Model save

README.md

@@ -1,13 +1,10 @@
 ---
 library_name: peft
 tags:
-- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
 base_model: NbAiLab/nb-gpt-j-6B-v2
-datasets:
-- hugodk-sch/aftonposten_title_prefs
 model-index:
 - name: aftonposten-6b-align-scan
   results: []
@@ -18,17 +15,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # aftonposten-6b-align-scan
 
-This model is a fine-tuned version of [data/ap-gpt-j-6b-sft-qlora-04-08](https://huggingface.co/data/ap-gpt-j-6b-sft-qlora-04-08) on the hugodk-sch/aftonposten_title_prefs dataset.
+This model is a fine-tuned version of [NbAiLab/nb-gpt-j-6B-v2](https://huggingface.co/NbAiLab/nb-gpt-j-6B-v2) on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.5753
-- Rewards/chosen: 0.0012
-- Rewards/rejected: 0.0139
-- Rewards/accuracies: 0.4576
-- Rewards/margins: -0.0127
-- Logps/rejected: -37.4968
-- Logps/chosen: -34.0329
-- Logits/rejected: -2.2389
-- Logits/chosen: -2.2438
+- Loss: 0.4345
+- Rewards/chosen: 0.0086
+- Rewards/rejected: 0.0130
+- Rewards/accuracies: 0.4809
+- Rewards/margins: -0.0044
+- Logps/rejected: -37.5004
+- Logps/chosen: -34.0238
+- Logits/rejected: -2.2387
+- Logits/chosen: -2.2436
 
 ## Model description
 
@@ -63,9 +60,9 @@ 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.5589        | 0.26  | 100  | 0.5107          | 0.0178         | 0.0027           | 0.5573             | 0.0151          | -37.5127       | -34.0091     | -2.2393         | -2.2442       |
-| 0.522         | 0.52  | 200  | 0.5423          | 0.0040         | 0.0035           | 0.4689             | 0.0005          | -37.5116       | -34.0288     | -2.2385         | -2.2433       |
-| 0.4616        | 0.78  | 300  | 0.5652          | 0.0047         | 0.0129           | 0.4809             | -0.0083         | -37.4981       | -34.0279     | -2.2391         | -2.2439       |
+| 0.3805        | 0.26  | 100  | 0.4061          | 0.0139         | 0.0017           | 0.5395             | 0.0122          | -37.5145       | -34.0171     | -2.2393         | -2.2441       |
+| 0.371         | 0.52  | 200  | 0.4323          | 0.0091         | 0.0118           | 0.5253             | -0.0027         | -37.5019       | -34.0231     | -2.2390         | -2.2439       |
+| 0.3839        | 0.78  | 300  | 0.4345          | 0.0086         | 0.0130           | 0.4809             | -0.0044         | -37.5004       | -34.0238     | -2.2387         | -2.2436       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8f7506d9d501fe5bf6a4f2b7a9ac06eb11dd866f4865777617896f075d6fc6a0
+oid sha256:dd4d4dffa9b30971c702f1f39924c707a8b1925a5d67f8de841d1128fe03486b
 size 176183216

all_results.json

@@ -1,21 +1,8 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -2.2437870502471924,
-    "eval_logits/rejected": -2.2389047145843506,
-    "eval_logps/chosen": -34.03290939331055,
-    "eval_logps/rejected": -37.49677658081055,
-    "eval_loss": 0.5752949118614197,
-    "eval_rewards/accuracies": 0.45764121413230896,
-    "eval_rewards/chosen": 0.001152246491983533,
-    "eval_rewards/margins": -0.012740416452288628,
-    "eval_rewards/rejected": 0.013892664574086666,
-    "eval_runtime": 145.7802,
-    "eval_samples": 343,
-    "eval_samples_per_second": 2.353,
-    "eval_steps_per_second": 0.295,
-    "train_loss": 0.5317366457604743,
-    "train_runtime": 3254.062,
+    "train_loss": 0.4092799360101873,
+    "train_runtime": 3256.9261,
     "train_samples": 3079,
-    "train_samples_per_second": 0.946,
+    "train_samples_per_second": 0.945,
     "train_steps_per_second": 0.118
 }

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.5317366457604743,
-    "train_runtime": 3254.062,
+    "train_loss": 0.4092799360101873,
+    "train_runtime": 3256.9261,
     "train_samples": 3079,
-    "train_samples_per_second": 0.946,
+    "train_samples_per_second": 0.945,
     "train_steps_per_second": 0.118
 }

trainer_state.json

@@ -15,7 +15,7 @@
       "logits/rejected": -1.7377450466156006,
       "logps/chosen": -29.553977966308594,
       "logps/rejected": -42.813133239746094,
-      "loss": 0.5102,
+      "loss": 0.3906,
       "rewards/accuracies": 0.0,
       "rewards/chosen": 0.0,
       "rewards/margins": 0.0,
@@ -25,590 +25,590 @@
     {
       "epoch": 0.03,
       "learning_rate": 1.2820512820512818e-07,
-      "logits/chosen": -1.8663597106933594,
-      "logits/rejected": -1.8706719875335693,
-      "logps/chosen": -36.990386962890625,
-      "logps/rejected": -33.658267974853516,
-      "loss": 0.4771,
+      "logits/chosen": -1.8662992715835571,
+      "logits/rejected": -1.8706117868423462,
+      "logps/chosen": -36.97681427001953,
+      "logps/rejected": -33.66523361206055,
+      "loss": 0.3447,
       "rewards/accuracies": 0.5416666865348816,
-      "rewards/chosen": 0.011347133666276932,
-      "rewards/margins": 0.031080076470971107,
-      "rewards/rejected": -0.019732946529984474,
+      "rewards/chosen": 0.023829741403460503,
+      "rewards/margins": 0.05195777863264084,
+      "rewards/rejected": -0.028128040954470634,
       "step": 10
     },
     {
