Model save

README.md

@@ -26,7 +26,7 @@ print(output["generated_text"])
 
 ## Training procedure
 
-[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/yifanwang/huggingface/runs/5tugty2f)
+[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/yifanwang/huggingface/runs/0h2cadaf)
 
 This model was trained with DPO, a method introduced in [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://huggingface.co/papers/2305.18290).
 

all_results.json

@@ -1,9 +1,9 @@
 {
     "epoch": 3.0,
     "total_flos": 0.0,
-    "train_loss": 0.011217064683810743,
-    "train_runtime": 5892.2099,
+    "train_loss": 0.014182947793543137,
+    "train_runtime": 4808.48,
     "train_samples": 10000,
-    "train_samples_per_second": 5.091,
-    "train_steps_per_second": 0.159
+    "train_samples_per_second": 6.239,
+    "train_steps_per_second": 0.195
 }

train_results.json

@@ -1,9 +1,9 @@
 {
     "epoch": 3.0,
     "total_flos": 0.0,
-    "train_loss": 0.011217064683810743,
-    "train_runtime": 5892.2099,
+    "train_loss": 0.014182947793543137,
+    "train_runtime": 4808.48,
     "train_samples": 10000,
-    "train_samples_per_second": 5.091,
-    "train_steps_per_second": 0.159
+    "train_samples_per_second": 6.239,
+    "train_steps_per_second": 0.195
 }

trainer_state.json

@@ -10,12 +10,12 @@
   "log_history": [
     {
       "epoch": 0.003194888178913738,
-      "grad_norm": 159.3779178763582,
+      "grad_norm": 116.0621472800408,
       "learning_rate": 5.3191489361702125e-09,
-      "logits/chosen": -3.640625,
-      "logits/rejected": -3.703125,
-      "logps/chosen": -173.0,
-      "logps/rejected": -142.0,
+      "logits/chosen": -3.921875,
+      "logits/rejected": -3.8125,
+      "logps/chosen": -179.0,
+      "logps/rejected": -120.0,
       "loss": 0.6914,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
@@ -25,1455 +25,1455 @@
     },
     {
       "epoch": 0.03194888178913738,
-      "grad_norm": 142.0171514848489,
+      "grad_norm": 123.707490939538,
       "learning_rate": 5.3191489361702123e-08,
-      "logits/chosen": -3.59375,
-      "logits/rejected": -3.59375,
-      "logps/chosen": -146.0,
-      "logps/rejected": -152.0,
-      "loss": 0.6656,
-      "rewards/accuracies": 0.4722222089767456,
-      "rewards/chosen": 0.005889892578125,
-      "rewards/margins": 0.060302734375,
-      "rewards/rejected": -0.05419921875,
+      "logits/chosen": -3.75,
+      "logits/rejected": -3.671875,
+      "logps/chosen": -156.0,
+      "logps/rejected": -125.0,
+      "loss": 0.6803,
+      "rewards/accuracies": 0.3611111044883728,
+      "rewards/chosen": 0.00592041015625,
+      "rewards/margins": 0.0264892578125,
+      "rewards/rejected": -0.0205078125,
       "step": 10
     },
     {
       "epoch": 0.06389776357827476,
-      "grad_norm": 41.93159336713555,
+      "grad_norm": 63.61099541251128,
       "learning_rate": 1.0638297872340425e-07,
-      "logits/chosen": -3.546875,
-      "logits/rejected": -3.53125,
-      "logps/chosen": -151.0,
-      "logps/rejected": -164.0,
-      "loss": 0.3441,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.1484375,
-      "rewards/margins": 1.09375,
-      "rewards/rejected": -0.94921875,
+      "logits/chosen": -3.71875,
+      "logits/rejected": -3.59375,
+      "logps/chosen": -161.0,
+      "logps/rejected": -128.0,
+      "loss": 0.5013,
+      "rewards/accuracies": 0.9750000238418579,
+      "rewards/chosen": 0.0294189453125,
+      "rewards/margins": 0.443359375,
+      "rewards/rejected": -0.4140625,
       "step": 20
     },
     {
       "epoch": 0.09584664536741214,
-      "grad_norm": 9.157588889007195,
+      "grad_norm": 16.3990461007188,
       "learning_rate": 1.5957446808510638e-07,
-      "logits/chosen": -3.546875,
-      "logits/rejected": -3.46875,
-      "logps/chosen": -157.0,
-      "logps/rejected": -208.0,
-      "loss": 0.0363,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.71484375,
-      "rewards/margins": 4.90625,
-      "rewards/rejected": -4.1875,
+      "logits/chosen": -3.640625,
+      "logits/rejected": -3.421875,
+      "logps/chosen": -172.0,
+      "logps/rejected": -155.0,
+      "loss": 0.1265,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.1396484375,
+      "rewards/margins": 2.6875,
+      "rewards/rejected": -2.546875,
       "step": 30
     },
     {
       "epoch": 0.12779552715654952,
-      "grad_norm": 0.17402017672333522,
+      "grad_norm": 0.671529242047923,
       "learning_rate": 2.127659574468085e-07,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -3.453125,
-      "logps/chosen": -138.0,
-      "logps/rejected": -237.0,
-      "loss": 0.0017,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.7265625,
-      "rewards/margins": 9.625,
-      "rewards/rejected": -8.875,
+      "logits/chosen": -3.53125,
+      "logits/rejected": -3.328125,
+      "logps/chosen": -152.0,
+      "logps/rejected": -185.0,
+      "loss": 0.0152,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.2265625,
+      "rewards/margins": 6.84375,
+      "rewards/rejected": -6.625,
       "step": 40
     },
     {
       "epoch": 0.1597444089456869,
-      "grad_norm": 0.5204944038679334,
+      "grad_norm": 1.123751636301978,
       "learning_rate": 2.659574468085106e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.390625,
-      "logps/chosen": -138.0,
-      "logps/rejected": -280.0,
-      "loss": 0.0004,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.796875,
-      "rewards/margins": 12.875,
-      "rewards/rejected": -12.125,
+      "logits/chosen": -3.421875,
+      "logits/rejected": -3.171875,
+      "logps/chosen": -150.0,
+      "logps/rejected": -228.0,
+      "loss": 0.0025,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.248046875,
+      "rewards/margins": 10.8125,
+      "rewards/rejected": -10.625,
       "step": 50
     },
     {
       "epoch": 0.19169329073482427,
-      "grad_norm": 1.0403500251971276,
+      "grad_norm": 0.08454507441807713,
       "learning_rate": 3.1914893617021275e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.359375,
-      "logps/chosen": -145.0,
-      "logps/rejected": -292.0,
-      "loss": 0.001,
-      "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.458984375,
-      "rewards/margins": 13.5625,
+      "logits/chosen": -3.3125,
+      "logits/rejected": -2.984375,
+      "logps/chosen": -155.0,
+      "logps/rejected": -254.0,
+      "loss": 0.0015,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": 0.1513671875,
+      "rewards/margins": 13.25,
       "rewards/rejected": -13.125,
       "step": 60
     },
     {
       "epoch": 0.22364217252396165,
-      "grad_norm": 0.03485477392790473,
+      "grad_norm": 0.012771788518112216,
       "learning_rate": 3.7234042553191484e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.40625,
-      "logps/chosen": -144.0,
-      "logps/rejected": -306.0,
-      "loss": 0.0003,
+      "logits/chosen": -3.234375,
+      "logits/rejected": -3.0,
+      "logps/chosen": -153.0,
+      "logps/rejected": -276.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.10791015625,
-      "rewards/margins": 15.9375,
-      "rewards/rejected": -15.875,
+      "rewards/chosen": -0.02685546875,
+      "rewards/margins": 15.625,
+      "rewards/rejected": -15.6875,
       "step": 70
     },
     {
       "epoch": 0.25559105431309903,
-      "grad_norm": 0.5699036915216166,
+      "grad_norm": 0.20303610510850278,
       "learning_rate": 4.25531914893617e-07,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -3.390625,
-      "logps/chosen": -159.0,
-      "logps/rejected": -336.0,
+      "logits/chosen": -3.125,
+      "logits/rejected": -2.90625,
+      "logps/chosen": -174.0,
+      "logps/rejected": -314.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.349609375,
+      "rewards/chosen": -0.431640625,
       "rewards/margins": 18.5,
-      "rewards/rejected": -18.125,
+      "rewards/rejected": -18.875,
       "step": 80
     },
     {
       "epoch": 0.28753993610223644,
-      "grad_norm": 0.000586271534556838,
+      "grad_norm": 0.006864178546485895,
       "learning_rate": 4.787234042553192e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.359375,
-      "logps/chosen": -141.0,
-      "logps/rejected": -342.0,
+      "logits/chosen": -3.046875,
+      "logits/rejected": -2.8125,
+      "logps/chosen": -166.0,
+      "logps/rejected": -330.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.53125,
+      "rewards/chosen": -1.1484375,
       "rewards/margins": 19.375,
-      "rewards/rejected": -18.75,
+      "rewards/rejected": -20.625,
       "step": 90
     },
     {
       "epoch": 0.3194888178913738,
