Model save

README.md

@@ -1,13 +1,11 @@
 ---
+license: apache-2.0
 library_name: peft
 tags:
-- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
 base_model: norallm/normistral-7b-warm
-datasets:
-- hugodk-sch/aftonposten_title_prefs
 model-index:
 - name: ap-normistral-7b-align-scan
   results: []
@@ -18,17 +16,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # ap-normistral-7b-align-scan
 
-This model is a fine-tuned version of [data/ap-normistral-7b-sft-qlora](https://huggingface.co/data/ap-normistral-7b-sft-qlora) on the hugodk-sch/aftonposten_title_prefs dataset.
+This model is a fine-tuned version of [norallm/normistral-7b-warm](https://huggingface.co/norallm/normistral-7b-warm) on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.7163
-- Rewards/chosen: -0.1039
-- Rewards/rejected: -0.1729
-- Rewards/accuracies: 0.5573
-- Rewards/margins: 0.0690
-- Logps/rejected: -36.3124
-- Logps/chosen: -32.6510
-- Logits/rejected: 98.5157
-- Logits/chosen: 98.5379
+- Loss: 0.6774
+- Rewards/chosen: -0.0728
+- Rewards/rejected: -0.1310
+- Rewards/accuracies: 0.5341
+- Rewards/margins: 0.0583
+- Logps/rejected: -36.6218
+- Logps/chosen: -32.8071
+- Logits/rejected: 98.2471
+- Logits/chosen: 98.2788
 
 ## Model description
 
@@ -63,9 +61,9 @@ 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 |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.6859        | 0.26  | 100  | 0.7229          | -0.0559        | -0.0850          | 0.5282             | 0.0290          | -36.1364       | -32.5550     | 98.6727         | 98.6826       |
-| 0.5453        | 0.52  | 200  | 0.7068          | -0.1291        | -0.2159          | 0.5332             | 0.0868          | -36.3984       | -32.7014     | 98.4732         | 98.4956       |
-| 0.5511        | 0.78  | 300  | 0.7290          | -0.1187        | -0.1591          | 0.5453             | 0.0404          | -36.2848       | -32.6805     | 98.5351         | 98.5572       |
+| 0.6551        | 0.26  | 100  | 0.6957          | -0.0035        | -0.0114          | 0.5224             | 0.0079          | -36.0235       | -32.4608     | 98.7375         | 98.7478       |
+| 0.5948        | 0.52  | 200  | 0.6720          | -0.0629        | -0.1290          | 0.5515             | 0.0661          | -36.6115       | -32.7579     | 98.3088         | 98.3381       |
+| 0.5903        | 0.78  | 300  | 0.6774          | -0.0728        | -0.1310          | 0.5341             | 0.0583          | -36.6218       | -32.8071     | 98.2471         | 98.2788       |
 
 
 ### Framework versions

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e520e2adfe8d72a46e9ed74b94ad1452a293b98cc6c277dee2b894031b0bc29a
+oid sha256:b8b3a19fd31eb8e786eae95541bfae26db68f26f062f9a217224e3be56f13830
 size 671150064

all_results.json

@@ -1,20 +1,7 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": 98.53792572021484,
-    "eval_logits/rejected": 98.51567077636719,
-    "eval_logps/chosen": -32.65095138549805,
-    "eval_logps/rejected": -36.31239318847656,
-    "eval_loss": 0.7163410782814026,
-    "eval_rewards/accuracies": 0.5573089718818665,
-    "eval_rewards/chosen": -0.10388783365488052,
-    "eval_rewards/margins": 0.06904496997594833,
-    "eval_rewards/rejected": -0.17293280363082886,
-    "eval_runtime": 103.6495,
-    "eval_samples": 343,
-    "eval_samples_per_second": 3.309,
-    "eval_steps_per_second": 0.415,
-    "train_loss": 0.5860598966672823,
-    "train_runtime": 2558.949,
+    "train_loss": 0.6168378699909557,
+    "train_runtime": 2560.1818,
     "train_samples": 3079,
     "train_samples_per_second": 1.203,
     "train_steps_per_second": 0.15

train_results.json

@@ -1,7 +1,7 @@
 {
     "epoch": 1.0,
-    "train_loss": 0.5860598966672823,
-    "train_runtime": 2558.949,
+    "train_loss": 0.6168378699909557,
+    "train_runtime": 2560.1818,
     "train_samples": 3079,
     "train_samples_per_second": 1.203,
     "train_steps_per_second": 0.15

trainer_state.json

@@ -10,7 +10,7 @@
   "log_history": [
     {
       "epoch": 0.0,
-      "grad_norm": 32.75,
+      "grad_norm": 13.0625,
       "learning_rate": 1.282051282051282e-07,
       "logits/chosen": 88.18099975585938,
       "logits/rejected": 88.25153350830078,
@@ -25,628 +25,628 @@
     },
     {
       "epoch": 0.03,
-      "grad_norm": 25.5,
+      "grad_norm": 10.6875,
       "learning_rate": 1.282051282051282e-06,
-      "logits/chosen": 81.08353424072266,
-      "logits/rejected": 80.7865219116211,
-      "logps/chosen": -34.20063781738281,
-      "logps/rejected": -33.02577590942383,
-      "loss": 0.684,
-      "rewards/accuracies": 0.5,
-      "rewards/chosen": 0.020544879138469696,
-      "rewards/margins": 0.05422975495457649,
