,
+      "logits/chosen": -1.307450294494629,
+      "logits/rejected": -1.2217823266983032,
+      "logps/chosen": -259.33782958984375,
+      "logps/rejected": -280.0881652832031,
+      "loss": 0.6246,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.15995870530605316,
+      "rewards/margins": 0.1672067791223526,
+      "rewards/rejected": -0.32716548442840576,
       "step": 220
     },
     {
       "epoch": 0.48142333856619574,
-      "grad_norm": 4.257014819767513,
+      "grad_norm": 5.844827244074732,
       "learning_rate": 3.0893973387735683e-07,
-      "logits/chosen": -1.349918246269226,
-      "logits/rejected": -1.2157604694366455,
-      "logps/chosen": -280.90447998046875,
-      "logps/rejected": -282.1334533691406,
-      "loss": 0.6217,
-      "rewards/accuracies": 0.7749999761581421,
-      "rewards/chosen": -0.12974441051483154,
-      "rewards/margins": 0.2553783059120178,
-      "rewards/rejected": -0.38512271642684937,
+      "logits/chosen": -1.3550784587860107,
+      "logits/rejected": -1.2227942943572998,
+      "logps/chosen": -283.80657958984375,
+      "logps/rejected": -284.75250244140625,
+      "loss": 0.6239,
+      "rewards/accuracies": 0.7562500238418579,
+      "rewards/chosen": -0.1670159548521042,
+      "rewards/margins": 0.25173139572143555,
+      "rewards/rejected": -0.41874733567237854,
       "step": 230
     },
     {
       "epoch": 0.5023547880690737,
-      "grad_norm": 6.060399302854027,
+      "grad_norm": 4.456680896304261,
       "learning_rate": 2.910060778827554e-07,
-      "logits/chosen": -1.3228087425231934,
-      "logits/rejected": -1.2334887981414795,
-      "logps/chosen": -307.2931213378906,
-      "logps/rejected": -309.89404296875,
-      "loss": 0.6087,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.1281367689371109,
-      "rewards/margins": 0.24148380756378174,
-      "rewards/rejected": -0.36962056159973145,
+      "logits/chosen": -1.3266417980194092,
+      "logits/rejected": -1.2376025915145874,
+      "logps/chosen": -305.8389587402344,
+      "logps/rejected": -309.94970703125,
+      "loss": 0.6105,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.12230654060840607,
+      "rewards/margins": 0.25457701086997986,
+      "rewards/rejected": -0.37688353657722473,
       "step": 240
     },
     {
       "epoch": 0.5232862375719518,
-      "grad_norm": 7.127346181224084,
+      "grad_norm": 6.235910945305986,
       "learning_rate": 2.7285261601056697e-07,
-      "logits/chosen": -1.355546236038208,
-      "logits/rejected": -1.2053163051605225,
-      "logps/chosen": -312.4363098144531,
-      "logps/rejected": -294.50927734375,
-      "loss": 0.612,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -0.22020766139030457,
-      "rewards/margins": 0.24413248896598816,
-      "rewards/rejected": -0.4643401503562927,
+      "logits/chosen": -1.3577561378479004,
+      "logits/rejected": -1.2086182832717896,
+      "logps/chosen": -304.21600341796875,
+      "logps/rejected": -284.5404052734375,
+      "loss": 0.6136,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -0.144565150141716,
+      "rewards/margins": 0.2273297756910324,
+      "rewards/rejected": -0.3718949556350708,
       "step": 250
     },
     {
       "epoch": 0.54421768707483,
-      "grad_norm": 5.651798436257719,
+      "grad_norm": 5.424990756269059,
       "learning_rate": 2.5457665670441937e-07,
-      "logits/chosen": -1.3042914867401123,
-      "logits/rejected": -1.2497762441635132,
-      "logps/chosen": -284.8901672363281,
-      "logps/rejected": -291.6590270996094,
-      "loss": 0.6058,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -0.18228834867477417,
-      "rewards/margins": 0.21573488414287567,
-      "rewards/rejected": -0.39802324771881104,
+      "logits/chosen": -1.3028128147125244,
+      "logits/rejected": -1.2477747201919556,
+      "logps/chosen": -286.72296142578125,
+      "logps/rejected": -292.54144287109375,
+      "loss": 0.6055,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.20892027020454407,
+      "rewards/margins": 0.20335116982460022,
+      "rewards/rejected": -0.4122714102268219,
       "step": 260
     },
     {
       "epoch": 0.565149136577708,
-      "grad_norm": 5.210409317687377,