-      "grad_norm": 0.6226854251495038,
+      "grad_norm": 0.039988698214440496,
       "learning_rate": 4.964497041420119e-07,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -3.375,
-      "logps/chosen": -143.0,
-      "logps/rejected": -338.0,
+      "logits/chosen": -3.015625,
+      "logits/rejected": -2.8125,
+      "logps/chosen": -167.0,
+      "logps/rejected": -334.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.451171875,
-      "rewards/margins": 19.125,
-      "rewards/rejected": -18.625,
+      "rewards/chosen": -1.1328125,
+      "rewards/margins": 20.0,
+      "rewards/rejected": -21.25,
       "step": 100
     },
     {
       "epoch": 0.3514376996805112,
-      "grad_norm": 0.010449160008542551,
+      "grad_norm": 0.08090795499045322,
       "learning_rate": 4.905325443786982e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -152.0,
-      "logps/rejected": -360.0,
-      "loss": 0.0004,
+      "logits/chosen": -3.0625,
+      "logits/rejected": -2.75,
+      "logps/chosen": -168.0,
+      "logps/rejected": -348.0,
+      "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.10205078125,
-      "rewards/margins": 20.5,
-      "rewards/rejected": -20.625,
+      "rewards/chosen": -0.9609375,
+      "rewards/margins": 21.625,
+      "rewards/rejected": -22.625,
       "step": 110
     },
     {
       "epoch": 0.38338658146964855,
-      "grad_norm": 0.0009548362087768612,
+      "grad_norm": 0.0005053190792494335,
       "learning_rate": 4.846153846153846e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.375,
-      "logps/chosen": -157.0,
-      "logps/rejected": -362.0,
-      "loss": 0.0,
+      "logits/chosen": -3.140625,
+      "logits/rejected": -2.953125,
+      "logps/chosen": -178.0,
+      "logps/rejected": -330.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.119140625,
-      "rewards/margins": 19.75,
-      "rewards/rejected": -19.75,
+      "rewards/chosen": -1.5,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -20.375,
       "step": 120
     },
     {
       "epoch": 0.41533546325878595,
-      "grad_norm": 0.00013646586994902306,
+      "grad_norm": 0.005793842249166428,
       "learning_rate": 4.78698224852071e-07,
-      "logits/chosen": -3.453125,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -160.0,
-      "logps/rejected": -380.0,
+      "logits/chosen": -3.140625,
+      "logits/rejected": -2.953125,
+      "logps/chosen": -185.0,
+      "logps/rejected": -328.0,
       "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.1240234375,
-      "rewards/margins": 21.625,
-      "rewards/rejected": -21.5,
+      "rewards/chosen": -1.3203125,
+      "rewards/margins": 18.875,
+      "rewards/rejected": -20.125,
       "step": 130
     },
     {
       "epoch": 0.4472843450479233,
-      "grad_norm": 0.011640529634937136,
+      "grad_norm": 0.00547250399529398,
       "learning_rate": 4.727810650887574e-07,
-      "logits/chosen": -3.453125,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -133.0,
-      "logps/rejected": -404.0,
-      "loss": 0.0,
+      "logits/chosen": -3.265625,
+      "logits/rejected": -2.96875,
+      "logps/chosen": -142.0,
+      "logps/rejected": -304.0,
+      "loss": 0.0001,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.466796875,
-      "rewards/margins": 25.875,
-      "rewards/rejected": -25.375,
+      "rewards/chosen": 0.1337890625,
+      "rewards/margins": 18.25,
+      "rewards/rejected": -18.125,
       "step": 140
     },
     {
       "epoch": 0.4792332268370607,
-      "grad_norm": 3.475718964256916e-05,
+      "grad_norm": 0.0009553781516234761,
       "learning_rate": 4.668639053254438e-07,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -3.265625,
-      "logps/chosen": -148.0,
-      "logps/rejected": -388.0,
+      "logits/chosen": -3.125,
+      "logits/rejected": -2.84375,
+      "logps/chosen": -164.0,
+      "logps/rejected": -300.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.42578125,
-      "rewards/margins": 23.625,
-      "rewards/rejected": -23.25,
+      "rewards/chosen": -0.11328125,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -17.75,
       "step": 150
     },
     {
       "epoch": 0.5111821086261981,
-      "grad_norm": 1.9692663945470987,
+      "grad_norm": 0.15660591509388516,
       "learning_rate": 4.6094674556213014e-07,
-      "logits/chosen": -3.4375,
-      "logits/rejected": -3.296875,
-      "logps/chosen": -149.0,
-      "logps/rejected": -416.0,
-      "loss": 0.0001,
+      "logits/chosen": -3.0625,
+      "logits/rejected": -2.84375,
+      "logps/chosen": -164.0,
+      "logps/rejected": -320.0,
+      "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.396484375,
-      "rewards/margins": 26.375,
-      "rewards/rejected": -26.0,
+      "rewards/chosen": -0.2392578125,
+      "rewards/margins": 19.25,
+      "rewards/rejected": -19.5,
       "step": 160
     },
     {
       "epoch": 0.5431309904153354,
-      "grad_norm": 0.08170048165908884,
+      "grad_norm": 0.07414583594558251,
       "learning_rate": 4.5502958579881655e-07,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -166.0,
-      "logps/rejected": -422.0,
-      "loss": 0.0001,
+      "logits/chosen": -3.296875,
+      "logits/rejected": -3.078125,
+      "logps/chosen": -150.0,
+      "logps/rejected": -302.0,
+      "loss": 0.0002,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.140625,
-      "rewards/margins": 25.0,
-      "rewards/rejected": -27.125,
+      "rewards/chosen": 0.134765625,
+      "rewards/margins": 18.5,
+      "rewards/rejected": -18.375,
       "step": 170
     },
     {
       "epoch": 0.5750798722044729,
-      "grad_norm": 0.0035499679004452286,
+      "grad_norm": 0.01835763457343932,
       "learning_rate": 4.491124260355029e-07,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.375,
-      "logps/chosen": -175.0,
-      "logps/rejected": -422.0,
-      "loss": 0.0003,
+      "logits/chosen": -3.25,
+      "logits/rejected": -3.046875,
+      "logps/chosen": -162.0,
+      "logps/rejected": -304.0,
+      "loss": 0.0023,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.21875,
-      "rewards/margins": 23.75,
-      "rewards/rejected": -26.0,
+      "rewards/chosen": -0.02294921875,
+      "rewards/margins": 17.625,
+      "rewards/rejected": -17.625,
       "step": 180
     },
     {
       "epoch": 0.6070287539936102,
-      "grad_norm": 0.04104575399817043,
+      "grad_norm": 0.019396673692171084,
       "learning_rate": 4.4319526627218936e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -176.0,
-      "logps/rejected": -440.0,
+      "logits/chosen": -3.296875,
+      "logits/rejected": -3.125,
+      "logps/chosen": -162.0,
+      "logps/rejected": -304.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -2.3125,
-      "rewards/margins": 26.25,
-      "rewards/rejected": -28.5,
+      "rewards/chosen": -0.0732421875,
+      "rewards/margins": 18.375,
+      "rewards/rejected": -18.375,
       "step": 190
     },
     {
       "epoch": 0.6389776357827476,
-      "grad_norm": 3.7685629191447305e-05,
+      "grad_norm": 0.004950996267230264,
       "learning_rate": 4.372781065088757e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -171.0,
-      "logps/rejected": -448.0,
+      "logits/chosen": -3.21875,
+      "logits/rejected": -3.015625,
+      "logps/chosen": -165.0,
+      "logps/rejected": -316.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -1.4453125,
-      "rewards/margins": 26.375,
-      "rewards/rejected": -27.875,
+      "rewards/chosen": -0.05419921875,
+      "rewards/margins": 18.625,
+      "rewards/rejected": -18.625,
       "step": 200
     },
     {
       "epoch": 0.670926517571885,
-      "grad_norm": 9.004933203527336e-06,
+      "grad_norm": 0.004109028420959876,
       "learning_rate": 4.313609467455621e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.359375,
-      "logps/chosen": -161.0,
-      "logps/rejected": -428.0,
+      "logits/chosen": -3.25,
+      "logits/rejected": -3.03125,
+      "logps/chosen": -164.0,
+      "logps/rejected": -312.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.388671875,
-      "rewards/margins": 27.5,
-      "rewards/rejected": -28.0,
+      "rewards/chosen": 0.02001953125,
+      "rewards/margins": 19.375,
+      "rewards/rejected": -19.375,
       "step": 210
     },
     {
       "epoch": 0.7028753993610224,
-      "grad_norm": 0.0029222087631083824,
+      "grad_norm": 0.03343721361256651,
       "learning_rate": 4.2544378698224847e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -147.0,
-      "logps/rejected": -460.0,
+      "logits/chosen": -3.21875,