-      "rewards/rejected": -0.033684875816106796,
+      "logits/chosen": 81.09551239013672,
+      "logits/rejected": 80.80068969726562,
+      "logps/chosen": -34.28557586669922,
+      "logps/rejected": -33.081260681152344,
+      "loss": 0.6879,
+      "rewards/accuracies": 0.5277777910232544,
+      "rewards/chosen": -0.008770154789090157,
+      "rewards/margins": 0.01580061763525009,
+      "rewards/rejected": -0.0245707705616951,
       "step": 10
     },
     {
       "epoch": 0.05,
-      "grad_norm": 25.0,
+      "grad_norm": 11.8125,
       "learning_rate": 2.564102564102564e-06,
-      "logits/chosen": 80.67228698730469,
-      "logits/rejected": 80.55956268310547,
-      "logps/chosen": -33.523399353027344,
-      "logps/rejected": -30.766698837280273,
-      "loss": 0.6867,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.0832558423280716,
-      "rewards/margins": 0.0778605192899704,
-      "rewards/rejected": 0.00539533281698823,
+      "logits/chosen": 80.68476867675781,
+      "logits/rejected": 80.57356262207031,
+      "logps/chosen": -33.56308364868164,
+      "logps/rejected": -30.81569480895996,
+      "loss": 0.6814,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.02536492981016636,
+      "rewards/margins": 0.03300638496875763,
+      "rewards/rejected": -0.007641455624252558,
       "step": 20
     },
     {
       "epoch": 0.08,
-      "grad_norm": 29.375,
+      "grad_norm": 12.4375,
       "learning_rate": 3.846153846153847e-06,
-      "logits/chosen": 82.52102661132812,
-      "logits/rejected": 82.54798889160156,
-      "logps/chosen": -33.94279861450195,
-      "logps/rejected": -31.214736938476562,
-      "loss": 0.7575,
-      "rewards/accuracies": 0.4375,
-      "rewards/chosen": 0.035010773688554764,
-      "rewards/margins": -0.06953731179237366,
-      "rewards/rejected": 0.10454808175563812,
+      "logits/chosen": 82.52334594726562,
+      "logits/rejected": 82.55731964111328,
+      "logps/chosen": -33.843528747558594,
+      "logps/rejected": -31.111019134521484,
+      "loss": 0.7142,
+      "rewards/accuracies": 0.4124999940395355,
+      "rewards/chosen": 0.03385842964053154,
+      "rewards/margins": -0.028703877702355385,
+      "rewards/rejected": 0.06256230920553207,
       "step": 30
     },
     {
       "epoch": 0.1,
-      "grad_norm": 24.25,
+      "grad_norm": 11.0,
       "learning_rate": 4.999896948438434e-06,
-      "logits/chosen": 81.06597137451172,
-      "logits/rejected": 81.06147766113281,
-      "logps/chosen": -32.76929473876953,
-      "logps/rejected": -33.212310791015625,
-      "loss": 0.6836,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.20238244533538818,
-      "rewards/margins": 0.12427721172571182,
-      "rewards/rejected": 0.07810524851083755,
+      "logits/chosen": 81.0788803100586,
+      "logits/rejected": 81.07527923583984,
+      "logps/chosen": -32.68939971923828,
+      "logps/rejected": -33.106483459472656,
+      "loss": 0.6804,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": 0.0969332829117775,
+      "rewards/margins": 0.044524990022182465,
+      "rewards/rejected": 0.05240829661488533,
       "step": 40
     },
     {
       "epoch": 0.13,
-      "grad_norm": 18.75,
+      "grad_norm": 9.25,
       "learning_rate": 4.987541037542187e-06,
-      "logits/chosen": 78.68827819824219,
-      "logits/rejected": 78.70021057128906,
-      "logps/chosen": -30.579212188720703,
-      "logps/rejected": -30.904521942138672,
-      "loss": 0.6716,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.3130193054676056,
-      "rewards/margins": 0.22863590717315674,
-      "rewards/rejected": 0.08438339829444885,
+      "logits/chosen": 78.8182373046875,
+      "logits/rejected": 78.82775115966797,
+      "logps/chosen": -30.46297836303711,
+      "logps/rejected": -30.7069091796875,
+      "loss": 0.6746,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.14845404028892517,
+      "rewards/margins": 0.07517841458320618,
+      "rewards/rejected": 0.073275625705719,
       "step": 50
     },
     {
       "epoch": 0.16,
-      "grad_norm": 25.375,
+      "grad_norm": 10.0,
       "learning_rate": 4.954691471941119e-06,
-      "logits/chosen": 83.18531799316406,
-      "logits/rejected": 83.2416763305664,
-      "logps/chosen": -30.874561309814453,
-      "logps/rejected": -29.523483276367188,
-      "loss": 0.6904,
-      "rewards/accuracies": 0.5625,
-      "rewards/chosen": 0.15030017495155334,
-      "rewards/margins": 0.11678256839513779,
-      "rewards/rejected": 0.03351757675409317,
+      "logits/chosen": 83.39994812011719,
+      "logits/rejected": 83.45416259765625,
+      "logps/chosen": -30.7622013092041,
