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: 0.6052
-- Rewards/chosen: -0.5967
-- Rewards/rejected: -1.0277
-- Rewards/accuracies: 0.7031
-- Rewards/margins: 0.4310
-- Logps/rejected: -1305.1270
-- Logps/chosen: -872.1711
-- Logits/rejected: -0.6640
-- Logits/chosen: -0.7053
+- Loss: 0.5987
+- Rewards/chosen: -0.6320
+- Rewards/rejected: -1.1187
+- Rewards/accuracies: 0.7070
+- Rewards/margins: 0.4867
+- Logps/rejected: -1395.8809
+- Logps/chosen: -907.7229
+- Logits/rejected: -0.5772
+- Logits/chosen: -0.6368
 
 ## 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 |
 |:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.5914        | 0.9984 | 477  | 0.6052          | -0.5967        | -1.0277          | 0.7031             | 0.4310          | -1305.1270     | -872.1711    | -0.6640         | -0.7053       |
+| 0.5855        | 0.9984 | 477  | 0.5987          | -0.6320        | -1.1187          | 0.7070             | 0.4867          | -1395.8809     | -907.7229    | -0.5772         | -0.6368       |
 
 
 ### Framework versions

all_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 0.9984301412872841,
     "total_flos": 0.0,
-    "train_loss": 0.6380661858452691,
-    "train_runtime": 13550.8489,
+    "train_loss": 0.6339174956395691,
+    "train_runtime": 13552.9806,
     "train_samples": 61134,
     "train_samples_per_second": 4.511,
     "train_steps_per_second": 0.035

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "/scratch/pawsey0935/tliu1/alignment-handbook/data/zephyr-8b-sft-1e/checkpoint-300",
+  "_name_or_path": "/scratch/pawsey0935/tliu1/alignment-handbook/data/zephyr-8b-sft-1e/checkpoint-200",
   "architectures": [
     "LlamaForCausalLM"
   ],

model-00001-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:486a40fc4174fc8a41e86292c06e78868bf4a558dab887135e53e736d17a3ae9
+oid sha256:78365fa3381a8af16010371ce29b8dfa478d52cb1bfab568d075dca246eacded
 size 4976698672

model-00002-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f2633d08fc6f681079dadef27f42fd6dfb388fd87d5532b2c0c337330a2f3364
+oid sha256:41668989e79c17a1de048d670ee14ca7f1004e6b5f52c4e3b1e351a002135936
 size 4999802720

model-00003-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9f7a3bb35b0d15b16f1eece5512e9ffb14312aa45ea2e76c9fc284b1c5c364f0
+oid sha256:60c4d41e6f5e9de2a44e720b15af055f99d3b9ec6e3759d215c9b8b209fbb1ef
 size 4915916176

model-00004-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d6dccf26c9807826afbddfe1d689eda5347eb2f1a456f124df52fcc0b7f81915
+oid sha256:fbb846ebf90ebc765d24b01007039ab25e0c43ea0f6a7a7865efba4ef9dac2dd
 size 1168138808

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 0.9984301412872841,
     "total_flos": 0.0,
-    "train_loss": 0.6380661858452691,
-    "train_runtime": 13550.8489,
+    "train_loss": 0.6339174956395691,
+    "train_runtime": 13552.9806,
     "train_samples": 61134,
     "train_samples_per_second": 4.511,
     "train_steps_per_second": 0.035

trainer_state.json

@@ -10,12 +10,12 @@
   "log_history": [
     {
       "epoch": 0.0020931449502878076,
-      "grad_norm": 0.31482814529963843,
+      "grad_norm": 0.3233749908780064,
       "learning_rate": 1.0416666666666666e-08,
-      "logits/chosen": -1.1032249927520752,
-      "logits/rejected": -1.097320795059204,
-      "logps/chosen": -361.6676940917969,
-      "logps/rejected": -360.52056884765625,
+      "logits/chosen": -0.9934180974960327,
+      "logits/rejected": -0.9732793569564819,
+      "logps/chosen": -362.2581481933594,
+      "logps/rejected": -361.5176696777344,
       "loss": 0.6931,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
@@ -25,731 +25,731 @@
     },
     {
       "epoch": 0.020931449502878074,
-      "grad_norm": 0.3101159594537847,
+      "grad_norm": 0.31818436563386754,
       "learning_rate": 1.0416666666666667e-07,
-      "logits/chosen": -1.0542187690734863,
-      "logits/rejected": -1.0401158332824707,
-      "logps/chosen": -301.57366943359375,
-      "logps/rejected": -278.6243896484375,
+      "logits/chosen": -0.9403506517410278,
+      "logits/rejected": -0.920226514339447,
+      "logps/chosen": -300.6136474609375,
+      "logps/rejected": -277.24029541015625,
       "loss": 0.6931,
-      "rewards/accuracies": 0.4097222089767456,
-      "rewards/chosen": 4.0215920307673514e-05,
-      "rewards/margins": 7.037212526483927e-06,
-      "rewards/rejected": 3.317870869068429e-05,
+      "rewards/accuracies": 0.4583333432674408,
+      "rewards/chosen": 3.990698678535409e-05,
+      "rewards/margins": 2.8463546186685562e-05,
+      "rewards/rejected": 1.144344059866853e-05,
       "step": 10
     },
     {
       "epoch": 0.04186289900575615,
-      "grad_norm": 0.2988255948509268,
+      "grad_norm": 0.32066911516768193,
       "learning_rate": 2.0833333333333333e-07,
-      "logits/chosen": -1.0230003595352173,
-      "logits/rejected": -1.013373613357544,
-      "logps/chosen": -313.2471618652344,
-      "logps/rejected": -280.4976806640625,
+      "logits/chosen": -0.9091697931289673,
+      "logits/rejected": -0.8898228406906128,
+      "logps/chosen": -313.4816589355469,
+      "logps/rejected": -279.7200012207031,
       "loss": 0.6931,
-      "rewards/accuracies": 0.5562499761581421,
-      "rewards/chosen": 7.316504343179986e-05,
-      "rewards/margins": 6.343351560644805e-05,
-      "rewards/rejected": 9.731529644341208e-06,
+      "rewards/accuracies": 0.53125,
+      "rewards/chosen": 9.299140947405249e-05,
+      "rewards/margins": 0.00010705796012189239,
+      "rewards/rejected": -1.4066553376324009e-05,
       "step": 20
     },
     {
       "epoch": 0.06279434850863422,
-      "grad_norm": 0.29546258917162005,
+      "grad_norm": 0.29725853737386143,
       "learning_rate": 3.1249999999999997e-07,
-      "logits/chosen": -1.0373719930648804,
-      "logits/rejected": -1.0336660146713257,
-      "logps/chosen": -289.7471618652344,
-      "logps/rejected": -249.1915283203125,
+      "logits/chosen": -0.9132905006408691,
+      "logits/rejected": -0.9013525247573853,
+      "logps/chosen": -290.26397705078125,
+      "logps/rejected": -249.2030792236328,
       "loss": 0.6931,
-      "rewards/accuracies": 0.518750011920929,