+      "logits/rejected": -2.921875,
+      "logps/chosen": -149.0,
+      "logps/rejected": -320.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.326171875,
-      "rewards/margins": 30.5,
-      "rewards/rejected": -30.875,
+      "rewards/chosen": -0.07177734375,
+      "rewards/margins": 19.875,
+      "rewards/rejected": -20.0,
       "step": 220
     },
     {
       "epoch": 0.7348242811501597,
-      "grad_norm": 0.00044147141232812285,
+      "grad_norm": 0.0009315852968474835,
       "learning_rate": 4.195266272189349e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -146.0,
-      "logps/rejected": -456.0,
+      "logits/chosen": -3.15625,
+      "logits/rejected": -2.921875,
+      "logps/chosen": -147.0,
+      "logps/rejected": -330.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.51953125,
-      "rewards/margins": 29.5,
-      "rewards/rejected": -30.125,
+      "rewards/chosen": -0.019775390625,
+      "rewards/margins": 20.875,
+      "rewards/rejected": -20.875,
       "step": 230
     },
     {
       "epoch": 0.7667731629392971,
-      "grad_norm": 0.010403918315756865,
+      "grad_norm": 0.006975720705464563,
       "learning_rate": 4.1360946745562133e-07,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -152.0,
-      "logps/rejected": -436.0,
+      "logits/chosen": -3.1875,
+      "logits/rejected": -2.9375,
+      "logps/chosen": -158.0,
+      "logps/rejected": -318.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.2294921875,
-      "rewards/margins": 28.0,
-      "rewards/rejected": -28.25,
+      "rewards/chosen": 0.0037841796875,
+      "rewards/margins": 19.5,
+      "rewards/rejected": -19.5,
       "step": 240
     },
     {
       "epoch": 0.7987220447284346,
-      "grad_norm": 0.015941514809629027,
+      "grad_norm": 0.007054899693969346,
       "learning_rate": 4.076923076923077e-07,
-      "logits/chosen": -3.546875,
-      "logits/rejected": -3.390625,
-      "logps/chosen": -164.0,
-      "logps/rejected": -452.0,
+      "logits/chosen": -3.234375,
+      "logits/rejected": -3.046875,
+      "logps/chosen": -169.0,
+      "logps/rejected": -330.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.5546875,
-      "rewards/margins": 28.875,
-      "rewards/rejected": -29.5,
+      "rewards/chosen": -0.09228515625,
+      "rewards/margins": 20.5,
+      "rewards/rejected": -20.625,
       "step": 250
     },
     {
       "epoch": 0.8306709265175719,
-      "grad_norm": 5.9135229813990824e-05,
+      "grad_norm": 0.00029837762630587707,
       "learning_rate": 4.017751479289941e-07,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -138.0,
-      "logps/rejected": -446.0,
+      "logits/chosen": -3.140625,
+      "logits/rejected": -2.90625,
+      "logps/chosen": -142.0,
+      "logps/rejected": -324.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.00848388671875,
-      "rewards/margins": 28.75,
-      "rewards/rejected": -28.75,
+      "rewards/chosen": 0.0791015625,
+      "rewards/margins": 20.25,
+      "rewards/rejected": -20.125,
       "step": 260
     },
     {
       "epoch": 0.8626198083067093,
-      "grad_norm": 3.054828362608032e-05,
+      "grad_norm": 0.0018546472858726839,
       "learning_rate": 3.9585798816568044e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -164.0,
-      "logps/rejected": -452.0,
+      "logits/chosen": -3.125,
+      "logits/rejected": -2.84375,
+      "logps/chosen": -172.0,
+      "logps/rejected": -332.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.365234375,
-      "rewards/margins": 29.0,
-      "rewards/rejected": -29.375,
+      "rewards/chosen": -0.2119140625,
+      "rewards/margins": 20.75,
+      "rewards/rejected": -21.0,
       "step": 270
     },
     {
       "epoch": 0.8945686900958466,
-      "grad_norm": 0.00011745687765947919,
+      "grad_norm": 0.0007398452733728992,
       "learning_rate": 3.8994082840236685e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.359375,
-      "logps/chosen": -186.0,
-      "logps/rejected": -464.0,
-      "loss": 0.0001,
+      "logits/chosen": -3.078125,
+      "logits/rejected": -2.890625,
+      "logps/chosen": -192.0,
+      "logps/rejected": -346.0,
+      "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.515625,
-      "rewards/margins": 30.5,
-      "rewards/rejected": -31.0,
+      "rewards/chosen": -0.10986328125,
+      "rewards/margins": 22.375,
+      "rewards/rejected": -22.5,
       "step": 280
     },
     {
       "epoch": 0.9265175718849841,
-      "grad_norm": 1.417074129799485e-06,
+      "grad_norm": 0.007250178415054528,
       "learning_rate": 3.840236686390532e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -149.0,
-      "logps/rejected": -478.0,
+      "logits/chosen": -3.09375,
+      "logits/rejected": -2.90625,
+      "logps/chosen": -156.0,
+      "logps/rejected": -338.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.05810546875,
-      "rewards/margins": 31.75,
-      "rewards/rejected": -31.75,
+      "rewards/chosen": -0.1318359375,
+      "rewards/margins": 21.125,
+      "rewards/rejected": -21.25,
       "step": 290
     },
     {
       "epoch": 0.9584664536741214,
-      "grad_norm": 9.730034757690613e-07,
+      "grad_norm": 0.0017287529388297464,
       "learning_rate": 3.7810650887573966e-07,
-      "logits/chosen": -3.546875,
-      "logits/rejected": -3.359375,
-      "logps/chosen": -174.0,
-      "logps/rejected": -466.0,
+      "logits/chosen": -3.140625,
+      "logits/rejected": -2.890625,
+      "logps/chosen": -180.0,
+      "logps/rejected": -338.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.58203125,
-      "rewards/margins": 30.5,
-      "rewards/rejected": -31.0,
+      "rewards/chosen": -0.3125,
+      "rewards/margins": 21.25,
+      "rewards/rejected": -21.625,
       "step": 300
     },
     {
       "epoch": 0.9904153354632588,
-      "grad_norm": 0.0022653315379438928,
+      "grad_norm": 0.013205636145569603,
       "learning_rate": 3.72189349112426e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.375,
-      "logps/chosen": -182.0,
-      "logps/rejected": -462.0,
+      "logits/chosen": -3.109375,
+      "logits/rejected": -2.921875,
+      "logps/chosen": -192.0,
+      "logps/rejected": -340.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.2177734375,
-      "rewards/margins": 30.0,
-      "rewards/rejected": -30.125,
+      "rewards/chosen": -0.271484375,
+      "rewards/margins": 21.0,
+      "rewards/rejected": -21.25,
       "step": 310
     },
     {
       "epoch": 1.0,
-      "eval_logits/chosen": -3.53125,
-      "eval_logits/rejected": -3.296875,
-      "eval_logps/chosen": -150.0,
-      "eval_logps/rejected": -476.0,
-      "eval_loss": 3.5477933124639094e-07,
+      "eval_logits/chosen": -3.21875,
+      "eval_logits/rejected": -2.921875,
+      "eval_logps/chosen": -151.0,
+      "eval_logps/rejected": -340.0,
+      "eval_loss": 7.380839633697178e-06,
       "eval_rewards/accuracies": 1.0,
-      "eval_rewards/chosen": -0.275390625,
-      "eval_rewards/margins": 33.0,
-      "eval_rewards/rejected": -33.25,
-      "eval_runtime": 15.4158,
-      "eval_samples_per_second": 12.974,
-      "eval_steps_per_second": 0.454,
+      "eval_rewards/chosen": -0.031982421875,
+      "eval_rewards/margins": 22.0,
+      "eval_rewards/rejected": -22.125,
+      "eval_runtime": 12.3794,
+      "eval_samples_per_second": 16.156,
+      "eval_steps_per_second": 0.565,
       "step": 313
     },
     {
       "epoch": 1.0223642172523961,
-      "grad_norm": 0.0003356864204729289,
+      "grad_norm": 0.001081863238066753,
       "learning_rate": 3.662721893491124e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.296875,
-      "logps/chosen": -169.0,
-      "logps/rejected": -478.0,
+      "logits/chosen": -3.09375,
+      "logits/rejected": -2.875,
+      "logps/chosen": -178.0,
+      "logps/rejected": -340.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.498046875,
-      "rewards/margins": 31.125,
-      "rewards/rejected": -31.625,
+      "rewards/chosen": -0.228515625,
+      "rewards/margins": 21.375,
+      "rewards/rejected": -21.5,
       "step": 320
     },
     {
       "epoch": 1.0543130990415335,
-      "grad_norm": 5.8557721184616675e-06,
+      "grad_norm": 0.0014656429504141843,
       "learning_rate": 3.603550295857988e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -156.0,
-      "logps/rejected": -480.0,
+      "logits/chosen": -3.171875,
+      "logits/rejected": -2.90625,
+      "logps/chosen": -164.0,
+      "logps/rejected": -344.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.1435546875,
-      "rewards/margins": 32.5,