+      "logps/rejected": -29.376235961914062,
+      "loss": 0.6858,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": 0.08259173482656479,
+      "rewards/margins": 0.03973449766635895,
+      "rewards/rejected": 0.04285724088549614,
       "step": 60
     },
     {
       "epoch": 0.18,
-      "grad_norm": 40.75,
+      "grad_norm": 14.3125,
       "learning_rate": 4.901618883413549e-06,
-      "logits/chosen": 83.8118896484375,
-      "logits/rejected": 83.84835815429688,
-      "logps/chosen": -30.521596908569336,
-      "logps/rejected": -33.058387756347656,
-      "loss": 0.7083,
-      "rewards/accuracies": 0.512499988079071,
-      "rewards/chosen": 0.05371224880218506,
-      "rewards/margins": 0.06404958665370941,
-      "rewards/rejected": -0.010337340645492077,
+      "logits/chosen": 84.03162384033203,
+      "logits/rejected": 84.06212615966797,
+      "logps/chosen": -30.389270782470703,
+      "logps/rejected": -32.93138885498047,
+      "loss": 0.6894,
+      "rewards/accuracies": 0.4625000059604645,
+      "rewards/chosen": 0.04795076698064804,
+      "rewards/margins": 0.02668655477464199,
+      "rewards/rejected": 0.02126421593129635,
       "step": 70
     },
     {
       "epoch": 0.21,
-      "grad_norm": 25.625,
+      "grad_norm": 12.0,
       "learning_rate": 4.828760511501322e-06,
-      "logits/chosen": 81.4435043334961,
-      "logits/rejected": 81.42585754394531,
-      "logps/chosen": -31.408594131469727,
-      "logps/rejected": -31.023513793945312,
-      "loss": 0.6594,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.04817419499158859,
-      "rewards/margins": 0.18110866844654083,
-      "rewards/rejected": -0.13293446600437164,
+      "logits/chosen": 81.65167236328125,
+      "logits/rejected": 81.63506317138672,
+      "logps/chosen": -31.229572296142578,
+      "logps/rejected": -30.818950653076172,
+      "loss": 0.6715,
+      "rewards/accuracies": 0.574999988079071,
+      "rewards/chosen": 0.05507441610097885,
+      "rewards/margins": 0.06733567267656326,
+      "rewards/rejected": -0.012261250987648964,
       "step": 80
     },
     {
       "epoch": 0.23,
-      "grad_norm": 31.75,
+      "grad_norm": 13.875,
       "learning_rate": 4.7367166013034295e-06,
-      "logits/chosen": 78.11305236816406,
-      "logits/rejected": 78.08197784423828,
-      "logps/chosen": -32.596641540527344,
-      "logps/rejected": -31.254547119140625,
-      "loss": 0.677,
-      "rewards/accuracies": 0.6000000238418579,
-      "rewards/chosen": 0.02076822891831398,
-      "rewards/margins": 0.17230577766895294,
-      "rewards/rejected": -0.15153756737709045,
+      "logits/chosen": 78.45601654052734,
+      "logits/rejected": 78.42174530029297,
+      "logps/chosen": -32.425289154052734,
+      "logps/rejected": -31.132177352905273,
+      "loss": 0.6671,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": 0.04257836565375328,
+      "rewards/margins": 0.07871954143047333,
+      "rewards/rejected": -0.03614116832613945,
       "step": 90
     },
     {
       "epoch": 0.26,
-      "grad_norm": 24.25,
+      "grad_norm": 11.4375,
       "learning_rate": 4.626245458345211e-06,
-      "logits/chosen": 83.37003326416016,
-      "logits/rejected": 83.38694763183594,
-      "logps/chosen": -34.16094207763672,
-      "logps/rejected": -31.8507137298584,
-      "loss": 0.6859,
+      "logits/chosen": 83.59445190429688,
+      "logits/rejected": 83.61743927001953,
+      "logps/chosen": -33.955238342285156,
+      "logps/rejected": -31.89117431640625,
+      "loss": 0.6551,
       "rewards/accuracies": 0.5874999761581421,
-      "rewards/chosen": 0.0720243752002716,
-      "rewards/margins": 0.12210811674594879,
-      "rewards/rejected": -0.05008373409509659,
+      "rewards/chosen": 0.06995107233524323,
+      "rewards/margins": 0.09807688742876053,
+      "rewards/rejected": -0.028125811368227005,
       "step": 100
     },
     {
       "epoch": 0.26,
-      "eval_logits/chosen": 98.68260955810547,
-      "eval_logits/rejected": 98.67269134521484,
-      "eval_logps/chosen": -32.555023193359375,
-      "eval_logps/rejected": -36.136436462402344,
-      "eval_loss": 0.7229137420654297,
-      "eval_rewards/accuracies": 0.5282392501831055,
-      "eval_rewards/chosen": -0.05592246353626251,
-      "eval_rewards/margins": 0.029031021520495415,
-      "eval_rewards/rejected": -0.08495348691940308,
-      "eval_runtime": 104.2675,
-      "eval_samples_per_second": 3.29,
+      "eval_logits/chosen": 98.74779510498047,
+      "eval_logits/rejected": 98.73747253417969,
+      "eval_logps/chosen": -32.46078872680664,
+      "eval_logps/rejected": -36.02348327636719,
+      "eval_loss": 0.6956869959831238,