       "epoch": 0.05,
       "learning_rate": 2.5641025641025636e-07,
-      "logits/chosen": -1.9978214502334595,
-      "logits/rejected": -2.000457286834717,
-      "logps/chosen": -29.62484359741211,
-      "logps/rejected": -29.059850692749023,
-      "loss": 0.5261,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.012144992128014565,
-      "rewards/margins": 0.009631244465708733,
-      "rewards/rejected": 0.0025137457996606827,
+      "logits/chosen": -1.997193694114685,
+      "logits/rejected": -1.9998573064804077,
+      "logps/chosen": -29.65359878540039,
+      "logps/rejected": -29.054311752319336,
+      "loss": 0.4534,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": -0.009124360978603363,
+      "rewards/margins": -0.016431041061878204,
+      "rewards/rejected": 0.007306680083274841,
       "step": 20
     },
     {
       "epoch": 0.08,
       "learning_rate": 3.8461538461538463e-07,
-      "logits/chosen": -1.9197276830673218,
-      "logits/rejected": -1.9170366525650024,
-      "logps/chosen": -31.404308319091797,
-      "logps/rejected": -33.229034423828125,
-      "loss": 0.5076,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.008251860737800598,
-      "rewards/margins": 0.015654325485229492,
-      "rewards/rejected": -0.007402463350445032,
+      "logits/chosen": -1.9197280406951904,
+      "logits/rejected": -1.9170429706573486,
+      "logps/chosen": -31.414775848388672,
+      "logps/rejected": -33.24064254760742,
+      "loss": 0.4025,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.0010542668169364333,
+      "rewards/margins": 0.018799806013703346,
+      "rewards/rejected": -0.01774553768336773,
       "step": 30
     },
     {
       "epoch": 0.1,
       "learning_rate": 4.999896948438433e-07,
-      "logits/chosen": -2.0169289112091064,
-      "logits/rejected": -2.008176803588867,
-      "logps/chosen": -32.55079650878906,
-      "logps/rejected": -32.502708435058594,
-      "loss": 0.535,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.018207356333732605,
-      "rewards/margins": 0.011235545389354229,
-      "rewards/rejected": 0.006971807684749365,
+      "logits/chosen": -2.0171010494232178,
+      "logits/rejected": -2.008350372314453,
+      "logps/chosen": -32.59648513793945,
+      "logps/rejected": -32.50862121582031,
+      "loss": 0.4532,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": -0.015740757808089256,
+      "rewards/margins": -0.018979396671056747,
+      "rewards/rejected": 0.0032386414241045713,
       "step": 40
     },
     {
       "epoch": 0.13,
       "learning_rate": 4.987541037542186e-07,
-      "logits/chosen": -1.86395263671875,
-      "logits/rejected": -1.8531732559204102,
-      "logps/chosen": -33.559814453125,
-      "logps/rejected": -35.43522262573242,
-      "loss": 0.5718,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": -0.0016335565596818924,
-      "rewards/margins": -0.013616559095680714,
-      "rewards/rejected": 0.011983001604676247,
+      "logits/chosen": -1.8645904064178467,
+      "logits/rejected": -1.8538070917129517,
+      "logps/chosen": -33.542823791503906,
+      "logps/rejected": -35.43744659423828,
+      "loss": 0.4304,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": 0.011724205687642097,
+      "rewards/margins": -0.00019102543592453003,
+      "rewards/rejected": 0.011915231123566628,
       "step": 50
     },
     {
       "epoch": 0.16,
       "learning_rate": 4.954691471941118e-07,
-      "logits/chosen": -1.9451929330825806,
-      "logits/rejected": -1.9471458196640015,
-      "logps/chosen": -32.59247589111328,
-      "logps/rejected": -33.19312286376953,
-      "loss": 0.4982,
-      "rewards/accuracies": 0.574999988079071,
-      "rewards/chosen": 0.005393369123339653,
-      "rewards/margins": 0.020615221932530403,
-      "rewards/rejected": -0.015221851877868176,
+      "logits/chosen": -1.9461469650268555,
+      "logits/rejected": -1.948094129562378,
+      "logps/chosen": -32.587215423583984,
+      "logps/rejected": -33.21445846557617,
+      "loss": 0.3539,
+      "rewards/accuracies": 0.637499988079071,
+      "rewards/chosen": 0.010374858975410461,
+      "rewards/margins": 0.04483799636363983,
+      "rewards/rejected": -0.03446313738822937,
       "step": 60
     },
     {
       "epoch": 0.18,
       "learning_rate": 4.901618883413548e-07,
-      "logits/chosen": -2.0797886848449707,
-      "logits/rejected": -2.08477783203125,
-      "logps/chosen": -33.99976348876953,
-      "logps/rejected": -36.57415771484375,
-      "loss": 0.575,
-      "rewards/accuracies": 0.44999998807907104,
-      "rewards/chosen": -0.007893012836575508,
-      "rewards/margins": -0.004194633569568396,
-      "rewards/rejected": -0.0036983792670071125,
+      "logits/chosen": -2.079591989517212,
+      "logits/rejected": -2.0845742225646973,
+      "logps/chosen": -34.00908279418945,
+      "logps/rejected": -36.58150863647461,
+      "loss": 0.4506,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.01648041605949402,
+      "rewards/margins": -0.006370754446834326,