-      "rewards/chosen": 0.0005304820952005684,
-      "rewards/margins": 0.0001712786324787885,
-      "rewards/rejected": 0.00035920340451411903,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.0005759169580414891,
+      "rewards/margins": 0.00012212179717607796,
+      "rewards/rejected": 0.0004537951317615807,
       "step": 30
     },
     {
       "epoch": 0.0837257980115123,
-      "grad_norm": 0.3027383129898098,
+      "grad_norm": 0.30422594886075766,
       "learning_rate": 4.1666666666666667e-07,
-      "logits/chosen": -1.0248687267303467,
-      "logits/rejected": -1.0122233629226685,
-      "logps/chosen": -258.7686767578125,
-      "logps/rejected": -260.17950439453125,
-      "loss": 0.6929,
+      "logits/chosen": -0.9063602685928345,
+      "logits/rejected": -0.8846728205680847,
+      "logps/chosen": -258.3272705078125,
+      "logps/rejected": -258.79901123046875,
+      "loss": 0.6928,
       "rewards/accuracies": 0.65625,
-      "rewards/chosen": 0.0014846790581941605,
-      "rewards/margins": 0.0004175625799689442,
-      "rewards/rejected": 0.0010671165073290467,
+      "rewards/chosen": 0.0017278010491281748,
+      "rewards/margins": 0.0005969521589577198,
+      "rewards/rejected": 0.0011308487737551332,
       "step": 40
     },
     {
       "epoch": 0.10465724751439037,
-      "grad_norm": 0.30526828593535077,
+      "grad_norm": 0.3141653560965076,
       "learning_rate": 4.999731868769026e-07,
-      "logits/chosen": -1.0067716836929321,
-      "logits/rejected": -0.9986906051635742,
-      "logps/chosen": -286.8924255371094,
-      "logps/rejected": -271.6407775878906,
-      "loss": 0.6925,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.0033252774737775326,
-      "rewards/margins": 0.0013752836966887116,
-      "rewards/rejected": 0.001949993660673499,
+      "logits/chosen": -0.888303279876709,
+      "logits/rejected": -0.8725374937057495,
+      "logps/chosen": -286.4285583496094,
+      "logps/rejected": -271.14837646484375,
+      "loss": 0.6924,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.0034768437035381794,
+      "rewards/margins": 0.0013709357008337975,
+      "rewards/rejected": 0.002105908002704382,
       "step": 50
     },
     {
       "epoch": 0.12558869701726844,
-      "grad_norm": 0.32324516495717,
+      "grad_norm": 0.32864602203785215,
       "learning_rate": 4.990353313429303e-07,
-      "logits/chosen": -1.0197694301605225,
-      "logits/rejected": -1.0113787651062012,
-      "logps/chosen": -252.9259033203125,
-      "logps/rejected": -243.18826293945312,
-      "loss": 0.692,
+      "logits/chosen": -0.8980317115783691,
+      "logits/rejected": -0.883760929107666,
+      "logps/chosen": -252.4823760986328,
+      "logps/rejected": -242.98117065429688,
+      "loss": 0.6918,
       "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.005434063728898764,
-      "rewards/margins": 0.0027187608648091555,
-      "rewards/rejected": 0.002715302398428321,
+      "rewards/chosen": 0.0057683298364281654,
+      "rewards/margins": 0.003004656871780753,
+      "rewards/rejected": 0.002763673197478056,
       "step": 60
     },
     {
       "epoch": 0.14652014652014653,
-      "grad_norm": 0.3019183182107993,
+      "grad_norm": 0.30851772421969176,
       "learning_rate": 4.967625656594781e-07,
-      "logits/chosen": -1.01347017288208,
-      "logits/rejected": -1.0048184394836426,
-      "logps/chosen": -293.8519592285156,
-      "logps/rejected": -282.8028259277344,
-      "loss": 0.6912,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.0049653202295303345,
-      "rewards/margins": 0.00389581685885787,
-      "rewards/rejected": 0.0010695032542571425,
+      "logits/chosen": -0.8898143768310547,
+      "logits/rejected": -0.877031683921814,
+      "logps/chosen": -293.1795959472656,
+      "logps/rejected": -282.2380676269531,
+      "loss": 0.6909,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": 0.005099158734083176,
+      "rewards/margins": 0.004411149770021439,
+      "rewards/rejected": 0.0006880092551000416,
       "step": 70
     },
     {
       "epoch": 0.1674515960230246,
-      "grad_norm": 0.3508991474726197,
+      "grad_norm": 0.3603242703891539,
       "learning_rate": 4.93167072587771e-07,
-      "logits/chosen": -1.0369741916656494,
-      "logits/rejected": -1.0129553079605103,
-      "logps/chosen": -326.29803466796875,
-      "logps/rejected": -257.17633056640625,
-      "loss": 0.6904,
-      "rewards/accuracies": 0.78125,
-      "rewards/chosen": 0.004279050976037979,
-      "rewards/margins": 0.0068035004660487175,
-      "rewards/rejected": -0.0025244501885026693,
+      "logits/chosen": -0.9251786470413208,
+      "logits/rejected": -0.8793746829032898,
+      "logps/chosen": -325.86688232421875,
+      "logps/rejected": -257.28363037109375,
+      "loss": 0.69,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.0038853283040225506,
+      "rewards/margins": 0.00776954460889101,
+      "rewards/rejected": -0.003884216072037816,
       "step": 80
     },
     {
       "epoch": 0.18838304552590268,
-      "grad_norm": 0.381026390763756,
+      "grad_norm": 0.4186108478199868,
       "learning_rate": 4.882681251368548e-07,
-      "logits/chosen": -1.0689797401428223,
-      "logits/rejected": -1.0408366918563843,
-      "logps/chosen": -256.1031188964844,
-      "logps/rejected": -268.39068603515625,
-      "loss": 0.6887,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.00043721304973587394,
-      "rewards/margins": 0.00908609852194786,
-      "rewards/rejected": -0.009523311629891396,
+      "logits/chosen": -0.9337117075920105,
+      "logits/rejected": -0.9007359743118286,
+      "logps/chosen": -257.81927490234375,
+      "logps/rejected": -270.7668151855469,
+      "loss": 0.6881,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.002065305132418871,
+      "rewards/margins": 0.010189032182097435,
+      "rewards/rejected": -0.012254336848855019,
       "step": 90
     },
     {
       "epoch": 0.20931449502878074,
-      "grad_norm": 0.4209389221734228,
+      "grad_norm": 0.4458182507297675,
       "learning_rate": 4.820919832540181e-07,
-      "logits/chosen": -1.0852195024490356,