+      "eval_rewards/accuracies": 0.5224252939224243,
+      "eval_rewards/chosen": -0.003522348590195179,
+      "eval_rewards/margins": 0.00786901917308569,
+      "eval_rewards/rejected": -0.011391367763280869,
+      "eval_runtime": 104.3728,
+      "eval_samples_per_second": 3.286,
       "eval_steps_per_second": 0.412,
       "step": 100
     },
     {
       "epoch": 0.29,
-      "grad_norm": 35.25,
+      "grad_norm": 14.0,
       "learning_rate": 4.498257201263691e-06,
-      "logits/chosen": 83.4867172241211,
-      "logits/rejected": 83.372802734375,
-      "logps/chosen": -32.426979064941406,
-      "logps/rejected": -32.882118225097656,
-      "loss": 0.5808,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.2979138493537903,
-      "rewards/margins": 0.4110265374183655,
-      "rewards/rejected": -0.11311274766921997,
+      "logits/chosen": 83.79670715332031,
+      "logits/rejected": 83.66975402832031,
+      "logps/chosen": -32.42261505126953,
+      "logps/rejected": -32.766204833984375,
+      "loss": 0.6376,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.12003795802593231,
+      "rewards/margins": 0.1421007364988327,
+      "rewards/rejected": -0.02206278033554554,
       "step": 110
     },
     {
       "epoch": 0.31,
-      "grad_norm": 38.5,
+      "grad_norm": 13.4375,
       "learning_rate": 4.353806263777678e-06,
-      "logits/chosen": 83.65151977539062,
-      "logits/rejected": 83.7606201171875,
-      "logps/chosen": -28.169925689697266,
-      "logps/rejected": -35.475196838378906,
-      "loss": 0.6054,
-      "rewards/accuracies": 0.612500011920929,
-      "rewards/chosen": 0.37982815504074097,
-      "rewards/margins": 0.3878515660762787,
-      "rewards/rejected": -0.008023401722311974,
+      "logits/chosen": 83.85148620605469,
+      "logits/rejected": 83.95706939697266,
+      "logps/chosen": -28.113927841186523,
+      "logps/rejected": -35.665374755859375,
+      "loss": 0.6109,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.1631302535533905,
+      "rewards/margins": 0.20437581837177277,
+      "rewards/rejected": -0.04124556854367256,
       "step": 120
     },
     {
       "epoch": 0.34,
-      "grad_norm": 24.0,
+      "grad_norm": 9.5,
       "learning_rate": 4.1940827077152755e-06,
-      "logits/chosen": 80.93061065673828,
-      "logits/rejected": 80.95037841796875,
-      "logps/chosen": -30.428030014038086,
-      "logps/rejected": -32.12748336791992,
-      "loss": 0.6058,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.23966185748577118,
-      "rewards/margins": 0.33758825063705444,
-      "rewards/rejected": -0.09792639315128326,
+      "logits/chosen": 80.99261474609375,
+      "logits/rejected": 81.01200866699219,
+      "logps/chosen": -30.389972686767578,
+      "logps/rejected": -32.014381408691406,
+      "loss": 0.6511,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.10347676277160645,
+      "rewards/margins": 0.120028056204319,
+      "rewards/rejected": -0.0165513064712286,
       "step": 130
     },
     {
       "epoch": 0.36,
-      "grad_norm": 17.75,
+      "grad_norm": 9.3125,
       "learning_rate": 4.0204024186666215e-06,
-      "logits/chosen": 82.01722717285156,
-      "logits/rejected": 82.02776336669922,
-      "logps/chosen": -27.013805389404297,
-      "logps/rejected": -33.108734130859375,
-      "loss": 0.5348,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.2167574167251587,
-      "rewards/margins": 0.5594369173049927,
-      "rewards/rejected": -0.3426794409751892,
+      "logits/chosen": 82.00664520263672,
+      "logits/rejected": 82.01942443847656,
+      "logps/chosen": -27.06869125366211,
+      "logps/rejected": -33.10210418701172,
+      "loss": 0.611,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.07572650164365768,
+      "rewards/margins": 0.21147167682647705,
+      "rewards/rejected": -0.13574519753456116,
       "step": 140
     },
     {
       "epoch": 0.39,
-      "grad_norm": 26.125,
+      "grad_norm": 10.0625,
       "learning_rate": 3.834196265035119e-06,
-      "logits/chosen": 80.49574279785156,
-      "logits/rejected": 80.46543884277344,
-      "logps/chosen": -28.96371078491211,
-      "logps/rejected": -33.28276443481445,
-      "loss": 0.5393,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.21269837021827698,
-      "rewards/margins": 0.5708296895027161,
-      "rewards/rejected": -0.3581313192844391,
+      "logits/chosen": 80.3797378540039,
+      "logits/rejected": 80.35098266601562,
+      "logps/chosen": -28.888397216796875,
+      "logps/rejected": -33.30742645263672,
+      "loss": 0.596,
+      "rewards/accuracies": 0.7250000238418579,
+      "rewards/chosen": 0.10014281421899796,