+      "rewards/rejected": -0.010109663009643555,
       "step": 70
     },
     {
       "epoch": 0.21,
       "learning_rate": 4.828760511501322e-07,
-      "logits/chosen": -1.9419677257537842,
-      "logits/rejected": -1.94512939453125,
-      "logps/chosen": -34.39982986450195,
-      "logps/rejected": -34.590721130371094,
-      "loss": 0.5441,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.019587691873311996,
-      "rewards/margins": 0.01840476132929325,
-      "rewards/rejected": 0.0011829293798655272,
+      "logits/chosen": -1.9423307180404663,
+      "logits/rejected": -1.9454963207244873,
+      "logps/chosen": -34.39698028564453,
+      "logps/rejected": -34.59761428833008,
+      "loss": 0.3889,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.024663064628839493,
+      "rewards/margins": 0.028829574584960938,
+      "rewards/rejected": -0.00416650902479887,
       "step": 80
     },
     {
       "epoch": 0.23,
       "learning_rate": 4.736716601303429e-07,
-      "logits/chosen": -1.9507859945297241,
-      "logits/rejected": -1.9552862644195557,
-      "logps/chosen": -32.467891693115234,
-      "logps/rejected": -32.358272552490234,
-      "loss": 0.5391,
-      "rewards/accuracies": 0.5375000238418579,
-      "rewards/chosen": -0.00033024101867340505,
-      "rewards/margins": -0.0007759220898151398,
-      "rewards/rejected": 0.00044568348675966263,
+      "logits/chosen": -1.9508873224258423,
+      "logits/rejected": -1.9553953409194946,
+      "logps/chosen": -32.48583221435547,
+      "logps/rejected": -32.35867691040039,
+      "loss": 0.4445,
+      "rewards/accuracies": 0.4625000059604645,
+      "rewards/chosen": -0.014726865105330944,
+      "rewards/margins": -0.014909917488694191,
+      "rewards/rejected": 0.00018304325931239873,
       "step": 90
     },
     {
       "epoch": 0.26,
       "learning_rate": 4.62624545834521e-07,
-      "logits/chosen": -2.049070358276367,
-      "logits/rejected": -2.047074794769287,
-      "logps/chosen": -32.25985336303711,
-      "logps/rejected": -31.274398803710938,
-      "loss": 0.5589,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": -0.013313899748027325,
-      "rewards/margins": -0.007583809085190296,
-      "rewards/rejected": -0.005730087868869305,
+      "logits/chosen": -2.0491719245910645,
+      "logits/rejected": -2.047179937362671,
+      "logps/chosen": -32.22635269165039,
+      "logps/rejected": -31.287487030029297,
+      "loss": 0.3805,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.011584864929318428,
+      "rewards/margins": 0.02860759198665619,
+      "rewards/rejected": -0.01702272891998291,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": -2.244201421737671,
-      "eval_logits/rejected": -2.239315986633301,
-      "eval_logps/chosen": -34.009124755859375,
-      "eval_logps/rejected": -37.5127067565918,
-      "eval_loss": 0.5106843709945679,
-      "eval_rewards/accuracies": 0.5573089718818665,
-      "eval_rewards/chosen": 0.017796913161873817,
-      "eval_rewards/margins": 0.015057443641126156,
-      "eval_rewards/rejected": 0.0027394662611186504,
-      "eval_runtime": 146.0293,
-      "eval_samples_per_second": 2.349,
+      "eval_logits/chosen": -2.244148015975952,
+      "eval_logits/rejected": -2.2392663955688477,
+      "eval_logps/chosen": -34.01713943481445,
+      "eval_logps/rejected": -37.514495849609375,
+      "eval_loss": 0.4061162769794464,
+      "eval_rewards/accuracies": 0.5394518375396729,
+      "eval_rewards/chosen": 0.013928660191595554,
+      "eval_rewards/margins": 0.012230273336172104,
+      "eval_rewards/rejected": 0.0016983875539153814,
+      "eval_runtime": 146.1091,
+      "eval_samples_per_second": 2.348,
       "eval_steps_per_second": 0.294,
       "step": 100
     },
     {
       "epoch": 0.29,
       "learning_rate": 4.4982572012636904e-07,
-      "logits/chosen": -2.005356550216675,
-      "logits/rejected": -2.0029516220092773,
-      "logps/chosen": -33.2292366027832,
-      "logps/rejected": -34.054996490478516,
-      "loss": 0.4957,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.01020804513245821,
-      "rewards/margins": 0.028141701593995094,
-      "rewards/rejected": -0.01793365553021431,
+      "logits/chosen": -2.005580186843872,
+      "logits/rejected": -2.0031564235687256,
+      "logps/chosen": -33.24415588378906,
+      "logps/rejected": -33.99993133544922,
+      "loss": 0.4792,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": -0.00026987865567207336,
+      "rewards/margins": -0.02382112666964531,
+      "rewards/rejected": 0.023551244288682938,
       "step": 110
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.353806263777677e-07,
-      "logits/chosen": -2.0166029930114746,
-      "logits/rejected": -2.0082242488861084,
-      "logps/chosen": -32.45597457885742,
-      "logps/rejected": -32.18632507324219,
-      "loss": 0.5282,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": -0.007746054325252771,
-      "rewards/margins": 0.009627602994441986,
-      "rewards/rejected": -0.01737365685403347,