+      "grad_norm": 4.597809166714594,
       "learning_rate": 2.3627616503391812e-07,
-      "logits/chosen": -1.3350675106048584,
-      "logits/rejected": -1.2428313493728638,
-      "logps/chosen": -326.89739990234375,
-      "logps/rejected": -318.1094055175781,
-      "loss": 0.6061,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -0.16515155136585236,
-      "rewards/margins": 0.25075724720954895,
-      "rewards/rejected": -0.4159087538719177,
+      "logits/chosen": -1.3325413465499878,
+      "logits/rejected": -1.2403695583343506,
+      "logps/chosen": -327.7884826660156,
+      "logps/rejected": -317.57977294921875,
+      "loss": 0.6076,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": -0.18088562786579132,
+      "rewards/margins": 0.2384168803691864,
+      "rewards/rejected": -0.4193025231361389,
       "step": 270
     },
     {
       "epoch": 0.5860805860805861,
-      "grad_norm": 5.819241913840486,
+      "grad_norm": 4.579037025916175,
       "learning_rate": 2.1804923757009882e-07,
-      "logits/chosen": -1.290400743484497,
-      "logits/rejected": -1.167297601699829,
-      "logps/chosen": -280.88848876953125,
-      "logps/rejected": -270.7590637207031,
-      "loss": 0.6149,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -0.11809780448675156,
-      "rewards/margins": 0.21220822632312775,
-      "rewards/rejected": -0.3303060233592987,
+      "logits/chosen": -1.2911279201507568,
+      "logits/rejected": -1.1682571172714233,
+      "logps/chosen": -277.64385986328125,
+      "logps/rejected": -266.3356018066406,
+      "loss": 0.6168,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.09358052909374237,
+      "rewards/margins": 0.19994844496250153,
+      "rewards/rejected": -0.2935289740562439,
       "step": 280
     },
     {
       "epoch": 0.6070120355834642,
-      "grad_norm": 5.1788939700785495,
+      "grad_norm": 6.558747896143247,
       "learning_rate": 1.9999357655598891e-07,
-      "logits/chosen": -1.2771365642547607,
-      "logits/rejected": -1.242760419845581,
-      "logps/chosen": -276.59552001953125,
-      "logps/rejected": -299.53179931640625,
-      "loss": 0.609,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.20070350170135498,
-      "rewards/margins": 0.21730685234069824,
-      "rewards/rejected": -0.41801032423973083,
+      "logits/chosen": -1.2763615846633911,
+      "logits/rejected": -1.241579294204712,
+      "logps/chosen": -274.0447998046875,
+      "logps/rejected": -297.26043701171875,
+      "loss": 0.61,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -0.18163642287254333,
+      "rewards/margins": 0.2189778983592987,
+      "rewards/rejected": -0.40061432123184204,
       "step": 290
     },
     {
       "epoch": 0.6279434850863422,
-      "grad_norm": 7.617597936498968,
+      "grad_norm": 5.737906800464494,
       "learning_rate": 1.8220596619089573e-07,
-      "logits/chosen": -1.3590519428253174,
-      "logits/rejected": -1.2383599281311035,
-      "logps/chosen": -364.2522888183594,
-      "logps/rejected": -331.903076171875,
-      "loss": 0.6013,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -0.20905812084674835,
-      "rewards/margins": 0.2222755402326584,
-      "rewards/rejected": -0.4313336908817291,
+      "logits/chosen": -1.3600513935089111,
+      "logits/rejected": -1.2401748895645142,
+      "logps/chosen": -360.7850341796875,
+      "logps/rejected": -327.5794982910156,
+      "loss": 0.6022,
+      "rewards/accuracies": 0.65625,
+      "rewards/chosen": -0.18194793164730072,
+      "rewards/margins": 0.21308335661888123,
+      "rewards/rejected": -0.39503127336502075,
       "step": 300
     },
     {
       "epoch": 0.6488749345892203,
-      "grad_norm": 4.30066303052364,
+      "grad_norm": 4.739990566152762,
       "learning_rate": 1.647817538357072e-07,
-      "logits/chosen": -1.3293473720550537,
-      "logits/rejected": -1.244685411453247,
-      "logps/chosen": -336.703125,
-      "logps/rejected": -305.4994812011719,
-      "loss": 0.5978,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -0.17943550646305084,
-      "rewards/margins": 0.30642929673194885,
-      "rewards/rejected": -0.4858648180961609,
+      "logits/chosen": -1.330038070678711,