+      "rewards/margins": 0.24832835793495178,
+      "rewards/rejected": -0.14818556606769562,
       "step": 150
     },
     {
       "epoch": 0.42,
-      "grad_norm": 36.75,
+      "grad_norm": 13.9375,
       "learning_rate": 3.636998309800573e-06,
-      "logits/chosen": 82.2945556640625,
-      "logits/rejected": 82.3069839477539,
-      "logps/chosen": -33.60367202758789,
-      "logps/rejected": -30.597631454467773,
-      "loss": 0.5835,
-      "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.266539990901947,
-      "rewards/margins": 0.5890872478485107,
-      "rewards/rejected": -0.3225473165512085,
+      "logits/chosen": 82.10334014892578,
+      "logits/rejected": 82.10926818847656,
+      "logps/chosen": -33.88823699951172,
+      "logps/rejected": -30.74798011779785,
+      "loss": 0.6209,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.04970277473330498,
+      "rewards/margins": 0.2087908238172531,
+      "rewards/rejected": -0.15908804535865784,
       "step": 160
     },
     {
       "epoch": 0.44,
-      "grad_norm": 26.25,
+      "grad_norm": 12.4375,
       "learning_rate": 3.4304331721118078e-06,
-      "logits/chosen": 83.09800720214844,
-      "logits/rejected": 83.04072570800781,
-      "logps/chosen": -30.741390228271484,
-      "logps/rejected": -32.57554626464844,
-      "loss": 0.5698,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.25894853472709656,
-      "rewards/margins": 0.5366927981376648,
-      "rewards/rejected": -0.27774426341056824,
+      "logits/chosen": 82.82877349853516,
+      "logits/rejected": 82.77672576904297,
+      "logps/chosen": -30.899311065673828,
+      "logps/rejected": -32.958866119384766,
+      "loss": 0.598,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.0719953253865242,
+      "rewards/margins": 0.2597564160823822,
+      "rewards/rejected": -0.1877611130475998,
       "step": 170
     },
     {
       "epoch": 0.47,
-      "grad_norm": 23.625,
+      "grad_norm": 8.75,
       "learning_rate": 3.2162026428305436e-06,
-      "logits/chosen": 80.62547302246094,
-      "logits/rejected": 80.6036605834961,
-      "logps/chosen": -30.658870697021484,
-      "logps/rejected": -31.662927627563477,
-      "loss": 0.55,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.2691388726234436,
-      "rewards/margins": 0.5219991207122803,
-      "rewards/rejected": -0.25286024808883667,
+      "logits/chosen": 80.19332885742188,
+      "logits/rejected": 80.17335510253906,
+      "logps/chosen": -30.60544204711914,
+      "logps/rejected": -31.953319549560547,
+      "loss": 0.5849,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.1183418408036232,
+      "rewards/margins": 0.2775646448135376,
+      "rewards/rejected": -0.159222811460495,
       "step": 180
     },
     {
       "epoch": 0.49,
-      "grad_norm": 12.4375,
+      "grad_norm": 6.59375,
       "learning_rate": 2.996071664294641e-06,
-      "logits/chosen": 82.23011779785156,
-      "logits/rejected": 82.23200225830078,
-      "logps/chosen": -30.246551513671875,
-      "logps/rejected": -30.743694305419922,
-      "loss": 0.6343,
-      "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.25993233919143677,
-      "rewards/margins": 0.3977679908275604,
-      "rewards/rejected": -0.13783565163612366,
+      "logits/chosen": 81.75331115722656,
+      "logits/rejected": 81.74398040771484,
+      "logps/chosen": -30.514236450195312,
+      "logps/rejected": -30.920120239257812,
+      "loss": 0.6564,
+      "rewards/accuracies": 0.625,
+      "rewards/chosen": 0.050435226410627365,
+      "rewards/margins": 0.14085446298122406,
+      "rewards/rejected": -0.090419240295887,
       "step": 190
     },
     {
       "epoch": 0.52,
-      "grad_norm": 14.625,
+      "grad_norm": 8.9375,
       "learning_rate": 2.7718537898066833e-06,
-      "logits/chosen": 77.68287658691406,
-      "logits/rejected": 77.6307144165039,
-      "logps/chosen": -33.85746383666992,
-      "logps/rejected": -32.90498352050781,
-      "loss": 0.5453,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.44763606786727905,
-      "rewards/margins": 0.6534249186515808,
-      "rewards/rejected": -0.20578882098197937,
+      "logits/chosen": 77.08003997802734,
+      "logits/rejected": 77.01986694335938,
+      "logps/chosen": -33.93625259399414,
+      "logps/rejected": -33.12162780761719,
+      "loss": 0.5948,
+      "rewards/accuracies": 0.6875,
+      "rewards/chosen": 0.16329774260520935,
+      "rewards/margins": 0.28894227743148804,
+      "rewards/rejected": -0.1256445348262787,
       "step": 200
     },
     {
       "epoch": 0.52,
-      "eval_logits/chosen": 98.49555206298828,