+      "logits/chosen": -2.01680326461792,
+      "logits/rejected": -2.0084121227264404,
+      "logps/chosen": -32.45465850830078,
+      "logps/rejected": -32.1729850769043,
+      "loss": 0.4286,
+      "rewards/accuracies": 0.4375,
+      "rewards/chosen": -0.007803081069141626,
+      "rewards/margins": 0.0013800703454762697,
+      "rewards/rejected": -0.009183152578771114,
       "step": 120
     },
     {
       "epoch": 0.34,
       "learning_rate": 4.194082707715275e-07,
-      "logits/chosen": -2.046288251876831,
-      "logits/rejected": -2.038238525390625,
-      "logps/chosen": -30.492712020874023,
-      "logps/rejected": -32.042259216308594,
-      "loss": 0.5777,
-      "rewards/accuracies": 0.44999998807907104,
-      "rewards/chosen": -0.011898026801645756,
-      "rewards/margins": -0.01942119374871254,
-      "rewards/rejected": 0.007523166947066784,
+      "logits/chosen": -2.0466856956481934,
+      "logits/rejected": -2.0386409759521484,
+      "logps/chosen": -30.494335174560547,
+      "logps/rejected": -32.051963806152344,
+      "loss": 0.4497,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.014895597472786903,
+      "rewards/margins": -0.015729816630482674,
+      "rewards/rejected": 0.0008342192741110921,
       "step": 130
     },
     {
       "epoch": 0.36,
       "learning_rate": 4.020402418666621e-07,
-      "logits/chosen": -1.976900339126587,
-      "logits/rejected": -1.9871864318847656,
-      "logps/chosen": -31.39472007751465,
-      "logps/rejected": -32.55016326904297,
-      "loss": 0.4864,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.01745815947651863,
-      "rewards/margins": 0.028155237436294556,
-      "rewards/rejected": -0.01069707702845335,
+      "logits/chosen": -1.977065086364746,
+      "logits/rejected": -1.987343430519104,
+      "logps/chosen": -31.37868309020996,
+      "logps/rejected": -32.54730224609375,
+      "loss": 0.3589,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.03278205543756485,
+      "rewards/margins": 0.04271895810961723,
+      "rewards/rejected": -0.009936909191310406,
       "step": 140
     },
     {
       "epoch": 0.39,
       "learning_rate": 3.8341962650351185e-07,
-      "logits/chosen": -1.8908016681671143,
-      "logits/rejected": -1.8918870687484741,
-      "logps/chosen": -34.199378967285156,
-      "logps/rejected": -34.751861572265625,
-      "loss": 0.5621,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": -0.004678909666836262,
-      "rewards/margins": -0.008454290218651295,
-      "rewards/rejected": 0.0037753782235085964,
+      "logits/chosen": -1.8916466236114502,
+      "logits/rejected": -1.8927490711212158,
+      "logps/chosen": -34.209651947021484,
+      "logps/rejected": -34.766143798828125,
+      "loss": 0.4325,
+      "rewards/accuracies": 0.42500001192092896,
+      "rewards/chosen": -0.013569498434662819,
+      "rewards/margins": -0.00645996630191803,
+      "rewards/rejected": -0.007109532598406076,
       "step": 150
     },
     {
       "epoch": 0.42,
       "learning_rate": 3.636998309800572e-07,
-      "logits/chosen": -1.9425878524780273,
-      "logits/rejected": -1.9391052722930908,
-      "logps/chosen": -36.136619567871094,
-      "logps/rejected": -32.712371826171875,
-      "loss": 0.5148,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.02454109489917755,
-      "rewards/margins": 0.01727622002363205,
-      "rewards/rejected": 0.007264876272529364,
+      "logits/chosen": -1.9427303075790405,
+      "logits/rejected": -1.939252495765686,
+      "logps/chosen": -36.14452362060547,
+      "logps/rejected": -32.73284149169922,
+      "loss": 0.3796,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.021723434329032898,
+      "rewards/margins": 0.02979486621916294,
+      "rewards/rejected": -0.008071433752775192,
       "step": 160
     },
     {
       "epoch": 0.44,
       "learning_rate": 3.430433172111807e-07,
-      "logits/chosen": -2.042212963104248,
-      "logits/rejected": -2.0348353385925293,
-      "logps/chosen": -33.78378677368164,
-      "logps/rejected": -31.365230560302734,
-      "loss": 0.5133,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.009854594245553017,
-      "rewards/margins": 0.01922372169792652,
-      "rewards/rejected": -0.009369125589728355,
+      "logits/chosen": -2.0424513816833496,
+      "logits/rejected": -2.035060405731201,
+      "logps/chosen": -33.786170959472656,
+      "logps/rejected": -31.34820556640625,
+      "loss": 0.4172,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": 0.009350698441267014,
+      "rewards/margins": 0.006434415467083454,
+      "rewards/rejected": 0.0029162843711674213,
       "step": 170
     },
     {
       "epoch": 0.47,
       "learning_rate": 3.216202642830543e-07,
-      "logits/chosen": -2.047489881515503,
-      "logits/rejected": -2.0527498722076416,
-      "logps/chosen": -32.54710388183594,
-      "logps/rejected": -32.50310134887695,
-      "loss": 0.5261,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": -0.001622129580937326,