+      "logits/rejected": -1.2463994026184082,
+      "logps/chosen": -340.2279357910156,
+      "logps/rejected": -306.41937255859375,
+      "loss": 0.6001,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -0.22187058627605438,
+      "rewards/margins": 0.27941471338272095,
+      "rewards/rejected": -0.5012853145599365,
       "step": 310
     },
     {
       "epoch": 0.6698063840920984,
-      "grad_norm": 4.640582385691058,
+      "grad_norm": 5.000423576550655,
       "learning_rate": 1.478143389201113e-07,
-      "logits/chosen": -1.2704174518585205,
-      "logits/rejected": -1.1587920188903809,
-      "logps/chosen": -282.0940246582031,
-      "logps/rejected": -275.9901428222656,
-      "loss": 0.5917,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.1683064103126526,
-      "rewards/margins": 0.27890798449516296,
-      "rewards/rejected": -0.44721442461013794,
+      "logits/chosen": -1.2705892324447632,
+      "logits/rejected": -1.160201072692871,
+      "logps/chosen": -279.822509765625,
+      "logps/rejected": -274.10589599609375,
+      "loss": 0.5929,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.15409648418426514,
+      "rewards/margins": 0.2819980978965759,
+      "rewards/rejected": -0.43609458208084106,
       "step": 320
     },
     {
       "epoch": 0.6907378335949764,
-      "grad_norm": 7.435137951304479,
+      "grad_norm": 7.610542933399344,
       "learning_rate": 1.3139467229135998e-07,
-      "logits/chosen": -1.323042392730713,
-      "logits/rejected": -1.2895574569702148,
-      "logps/chosen": -296.9349670410156,
-      "logps/rejected": -311.9447021484375,
-      "loss": 0.6064,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -0.10371486842632294,
-      "rewards/margins": 0.2581017315387726,
-      "rewards/rejected": -0.3618166148662567,
+      "logits/chosen": -1.3198235034942627,
+      "logits/rejected": -1.2864559888839722,
+      "logps/chosen": -295.50067138671875,
+      "logps/rejected": -309.83233642578125,
+      "loss": 0.6089,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -0.09709228575229645,
+      "rewards/margins": 0.24964985251426697,
+      "rewards/rejected": -0.346742182970047,
       "step": 330
     },
     {
       "epoch": 0.7116692830978545,
-      "grad_norm": 7.696347159762246,
+      "grad_norm": 5.167367050897432,
       "learning_rate": 1.1561076868822755e-07,
-      "logits/chosen": -1.3477349281311035,
-      "logits/rejected": -1.298015832901001,
-      "logps/chosen": -318.29705810546875,
-      "logps/rejected": -308.1880798339844,
-      "loss": 0.6015,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -0.12474183738231659,
-      "rewards/margins": 0.20875398814678192,
-      "rewards/rejected": -0.3334958851337433,
+      "logits/chosen": -1.3470569849014282,
+      "logits/rejected": -1.2970125675201416,
+      "logps/chosen": -316.94403076171875,
+      "logps/rejected": -306.01544189453125,
+      "loss": 0.6032,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -0.11861377954483032,
+      "rewards/margins": 0.19918926060199738,
+      "rewards/rejected": -0.3178029954433441,
       "step": 340
     },
     {
       "epoch": 0.7326007326007326,
-      "grad_norm": 5.896238550072539,
+      "grad_norm": 5.559046106380532,
       "learning_rate": 1.0054723495346482e-07,
-      "logits/chosen": -1.3026797771453857,
-      "logits/rejected": -1.1723350286483765,
-      "logps/chosen": -286.9999084472656,
-      "logps/rejected": -284.67120361328125,
-      "loss": 0.5893,
-      "rewards/accuracies": 0.6812499761581421,
-      "rewards/chosen": -0.16782598197460175,
-      "rewards/margins": 0.29913097620010376,
-      "rewards/rejected": -0.4669569134712219,
+      "logits/chosen": -1.3032571077346802,
+      "logits/rejected": -1.1714508533477783,
+      "logps/chosen": -287.7757263183594,
+      "logps/rejected": -283.84466552734375,
+      "loss": 0.5915,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.1826373040676117,
+      "rewards/margins": 0.2831045389175415,
+      "rewards/rejected": -0.4657418131828308,
       "step": 350
     },
     {
       "epoch": 0.7535321821036107,
-      "grad_norm": 9.86002444391597,
+      "grad_norm": 6.409390436611249,