-      "eval_logits/rejected": 98.47319030761719,
-      "eval_logps/chosen": -32.7014274597168,
-      "eval_logps/rejected": -36.39837646484375,
-      "eval_loss": 0.7067598700523376,
-      "eval_rewards/accuracies": 0.5332226157188416,
-      "eval_rewards/chosen": -0.12912562489509583,
-      "eval_rewards/margins": 0.0867983028292656,
-      "eval_rewards/rejected": -0.21592393517494202,
-      "eval_runtime": 103.9124,
-      "eval_samples_per_second": 3.301,
-      "eval_steps_per_second": 0.414,
+      "eval_logits/chosen": 98.33805847167969,
+      "eval_logits/rejected": 98.30875396728516,
+      "eval_logps/chosen": -32.75787353515625,
+      "eval_logps/rejected": -36.611534118652344,
+      "eval_loss": 0.672012209892273,
+      "eval_rewards/accuracies": 0.5514950156211853,
+      "eval_rewards/chosen": -0.06293957680463791,
+      "eval_rewards/margins": 0.06606128066778183,
+      "eval_rewards/rejected": -0.12900085747241974,
+      "eval_runtime": 104.1536,
+      "eval_samples_per_second": 3.293,
+      "eval_steps_per_second": 0.413,
       "step": 200
     },
     {
       "epoch": 0.55,
-      "grad_norm": 46.5,
+      "grad_norm": 15.0625,
       "learning_rate": 2.5453962426402006e-06,
-      "logits/chosen": 80.28144836425781,
-      "logits/rejected": 80.18412780761719,
-      "logps/chosen": -33.1484489440918,
-      "logps/rejected": -35.34803009033203,
-      "loss": 0.5401,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.39863476157188416,
-      "rewards/margins": 0.5796052813529968,
-      "rewards/rejected": -0.18097054958343506,
+      "logits/chosen": 79.71280670166016,
+      "logits/rejected": 79.62548065185547,
+      "logps/chosen": -33.420021057128906,
+      "logps/rejected": -35.76679229736328,
+      "loss": 0.6026,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.10514093935489655,
+      "rewards/margins": 0.2612799406051636,
+      "rewards/rejected": -0.15613897144794464,
       "step": 210
     },
     {
       "epoch": 0.57,
-      "grad_norm": 18.875,
+      "grad_norm": 10.375,
       "learning_rate": 2.3185646976551794e-06,
-      "logits/chosen": 82.44231414794922,
-      "logits/rejected": 82.5237808227539,
-      "logps/chosen": -31.035247802734375,
-      "logps/rejected": -31.218652725219727,
-      "loss": 0.5206,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.45685210824012756,
-      "rewards/margins": 0.7017010450363159,
-      "rewards/rejected": -0.24484892189502716,
+      "logits/chosen": 81.73656463623047,
+      "logits/rejected": 81.82989501953125,
+      "logps/chosen": -31.172740936279297,
+      "logps/rejected": -31.53964614868164,
+      "loss": 0.5726,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": 0.15524254739284515,
+      "rewards/margins": 0.3173803985118866,
+      "rewards/rejected": -0.16213785111904144,
       "step": 220
     },
     {
       "epoch": 0.6,
-      "grad_norm": 27.5,
+      "grad_norm": 11.5625,
       "learning_rate": 2.0932279108998323e-06,
-      "logits/chosen": 79.57551574707031,
-      "logits/rejected": 79.63117218017578,
-      "logps/chosen": -32.342308044433594,
-      "logps/rejected": -34.377655029296875,
-      "loss": 0.6337,
+      "logits/chosen": 78.81214904785156,
+      "logits/rejected": 78.8668212890625,
+      "logps/chosen": -32.4487190246582,
+      "logps/rejected": -34.767303466796875,
+      "loss": 0.6172,
       "rewards/accuracies": 0.637499988079071,
-      "rewards/chosen": 0.23171910643577576,
-      "rewards/margins": 0.4144717752933502,
-      "rewards/rejected": -0.18275271356105804,
+      "rewards/chosen": 0.07140493392944336,
+      "rewards/margins": 0.22243526577949524,
+      "rewards/rejected": -0.15103033185005188,
       "step": 230
     },
     {
       "epoch": 0.62,
-      "grad_norm": 29.375,
+      "grad_norm": 14.25,
       "learning_rate": 1.8712423238279358e-06,
-      "logits/chosen": 81.96827697753906,
-      "logits/rejected": 82.25470733642578,
-      "logps/chosen": -30.665319442749023,
-      "logps/rejected": -31.84256362915039,
-      "loss": 0.5035,
+      "logits/chosen": 81.07857513427734,
+      "logits/rejected": 81.38665771484375,
+      "logps/chosen": -31.030254364013672,
+      "logps/rejected": -32.145328521728516,
+      "loss": 0.5977,
       "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.5113734006881714,
-      "rewards/margins": 0.6666342616081238,
-      "rewards/rejected": -0.1552608758211136,
+      "rewards/chosen": 0.13156220316886902,
+      "rewards/margins": 0.254219114780426,
+      "rewards/rejected": -0.12265688180923462,
       "step": 240
     },
     {
       "epoch": 0.65,
-      "grad_norm": 27.125,