-      "rewards/rejected": -32.75,
+      "rewards/chosen": -0.0927734375,
+      "rewards/margins": 22.0,
+      "rewards/rejected": -22.0,
       "step": 330
     },
     {
       "epoch": 1.0862619808306708,
-      "grad_norm": 3.5777121751233287e-06,
+      "grad_norm": 9.752625657144038e-05,
       "learning_rate": 3.544378698224852e-07,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.296875,
-      "logps/chosen": -144.0,
-      "logps/rejected": -478.0,
+      "logits/chosen": -3.171875,
+      "logits/rejected": -2.90625,
+      "logps/chosen": -153.0,
+      "logps/rejected": -338.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.03955078125,
-      "rewards/margins": 30.75,
-      "rewards/rejected": -30.75,
+      "rewards/chosen": -0.0031585693359375,
+      "rewards/margins": 20.875,
+      "rewards/rejected": -20.875,
       "step": 340
     },
     {
       "epoch": 1.1182108626198084,
-      "grad_norm": 6.936406729828301e-07,
+      "grad_norm": 5.907578925715793e-05,
       "learning_rate": 3.485207100591716e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -154.0,
-      "logps/rejected": -492.0,
+      "logits/chosen": -3.140625,
+      "logits/rejected": -2.90625,
+      "logps/chosen": -158.0,
+      "logps/rejected": -354.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.220703125,
-      "rewards/margins": 32.75,
-      "rewards/rejected": -33.0,
+      "rewards/chosen": -0.1630859375,
+      "rewards/margins": 22.75,
+      "rewards/rejected": -22.875,
       "step": 350
     },
     {
       "epoch": 1.1501597444089458,
-      "grad_norm": 0.009589555434561007,
+      "grad_norm": 0.008399378571967797,
       "learning_rate": 3.42603550295858e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -168.0,
-      "logps/rejected": -480.0,
+      "logits/chosen": -3.140625,
+      "logits/rejected": -2.90625,
+      "logps/chosen": -175.0,
+      "logps/rejected": -348.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.146484375,
-      "rewards/margins": 32.0,
-      "rewards/rejected": -32.25,
+      "rewards/chosen": -0.0084228515625,
+      "rewards/margins": 22.125,
+      "rewards/rejected": -22.125,
       "step": 360
     },
     {
       "epoch": 1.182108626198083,
-      "grad_norm": 2.378099652849891e-05,
+      "grad_norm": 0.00096142832181881,
       "learning_rate": 3.366863905325444e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -137.0,
-      "logps/rejected": -472.0,
+      "logits/chosen": -3.109375,
+      "logits/rejected": -2.875,
+      "logps/chosen": -146.0,
+      "logps/rejected": -356.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.1630859375,
-      "rewards/margins": 31.5,
-      "rewards/rejected": -31.375,
+      "rewards/chosen": -0.034912109375,
+      "rewards/margins": 22.875,
+      "rewards/rejected": -22.875,
       "step": 370
     },
     {
       "epoch": 1.2140575079872205,
-      "grad_norm": 5.514648329432445e-05,
+      "grad_norm": 1.117707679199626e-05,
       "learning_rate": 3.3076923076923075e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.390625,
-      "logps/chosen": -151.0,
-      "logps/rejected": -484.0,
+      "logits/chosen": -3.140625,
+      "logits/rejected": -3.0,
+      "logps/chosen": -158.0,
+      "logps/rejected": -352.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.03662109375,
-      "rewards/margins": 32.75,
-      "rewards/rejected": -32.75,
+      "rewards/chosen": -0.10986328125,
+      "rewards/margins": 22.75,
+      "rewards/rejected": -22.75,
       "step": 380
     },
     {
       "epoch": 1.2460063897763578,
-      "grad_norm": 5.280407319654463e-07,
+      "grad_norm": 3.391687654228242e-05,
       "learning_rate": 3.2485207100591716e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -127.0,
-      "logps/rejected": -480.0,
+      "logits/chosen": -3.078125,
+      "logits/rejected": -2.84375,
+      "logps/chosen": -134.0,
+      "logps/rejected": -352.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.23046875,
-      "rewards/margins": 33.0,
-      "rewards/rejected": -32.75,
+      "rewards/chosen": -0.0036773681640625,
+      "rewards/margins": 23.0,
+      "rewards/rejected": -23.0,
       "step": 390
     },
     {
       "epoch": 1.2779552715654952,
-      "grad_norm": 0.00046164498605506235,
+      "grad_norm": 0.00015741926047163637,
       "learning_rate": 3.189349112426035e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.359375,
-      "logps/chosen": -132.0,
-      "logps/rejected": -486.0,
+      "logits/chosen": -3.0,
+      "logits/rejected": -2.890625,
+      "logps/chosen": -141.0,
+      "logps/rejected": -358.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.1435546875,
-      "rewards/margins": 33.0,
-      "rewards/rejected": -33.0,
+      "rewards/chosen": -0.318359375,
+      "rewards/margins": 23.5,
+      "rewards/rejected": -23.75,
       "step": 400
     },
     {
       "epoch": 1.3099041533546325,
-      "grad_norm": 0.0001779209961052298,
+      "grad_norm": 0.001381822064334206,
       "learning_rate": 3.130177514792899e-07,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.265625,
-      "logps/chosen": -144.0,
-      "logps/rejected": -496.0,
+      "logits/chosen": -3.09375,
+      "logits/rejected": -2.765625,
+      "logps/chosen": -150.0,
+      "logps/rejected": -360.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.07177734375,
-      "rewards/margins": 35.0,
-      "rewards/rejected": -35.0,
+      "rewards/chosen": -0.07568359375,
+      "rewards/margins": 24.125,
+      "rewards/rejected": -24.25,
       "step": 410
     },
     {
       "epoch": 1.34185303514377,
-      "grad_norm": 4.9071818579178655e-05,
+      "grad_norm": 0.0008640673608836212,
       "learning_rate": 3.071005917159763e-07,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.375,
-      "logps/chosen": -160.0,
-      "logps/rejected": -478.0,
+      "logits/chosen": -3.078125,
+      "logits/rejected": -2.9375,
+      "logps/chosen": -169.0,
+      "logps/rejected": -356.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.12158203125,
-      "rewards/margins": 32.5,
-      "rewards/rejected": -32.5,
+      "rewards/chosen": -0.244140625,
+      "rewards/margins": 23.25,
+      "rewards/rejected": -23.5,
       "step": 420
     },
     {
       "epoch": 1.3738019169329074,
-      "grad_norm": 4.25683016680881e-06,
+      "grad_norm": 0.00010813946327226968,
       "learning_rate": 3.011834319526627e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -149.0,
-      "logps/rejected": -490.0,
+      "logits/chosen": -3.125,
+      "logits/rejected": -2.859375,
+      "logps/chosen": -157.0,
+      "logps/rejected": -364.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.1220703125,
-      "rewards/margins": 33.75,
-      "rewards/rejected": -33.75,
+      "rewards/chosen": -0.2255859375,
+      "rewards/margins": 23.875,
+      "rewards/rejected": -24.125,
       "step": 430
     },
     {
       "epoch": 1.4057507987220448,
-      "grad_norm": 0.0009887420156956786,
+      "grad_norm": 0.0001540987124577923,
       "learning_rate": 2.952662721893491e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.28125,
-      "logps/chosen": -157.0,
-      "logps/rejected": -482.0,
+      "logits/chosen": -3.078125,
+      "logits/rejected": -2.859375,
+      "logps/chosen": -164.0,
+      "logps/rejected": -360.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.130859375,
-      "rewards/margins": 32.5,
-      "rewards/rejected": -32.5,
+      "rewards/chosen": -0.13671875,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -23.5,
       "step": 440
     },
     {
       "epoch": 1.4376996805111821,
-      "grad_norm": 1.4837415298689954e-05,
+      "grad_norm": 0.0003904163069516865,
       "learning_rate": 2.893491124260355e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -156.0,
-      "logps/rejected": -488.0,
+      "logits/chosen": -3.0625,
+      "logits/rejected": -2.84375,
+      "logps/chosen": -166.0,
+      "logps/rejected": -364.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.12890625,
-      "rewards/margins": 34.75,
-      "rewards/rejected": -34.5,
+      "rewards/chosen": -0.0277099609375,
+      "rewards/margins": 24.625,
+      "rewards/rejected": -24.75,
       "step": 450
     },
     {
       "epoch": 1.4696485623003195,
-      "grad_norm": 8.248177987361765e-06,
+      "grad_norm": 0.00010719442458275845,
       "learning_rate": 2.834319526627219e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.375,
-      "logps/chosen": -148.0,
-      "logps/rejected": -498.0,
+      "logits/chosen": -3.03125,