       "learning_rate": 8.628481651367875e-08,
-      "logits/chosen": -1.3053710460662842,
-      "logits/rejected": -1.2350800037384033,
-      "logps/chosen": -342.7742919921875,
-      "logps/rejected": -325.7217712402344,
-      "loss": 0.6119,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -0.18869134783744812,
-      "rewards/margins": 0.29812943935394287,
-      "rewards/rejected": -0.486820787191391,
+      "logits/chosen": -1.3029541969299316,
+      "logits/rejected": -1.2337419986724854,
+      "logps/chosen": -342.60174560546875,
+      "logps/rejected": -325.2986755371094,
+      "loss": 0.6136,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.19477759301662445,
+      "rewards/margins": 0.2947302460670471,
+      "rewards/rejected": -0.48950785398483276,
       "step": 360
     },
     {
       "epoch": 0.7744636316064888,
-      "grad_norm": 6.949682931582093,
+      "grad_norm": 6.433319076324727,
       "learning_rate": 7.289996455765748e-08,
-      "logits/chosen": -1.2723801136016846,
-      "logits/rejected": -1.1740524768829346,
-      "logps/chosen": -296.9388427734375,
-      "logps/rejected": -285.5877990722656,
-      "loss": 0.5982,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -0.21268990635871887,
-      "rewards/margins": 0.2266589105129242,
-      "rewards/rejected": -0.43934884667396545,
+      "logits/chosen": -1.271382212638855,
+      "logits/rejected": -1.172966718673706,
+      "logps/chosen": -296.20269775390625,
+      "logps/rejected": -285.0596923828125,
+      "loss": 0.5995,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.21164819598197937,
+      "rewards/margins": 0.22872471809387207,
+      "rewards/rejected": -0.44037294387817383,
       "step": 370
     },
     {
       "epoch": 0.7953950811093669,
-      "grad_norm": 5.533238737145182,
+      "grad_norm": 5.305253750346113,
       "learning_rate": 6.046442623320145e-08,
-      "logits/chosen": -1.327059030532837,
-      "logits/rejected": -1.2507330179214478,
-      "logps/chosen": -278.5428771972656,
-      "logps/rejected": -326.29437255859375,
-      "loss": 0.5957,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.2407039850950241,
-      "rewards/margins": 0.3131566047668457,
-      "rewards/rejected": -0.5538605451583862,
+      "logits/chosen": -1.3278123140335083,
+      "logits/rejected": -1.2512152194976807,
+      "logps/chosen": -277.59759521484375,
+      "logps/rejected": -325.248046875,
+      "loss": 0.597,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -0.23936133086681366,
+      "rewards/margins": 0.31056278944015503,
+      "rewards/rejected": -0.5499240756034851,
       "step": 380
     },
     {
       "epoch": 0.8163265306122449,
-      "grad_norm": 7.095886835068374,
+      "grad_norm": 6.870117914717364,
       "learning_rate": 4.904486005914027e-08,
-      "logits/chosen": -1.4183173179626465,
-      "logits/rejected": -1.3427860736846924,
-      "logps/chosen": -360.0533142089844,
-      "logps/rejected": -353.25457763671875,
-      "loss": 0.5985,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -0.1221286877989769,
-      "rewards/margins": 0.29653307795524597,
-      "rewards/rejected": -0.41866177320480347,
+      "logits/chosen": -1.4171468019485474,
+      "logits/rejected": -1.341071367263794,
+      "logps/chosen": -358.9019470214844,
+      "logps/rejected": -351.606689453125,
+      "loss": 0.6002,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.11978839337825775,
+      "rewards/margins": 0.2912915349006653,
+      "rewards/rejected": -0.41107994318008423,
       "step": 390
     },
     {
       "epoch": 0.837257980115123,
-      "grad_norm": 8.514141041336558,
+      "grad_norm": 6.335936579705355,
       "learning_rate": 3.8702478614051345e-08,
-      "logits/chosen": -1.3505162000656128,
-      "logits/rejected": -1.2189605236053467,
-      "logps/chosen": -286.3233947753906,
-      "logps/rejected": -293.57025146484375,
-      "loss": 0.6044,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.14968815445899963,
-      "rewards/margins": 0.3114926218986511,
-      "rewards/rejected": -0.46118077635765076,
+      "logits/chosen": -1.35098135471344,
+      "logits/rejected": -1.2197775840759277,
+      "logps/chosen": -285.03143310546875,