-      "rewards/margins": 0.010947163216769695,
-      "rewards/rejected": -0.01256929337978363,
+      "logits/chosen": -2.0475661754608154,
+      "logits/rejected": -2.052834987640381,
+      "logps/chosen": -32.528114318847656,
+      "logps/rejected": -32.50902557373047,
+      "loss": 0.3766,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": 0.013338183984160423,
+      "rewards/margins": 0.03244578838348389,
+      "rewards/rejected": -0.019107606261968613,
       "step": 180
     },
     {
       "epoch": 0.49,
       "learning_rate": 2.9960716642946403e-07,
-      "logits/chosen": -2.0485682487487793,
-      "logits/rejected": -2.04580020904541,
-      "logps/chosen": -31.48910903930664,
-      "logps/rejected": -31.318958282470703,
-      "loss": 0.5482,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": -0.00010873023711610585,
-      "rewards/margins": 0.0004080161452293396,
-      "rewards/rejected": -0.000516746542416513,
+      "logits/chosen": -2.048661231994629,
+      "logits/rejected": -2.045872926712036,
+      "logps/chosen": -31.496755599975586,
+      "logps/rejected": -31.314464569091797,
+      "loss": 0.4475,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.00624021515250206,
+      "rewards/margins": -0.009242123924195766,
+      "rewards/rejected": 0.003001909703016281,
       "step": 190
     },
     {
       "epoch": 0.52,
       "learning_rate": 2.771853789806683e-07,
-      "logits/chosen": -1.9184706211090088,
-      "logits/rejected": -1.9231430292129517,
-      "logps/chosen": -31.597286224365234,
-      "logps/rejected": -32.80516815185547,
-      "loss": 0.522,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.006296842359006405,
-      "rewards/margins": 0.016262350603938103,
-      "rewards/rejected": -0.009965506382286549,
+      "logits/chosen": -1.9193611145019531,
+      "logits/rejected": -1.9240529537200928,
+      "logps/chosen": -31.579212188720703,
+      "logps/rejected": -32.80603790283203,
+      "loss": 0.371,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.02165621519088745,
+      "rewards/margins": 0.033739686012268066,
+      "rewards/rejected": -0.012083468958735466,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": -2.2433319091796875,
-      "eval_logits/rejected": -2.238459825515747,
-      "eval_logps/chosen": -34.028812408447266,
-      "eval_logps/rejected": -37.51158905029297,
-      "eval_loss": 0.5422906875610352,
-      "eval_rewards/accuracies": 0.46885380148887634,
-      "eval_rewards/chosen": 0.00401716772466898,
-      "eval_rewards/margins": 0.0004931418807245791,
-      "eval_rewards/rejected": 0.0035240259021520615,
-      "eval_runtime": 145.7818,
-      "eval_samples_per_second": 2.353,
+      "eval_logits/chosen": -2.243856191635132,
+      "eval_logits/rejected": -2.2389791011810303,
+      "eval_logps/chosen": -34.023136138916016,
+      "eval_logps/rejected": -37.501853942871094,
+      "eval_loss": 0.4322855770587921,
+      "eval_rewards/accuracies": 0.5253322720527649,
+      "eval_rewards/chosen": 0.009133166633546352,
+      "eval_rewards/margins": -0.002679171971976757,
+      "eval_rewards/rejected": 0.01181233860552311,
+      "eval_runtime": 145.9618,
+      "eval_samples_per_second": 2.35,
       "eval_steps_per_second": 0.295,
       "step": 200
     },
     {
       "epoch": 0.55,
       "learning_rate": 2.5453962426402e-07,
-      "logits/chosen": -2.0319957733154297,
-      "logits/rejected": -2.0426838397979736,
-      "logps/chosen": -31.943634033203125,
-      "logps/rejected": -33.902008056640625,
-      "loss": 0.4594,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.01936788484454155,
-      "rewards/margins": 0.046881090849637985,
-      "rewards/rejected": -0.027513209730386734,
+      "logits/chosen": -2.0320448875427246,
+      "logits/rejected": -2.0427393913269043,
+      "logps/chosen": -31.947372436523438,
+      "logps/rejected": -33.899864196777344,
+      "loss": 0.3622,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": 0.019143681973218918,
+      "rewards/margins": 0.04887578636407852,
+      "rewards/rejected": -0.029732098802924156,
       "step": 210
     },
     {
       "epoch": 0.57,
       "learning_rate": 2.318564697655179e-07,
-      "logits/chosen": -1.9252302646636963,
-      "logits/rejected": -1.9400790929794312,
-      "logps/chosen": -30.092998504638672,
-      "logps/rejected": -31.55881690979004,
-      "loss": 0.5269,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.012976284138858318,
-      "rewards/margins": 0.013424187898635864,
-      "rewards/rejected": -0.0004478988121263683,
+      "logits/chosen": -1.9257709980010986,
+      "logits/rejected": -1.940639853477478,
+      "logps/chosen": -30.079341888427734,
+      "logps/rejected": -31.5616397857666,
+      "loss": 0.3884,
+      "rewards/accuracies": 0.612500011920929,
+      "rewards/chosen": 0.025758206844329834,
+      "rewards/margins": 0.028527002781629562,
+      "rewards/rejected": -0.0027687971014529467,
       "step": 220
     },
     {
       "epoch": 0.6,