+      "logits/rejected": -2.84375,
+      "logps/chosen": -156.0,
+      "logps/rejected": -348.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.150390625,
-      "rewards/margins": 34.5,
-      "rewards/rejected": -34.5,
+      "rewards/chosen": -0.431640625,
+      "rewards/margins": 22.25,
+      "rewards/rejected": -22.75,
       "step": 460
     },
     {
       "epoch": 1.5015974440894568,
-      "grad_norm": 1.9806936842810454e-05,
+      "grad_norm": 4.819727273913693e-05,
       "learning_rate": 2.7751479289940824e-07,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.375,
-      "logps/chosen": -157.0,
-      "logps/rejected": -478.0,
+      "logits/chosen": -3.03125,
+      "logits/rejected": -2.890625,
+      "logps/chosen": -164.0,
+      "logps/rejected": -362.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.396484375,
-      "rewards/margins": 32.25,
-      "rewards/rejected": -32.5,
+      "rewards/chosen": -0.33203125,
+      "rewards/margins": 24.0,
+      "rewards/rejected": -24.375,
       "step": 470
     },
     {
       "epoch": 1.5335463258785942,
-      "grad_norm": 5.907801581150698e-07,
+      "grad_norm": 8.148529465105985e-05,
       "learning_rate": 2.715976331360947e-07,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.390625,
-      "logps/chosen": -153.0,
-      "logps/rejected": -460.0,
+      "logits/chosen": -3.03125,
+      "logits/rejected": -2.875,
+      "logps/chosen": -159.0,
+      "logps/rejected": -356.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.17578125,
-      "rewards/margins": 30.5,
-      "rewards/rejected": -30.75,
+      "rewards/chosen": -0.318359375,
+      "rewards/margins": 22.875,
+      "rewards/rejected": -23.25,
       "step": 480
     },
     {
       "epoch": 1.5654952076677318,
-      "grad_norm": 0.00014451950487479103,
+      "grad_norm": 0.00023912166796823171,
       "learning_rate": 2.6568047337278106e-07,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -158.0,
-      "logps/rejected": -470.0,
+      "logits/chosen": -3.0625,
+      "logits/rejected": -2.84375,
+      "logps/chosen": -167.0,
+      "logps/rejected": -364.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.0057373046875,
-      "rewards/margins": 32.25,
-      "rewards/rejected": -32.25,
+      "rewards/chosen": -0.33203125,
+      "rewards/margins": 23.875,
+      "rewards/rejected": -24.25,
       "step": 490
     },
     {
       "epoch": 1.5974440894568689,
-      "grad_norm": 0.0012272802134721346,
+      "grad_norm": 0.00038137779946660004,
       "learning_rate": 2.5976331360946746e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -148.0,
-      "logps/rejected": -482.0,
+      "logits/chosen": -3.03125,
+      "logits/rejected": -2.765625,
+      "logps/chosen": -161.0,
+      "logps/rejected": -358.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.19921875,
-      "rewards/margins": 33.0,
-      "rewards/rejected": -32.75,
+      "rewards/chosen": -0.09912109375,
+      "rewards/margins": 23.375,
+      "rewards/rejected": -23.5,
       "step": 500
     },
     {
       "epoch": 1.6293929712460065,
-      "grad_norm": 2.9833293563020383e-05,
+      "grad_norm": 0.00020618761735157146,
       "learning_rate": 2.538461538461538e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.28125,
-      "logps/chosen": -150.0,
-      "logps/rejected": -494.0,
+      "logits/chosen": -2.96875,
+      "logits/rejected": -2.75,
+      "logps/chosen": -164.0,
+      "logps/rejected": -374.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.1962890625,
-      "rewards/margins": 34.75,
-      "rewards/rejected": -34.75,
+      "rewards/chosen": -0.26953125,
+      "rewards/margins": 25.125,
+      "rewards/rejected": -25.375,
       "step": 510
     },
     {
       "epoch": 1.6613418530351438,
-      "grad_norm": 0.001356611209656493,
+      "grad_norm": 0.00173536920939462,
       "learning_rate": 2.479289940828402e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -135.0,
-      "logps/rejected": -500.0,
+      "logits/chosen": -2.984375,
+      "logits/rejected": -2.828125,
+      "logps/chosen": -143.0,
+      "logps/rejected": -378.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.035400390625,
-      "rewards/margins": 34.75,
-      "rewards/rejected": -34.75,
+      "rewards/chosen": -0.2431640625,
+      "rewards/margins": 25.125,
+      "rewards/rejected": -25.375,
       "step": 520
     },
     {
       "epoch": 1.6932907348242812,
-      "grad_norm": 0.00022456450452624014,
+      "grad_norm": 0.00036883452678222923,
       "learning_rate": 2.420118343195266e-07,
-      "logits/chosen": -3.53125,
-      "logits/rejected": -3.359375,
-      "logps/chosen": -166.0,
-      "logps/rejected": -484.0,
+      "logits/chosen": -2.96875,
+      "logits/rejected": -2.8125,
+      "logps/chosen": -179.0,
+      "logps/rejected": -374.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.35546875,
-      "rewards/margins": 31.5,
-      "rewards/rejected": -31.75,
+      "rewards/chosen": -0.73828125,
+      "rewards/margins": 23.75,
+      "rewards/rejected": -24.5,
       "step": 530
     },
     {
       "epoch": 1.7252396166134185,
-      "grad_norm": 6.820094977898695e-06,
+      "grad_norm": 0.00046710302845877187,
       "learning_rate": 2.36094674556213e-07,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -169.0,
-      "logps/rejected": -478.0,
+      "logits/chosen": -2.953125,
+      "logits/rejected": -2.78125,
+      "logps/chosen": -183.0,
+      "logps/rejected": -364.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.298828125,
-      "rewards/margins": 33.25,
-      "rewards/rejected": -33.5,
+      "rewards/chosen": -0.54296875,
+      "rewards/margins": 24.25,
+      "rewards/rejected": -24.75,
       "step": 540
     },
     {
       "epoch": 1.7571884984025559,
-      "grad_norm": 0.001992382751935317,
+      "grad_norm": 0.0012311707501514388,
       "learning_rate": 2.301775147928994e-07,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.296875,
-      "logps/chosen": -171.0,
-      "logps/rejected": -492.0,
+      "logits/chosen": -2.984375,
+      "logits/rejected": -2.765625,
+      "logps/chosen": -184.0,
+      "logps/rejected": -370.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.10107421875,
-      "rewards/margins": 33.5,
-      "rewards/rejected": -33.5,
+      "rewards/chosen": -0.349609375,
+      "rewards/margins": 24.5,
+      "rewards/rejected": -24.875,
       "step": 550
     },
     {
       "epoch": 1.7891373801916934,
-      "grad_norm": 1.115393663415116e-05,
+      "grad_norm": 6.751851643657405e-05,
       "learning_rate": 2.242603550295858e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.375,
-      "logps/chosen": -142.0,
-      "logps/rejected": -512.0,
+      "logits/chosen": -2.9375,
+      "logits/rejected": -2.8125,
+      "logps/chosen": -155.0,
+      "logps/rejected": -360.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.16796875,
-      "rewards/margins": 35.5,
-      "rewards/rejected": -35.5,
+      "rewards/chosen": -0.326171875,
+      "rewards/margins": 23.75,
+      "rewards/rejected": -24.0,
       "step": 560
     },
     {
       "epoch": 1.8210862619808306,
-      "grad_norm": 1.489072794590662e-05,
+      "grad_norm": 0.00011924326319637662,
       "learning_rate": 2.1834319526627217e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.359375,
-      "logps/chosen": -154.0,
-      "logps/rejected": -476.0,
+      "logits/chosen": -3.0,
+      "logits/rejected": -2.8125,
+      "logps/chosen": -164.0,
+      "logps/rejected": -358.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.06640625,
-      "rewards/margins": 32.0,
-      "rewards/rejected": -32.0,
+      "rewards/chosen": -0.40625,
+      "rewards/margins": 23.25,
+      "rewards/rejected": -23.625,
       "step": 570
     },
     {
       "epoch": 1.8530351437699681,
-      "grad_norm": 8.145269082773688e-06,
+      "grad_norm": 1.591754779283049e-05,
       "learning_rate": 2.1242603550295858e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.359375,
-      "logps/chosen": -154.0,
-      "logps/rejected": -472.0,
+      "logits/chosen": -3.078125,
+      "logits/rejected": -2.8125,
+      "logps/chosen": -162.0,
+      "logps/rejected": -364.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.10400390625,
-      "rewards/margins": 32.25,
-      "rewards/rejected": -32.25,
+      "rewards/chosen": -0.369140625,
+      "rewards/margins": 24.0,
+      "rewards/rejected": -24.375,
       "step": 580
     },
     {
       "epoch": 1.8849840255591053,
-      "grad_norm": 0.0005030582745028686,