+      "logps/rejected": -291.09283447265625,
+      "loss": 0.6057,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -0.14426860213279724,
+      "rewards/margins": 0.29942744970321655,
+      "rewards/rejected": -0.4436960220336914,
       "step": 400
     },
     {
       "epoch": 0.858189429618001,
-      "grad_norm": 5.4739645463965925,
+      "grad_norm": 5.018088857212084,
       "learning_rate": 2.9492720416985e-08,
-      "logits/chosen": -1.3487937450408936,
-      "logits/rejected": -1.1910629272460938,
-      "logps/chosen": -342.5664367675781,
-      "logps/rejected": -320.82440185546875,
-      "loss": 0.6091,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.2034163773059845,
-      "rewards/margins": 0.2628234028816223,
-      "rewards/rejected": -0.4662397801876068,
+      "logits/chosen": -1.347543478012085,
+      "logits/rejected": -1.187715768814087,
+      "logps/chosen": -339.9741516113281,
+      "logps/rejected": -317.2913818359375,
+      "loss": 0.6128,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.18554477393627167,
+      "rewards/margins": 0.2517685294151306,
+      "rewards/rejected": -0.4373132586479187,
       "step": 410
     },
     {
       "epoch": 0.8791208791208791,
-      "grad_norm": 5.040150062033603,
+      "grad_norm": 4.932156797184576,
       "learning_rate": 2.1464952759020856e-08,
-      "logits/chosen": -1.3143905401229858,
-      "logits/rejected": -1.221419334411621,
-      "logps/chosen": -296.86175537109375,
-      "logps/rejected": -331.79815673828125,
-      "loss": 0.5911,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -0.2642979025840759,
-      "rewards/margins": 0.307772696018219,
-      "rewards/rejected": -0.5720705986022949,
+      "logits/chosen": -1.3127918243408203,
+      "logits/rejected": -1.2197153568267822,
+      "logps/chosen": -293.56854248046875,
+      "logps/rejected": -327.4413146972656,
+      "loss": 0.5924,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -0.23692865669727325,
+      "rewards/margins": 0.3008675277233124,
+      "rewards/rejected": -0.5377961993217468,
       "step": 420
     },
     {
       "epoch": 0.9000523286237572,
-      "grad_norm": 6.216697819153223,
+      "grad_norm": 6.387390469417908,
       "learning_rate": 1.4662207078575684e-08,
-      "logits/chosen": -1.2935734987258911,
-      "logits/rejected": -1.2418222427368164,
-      "logps/chosen": -325.4705810546875,
-      "logps/rejected": -326.40643310546875,
-      "loss": 0.5903,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -0.2370009422302246,
-      "rewards/margins": 0.28985247015953064,
-      "rewards/rejected": -0.5268534421920776,
+      "logits/chosen": -1.2895276546478271,
+      "logits/rejected": -1.2380428314208984,
+      "logps/chosen": -323.2501220703125,
+      "logps/rejected": -324.5555419921875,
+      "loss": 0.5889,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.22188372910022736,
+      "rewards/margins": 0.2940290570259094,
+      "rewards/rejected": -0.5159127712249756,
       "step": 430
     },
     {
       "epoch": 0.9209837781266352,
-      "grad_norm": 7.18871332609168,
+      "grad_norm": 8.06853042869446,
       "learning_rate": 9.12094829893642e-09,
-      "logits/chosen": -1.3459361791610718,
-      "logits/rejected": -1.2779524326324463,
-      "logps/chosen": -295.43780517578125,
-      "logps/rejected": -275.8896789550781,
-      "loss": 0.5926,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": -0.19762298464775085,
-      "rewards/margins": 0.2516266703605652,
-      "rewards/rejected": -0.44924965500831604,
+      "logits/chosen": -1.3438984155654907,
+      "logits/rejected": -1.2755773067474365,
+      "logps/chosen": -293.9847412109375,
+      "logps/rejected": -275.2344665527344,
+      "loss": 0.5945,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.1894671469926834,
+      "rewards/margins": 0.2590894103050232,
+      "rewards/rejected": -0.4485565721988678,
       "step": 440
     },
     {
       "epoch": 0.9419152276295133,
-      "grad_norm": 8.124991946778033,
+      "grad_norm": 7.968122860306462,
       "learning_rate": 4.8708793644441086e-09,
-      "logits/chosen": -1.3958879709243774,
-      "logits/rejected": -1.3034255504608154,