-      "logits/rejected": -1.0748122930526733,
-      "logps/chosen": -310.9391174316406,
-      "logps/rejected": -309.0555419921875,
-      "loss": 0.6867,
-      "rewards/accuracies": 0.7562500238418579,
-      "rewards/chosen": -0.009355477057397366,
-      "rewards/margins": 0.017127614468336105,
-      "rewards/rejected": -0.026483092457056046,
+      "logits/chosen": -0.9627386927604675,
+      "logits/rejected": -0.9394882917404175,
+      "logps/chosen": -312.60650634765625,
+      "logps/rejected": -312.58447265625,
+      "loss": 0.6857,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.01190097164362669,
+      "rewards/margins": 0.019792668521404266,
+      "rewards/rejected": -0.03169363737106323,
       "step": 100
     },
     {
       "epoch": 0.2302459445316588,
-      "grad_norm": 0.5725040580824025,
+      "grad_norm": 0.7401018046473335,
       "learning_rate": 4.7467175306295647e-07,
-      "logits/chosen": -1.0323398113250732,
-      "logits/rejected": -1.0182864665985107,
-      "logps/chosen": -308.17236328125,
-      "logps/rejected": -310.19012451171875,
-      "loss": 0.6845,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.01750779151916504,
-      "rewards/margins": 0.020053904503583908,
-      "rewards/rejected": -0.03756169229745865,
+      "logits/chosen": -0.9153742790222168,
+      "logits/rejected": -0.8901848793029785,
+      "logps/chosen": -313.6246032714844,
+      "logps/rejected": -317.1287536621094,
+      "loss": 0.6836,
+      "rewards/accuracies": 0.668749988079071,
+      "rewards/chosen": -0.023393774405121803,
+      "rewards/margins": 0.021103577688336372,
+      "rewards/rejected": -0.04449734836816788,
       "step": 110
     },
     {
       "epoch": 0.25117739403453687,
-      "grad_norm": 0.7834521530338131,
+      "grad_norm": 0.8221557536673391,
       "learning_rate": 4.6604720940421207e-07,
-      "logits/chosen": -0.965900719165802,
-      "logits/rejected": -0.980473518371582,
-      "logps/chosen": -308.4568786621094,
-      "logps/rejected": -334.7530822753906,
-      "loss": 0.6787,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -0.03917182609438896,
-      "rewards/margins": 0.03068702295422554,
-      "rewards/rejected": -0.0698588490486145,
+      "logits/chosen": -0.8409837484359741,
+      "logits/rejected": -0.8502853512763977,
+      "logps/chosen": -308.5286560058594,
+      "logps/rejected": -337.5436706542969,
+      "loss": 0.6781,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -0.039570122957229614,
+      "rewards/margins": 0.03355463594198227,
+      "rewards/rejected": -0.07312475144863129,
       "step": 120
     },
     {
       "epoch": 0.272108843537415,
-      "grad_norm": 1.576846666512127,
+      "grad_norm": 1.1482922478086675,
       "learning_rate": 4.5626458262912735e-07,
-      "logits/chosen": -1.003523826599121,
-      "logits/rejected": -0.9894965887069702,
-      "logps/chosen": -349.37078857421875,
-      "logps/rejected": -380.44793701171875,
-      "loss": 0.6713,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -0.06744460016489029,
-      "rewards/margins": 0.039642252027988434,
-      "rewards/rejected": -0.10708685219287872,
+      "logits/chosen": -0.8868376612663269,
+      "logits/rejected": -0.8650667071342468,
+      "logps/chosen": -354.9898986816406,
+      "logps/rejected": -386.808349609375,
+      "loss": 0.6707,
+      "rewards/accuracies": 0.6812499761581421,
+      "rewards/chosen": -0.07423487305641174,
+      "rewards/margins": 0.040343888103961945,
+      "rewards/rejected": -0.11457876116037369,
       "step": 130
     },
     {
       "epoch": 0.29304029304029305,
-      "grad_norm": 1.9102475195415156,
+      "grad_norm": 1.8202870864883658,
       "learning_rate": 4.453763107901675e-07,
-      "logits/chosen": -1.046918511390686,
-      "logits/rejected": -1.0205754041671753,
-      "logps/chosen": -382.21759033203125,
-      "logps/rejected": -396.6845397949219,
-      "loss": 0.6661,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": -0.06044614315032959,
-      "rewards/margins": 0.06777481734752655,
-      "rewards/rejected": -0.12822096049785614,
+      "logits/chosen": -0.9577043652534485,
+      "logits/rejected": -0.9177180528640747,
+      "logps/chosen": -401.80108642578125,
+      "logps/rejected": -426.27777099609375,
+      "loss": 0.6643,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.08050010353326797,
+      "rewards/margins": 0.07808584719896317,
+      "rewards/rejected": -0.15858595073223114,
       "step": 140
     },
     {
       "epoch": 0.3139717425431711,
-      "grad_norm": 1.539797438414403,
+      "grad_norm": 1.8127314418328795,
       "learning_rate": 4.3344075855595097e-07,
-      "logits/chosen": -1.0101072788238525,
-      "logits/rejected": -0.9913681745529175,
-      "logps/chosen": -400.263427734375,
-      "logps/rejected": -428.4429626464844,
-      "loss": 0.6643,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -0.11878516525030136,
-      "rewards/margins": 0.06145761162042618,
-      "rewards/rejected": -0.18024277687072754,
+      "logits/chosen": -0.9365851283073425,
+      "logits/rejected": -0.9067959785461426,
+      "logps/chosen": -408.94659423828125,
+      "logps/rejected": -440.8140563964844,
+      "loss": 0.6619,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": -0.12829507887363434,
+      "rewards/margins": 0.06538741290569305,
+      "rewards/rejected": -0.1936824768781662,
       "step": 150
     },
     {
       "epoch": 0.3349031920460492,
-      "grad_norm": 2.179277282235682,
+      "grad_norm": 1.8472665134009132,
       "learning_rate": 4.2052190435769554e-07,
-      "logits/chosen": -1.017316460609436,
-      "logits/rejected": -1.0034241676330566,
-      "logps/chosen": -415.86553955078125,
-      "logps/rejected": -473.75213623046875,
-      "loss": 0.6583,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.13567176461219788,
-      "rewards/margins": 0.08466406166553497,
-      "rewards/rejected": -0.22033581137657166,
+      "logits/chosen": -0.9482985734939575,
+      "logits/rejected": -0.9125078320503235,
+      "logps/chosen": -465.48223876953125,
+      "logps/rejected": -529.677001953125,