+      "grad_norm": 0.001273109125943514,
       "learning_rate": 2.0650887573964496e-07,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -154.0,
-      "logps/rejected": -488.0,
+      "logits/chosen": -3.03125,
+      "logits/rejected": -2.75,
+      "logps/chosen": -163.0,
+      "logps/rejected": -366.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.003448486328125,
-      "rewards/margins": 32.0,
-      "rewards/rejected": -32.0,
+      "rewards/chosen": -0.3828125,
+      "rewards/margins": 23.25,
+      "rewards/rejected": -23.625,
       "step": 590
     },
     {
       "epoch": 1.9169329073482428,
-      "grad_norm": 9.704683387564733e-06,
+      "grad_norm": 5.600029962176166e-05,
       "learning_rate": 2.0059171597633133e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -150.0,
-      "logps/rejected": -474.0,
+      "logits/chosen": -2.953125,
+      "logits/rejected": -2.78125,
+      "logps/chosen": -158.0,
+      "logps/rejected": -368.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.1357421875,
-      "rewards/margins": 31.375,
-      "rewards/rejected": -31.5,
+      "rewards/chosen": -0.283203125,
+      "rewards/margins": 24.125,
+      "rewards/rejected": -24.5,
       "step": 600
     },
     {
       "epoch": 1.9488817891373802,
-      "grad_norm": 1.7369899748244282e-05,
+      "grad_norm": 2.143142202843072e-05,
       "learning_rate": 1.9467455621301774e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -156.0,
-      "logps/rejected": -492.0,
+      "logits/chosen": -2.9375,
+      "logits/rejected": -2.734375,
+      "logps/chosen": -165.0,
+      "logps/rejected": -382.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.470703125,
-      "rewards/margins": 33.5,
-      "rewards/rejected": -33.75,
+      "rewards/chosen": -0.68359375,
+      "rewards/margins": 25.125,
+      "rewards/rejected": -25.75,
       "step": 610
     },
     {
       "epoch": 1.9808306709265175,
-      "grad_norm": 1.8988577382856775e-06,
+      "grad_norm": 0.00015877860321082202,
       "learning_rate": 1.8875739644970412e-07,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -146.0,
-      "logps/rejected": -516.0,
+      "logits/chosen": -2.921875,
+      "logits/rejected": -2.734375,
+      "logps/chosen": -153.0,
+      "logps/rejected": -380.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.326171875,
-      "rewards/margins": 34.5,
-      "rewards/rejected": -34.75,
+      "rewards/chosen": -0.6015625,
+      "rewards/margins": 24.625,
+      "rewards/rejected": -25.25,
       "step": 620
     },
     {
       "epoch": 2.0,
-      "eval_logits/chosen": -3.53125,
-      "eval_logits/rejected": -3.296875,
-      "eval_logps/chosen": -150.0,
-      "eval_logps/rejected": -498.0,
-      "eval_loss": 3.6485798204921593e-07,
+      "eval_logits/chosen": -3.0,
+      "eval_logits/rejected": -2.75,
+      "eval_logps/chosen": -155.0,
+      "eval_logps/rejected": -376.0,
+      "eval_loss": 1.6905501070141327e-06,
       "eval_rewards/accuracies": 1.0,
-      "eval_rewards/chosen": -0.265625,
-      "eval_rewards/margins": 35.25,
-      "eval_rewards/rejected": -35.5,
-      "eval_runtime": 15.435,
-      "eval_samples_per_second": 12.958,
-      "eval_steps_per_second": 0.454,
+      "eval_rewards/chosen": -0.443359375,
+      "eval_rewards/margins": 25.25,
+      "eval_rewards/rejected": -25.625,
+      "eval_runtime": 12.4078,
+      "eval_samples_per_second": 16.119,
+      "eval_steps_per_second": 0.564,
       "step": 626
     },
     {
       "epoch": 2.012779552715655,
-      "grad_norm": 0.00013898470541843215,
+      "grad_norm": 0.0002995311735215579,
       "learning_rate": 1.8284023668639053e-07,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -167.0,
-      "logps/rejected": -504.0,
+      "logits/chosen": -3.0,
+      "logits/rejected": -2.78125,
+      "logps/chosen": -179.0,
+      "logps/rejected": -382.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.208984375,
-      "rewards/margins": 34.75,
-      "rewards/rejected": -34.75,
+      "rewards/chosen": -0.66796875,
+      "rewards/margins": 25.5,
+      "rewards/rejected": -26.125,
       "step": 630
     },
     {
       "epoch": 2.0447284345047922,
-      "grad_norm": 0.0012555863778602044,
+      "grad_norm": 0.0007360144072745451,
       "learning_rate": 1.7692307692307693e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -141.0,
-      "logps/rejected": -496.0,
+      "logits/chosen": -3.0,
+      "logits/rejected": -2.84375,
+      "logps/chosen": -149.0,
+      "logps/rejected": -380.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.047607421875,
-      "rewards/margins": 35.0,
-      "rewards/rejected": -35.0,
+      "rewards/chosen": -0.26171875,
+      "rewards/margins": 26.0,
+      "rewards/rejected": -26.25,
       "step": 640
     },
     {
       "epoch": 2.07667731629393,
-      "grad_norm": 1.8154853505391357e-06,
+      "grad_norm": 7.200122629303882e-05,
       "learning_rate": 1.710059171597633e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -157.0,
-      "logps/rejected": -496.0,
+      "logits/chosen": -2.9375,
+      "logits/rejected": -2.71875,
+      "logps/chosen": -170.0,
+      "logps/rejected": -382.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.1689453125,
-      "rewards/margins": 33.5,
-      "rewards/rejected": -33.5,
+      "rewards/chosen": -0.66796875,
+      "rewards/margins": 25.0,
+      "rewards/rejected": -25.75,
       "step": 650
     },
     {
       "epoch": 2.108626198083067,
-      "grad_norm": 2.4286044775256908e-05,
+      "grad_norm": 0.0005337190402402807,
       "learning_rate": 1.650887573964497e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.28125,
-      "logps/chosen": -162.0,
-      "logps/rejected": -500.0,
+      "logits/chosen": -2.9375,
+      "logits/rejected": -2.703125,
+      "logps/chosen": -175.0,
+      "logps/rejected": -388.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.177734375,
-      "rewards/margins": 35.0,
-      "rewards/rejected": -35.0,
+      "rewards/chosen": -0.5703125,
+      "rewards/margins": 26.0,
+      "rewards/rejected": -26.625,
       "step": 660
     },
     {
       "epoch": 2.1405750798722045,
-      "grad_norm": 9.035767528821231e-06,
+      "grad_norm": 9.639763045686634e-05,
       "learning_rate": 1.591715976331361e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -153.0,
-      "logps/rejected": -496.0,
+      "logits/chosen": -2.90625,
+      "logits/rejected": -2.796875,
+      "logps/chosen": -167.0,
+      "logps/rejected": -376.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.11376953125,
-      "rewards/margins": 34.5,
-      "rewards/rejected": -34.75,
+      "rewards/chosen": -0.609375,
+      "rewards/margins": 25.25,
+      "rewards/rejected": -25.875,
       "step": 670
     },
     {
       "epoch": 2.1725239616613417,
-      "grad_norm": 9.554673179594124e-06,
+      "grad_norm": 6.995868636338041e-05,
       "learning_rate": 1.5325443786982248e-07,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -3.296875,
-      "logps/chosen": -147.0,
-      "logps/rejected": -488.0,
+      "logits/chosen": -2.9375,
+      "logits/rejected": -2.734375,
+      "logps/chosen": -157.0,
+      "logps/rejected": -376.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.050537109375,
-      "rewards/margins": 33.0,
-      "rewards/rejected": -33.0,
+      "rewards/chosen": -0.453125,
+      "rewards/margins": 25.125,
+      "rewards/rejected": -25.625,
       "step": 680
     },
     {
       "epoch": 2.2044728434504792,
-      "grad_norm": 9.479403806459367e-08,
+      "grad_norm": 2.902935917834795e-06,
       "learning_rate": 1.4733727810650885e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.359375,
-      "logps/chosen": -144.0,
-      "logps/rejected": -490.0,
+      "logits/chosen": -2.90625,
+      "logits/rejected": -2.75,
+      "logps/chosen": -157.0,
+      "logps/rejected": -380.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.07177734375,
-      "rewards/margins": 33.5,
-      "rewards/rejected": -33.5,
+      "rewards/chosen": -0.5546875,
+      "rewards/margins": 25.25,
+      "rewards/rejected": -25.875,
       "step": 690
     },
     {
       "epoch": 2.236421725239617,
-      "grad_norm": 1.1818574031797608e-05,
+      "grad_norm": 2.496039811099546e-05,
       "learning_rate": 1.4142011834319526e-07,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -161.0,
-      "logps/rejected": -524.0,
+      "logits/chosen": -2.875,
+      "logits/rejected": -2.734375,
+      "logps/chosen": -176.0,
+      "logps/rejected": -384.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.2412109375,