-      "logps/chosen": -303.18890380859375,
-      "logps/rejected": -319.37005615234375,
-      "loss": 0.5964,
+      "logits/chosen": -1.3937358856201172,
+      "logits/rejected": -1.3009790182113647,
+      "logps/chosen": -303.0423889160156,
+      "logps/rejected": -318.12139892578125,
+      "loss": 0.5972,
       "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.2189836949110031,
-      "rewards/margins": 0.29161643981933594,
-      "rewards/rejected": -0.5106001496315002,
+      "rewards/chosen": -0.22540982067584991,
+      "rewards/margins": 0.2786172330379486,
+      "rewards/rejected": -0.5040270686149597,
       "step": 450
     },
     {
       "epoch": 0.9628466771323915,
-      "grad_norm": 5.659774218029892,
+      "grad_norm": 6.195706260723952,
       "learning_rate": 1.9347820230782295e-09,
-      "logits/chosen": -1.3608148097991943,
-      "logits/rejected": -1.2196918725967407,
-      "logps/chosen": -309.38525390625,
-      "logps/rejected": -287.00640869140625,
-      "loss": 0.6029,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.2412925511598587,
-      "rewards/margins": 0.2627820372581482,
-      "rewards/rejected": -0.5040745735168457,
+      "logits/chosen": -1.35757577419281,
+      "logits/rejected": -1.2171084880828857,
+      "logps/chosen": -308.3154296875,
+      "logps/rejected": -285.93939208984375,
+      "loss": 0.6025,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.2377847135066986,
+      "rewards/margins": 0.2613809406757355,
+      "rewards/rejected": -0.4991656243801117,
       "step": 460
     },
     {
       "epoch": 0.9837781266352695,
-      "grad_norm": 5.896494339712462,
+      "grad_norm": 5.551955093374016,
       "learning_rate": 3.2839470889836627e-10,
-      "logits/chosen": -1.3176569938659668,
-      "logits/rejected": -1.2591168880462646,
-      "logps/chosen": -333.4380798339844,
-      "logps/rejected": -338.2153015136719,
-      "loss": 0.5895,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.21501226723194122,
-      "rewards/margins": 0.2965441048145294,
-      "rewards/rejected": -0.5115563273429871,
+      "logits/chosen": -1.3152616024017334,
+      "logits/rejected": -1.2568466663360596,
+      "logps/chosen": -332.40350341796875,
+      "logps/rejected": -336.4817199707031,
+      "loss": 0.5913,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.21322469413280487,
+      "rewards/margins": 0.2868717312812805,
+      "rewards/rejected": -0.5000964403152466,
       "step": 470
     },
     {
       "epoch": 0.9984301412872841,
-      "eval_logits/chosen": -1.2640103101730347,
-      "eval_logits/rejected": -1.166429877281189,
-      "eval_logps/chosen": -301.121826171875,
-      "eval_logps/rejected": -331.7664489746094,
-      "eval_loss": 0.5993745923042297,
-      "eval_rewards/accuracies": 0.71875,
-      "eval_rewards/chosen": -0.19436991214752197,
-      "eval_rewards/margins": 0.3128090500831604,
-      "eval_rewards/rejected": -0.5071789026260376,
-      "eval_runtime": 140.5014,
-      "eval_samples_per_second": 14.235,
+      "eval_logits/chosen": -1.258691430091858,
+      "eval_logits/rejected": -1.161324381828308,
+      "eval_logps/chosen": -300.70684814453125,
+      "eval_logps/rejected": -331.0023193359375,
+      "eval_loss": 0.6002897024154663,
+      "eval_rewards/accuracies": 0.69921875,
+      "eval_rewards/chosen": -0.19763432443141937,
+      "eval_rewards/margins": 0.30977627635002136,
+      "eval_rewards/rejected": -0.5074105858802795,
+      "eval_runtime": 140.0764,
+      "eval_samples_per_second": 14.278,
       "eval_steps_per_second": 0.228,
       "step": 477
     },
@@ -748,9 +748,9 @@
       "epoch": 0.9984301412872841,
       "step": 477,
       "total_flos": 0.0,
-      "train_loss": 0.6282720775724208,
-      "train_runtime": 7595.833,
-      "train_samples_per_second": 8.048,
+      "train_loss": 0.6291670099494342,
+      "train_runtime": 7568.1764,
+      "train_samples_per_second": 8.078,
       "train_steps_per_second": 0.063
     }
   ],

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c08e75c6e23e1f59e3a231ee35bc257f0b62f1e50421ddf5454dfe0f26833cca
+oid sha256:672ef41dcf8226eca70e4ed06917665de5004c28bcb9d5ac3ecd34662eeef742
 size 7672