       "learning_rate": 2.093227910899832e-07,
-      "logits/chosen": -1.983025312423706,
-      "logits/rejected": -1.9869884252548218,
-      "logps/chosen": -33.37274932861328,
-      "logps/rejected": -31.550548553466797,
-      "loss": 0.496,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.03130333498120308,
-      "rewards/margins": 0.03217558190226555,
-      "rewards/rejected": -0.0008722454076632857,
+      "logits/chosen": -1.9829959869384766,
+      "logits/rejected": -1.9869670867919922,
+      "logps/chosen": -33.4053955078125,
+      "logps/rejected": -31.562353134155273,
+      "loss": 0.3918,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.009657363407313824,
+      "rewards/margins": 0.02010050043463707,
+      "rewards/rejected": -0.010443134233355522,
       "step": 230
     },
     {
       "epoch": 0.62,
       "learning_rate": 1.8712423238279356e-07,
-      "logits/chosen": -1.983493447303772,
-      "logits/rejected": -1.9615182876586914,
-      "logps/chosen": -34.13421630859375,
-      "logps/rejected": -34.97159957885742,
-      "loss": 0.4915,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.01367081981152296,
-      "rewards/margins": 0.02823723293840885,
-      "rewards/rejected": -0.014566412195563316,
+      "logits/chosen": -1.9836658239364624,
+      "logits/rejected": -1.9616800546646118,
+      "logps/chosen": -34.175201416015625,
+      "logps/rejected": -34.966102600097656,
+      "loss": 0.4346,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.017159918323159218,
+      "rewards/margins": -0.004913450218737125,
+      "rewards/rejected": -0.012246469967067242,
       "step": 240
     },
     {
       "epoch": 0.65,
       "learning_rate": 1.654436768970182e-07,
-      "logits/chosen": -2.02471661567688,
-      "logits/rejected": -2.0214104652404785,
-      "logps/chosen": -32.94930648803711,
-      "logps/rejected": -36.21381759643555,
-      "loss": 0.6084,
-      "rewards/accuracies": 0.4000000059604645,
-      "rewards/chosen": -0.016968127340078354,
-      "rewards/margins": -0.02474220283329487,
-      "rewards/rejected": 0.007774075958877802,
+      "logits/chosen": -2.0244596004486084,
+      "logits/rejected": -2.021144390106201,
+      "logps/chosen": -32.932373046875,
+      "logps/rejected": -36.2264289855957,
+      "loss": 0.4397,
+      "rewards/accuracies": 0.512499988079071,
+      "rewards/chosen": -0.005845203995704651,
+      "rewards/margins": -0.004640025552362204,
+      "rewards/rejected": -0.001205177279189229,
       "step": 250
     },
     {
       "epoch": 0.68,
       "learning_rate": 1.444597403062196e-07,
-      "logits/chosen": -1.8912674188613892,
-      "logits/rejected": -1.8888145685195923,
-      "logps/chosen": -34.208099365234375,
-      "logps/rejected": -35.50531005859375,
-      "loss": 0.5753,
-      "rewards/accuracies": 0.44999998807907104,
-      "rewards/chosen": -0.013791380450129509,
-      "rewards/margins": -0.013306483626365662,
-      "rewards/rejected": -0.00048489533946849406,
+      "logits/chosen": -1.891283392906189,
+      "logits/rejected": -1.8888483047485352,
+      "logps/chosen": -34.20048522949219,
+      "logps/rejected": -35.507022857666016,
+      "loss": 0.4441,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": -0.009669994935393333,
+      "rewards/margins": -0.0077504729852080345,
+      "rewards/rejected": -0.0019195213681086898,
       "step": 260
     },
     {
       "epoch": 0.7,
       "learning_rate": 1.2434529917578887e-07,
-      "logits/chosen": -1.8766494989395142,
-      "logits/rejected": -1.8741207122802734,
-      "logps/chosen": -34.38150405883789,
-      "logps/rejected": -31.744686126708984,
-      "loss": 0.5631,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": 0.011268051341176033,
-      "rewards/margins": -0.007382377982139587,
-      "rewards/rejected": 0.018650425598025322,
+      "logits/chosen": -1.8760111331939697,
+      "logits/rejected": -1.8734772205352783,
+      "logps/chosen": -34.39020919799805,
+      "logps/rejected": -31.75579261779785,
+      "loss": 0.4295,
+      "rewards/accuracies": 0.42500001192092896,
+      "rewards/chosen": 0.0059113698080182076,
+      "rewards/margins": -0.006519269198179245,
+      "rewards/rejected": 0.012430639937520027,
       "step": 270
     },
     {
       "epoch": 0.73,
       "learning_rate": 1.0526606671603521e-07,
-      "logits/chosen": -1.9800630807876587,
-      "logits/rejected": -1.9694359302520752,
-      "logps/chosen": -35.32728958129883,
-      "logps/rejected": -31.843835830688477,
-      "loss": 0.4967,
-      "rewards/accuracies": 0.550000011920929,
-      "rewards/chosen": 0.02054680697619915,
-      "rewards/margins": 0.022537903860211372,
-      "rewards/rejected": -0.001991095719859004,
+      "logits/chosen": -1.9797089099884033,
+      "logits/rejected": -1.9690707921981812,
+      "logps/chosen": -35.31258773803711,
+      "logps/rejected": -31.837697982788086,
+      "loss": 0.3868,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.03524213656783104,
+      "rewards/margins": 0.0326063297688961,