-      "rewards/margins": 37.25,
-      "rewards/rejected": -37.25,
+      "rewards/chosen": -0.52734375,
+      "rewards/margins": 25.875,
+      "rewards/rejected": -26.375,
       "step": 700
     },
     {
       "epoch": 2.268370607028754,
-      "grad_norm": 0.0011652593057304907,
+      "grad_norm": 0.0007535369831177368,
       "learning_rate": 1.3550295857988164e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -127.5,
-      "logps/rejected": -510.0,
+      "logits/chosen": -2.90625,
+      "logits/rejected": -2.734375,
+      "logps/chosen": -137.0,
+      "logps/rejected": -390.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.043212890625,
-      "rewards/margins": 34.75,
-      "rewards/rejected": -34.75,
+      "rewards/chosen": -0.52734375,
+      "rewards/margins": 25.75,
+      "rewards/rejected": -26.375,
       "step": 710
     },
     {
       "epoch": 2.3003194888178915,
-      "grad_norm": 8.823428133708636e-06,
+      "grad_norm": 3.433513476668793e-05,
       "learning_rate": 1.2958579881656802e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -158.0,
-      "logps/rejected": -488.0,
+      "logits/chosen": -2.921875,
+      "logits/rejected": -2.75,
+      "logps/chosen": -170.0,
+      "logps/rejected": -378.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.018310546875,
-      "rewards/margins": 32.0,
-      "rewards/rejected": -32.0,
+      "rewards/chosen": -0.515625,
+      "rewards/margins": 24.5,
+      "rewards/rejected": -25.125,
       "step": 720
     },
     {
       "epoch": 2.3322683706070286,
-      "grad_norm": 0.00016743392421582047,
+      "grad_norm": 6.19179422080526e-05,
       "learning_rate": 1.2366863905325443e-07,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -165.0,
-      "logps/rejected": -488.0,
+      "logits/chosen": -2.953125,
+      "logits/rejected": -2.6875,
+      "logps/chosen": -176.0,
+      "logps/rejected": -386.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.00885009765625,
-      "rewards/margins": 34.25,
-      "rewards/rejected": -34.25,
+      "rewards/chosen": -0.59375,
+      "rewards/margins": 26.0,
+      "rewards/rejected": -26.625,
       "step": 730
     },
     {
       "epoch": 2.364217252396166,
-      "grad_norm": 0.00012209667004203144,
+      "grad_norm": 7.621063491496295e-05,
       "learning_rate": 1.1775147928994082e-07,
-      "logits/chosen": -3.46875,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -152.0,
-      "logps/rejected": -498.0,
+      "logits/chosen": -2.890625,
+      "logits/rejected": -2.703125,
+      "logps/chosen": -166.0,
+      "logps/rejected": -386.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.1904296875,
-      "rewards/margins": 34.75,
-      "rewards/rejected": -35.0,
+      "rewards/chosen": -0.765625,
+      "rewards/margins": 25.75,
+      "rewards/rejected": -26.5,
       "step": 740
     },
     {
       "epoch": 2.3961661341853033,
-      "grad_norm": 0.002623780699506834,
+      "grad_norm": 0.010658960126496895,
       "learning_rate": 1.1183431952662721e-07,
-      "logits/chosen": -3.53125,
-      "logits/rejected": -3.375,
-      "logps/chosen": -167.0,
-      "logps/rejected": -508.0,
+      "logits/chosen": -2.953125,
+      "logits/rejected": -2.75,
+      "logps/chosen": -175.0,
+      "logps/rejected": -394.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.40234375,
-      "rewards/margins": 34.75,
-      "rewards/rejected": -35.25,
+      "rewards/chosen": -0.7578125,
+      "rewards/margins": 26.75,
+      "rewards/rejected": -27.375,
       "step": 750
     },
     {
       "epoch": 2.428115015974441,
-      "grad_norm": 5.726021462588604e-05,
+      "grad_norm": 1.9886464080428738e-05,
       "learning_rate": 1.059171597633136e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -149.0,
-      "logps/rejected": -516.0,
+      "logits/chosen": -2.90625,
+      "logits/rejected": -2.71875,
+      "logps/chosen": -158.0,
+      "logps/rejected": -390.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.482421875,
-      "rewards/margins": 34.5,
-      "rewards/rejected": -35.0,
+      "rewards/chosen": -0.7109375,
+      "rewards/margins": 25.375,
+      "rewards/rejected": -26.125,
       "step": 760
     },
     {
       "epoch": 2.460063897763578,
-      "grad_norm": 0.0013698565362618975,
+      "grad_norm": 0.0002488714626064026,
       "learning_rate": 1e-07,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.359375,
-      "logps/chosen": -141.0,
-      "logps/rejected": -502.0,
+      "logits/chosen": -2.890625,
+      "logits/rejected": -2.75,
+      "logps/chosen": -150.0,
+      "logps/rejected": -400.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.216796875,
-      "rewards/margins": 33.5,
-      "rewards/rejected": -33.75,
+      "rewards/chosen": -0.86328125,
+      "rewards/margins": 26.25,
+      "rewards/rejected": -27.125,
       "step": 770
     },
     {
       "epoch": 2.4920127795527156,
-      "grad_norm": 2.7783510966803164e-07,
+      "grad_norm": 0.00016121672886946543,
       "learning_rate": 9.408284023668639e-08,
-      "logits/chosen": -3.515625,
-      "logits/rejected": -3.375,
-      "logps/chosen": -158.0,
-      "logps/rejected": -516.0,
+      "logits/chosen": -2.890625,
+      "logits/rejected": -2.75,
+      "logps/chosen": -174.0,
+      "logps/rejected": -394.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.095703125,
-      "rewards/margins": 35.5,
-      "rewards/rejected": -35.5,
+      "rewards/chosen": -0.84375,
+      "rewards/margins": 25.75,
+      "rewards/rejected": -26.625,
       "step": 780
     },
     {
       "epoch": 2.523961661341853,
-      "grad_norm": 7.4293806602695065e-06,
+      "grad_norm": 5.202363139806019e-05,
       "learning_rate": 8.816568047337278e-08,
-      "logits/chosen": -3.453125,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -155.0,
-      "logps/rejected": -506.0,
+      "logits/chosen": -2.796875,
+      "logits/rejected": -2.640625,
+      "logps/chosen": -170.0,
+      "logps/rejected": -400.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.431640625,
-      "rewards/margins": 33.75,
-      "rewards/rejected": -34.0,
+      "rewards/chosen": -0.9921875,
+      "rewards/margins": 26.125,
+      "rewards/rejected": -27.25,
       "step": 790
     },
     {
       "epoch": 2.5559105431309903,
-      "grad_norm": 7.974176579773915e-05,
+      "grad_norm": 0.00026720074074842884,
       "learning_rate": 8.224852071005916e-08,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -170.0,
-      "logps/rejected": -510.0,
+      "logits/chosen": -2.84375,
+      "logits/rejected": -2.671875,
+      "logps/chosen": -180.0,
+      "logps/rejected": -394.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.376953125,
-      "rewards/margins": 34.75,
-      "rewards/rejected": -35.25,
+      "rewards/chosen": -0.91015625,
+      "rewards/margins": 25.875,
+      "rewards/rejected": -26.75,
       "step": 800
     },
     {
       "epoch": 2.587859424920128,
-      "grad_norm": 4.007176416439276e-07,
+      "grad_norm": 4.582941461064897e-06,
       "learning_rate": 7.633136094674555e-08,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -137.0,
-      "logps/rejected": -478.0,
+      "logits/chosen": -2.84375,
+      "logits/rejected": -2.640625,
+      "logps/chosen": -150.0,
+      "logps/rejected": -370.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.01239013671875,
-      "rewards/margins": 32.75,
-      "rewards/rejected": -32.75,
+      "rewards/chosen": -0.6328125,
+      "rewards/margins": 24.5,
+      "rewards/rejected": -25.125,
       "step": 810
     },
     {
       "epoch": 2.619808306709265,
-      "grad_norm": 4.7177786599929654e-05,
+      "grad_norm": 0.00013491541270218465,
       "learning_rate": 7.041420118343195e-08,
-      "logits/chosen": -3.484375,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -161.0,
-      "logps/rejected": -516.0,
+      "logits/chosen": -2.859375,
+      "logits/rejected": -2.6875,
+      "logps/chosen": -172.0,
+      "logps/rejected": -400.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.44140625,
-      "rewards/margins": 36.0,
-      "rewards/rejected": -36.25,
+      "rewards/chosen": -0.8828125,
+      "rewards/margins": 27.0,
+      "rewards/rejected": -27.875,
       "step": 820
     },
     {
       "epoch": 2.6517571884984026,
-      "grad_norm": 2.1394562820460947e-06,
+      "grad_norm": 9.158762056685416e-05,
       "learning_rate": 6.449704142011835e-08,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -163.0,
-      "logps/rejected": -500.0,
+      "logits/chosen": -2.890625,