+      "loss": 0.6572,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -0.18543408811092377,
+      "rewards/margins": 0.09176287055015564,
+      "rewards/rejected": -0.2771969437599182,
       "step": 160
     },
     {
       "epoch": 0.35583464154892724,
-      "grad_norm": 2.049593953433133,
+      "grad_norm": 2.8194220987690435,
       "learning_rate": 4.0668899744407567e-07,
-      "logits/chosen": -0.9147170186042786,
-      "logits/rejected": -0.9048551321029663,
-      "logps/chosen": -407.8157043457031,
-      "logps/rejected": -462.26416015625,
-      "loss": 0.657,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -0.14575687050819397,
-      "rewards/margins": 0.08090446144342422,
-      "rewards/rejected": -0.2266613245010376,
+      "logits/chosen": -0.8278239965438843,
+      "logits/rejected": -0.8063944578170776,
+      "logps/chosen": -486.142578125,
+      "logps/rejected": -554.540283203125,
+      "loss": 0.6541,
+      "rewards/accuracies": 0.6187499761581421,
+      "rewards/chosen": -0.22446854412555695,
+      "rewards/margins": 0.09519634395837784,
+      "rewards/rejected": -0.3196648955345154,
       "step": 170
     },
     {
       "epoch": 0.37676609105180536,
-      "grad_norm": 2.39579965219835,
+      "grad_norm": 2.97208521212178,
       "learning_rate": 3.920161866827889e-07,
-      "logits/chosen": -0.9678335189819336,
-      "logits/rejected": -0.9409612417221069,
-      "logps/chosen": -448.0008850097656,
-      "logps/rejected": -494.6175842285156,
-      "loss": 0.6464,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -0.17547427117824554,
-      "rewards/margins": 0.0715043693780899,
-      "rewards/rejected": -0.24697861075401306,
+      "logits/chosen": -0.8902044296264648,
+      "logits/rejected": -0.8527171015739441,
+      "logps/chosen": -530.2462158203125,
+      "logps/rejected": -592.8751831054688,
+      "loss": 0.6406,
+      "rewards/accuracies": 0.6312500238418579,
+      "rewards/chosen": -0.2585751712322235,
+      "rewards/margins": 0.087519571185112,
+      "rewards/rejected": -0.3460947573184967,
       "step": 180
     },
     {
       "epoch": 0.3976975405546834,
-      "grad_norm": 2.884921105496048,
+      "grad_norm": 2.702362177917078,
       "learning_rate": 3.765821230985757e-07,
-      "logits/chosen": -0.9461600184440613,
-      "logits/rejected": -0.9235378503799438,
-      "logps/chosen": -472.609130859375,
-      "logps/rejected": -562.6459350585938,
-      "loss": 0.6478,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -0.20605719089508057,
-      "rewards/margins": 0.09339461475610733,
-      "rewards/rejected": -0.2994517982006073,
+      "logits/chosen": -0.8619738817214966,
+      "logits/rejected": -0.8327099084854126,
+      "logps/chosen": -513.2433471679688,
+      "logps/rejected": -626.6792602539062,
+      "loss": 0.6462,
+      "rewards/accuracies": 0.643750011920929,
+      "rewards/chosen": -0.2469058483839035,
+      "rewards/margins": 0.1172383576631546,
+      "rewards/rejected": -0.3641442060470581,
       "step": 190
     },
     {
       "epoch": 0.4186289900575615,
-      "grad_norm": 3.124537723496524,
+      "grad_norm": 3.8087843322902124,
       "learning_rate": 3.604695382782159e-07,
-      "logits/chosen": -0.8811132311820984,
-      "logits/rejected": -0.8968720436096191,
-      "logps/chosen": -475.21173095703125,
-      "logps/rejected": -608.8429565429688,
-      "loss": 0.6424,
+      "logits/chosen": -0.7917501926422119,
+      "logits/rejected": -0.8016451001167297,
+      "logps/chosen": -555.9425659179688,
+      "logps/rejected": -720.2203979492188,
+      "loss": 0.6402,
       "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -0.2239048182964325,
-      "rewards/margins": 0.11250551044940948,
-      "rewards/rejected": -0.3364103436470032,
+      "rewards/chosen": -0.3054454028606415,
+      "rewards/margins": 0.14363829791545868,
+      "rewards/rejected": -0.44908371567726135,
       "step": 200
     },
     {
       "epoch": 0.43956043956043955,
-      "grad_norm": 3.352705934955811,
+      "grad_norm": 3.9347188229731374,
       "learning_rate": 3.4376480090239047e-07,
-      "logits/chosen": -0.8993983268737793,
-      "logits/rejected": -0.8476327657699585,
-      "logps/chosen": -613.8883666992188,
-      "logps/rejected": -693.7938232421875,
-      "loss": 0.6459,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.30135461688041687,
-      "rewards/margins": 0.13305452466011047,
-      "rewards/rejected": -0.43440908193588257,
+      "logits/chosen": -0.8363162279129028,
+      "logits/rejected": -0.7596572637557983,
+      "logps/chosen": -671.146728515625,
+      "logps/rejected": -783.2349243164062,
+      "loss": 0.6461,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.36001214385032654,
+      "rewards/margins": 0.16495202481746674,
+      "rewards/rejected": -0.5249641537666321,
       "step": 210
     },
     {
       "epoch": 0.4604918890633176,
-      "grad_norm": 3.483784715461117,
+      "grad_norm": 3.4218343739474046,
       "learning_rate": 3.265574537815398e-07,
-      "logits/chosen": -0.827401340007782,
-      "logits/rejected": -0.8360942602157593,
-      "logps/chosen": -618.6294555664062,
-      "logps/rejected": -796.0371704101562,
-      "loss": 0.6401,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.38492849469184875,
-      "rewards/margins": 0.16860195994377136,
-      "rewards/rejected": -0.5535304546356201,
+      "logits/chosen": -0.7804139256477356,
+      "logits/rejected": -0.7788348197937012,
+      "logps/chosen": -537.9508666992188,
+      "logps/rejected": -720.7735595703125,
+      "loss": 0.6365,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": -0.3049758970737457,
+      "rewards/margins": 0.17417363822460175,
+      "rewards/rejected": -0.4791494309902191,
       "step": 220
     },
     {
       "epoch": 0.48142333856619574,
-      "grad_norm": 3.68680579193991,
+      "grad_norm": 3.795962622718272,
       "learning_rate": 3.0893973387735683e-07,
-      "logits/chosen": -0.8449612855911255,
-      "logits/rejected": -0.815530002117157,
-      "logps/chosen": -661.7880249023438,