+      "rewards/rejected": 0.0026358035393059254,
       "step": 280
     },
     {
       "epoch": 0.75,
       "learning_rate": 8.737922755071453e-08,
-      "logits/chosen": -2.0754363536834717,
-      "logits/rejected": -2.060412645339966,
-      "logps/chosen": -30.902868270874023,
-      "logps/rejected": -32.63262939453125,
-      "loss": 0.5486,
-      "rewards/accuracies": 0.4625000059604645,
-      "rewards/chosen": 0.014878431335091591,
-      "rewards/margins": 0.0014350058045238256,
-      "rewards/rejected": 0.013443423435091972,
+      "logits/chosen": -2.0755209922790527,
+      "logits/rejected": -2.060497760772705,
+      "logps/chosen": -30.90865135192871,
+      "logps/rejected": -32.64521026611328,
+      "loss": 0.41,
+      "rewards/accuracies": 0.44999998807907104,
+      "rewards/chosen": 0.012379061430692673,
+      "rewards/margins": 0.007080032490193844,
+      "rewards/rejected": 0.005299028940498829,
       "step": 290
     },
     {
       "epoch": 0.78,
       "learning_rate": 7.08321427484816e-08,
-      "logits/chosen": -1.9468863010406494,
-      "logits/rejected": -1.9443439245224,
-      "logps/chosen": -32.88282775878906,
-      "logps/rejected": -30.835247039794922,
-      "loss": 0.4616,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.03099890425801277,
-      "rewards/margins": 0.04390609636902809,
-      "rewards/rejected": -0.012907189317047596,
+      "logits/chosen": -1.9463651180267334,
+      "logits/rejected": -1.9438308477401733,
+      "logps/chosen": -32.89823532104492,
+      "logps/rejected": -30.81850814819336,
+      "loss": 0.3839,
+      "rewards/accuracies": 0.5874999761581421,
+      "rewards/chosen": 0.02309919334948063,
+      "rewards/margins": 0.024458223953843117,
+      "rewards/rejected": -0.0013590289745479822,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": -2.243941307067871,
-      "eval_logits/rejected": -2.2390596866607666,
-      "eval_logps/chosen": -34.027870178222656,
-      "eval_logps/rejected": -37.49814224243164,
-      "eval_loss": 0.5651828050613403,
+      "eval_logits/chosen": -2.2435548305511475,
+      "eval_logits/rejected": -2.2386720180511475,
+      "eval_logps/chosen": -34.023826599121094,
+      "eval_logps/rejected": -37.50039291381836,
+      "eval_loss": 0.4344586431980133,
       "eval_rewards/accuracies": 0.4808970093727112,
-      "eval_rewards/chosen": 0.004676156677305698,
-      "eval_rewards/margins": -0.008257162757217884,
-      "eval_rewards/rejected": 0.012933320365846157,
-      "eval_runtime": 145.8956,
-      "eval_samples_per_second": 2.351,
+      "eval_rewards/chosen": 0.00858243927359581,
+      "eval_rewards/margins": -0.004399063065648079,
+      "eval_rewards/rejected": 0.012981505133211613,
+      "eval_runtime": 145.9473,
+      "eval_samples_per_second": 2.35,
       "eval_steps_per_second": 0.295,
       "step": 300
     },
     {
       "epoch": 0.81,
       "learning_rate": 5.576113578589034e-08,
-      "logits/chosen": -1.928396224975586,
-      "logits/rejected": -1.9251337051391602,
-      "logps/chosen": -31.597158432006836,
-      "logps/rejected": -33.751304626464844,
-      "loss": 0.5251,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.004333490040153265,
-      "rewards/margins": 0.01073872484266758,
-      "rewards/rejected": -0.006405232939869165,
+      "logits/chosen": -1.9286285638809204,
+      "logits/rejected": -1.9253568649291992,
+      "logps/chosen": -31.583232879638672,
+      "logps/rejected": -33.75123596191406,
+      "loss": 0.3881,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.016091840341687202,
+      "rewards/margins": 0.023359118029475212,
+      "rewards/rejected": -0.007267280016094446,
       "step": 310
     },
     {
       "epoch": 0.83,
       "learning_rate": 4.229036944380912e-08,
-      "logits/chosen": -1.9806029796600342,
-      "logits/rejected": -1.9683090448379517,
-      "logps/chosen": -34.563812255859375,
-      "logps/rejected": -33.55340576171875,
-      "loss": 0.4803,
-      "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.022582050412893295,
-      "rewards/margins": 0.03277861326932907,
-      "rewards/rejected": -0.010196560993790627,
+      "logits/chosen": -1.9803975820541382,
+      "logits/rejected": -1.9680900573730469,
+      "logps/chosen": -34.58079147338867,
+      "logps/rejected": -33.575767517089844,
+      "loss": 0.3778,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.012225466780364513,
+      "rewards/margins": 0.041770923882722855,
+      "rewards/rejected": -0.029545456171035767,
       "step": 320
     },
     {
       "epoch": 0.86,
       "learning_rate": 3.053082288996112e-08,
-      "logits/chosen": -2.0163512229919434,
-      "logits/rejected": -2.0149030685424805,
-      "logps/chosen": -33.46595764160156,
-      "logps/rejected": -32.46282196044922,
-      "loss": 0.5443,
-      "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": 0.013885289430618286,
-      "rewards/margins": 0.0021549216471612453,
-      "rewards/rejected": 0.011730367317795753,