+      "grad_norm": 12.75,
       "learning_rate": 1.6544367689701824e-06,
-      "logits/chosen": 80.69163513183594,
-      "logits/rejected": 80.74885559082031,
-      "logps/chosen": -26.948383331298828,
-      "logps/rejected": -30.213348388671875,
-      "loss": 0.6037,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.32375770807266235,
-      "rewards/margins": 0.48006218671798706,
-      "rewards/rejected": -0.1563045233488083,
+      "logits/chosen": 79.71233367919922,
+      "logits/rejected": 79.77317810058594,
+      "logps/chosen": -27.3262939453125,
+      "logps/rejected": -30.594249725341797,
+      "loss": 0.6333,
+      "rewards/accuracies": 0.5625,
+      "rewards/chosen": 0.053920842707157135,
+      "rewards/margins": 0.19262339174747467,
+      "rewards/rejected": -0.13870255649089813,
       "step": 250
     },
     {
       "epoch": 0.68,
-      "grad_norm": 27.75,
+      "grad_norm": 10.5,
       "learning_rate": 1.4445974030621963e-06,
-      "logits/chosen": 77.91614532470703,
-      "logits/rejected": 78.04708099365234,
-      "logps/chosen": -30.49196434020996,
-      "logps/rejected": -36.51909255981445,
-      "loss": 0.4936,
-      "rewards/accuracies": 0.7250000238418579,
-      "rewards/chosen": 0.5345155000686646,
-      "rewards/margins": 0.7432474493980408,
-      "rewards/rejected": -0.2087319791316986,
+      "logits/chosen": 76.8447036743164,
+      "logits/rejected": 76.98011779785156,
+      "logps/chosen": -30.586124420166016,
+      "logps/rejected": -37.12653350830078,
+      "loss": 0.5519,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.19497378170490265,
+      "rewards/margins": 0.3999541699886322,
+      "rewards/rejected": -0.20498037338256836,
       "step": 260
     },
     {
       "epoch": 0.7,
-      "grad_norm": 16.75,
+      "grad_norm": 9.3125,
       "learning_rate": 1.243452991757889e-06,
-      "logits/chosen": 77.25732421875,
-      "logits/rejected": 77.28456115722656,
-      "logps/chosen": -30.877111434936523,
-      "logps/rejected": -31.975589752197266,
-      "loss": 0.5167,
-      "rewards/accuracies": 0.737500011920929,
-      "rewards/chosen": 0.4509071707725525,
-      "rewards/margins": 0.685990035533905,
-      "rewards/rejected": -0.23508290946483612,
+      "logits/chosen": 76.1262435913086,
+      "logits/rejected": 76.15739440917969,
+      "logps/chosen": -31.1810245513916,
+      "logps/rejected": -32.34056854248047,
+      "loss": 0.5874,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.11957981437444687,
+      "rewards/margins": 0.28660956025123596,
+      "rewards/rejected": -0.1670297235250473,
       "step": 270
     },
     {
       "epoch": 0.73,
-      "grad_norm": 25.125,
+      "grad_norm": 15.75,
       "learning_rate": 1.0526606671603523e-06,
-      "logits/chosen": 80.03535461425781,
-      "logits/rejected": 79.8197250366211,
-      "logps/chosen": -31.083843231201172,
-      "logps/rejected": -29.844961166381836,
-      "loss": 0.6198,
-      "rewards/accuracies": 0.675000011920929,
-      "rewards/chosen": 0.3630252182483673,
-      "rewards/margins": 0.45200806856155396,
-      "rewards/rejected": -0.08898283541202545,
+      "logits/chosen": 78.95848846435547,
+      "logits/rejected": 78.7259750366211,
+      "logps/chosen": -31.400470733642578,
+      "logps/rejected": -30.376489639282227,
+      "loss": 0.6251,
+      "rewards/accuracies": 0.7749999761581421,
+      "rewards/chosen": 0.08188416808843613,
+      "rewards/margins": 0.22378277778625488,
+      "rewards/rejected": -0.14189860224723816,
       "step": 280
     },
     {
       "epoch": 0.75,
-      "grad_norm": 16.625,
+      "grad_norm": 9.3125,
       "learning_rate": 8.737922755071455e-07,
-      "logits/chosen": 80.13519287109375,
-      "logits/rejected": 80.05864715576172,
-      "logps/chosen": -33.047523498535156,
-      "logps/rejected": -32.68352127075195,
-      "loss": 0.4787,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.5283600091934204,
-      "rewards/margins": 0.8616873025894165,
-      "rewards/rejected": -0.3333272337913513,
+      "logits/chosen": 79.04127502441406,
+      "logits/rejected": 78.95845031738281,
+      "logps/chosen": -33.34519577026367,
+      "logps/rejected": -33.35126876831055,
+      "loss": 0.5388,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": 0.15181000530719757,
+      "rewards/margins": 0.4186907410621643,
+      "rewards/rejected": -0.26688069105148315,
       "step": 290
     },
     {
       "epoch": 0.78,
-      "grad_norm": 26.875,
+      "grad_norm": 12.8125,
       "learning_rate": 7.08321427484816e-07,
-      "logits/chosen": 75.82341003417969,
-      "logits/rejected": 75.91938781738281,