-      "logps/rejected": -855.1505737304688,
-      "loss": 0.6323,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.4026467204093933,
-      "rewards/margins": 0.21427354216575623,
-      "rewards/rejected": -0.6169203519821167,
+      "logits/chosen": -0.8238471150398254,
+      "logits/rejected": -0.7790335416793823,
+      "logps/chosen": -618.4681396484375,
+      "logps/rejected": -830.87109375,
+      "loss": 0.6268,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.3598284125328064,
+      "rewards/margins": 0.23437073826789856,
+      "rewards/rejected": -0.5941991209983826,
       "step": 230
     },
     {
       "epoch": 0.5023547880690737,
-      "grad_norm": 3.7270952978464242,
+      "grad_norm": 3.5058346085646144,
       "learning_rate": 2.910060778827554e-07,
-      "logits/chosen": -0.8660251498222351,
-      "logits/rejected": -0.8118718266487122,
-      "logps/chosen": -648.49755859375,
-      "logps/rejected": -792.0703125,
-      "loss": 0.6226,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -0.3631536066532135,
-      "rewards/margins": 0.16178566217422485,
-      "rewards/rejected": -0.524939239025116,
+      "logits/chosen": -0.8182986378669739,
+      "logits/rejected": -0.7530924081802368,
+      "logps/chosen": -669.210205078125,
+      "logps/rejected": -834.8556518554688,
+      "loss": 0.6181,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -0.3839070200920105,
+      "rewards/margins": 0.1838807910680771,
+      "rewards/rejected": -0.567787766456604,
       "step": 240
     },
     {
       "epoch": 0.5232862375719518,
-      "grad_norm": 3.5025939290918844,
+      "grad_norm": 3.784803079890303,
       "learning_rate": 2.7285261601056697e-07,
-      "logits/chosen": -0.8380582928657532,
-      "logits/rejected": -0.7766156792640686,
-      "logps/chosen": -687.4022216796875,
-      "logps/rejected": -870.7564697265625,
-      "loss": 0.6329,
-      "rewards/accuracies": 0.731249988079071,
-      "rewards/chosen": -0.405522882938385,
-      "rewards/margins": 0.223312109708786,
-      "rewards/rejected": -0.6288350224494934,
+      "logits/chosen": -0.803874671459198,
+      "logits/rejected": -0.7223009467124939,
+      "logps/chosen": -647.1034545898438,
+      "logps/rejected": -853.0436401367188,
+      "loss": 0.6268,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": -0.3654927909374237,
+      "rewards/margins": 0.24678358435630798,
+      "rewards/rejected": -0.6122764348983765,
       "step": 250
     },
     {
       "epoch": 0.54421768707483,
-      "grad_norm": 4.493991768115386,
+      "grad_norm": 5.477740532160938,
       "learning_rate": 2.5457665670441937e-07,
-      "logits/chosen": -0.8578090667724609,
-      "logits/rejected": -0.8512881994247437,
-      "logps/chosen": -660.48193359375,
-      "logps/rejected": -883.0379028320312,
-      "loss": 0.6206,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -0.3996823728084564,
-      "rewards/margins": 0.23860612511634827,
-      "rewards/rejected": -0.6382884979248047,
+      "logits/chosen": -0.8027833700180054,
+      "logits/rejected": -0.7885298728942871,
+      "logps/chosen": -674.1099853515625,
+      "logps/rejected": -937.3220825195312,
+      "loss": 0.6138,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.4139021039009094,
+      "rewards/margins": 0.27916184067726135,
+      "rewards/rejected": -0.6930639147758484,
       "step": 260
     },
     {
       "epoch": 0.565149136577708,
-      "grad_norm": 3.4403303240735252,
+      "grad_norm": 3.3580640957163905,
       "learning_rate": 2.3627616503391812e-07,
-      "logits/chosen": -0.7913758754730225,
-      "logits/rejected": -0.7701588869094849,
-      "logps/chosen": -734.1468505859375,
-      "logps/rejected": -964.8834228515625,
-      "loss": 0.6235,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -0.4329896569252014,
-      "rewards/margins": 0.261671245098114,
-      "rewards/rejected": -0.6946609020233154,
+      "logits/chosen": -0.7519720196723938,
+      "logits/rejected": -0.7142844796180725,
+      "logps/chosen": -751.9912719726562,
+      "logps/rejected": -1058.1048583984375,
+      "loss": 0.615,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -0.45110592246055603,
+      "rewards/margins": 0.3373827338218689,
+      "rewards/rejected": -0.7884886264801025,
       "step": 270
     },
     {
       "epoch": 0.5860805860805861,
-      "grad_norm": 4.102788465104312,
+      "grad_norm": 5.5177213152440565,
       "learning_rate": 2.1804923757009882e-07,
-      "logits/chosen": -0.7503592371940613,
-      "logits/rejected": -0.7338029742240906,
-      "logps/chosen": -688.9549560546875,
-      "logps/rejected": -883.1748046875,
-      "loss": 0.6273,
-      "rewards/accuracies": 0.643750011920929,
-      "rewards/chosen": -0.43057650327682495,
-      "rewards/margins": 0.2195829600095749,
-      "rewards/rejected": -0.650159478187561,
+      "logits/chosen": -0.6814185976982117,
+      "logits/rejected": -0.6468263864517212,
+      "logps/chosen": -705.2122192382812,
+      "logps/rejected": -920.8411865234375,
+      "loss": 0.6228,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.44767504930496216,
+      "rewards/margins": 0.24160179495811462,
+      "rewards/rejected": -0.6892768144607544,
       "step": 280
     },
     {
       "epoch": 0.6070120355834642,
-      "grad_norm": 3.636347519443132,
+      "grad_norm": 4.551630518528997,
       "learning_rate": 1.9999357655598891e-07,
-      "logits/chosen": -0.7760393023490906,
-      "logits/rejected": -0.7229740023612976,
-      "logps/chosen": -774.09423828125,
-      "logps/rejected": -982.6472778320312,
-      "loss": 0.626,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -0.5254346132278442,
-      "rewards/margins": 0.20538799464702606,
-      "rewards/rejected": -0.7308225631713867,
+      "logits/chosen": -0.7232012152671814,
+      "logits/rejected": -0.6624692678451538,
+      "logps/chosen": -756.8780517578125,
+      "logps/rejected": -1010.7252197265625,
+      "loss": 0.6241,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -0.5086268782615662,
+      "rewards/margins": 0.2510340213775635,
+      "rewards/rejected": -0.7596608996391296,