+      "logits/rejected": -2.703125,
+      "logps/chosen": -179.0,
+      "logps/rejected": -384.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.4765625,
-      "rewards/margins": 33.0,
-      "rewards/rejected": -33.5,
+      "rewards/chosen": -1.09375,
+      "rewards/margins": 24.875,
+      "rewards/rejected": -26.0,
       "step": 830
     },
     {
       "epoch": 2.68370607028754,
-      "grad_norm": 3.1691540566080763e-06,
+      "grad_norm": 0.00017437168798643718,
       "learning_rate": 5.857988165680473e-08,
-      "logits/chosen": -3.53125,
-      "logits/rejected": -3.296875,
-      "logps/chosen": -153.0,
-      "logps/rejected": -516.0,
+      "logits/chosen": -2.90625,
+      "logits/rejected": -2.703125,
+      "logps/chosen": -168.0,
+      "logps/rejected": -400.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.212890625,
-      "rewards/margins": 35.25,
-      "rewards/rejected": -35.5,
+      "rewards/chosen": -0.95703125,
+      "rewards/margins": 26.25,
+      "rewards/rejected": -27.25,
       "step": 840
     },
     {
       "epoch": 2.7156549520766773,
-      "grad_norm": 4.6438055915502857e-07,
+      "grad_norm": 5.8399803924560536e-05,
       "learning_rate": 5.266272189349112e-08,
-      "logits/chosen": -3.546875,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -175.0,
-      "logps/rejected": -532.0,
+      "logits/chosen": -2.90625,
+      "logits/rejected": -2.59375,
+      "logps/chosen": -189.0,
+      "logps/rejected": -396.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.384765625,
-      "rewards/margins": 37.25,
-      "rewards/rejected": -37.75,
+      "rewards/chosen": -0.890625,
+      "rewards/margins": 26.5,
+      "rewards/rejected": -27.5,
       "step": 850
     },
     {
       "epoch": 2.747603833865815,
-      "grad_norm": 1.200711849285529e-05,
+      "grad_norm": 3.8305510276667966e-05,
       "learning_rate": 4.674556213017751e-08,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.296875,
-      "logps/chosen": -157.0,
-      "logps/rejected": -516.0,
+      "logits/chosen": -2.84375,
+      "logits/rejected": -2.609375,
+      "logps/chosen": -174.0,
+      "logps/rejected": -400.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.2392578125,
-      "rewards/margins": 35.0,
-      "rewards/rejected": -35.25,
+      "rewards/chosen": -0.9921875,
+      "rewards/margins": 26.0,
+      "rewards/rejected": -27.0,
       "step": 860
     },
     {
       "epoch": 2.779552715654952,
-      "grad_norm": 6.246644777085319e-07,
+      "grad_norm": 4.711499537336409e-05,
       "learning_rate": 4.082840236686391e-08,
-      "logits/chosen": -3.53125,
-      "logits/rejected": -3.28125,
-      "logps/chosen": -143.0,
-      "logps/rejected": -506.0,
+      "logits/chosen": -2.890625,
+      "logits/rejected": -2.625,
+      "logps/chosen": -157.0,
+      "logps/rejected": -396.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.09375,
-      "rewards/margins": 35.0,
-      "rewards/rejected": -34.75,
+      "rewards/chosen": -0.86328125,
+      "rewards/margins": 26.375,
+      "rewards/rejected": -27.25,
       "step": 870
     },
     {
       "epoch": 2.8115015974440896,
-      "grad_norm": 0.0001921259561116637,
+      "grad_norm": 6.254389531377533e-05,
       "learning_rate": 3.4911242603550294e-08,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -134.0,
-      "logps/rejected": -528.0,
+      "logits/chosen": -2.84375,
+      "logits/rejected": -2.625,
+      "logps/chosen": -144.0,
+      "logps/rejected": -412.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": 0.031982421875,
-      "rewards/margins": 37.5,
-      "rewards/rejected": -37.5,
+      "rewards/chosen": -0.72265625,
+      "rewards/margins": 28.125,
+      "rewards/rejected": -28.875,
       "step": 880
     },
     {
       "epoch": 2.8434504792332267,
-      "grad_norm": 7.289079689251674e-07,
+      "grad_norm": 0.0003230797654674014,
       "learning_rate": 2.8994082840236687e-08,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -165.0,
-      "logps/rejected": -516.0,
+      "logits/chosen": -2.828125,
+      "logits/rejected": -2.671875,
+      "logps/chosen": -184.0,
+      "logps/rejected": -418.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.28125,
-      "rewards/margins": 35.0,
-      "rewards/rejected": -35.25,
+      "rewards/chosen": -1.1015625,
+      "rewards/margins": 28.0,
+      "rewards/rejected": -29.125,
       "step": 890
     },
     {
       "epoch": 2.8753993610223643,
-      "grad_norm": 0.00019988318278621426,
+      "grad_norm": 0.0003811987661985824,
       "learning_rate": 2.3076923076923076e-08,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -152.0,
-      "logps/rejected": -504.0,
+      "logits/chosen": -2.828125,
+      "logits/rejected": -2.609375,
+      "logps/chosen": -170.0,
+      "logps/rejected": -402.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.333984375,
-      "rewards/margins": 35.25,
-      "rewards/rejected": -35.5,
+      "rewards/chosen": -1.2265625,
+      "rewards/margins": 27.125,
+      "rewards/rejected": -28.375,
       "step": 900
     },
     {
       "epoch": 2.9073482428115014,
-      "grad_norm": 2.6903006524444206e-05,
+      "grad_norm": 0.00010610309145449586,
       "learning_rate": 1.7159763313609465e-08,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.3125,
-      "logps/chosen": -161.0,
-      "logps/rejected": -516.0,
+      "logits/chosen": -2.78125,
+      "logits/rejected": -2.625,
+      "logps/chosen": -179.0,
+      "logps/rejected": -404.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.435546875,
-      "rewards/margins": 34.75,
-      "rewards/rejected": -35.0,
+      "rewards/chosen": -1.1328125,
+      "rewards/margins": 26.375,
+      "rewards/rejected": -27.5,
       "step": 910
     },
     {
       "epoch": 2.939297124600639,
-      "grad_norm": 2.3081588109716523e-05,
+      "grad_norm": 1.1872477415615446e-05,
       "learning_rate": 1.1242603550295858e-08,
-      "logits/chosen": -3.5,
-      "logits/rejected": -3.328125,
-      "logps/chosen": -144.0,
-      "logps/rejected": -512.0,
+      "logits/chosen": -2.796875,
+      "logits/rejected": -2.65625,
+      "logps/chosen": -158.0,
+      "logps/rejected": -414.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.08154296875,
-      "rewards/margins": 35.75,
-      "rewards/rejected": -35.75,
+      "rewards/chosen": -0.83203125,
+      "rewards/margins": 28.25,
+      "rewards/rejected": -29.125,
       "step": 920
     },
     {
       "epoch": 2.9712460063897765,
-      "grad_norm": 9.334583705742485e-05,
+      "grad_norm": 0.00020819970285151538,
       "learning_rate": 5.325443786982248e-09,
-      "logits/chosen": -3.53125,
-      "logits/rejected": -3.34375,
-      "logps/chosen": -151.0,
-      "logps/rejected": -506.0,
+      "logits/chosen": -2.84375,
+      "logits/rejected": -2.671875,
+      "logps/chosen": -167.0,
+      "logps/rejected": -404.0,
       "loss": 0.0,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.330078125,
-      "rewards/margins": 35.0,
-      "rewards/rejected": -35.5,
+      "rewards/chosen": -1.1953125,
+      "rewards/margins": 27.125,
+      "rewards/rejected": -28.375,
       "step": 930
     },
     {
       "epoch": 3.0,
-      "eval_logits/chosen": -3.53125,
-      "eval_logits/rejected": -3.296875,
-      "eval_logps/chosen": -151.0,
-      "eval_logps/rejected": -512.0,
-      "eval_loss": 2.4679258103788015e-07,
+      "eval_logits/chosen": -2.890625,
+      "eval_logits/rejected": -2.640625,
+      "eval_logps/chosen": -161.0,
+      "eval_logps/rejected": -404.0,
+      "eval_loss": 7.691533596698719e-07,
       "eval_rewards/accuracies": 1.0,
-      "eval_rewards/chosen": -0.41015625,
-      "eval_rewards/margins": 36.75,
-      "eval_rewards/rejected": -37.25,
-      "eval_runtime": 18.6307,
-      "eval_samples_per_second": 10.735,
-      "eval_steps_per_second": 0.376,
+      "eval_rewards/chosen": -1.0234375,
+      "eval_rewards/margins": 27.25,
+      "eval_rewards/rejected": -28.375,
+      "eval_runtime": 15.3546,
+      "eval_samples_per_second": 13.025,
+      "eval_steps_per_second": 0.456,
       "step": 939
     },
     {
       "epoch": 3.0,
       "step": 939,
       "total_flos": 0.0,
-      "train_loss": 0.011217064683810743,
-      "train_runtime": 5892.2099,
-      "train_samples_per_second": 5.091,
-      "train_steps_per_second": 0.159
+      "train_loss": 0.014182947793543137,
+      "train_runtime": 4808.48,
+      "train_samples_per_second": 6.239,
+      "train_steps_per_second": 0.195
     }
   ],
   "logging_steps": 10,