+      "logits/chosen": -2.015634298324585,
+      "logits/rejected": -2.0141713619232178,
+      "logps/chosen": -33.45996856689453,
+      "logps/rejected": -32.48029327392578,
+      "loss": 0.4032,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.020660031586885452,
+      "rewards/margins": 0.021225089207291603,
+      "rewards/rejected": -0.0005650619277730584,
       "step": 330
     },
     {
       "epoch": 0.88,
       "learning_rate": 2.05793773749158e-08,
-      "logits/chosen": -2.1030476093292236,
-      "logits/rejected": -2.0872480869293213,
-      "logps/chosen": -34.173221588134766,
-      "logps/rejected": -33.08686447143555,
-      "loss": 0.5853,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.004321468528360128,
-      "rewards/margins": -0.01749541237950325,
-      "rewards/rejected": 0.021816883236169815,
+      "logits/chosen": -2.1035873889923096,
+      "logits/rejected": -2.0877768993377686,
+      "logps/chosen": -34.162208557128906,
+      "logps/rejected": -33.095733642578125,
+      "loss": 0.439,
+      "rewards/accuracies": 0.4625000059604645,
+      "rewards/chosen": 0.013746557757258415,
+      "rewards/margins": -0.004091509617865086,
+      "rewards/rejected": 0.017838066443800926,
       "step": 340
     },
     {
       "epoch": 0.91,
       "learning_rate": 1.251801807404168e-08,
-      "logits/chosen": -1.9748103618621826,
-      "logits/rejected": -1.9738647937774658,
-      "logps/chosen": -33.254066467285156,
-      "logps/rejected": -32.4422721862793,
-      "loss": 0.5603,
-      "rewards/accuracies": 0.38749998807907104,
-      "rewards/chosen": 0.019621744751930237,
-      "rewards/margins": 0.0036095953546464443,
-      "rewards/rejected": 0.016012147068977356,
+      "logits/chosen": -1.9746344089508057,
+      "logits/rejected": -1.9736888408660889,
+      "logps/chosen": -33.253448486328125,
+      "logps/rejected": -32.458499908447266,
+      "loss": 0.3954,
+      "rewards/accuracies": 0.5249999761581421,
+      "rewards/chosen": 0.02291743829846382,
+      "rewards/margins": 0.017600122839212418,
+      "rewards/rejected": 0.005317316390573978,
       "step": 350
     },
     {
       "epoch": 0.94,
       "learning_rate": 6.41315865106129e-09,
-      "logits/chosen": -1.9305530786514282,
-      "logits/rejected": -1.9409068822860718,
-      "logps/chosen": -32.2257080078125,
-      "logps/rejected": -35.29142379760742,
-      "loss": 0.5729,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": -0.0147119564935565,
-      "rewards/margins": -0.012842650525271893,
-      "rewards/rejected": -0.0018693048041313887,
+      "logits/chosen": -1.9309253692626953,
+      "logits/rejected": -1.941303014755249,
+      "logps/chosen": -32.21305465698242,
+      "logps/rejected": -35.309688568115234,
+      "loss": 0.4095,
+      "rewards/accuracies": 0.42500001192092896,
+      "rewards/chosen": -0.006690672133117914,
+      "rewards/margins": 0.010053041391074657,
+      "rewards/rejected": -0.01674371212720871,
       "step": 360
     },
     {
       "epoch": 0.96,
       "learning_rate": 2.3150941078050324e-09,
-      "logits/chosen": -2.069432497024536,
-      "logits/rejected": -2.0628814697265625,
-      "logps/chosen": -33.65316390991211,
-      "logps/rejected": -29.218481063842773,
-      "loss": 0.5602,
+      "logits/chosen": -2.069108009338379,
+      "logits/rejected": -2.0625574588775635,
+      "logps/chosen": -33.637176513671875,
+      "logps/rejected": -29.221187591552734,
+      "loss": 0.4059,
       "rewards/accuracies": 0.48750001192092896,
-      "rewards/chosen": -0.000836326158605516,
-      "rewards/margins": -0.006397420074790716,
-      "rewards/rejected": 0.005561096128076315,
+      "rewards/chosen": 0.011833530850708485,
+      "rewards/margins": 0.00764369685202837,
+      "rewards/rejected": 0.004189834464341402,
       "step": 370
     },
     {
       "epoch": 0.99,
       "learning_rate": 2.575864278703266e-10,
-      "logits/chosen": -1.9289767742156982,
-      "logits/rejected": -1.9311527013778687,
-      "logps/chosen": -34.243995666503906,
-      "logps/rejected": -30.892364501953125,
-      "loss": 0.5254,
-      "rewards/accuracies": 0.5249999761581421,
-      "rewards/chosen": -0.0034601963125169277,
-      "rewards/margins": 0.010354455560445786,
-      "rewards/rejected": -0.013814652338624,
+      "logits/chosen": -1.9292182922363281,
+      "logits/rejected": -1.9313886165618896,
+      "logps/chosen": -34.24732971191406,
+      "logps/rejected": -30.893402099609375,
+      "loss": 0.4086,
+      "rewards/accuracies": 0.4749999940395355,
+      "rewards/chosen": -0.006622877903282642,
+      "rewards/margins": 0.009996414184570312,
+      "rewards/rejected": -0.01661929115653038,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.5317366457604743,
-      "train_runtime": 3254.062,
-      "train_samples_per_second": 0.946,
+      "train_loss": 0.4092799360101873,
+      "train_runtime": 3256.9261,
+      "train_samples_per_second": 0.945,
       "train_steps_per_second": 0.118
     }
   ],