       "step": 290
     },
     {
       "epoch": 0.6279434850863422,
-      "grad_norm": 4.1262288937509926,
+      "grad_norm": 5.147689539765831,
       "learning_rate": 1.8220596619089573e-07,
-      "logits/chosen": -0.7782667279243469,
-      "logits/rejected": -0.737755298614502,
-      "logps/chosen": -827.1736450195312,
-      "logps/rejected": -975.5570068359375,
-      "loss": 0.6078,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -0.4986857771873474,
-      "rewards/margins": 0.19637663662433624,
-      "rewards/rejected": -0.6950623393058777,
+      "logits/chosen": -0.7237067818641663,
+      "logits/rejected": -0.6642863154411316,
+      "logps/chosen": -871.6720581054688,
+      "logps/rejected": -1069.462158203125,
+      "loss": 0.6007,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.542992115020752,
+      "rewards/margins": 0.24627108871936798,
+      "rewards/rejected": -0.7892631888389587,
       "step": 300
     },
     {
       "epoch": 0.6488749345892203,
-      "grad_norm": 4.993223384074502,
+      "grad_norm": 5.107775722831973,
       "learning_rate": 1.647817538357072e-07,
-      "logits/chosen": -0.7246193289756775,
-      "logits/rejected": -0.6662891507148743,
-      "logps/chosen": -836.2703247070312,
-      "logps/rejected": -1138.7655029296875,
-      "loss": 0.6066,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": -0.5288342237472534,
-      "rewards/margins": 0.35662880539894104,
-      "rewards/rejected": -0.8854629397392273,
+      "logits/chosen": -0.7091413736343384,
+      "logits/rejected": -0.6302168965339661,
+      "logps/chosen": -882.1735229492188,
+      "logps/rejected": -1218.36962890625,
+      "loss": 0.605,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -0.5749861598014832,
+      "rewards/margins": 0.3909863531589508,
+      "rewards/rejected": -0.9659725427627563,
       "step": 310
     },
     {
       "epoch": 0.6698063840920984,
-      "grad_norm": 3.91056577188791,
+      "grad_norm": 4.98218570447988,
       "learning_rate": 1.478143389201113e-07,
-      "logits/chosen": -0.7831446528434753,
-      "logits/rejected": -0.7029932141304016,
-      "logps/chosen": -742.99169921875,
-      "logps/rejected": -1062.3135986328125,
-      "loss": 0.6026,
+      "logits/chosen": -0.7340850830078125,
+      "logits/rejected": -0.6477845907211304,
+      "logps/chosen": -795.6258544921875,
+      "logps/rejected": -1128.2310791015625,
+      "loss": 0.5986,
       "rewards/accuracies": 0.7437499761581421,
-      "rewards/chosen": -0.4849509298801422,
-      "rewards/margins": 0.3511354327201843,
-      "rewards/rejected": -0.8360863924026489,
+      "rewards/chosen": -0.5385175943374634,
+      "rewards/margins": 0.3641910254955292,
+      "rewards/rejected": -0.9027086496353149,
       "step": 320
     },
     {
       "epoch": 0.6907378335949764,
-      "grad_norm": 6.654258947906022,
+      "grad_norm": 7.2646740452584195,
       "learning_rate": 1.3139467229135998e-07,
-      "logits/chosen": -0.7732152938842773,
-      "logits/rejected": -0.7264204621315002,
-      "logps/chosen": -827.3762817382812,
-      "logps/rejected": -1139.37890625,
-      "loss": 0.6167,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.549734890460968,
-      "rewards/margins": 0.3192141652107239,
-      "rewards/rejected": -0.8689490556716919,
+      "logits/chosen": -0.7201080322265625,
+      "logits/rejected": -0.6724112629890442,
+      "logps/chosen": -879.5479736328125,
+      "logps/rejected": -1196.298583984375,
+      "loss": 0.6091,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -0.6024604439735413,
+      "rewards/margins": 0.32484909892082214,
+      "rewards/rejected": -0.927309513092041,
       "step": 330
     },
     {
       "epoch": 0.7116692830978545,
-      "grad_norm": 5.714820319654088,
+      "grad_norm": 7.541765265306689,
       "learning_rate": 1.1561076868822755e-07,
-      "logits/chosen": -0.657776951789856,
-      "logits/rejected": -0.6362160444259644,
-      "logps/chosen": -892.25146484375,
-      "logps/rejected": -1171.9384765625,
-      "loss": 0.599,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": -0.5948387384414673,
-      "rewards/margins": 0.307717889547348,
-      "rewards/rejected": -0.9025565981864929,
+      "logits/chosen": -0.613195538520813,
+      "logits/rejected": -0.5763617753982544,
+      "logps/chosen": -933.0261840820312,
+      "logps/rejected": -1260.025634765625,
+      "loss": 0.5923,
+      "rewards/accuracies": 0.6937500238418579,
+      "rewards/chosen": -0.6348632574081421,
+      "rewards/margins": 0.35707950592041016,
+      "rewards/rejected": -0.9919427633285522,
       "step": 340
     },
     {
       "epoch": 0.7326007326007326,
-      "grad_norm": 4.543766057061079,
+      "grad_norm": 5.982906996429055,
       "learning_rate": 1.0054723495346482e-07,
-      "logits/chosen": -0.7216989994049072,
-      "logits/rejected": -0.6934664249420166,
-      "logps/chosen": -879.1535034179688,
-      "logps/rejected": -1134.7598876953125,
-      "loss": 0.5904,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.6154306530952454,
-      "rewards/margins": 0.2860354781150818,
-      "rewards/rejected": -0.9014661908149719,
+      "logits/chosen": -0.6862212419509888,
+      "logits/rejected": -0.6319141387939453,
+      "logps/chosen": -905.0227661132812,
+      "logps/rejected": -1189.1097412109375,
+      "loss": 0.5771,
+      "rewards/accuracies": 0.6499999761581421,
+      "rewards/chosen": -0.6419159173965454,
+      "rewards/margins": 0.3145480751991272,
+      "rewards/rejected": -0.9564639925956726,
       "step": 350
     },
     {
       "epoch": 0.7535321821036107,
-      "grad_norm": 5.229962296463252,
+      "grad_norm": 5.935343010634375,
       "learning_rate": 8.628481651367875e-08,
-      "logits/chosen": -0.7215043902397156,
-      "logits/rejected": -0.6787934303283691,
-      "logps/chosen": -978.87109375,
-      "logps/rejected": -1228.20849609375,
-      "loss": 0.6153,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.6645429730415344,
-      "rewards/margins": 0.2913895845413208,
-      "rewards/rejected": -0.9559324979782104,
+      "logits/chosen": -0.6817030310630798,
+      "logits/rejected": -0.6138208508491516,
+      "logps/chosen": -1023.2503662109375,
+      "logps/rejected": -1340.0045166015625,
+      "loss": 0.6097,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.7090341448783875,
+      "rewards/margins": 0.35953468084335327,
+      "rewards/rejected": -1.0685687065124512,
       "step": 360
     },
     {
       "epoch": 0.7744636316064888,
-      "grad_norm": 4.629531064191369,
+      "grad_norm": 4.786610680497262,
       "learning_rate": 7.289996455765748e-08,
-      "logits/chosen": -0.7984783053398132,
-      "logits/rejected": -0.7226461172103882,
-      "logps/chosen": -798.4375,
-      "logps/rejected": -1085.9097900390625,
-      "loss": 0.6017,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": -0.5339782238006592,
-      "rewards/margins": 0.3194805979728699,
-      "rewards/rejected": -0.8534588813781738,
+      "logits/chosen": -0.7248352766036987,
+      "logits/rejected": -0.6292211413383484,
+      "logps/chosen": -842.8912353515625,
+      "logps/rejected": -1210.052490234375,
+      "loss": 0.597,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.5791453123092651,
+      "rewards/margins": 0.39849239587783813,
+      "rewards/rejected": -0.977637767791748,
       "step": 370
     },
     {
       "epoch": 0.7953950811093669,
-      "grad_norm": 5.1848792854745955,
+      "grad_norm": 5.360927626743976,
       "learning_rate": 6.046442623320145e-08,
-      "logits/chosen": -0.667945146560669,
-      "logits/rejected": -0.6463702917098999,
-      "logps/chosen": -900.64111328125,
-      "logps/rejected": -1314.02685546875,
-      "loss": 0.6051,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.6532193422317505,
-      "rewards/margins": 0.39731550216674805,
-      "rewards/rejected": -1.0505348443984985,
+      "logits/chosen": -0.5932396650314331,
+      "logits/rejected": -0.5533714294433594,
+      "logps/chosen": -919.9244995117188,
+      "logps/rejected": -1385.126953125,
+      "loss": 0.5988,
+      "rewards/accuracies": 0.71875,
+      "rewards/chosen": -0.6735996007919312,
+      "rewards/margins": 0.44867175817489624,
+      "rewards/rejected": -1.1222712993621826,
       "step": 380
     },
     {
       "epoch": 0.8163265306122449,
-      "grad_norm": 5.269478675784282,
+      "grad_norm": 7.044779963713088,
       "learning_rate": 4.904486005914027e-08,
-      "logits/chosen": -0.7620413899421692,
-      "logits/rejected": -0.6749614477157593,
-      "logps/chosen": -971.4052734375,
-      "logps/rejected": -1313.494873046875,
-      "loss": 0.5981,
+      "logits/chosen": -0.713057279586792,
+      "logits/rejected": -0.6076905131340027,
+      "logps/chosen": -975.0090942382812,
+      "logps/rejected": -1386.0218505859375,
+      "loss": 0.5884,
       "rewards/accuracies": 0.6875,
-      "rewards/chosen": -0.6340218782424927,
-      "rewards/margins": 0.3738587498664856,
-      "rewards/rejected": -1.0078805685043335,
+      "rewards/chosen": -0.6378625631332397,
+      "rewards/margins": 0.4438837468624115,
+      "rewards/rejected": -1.0817463397979736,
       "step": 390
     },
     {
       "epoch": 0.837257980115123,
-      "grad_norm": 6.787786720961698,
+      "grad_norm": 6.339569576065322,
       "learning_rate": 3.8702478614051345e-08,
-      "logits/chosen": -0.7115162014961243,
-      "logits/rejected": -0.6728549003601074,
-      "logps/chosen": -830.5745849609375,
-      "logps/rejected": -1112.093505859375,
-      "loss": 0.6063,
-      "rewards/accuracies": 0.706250011920929,
-      "rewards/chosen": -0.5667731165885925,
-      "rewards/margins": 0.30460324883461,
-      "rewards/rejected": -0.8713763952255249,
+      "logits/chosen": -0.6517136693000793,
+      "logits/rejected": -0.589192807674408,
+      "logps/chosen": -841.0965576171875,
+      "logps/rejected": -1160.896484375,
+      "loss": 0.5985,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.577842652797699,
+      "rewards/margins": 0.3430059850215912,
+      "rewards/rejected": -0.9208486676216125,
       "step": 400
     },
     {
       "epoch": 0.858189429618001,
-      "grad_norm": 4.977795281165285,
+      "grad_norm": 5.268821551699033,
       "learning_rate": 2.9492720416985e-08,
-      "logits/chosen": -0.7629977464675903,
-      "logits/rejected": -0.7021812200546265,
-      "logps/chosen": -898.7531127929688,
-      "logps/rejected": -1238.019287109375,
-      "loss": 0.6036,
-      "rewards/accuracies": 0.6937500238418579,
-      "rewards/chosen": -0.5887300372123718,
-      "rewards/margins": 0.3835086226463318,
-      "rewards/rejected": -0.9722386598587036,
+      "logits/chosen": -0.7182366847991943,
+      "logits/rejected": -0.6271320581436157,
+      "logps/chosen": -902.9696044921875,
+      "logps/rejected": -1317.0338134765625,
+      "loss": 0.5968,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.5933186411857605,
+      "rewards/margins": 0.4583154320716858,
+      "rewards/rejected": -1.0516341924667358,
       "step": 410
     },
     {
       "epoch": 0.8791208791208791,
-      "grad_norm": 4.925386007685482,
+      "grad_norm": 6.427134267894669,
       "learning_rate": 2.1464952759020856e-08,
-      "logits/chosen": -0.612363874912262,
-      "logits/rejected": -0.6159146428108215,
-      "logps/chosen": -902.3529052734375,
-      "logps/rejected": -1285.909423828125,
-      "loss": 0.5883,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": -0.6378868222236633,
-      "rewards/margins": 0.3771875202655792,
-      "rewards/rejected": -1.015074372291565,
+      "logits/chosen": -0.5536683797836304,
+      "logits/rejected": -0.5348982810974121,
+      "logps/chosen": -916.4422607421875,
+      "logps/rejected": -1366.26171875,
+      "loss": 0.5804,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.6526989936828613,
+      "rewards/margins": 0.44348520040512085,
+      "rewards/rejected": -1.0961841344833374,
       "step": 420
     },
     {
       "epoch": 0.9000523286237572,
-      "grad_norm": 4.636474666716878,
+      "grad_norm": 5.223662175792827,
       "learning_rate": 1.4662207078575684e-08,
-      "logits/chosen": -0.6787495613098145,
-      "logits/rejected": -0.5969452857971191,
-      "logps/chosen": -951.3689575195312,
-      "logps/rejected": -1269.2164306640625,
-      "loss": 0.5936,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": -0.657620906829834,
-      "rewards/margins": 0.3409743010997772,
-      "rewards/rejected": -0.9985952377319336,
+      "logits/chosen": -0.6187500357627869,
+      "logits/rejected": -0.5157658457756042,
+      "logps/chosen": -987.8230590820312,
+      "logps/rejected": -1380.6385498046875,
+      "loss": 0.5858,
+      "rewards/accuracies": 0.706250011920929,
+      "rewards/chosen": -0.6942586898803711,
+      "rewards/margins": 0.41675519943237305,
+      "rewards/rejected": -1.1110140085220337,
       "step": 430
     },
     {
       "epoch": 0.9209837781266352,
-      "grad_norm": 5.865963437205565,
+      "grad_norm": 6.9904169822044855,
       "learning_rate": 9.12094829893642e-09,
-      "logits/chosen": -0.7566797137260437,
-      "logits/rejected": -0.6950474977493286,
-      "logps/chosen": -855.38525390625,
-      "logps/rejected": -1079.9859619140625,
-      "loss": 0.5893,
+      "logits/chosen": -0.6982117891311646,
+      "logits/rejected": -0.6187722682952881,
+      "logps/chosen": -875.1052856445312,
+      "logps/rejected": -1151.622314453125,
+      "loss": 0.5794,
       "rewards/accuracies": 0.65625,
-      "rewards/chosen": -0.5882030725479126,
-      "rewards/margins": 0.26657402515411377,
-      "rewards/rejected": -0.8547771573066711,
+      "rewards/chosen": -0.6084715127944946,
+      "rewards/margins": 0.3183104991912842,
+      "rewards/rejected": -0.9267820119857788,
       "step": 440
     },
     {
       "epoch": 0.9419152276295133,
-      "grad_norm": 4.156556279338722,
+      "grad_norm": 5.431378633125538,
       "learning_rate": 4.8708793644441086e-09,
-      "logits/chosen": -0.6016022562980652,
-      "logits/rejected": -0.5587543845176697,
-      "logps/chosen": -906.1237182617188,
-      "logps/rejected": -1271.012939453125,
-      "loss": 0.5946,
-      "rewards/accuracies": 0.71875,
-      "rewards/chosen": -0.6333063244819641,
-      "rewards/margins": 0.3771924376487732,
-      "rewards/rejected": -1.0104987621307373,
+      "logits/chosen": -0.5444275140762329,
+      "logits/rejected": -0.48638853430747986,
+      "logps/chosen": -933.2413940429688,
+      "logps/rejected": -1319.6456298828125,
+      "loss": 0.5906,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.6616878509521484,
+      "rewards/margins": 0.39823150634765625,
+      "rewards/rejected": -1.0599192380905151,
       "step": 450
     },
     {
       "epoch": 0.9628466771323915,
-      "grad_norm": 5.511218104411099,
+      "grad_norm": 6.561608804085118,
       "learning_rate": 1.9347820230782295e-09,
-      "logits/chosen": -0.6558966636657715,
-      "logits/rejected": -0.5925637483596802,
-      "logps/chosen": -901.0379638671875,
-      "logps/rejected": -1152.3828125,
-      "loss": 0.6073,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": -0.6263371706008911,
-      "rewards/margins": 0.2950023114681244,
-      "rewards/rejected": -0.9213393926620483,
+      "logits/chosen": -0.6087552905082703,
+      "logits/rejected": -0.5210331678390503,
+      "logps/chosen": -938.4767456054688,
+      "logps/rejected": -1216.3330078125,
+      "loss": 0.6007,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": -0.6643339395523071,
+      "rewards/margins": 0.32233306765556335,
+      "rewards/rejected": -0.9866669774055481,
       "step": 460
     },
     {
       "epoch": 0.9837781266352695,
-      "grad_norm": 6.982569839470116,
+      "grad_norm": 7.959100177075258,
       "learning_rate": 3.2839470889836627e-10,
-      "logits/chosen": -0.7134594321250916,
-      "logits/rejected": -0.6819964051246643,
-      "logps/chosen": -918.5238037109375,
-      "logps/rejected": -1149.1669921875,
-      "loss": 0.5914,
-      "rewards/accuracies": 0.668749988079071,
-      "rewards/chosen": -0.6207371950149536,
-      "rewards/margins": 0.24755153059959412,
-      "rewards/rejected": -0.8682886958122253,
+      "logits/chosen": -0.657787024974823,
+      "logits/rejected": -0.610327422618866,
+      "logps/chosen": -939.5802001953125,
+      "logps/rejected": -1180.6595458984375,
+      "loss": 0.5855,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": -0.6418840885162354,
+      "rewards/margins": 0.2594161033630371,
+      "rewards/rejected": -0.9013002514839172,
       "step": 470
     },
     {
       "epoch": 0.9984301412872841,
-      "eval_logits/chosen": -0.7052573561668396,
-      "eval_logits/rejected": -0.6639654636383057,
-      "eval_logps/chosen": -872.1710815429688,
-      "eval_logps/rejected": -1305.126953125,
-      "eval_loss": 0.6051552891731262,
-      "eval_rewards/accuracies": 0.703125,
-      "eval_rewards/chosen": -0.5966997146606445,
-      "eval_rewards/margins": 0.4309939742088318,
-      "eval_rewards/rejected": -1.0276936292648315,
-      "eval_runtime": 241.7409,
-      "eval_samples_per_second": 8.273,
-      "eval_steps_per_second": 0.132,
+      "eval_logits/chosen": -0.6367819905281067,
+      "eval_logits/rejected": -0.5771758556365967,
+      "eval_logps/chosen": -907.722900390625,
+      "eval_logps/rejected": -1395.880859375,
+      "eval_loss": 0.5986923575401306,
+      "eval_rewards/accuracies": 0.70703125,
+      "eval_rewards/chosen": -0.6320347189903259,
+      "eval_rewards/margins": 0.4867088794708252,
+      "eval_rewards/rejected": -1.118743658065796,
+      "eval_runtime": 222.6786,
+      "eval_samples_per_second": 8.982,
+      "eval_steps_per_second": 0.144,
       "step": 477
     },
     {
       "epoch": 0.9984301412872841,
       "step": 477,
       "total_flos": 0.0,
-      "train_loss": 0.6380661858452691,
-      "train_runtime": 13550.8489,
+      "train_loss": 0.6339174956395691,
+      "train_runtime": 13552.9806,
       "train_samples_per_second": 4.511,
       "train_steps_per_second": 0.035
     }

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1470c61b471fc2fc704fc3fc509669eaa4ce43112937a3f7b3e24835c7742159
+oid sha256:7d60eab87c985a5c7650684bc1ed6bd80433e9b31291eda50488d9bc6ce7211a
 size 7672