-      "logps/chosen": -32.24137878417969,
-      "logps/rejected": -29.33247947692871,
-      "loss": 0.5511,
-      "rewards/accuracies": 0.6625000238418579,
-      "rewards/chosen": 0.5821641087532043,
-      "rewards/margins": 0.6973918676376343,
-      "rewards/rejected": -0.11522769927978516,
+      "logits/chosen": 74.71063995361328,
+      "logits/rejected": 74.8211441040039,
+      "logps/chosen": -32.499046325683594,
+      "logps/rejected": -29.784500122070312,
+      "loss": 0.5903,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": 0.18133236467838287,
+      "rewards/margins": 0.3178277611732483,
+      "rewards/rejected": -0.13649536669254303,
       "step": 300
     },
     {
       "epoch": 0.78,
-      "eval_logits/chosen": 98.5571517944336,
-      "eval_logits/rejected": 98.53510284423828,
-      "eval_logps/chosen": -32.68053436279297,
-      "eval_logps/rejected": -36.284751892089844,
-      "eval_loss": 0.7289925813674927,
-      "eval_rewards/accuracies": 0.545265793800354,
-      "eval_rewards/chosen": -0.11868056654930115,
-      "eval_rewards/margins": 0.04043150320649147,
-      "eval_rewards/rejected": -0.1591120809316635,
-      "eval_runtime": 104.072,
-      "eval_samples_per_second": 3.296,
+      "eval_logits/chosen": 98.2787857055664,
+      "eval_logits/rejected": 98.24710083007812,
+      "eval_logps/chosen": -32.80708312988281,
+      "eval_logps/rejected": -36.62175750732422,
+      "eval_loss": 0.6774110198020935,
+      "eval_rewards/accuracies": 0.5340532064437866,
+      "eval_rewards/chosen": -0.07278140634298325,
+      "eval_rewards/margins": 0.05826449766755104,
+      "eval_rewards/rejected": -0.13104592263698578,
+      "eval_runtime": 104.1558,
+      "eval_samples_per_second": 3.293,
       "eval_steps_per_second": 0.413,
       "step": 300
     },
     {
       "epoch": 0.81,
-      "grad_norm": 21.625,
+      "grad_norm": 10.0,
       "learning_rate": 5.576113578589035e-07,
-      "logits/chosen": 82.95035552978516,
-      "logits/rejected": 82.97549438476562,
-      "logps/chosen": -30.2395076751709,
-      "logps/rejected": -32.58390808105469,
-      "loss": 0.5895,
+      "logits/chosen": 81.97877502441406,
+      "logits/rejected": 82.00166320800781,
+      "logps/chosen": -30.284442901611328,
+      "logps/rejected": -33.162872314453125,
+      "loss": 0.579,
       "rewards/accuracies": 0.7124999761581421,
-      "rewards/chosen": 0.31201133131980896,
-      "rewards/margins": 0.5040835738182068,
-      "rewards/rejected": -0.19207225739955902,
+      "rewards/chosen": 0.11581814289093018,
+      "rewards/margins": 0.30844008922576904,
+      "rewards/rejected": -0.19262196123600006,
       "step": 310
     },
     {
       "epoch": 0.83,
-      "grad_norm": 18.0,
+      "grad_norm": 9.8125,
       "learning_rate": 4.229036944380913e-07,
-      "logits/chosen": 80.41062927246094,
-      "logits/rejected": 80.41141510009766,
-      "logps/chosen": -30.43392562866211,
-      "logps/rejected": -29.273609161376953,
-      "loss": 0.4816,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.5823162794113159,
-      "rewards/margins": 0.7593742609024048,
-      "rewards/rejected": -0.1770579218864441,
+      "logits/chosen": 79.28789520263672,
+      "logits/rejected": 79.29541778564453,
+      "logps/chosen": -30.684707641601562,
+      "logps/rejected": -29.662694931030273,
+      "loss": 0.5734,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.1827705204486847,
+      "rewards/margins": 0.3314109444618225,
+      "rewards/rejected": -0.14864039421081543,
       "step": 320
     },
     {
       "epoch": 0.86,
-      "grad_norm": 16.625,
+      "grad_norm": 9.4375,
       "learning_rate": 3.053082288996112e-07,
-      "logits/chosen": 77.58735656738281,
-      "logits/rejected": 77.6371841430664,
-      "logps/chosen": -29.118694305419922,
-      "logps/rejected": -33.023460388183594,
-      "loss": 0.4851,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.6169556975364685,
-      "rewards/margins": 0.8112555742263794,
-      "rewards/rejected": -0.19429990649223328,
+      "logits/chosen": 76.40345764160156,
+      "logits/rejected": 76.45513916015625,
+      "logps/chosen": -29.180185317993164,
+      "logps/rejected": -33.6518440246582,
+      "loss": 0.5234,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": 0.2344847470521927,
+      "rewards/margins": 0.43788036704063416,
+      "rewards/rejected": -0.20339563488960266,
       "step": 330
     },
     {
       "epoch": 0.88,
-      "grad_norm": 33.5,
+      "grad_norm": 16.875,
       "learning_rate": 2.0579377374915805e-07,
-      "logits/chosen": 81.87162780761719,
-      "logits/rejected": 81.90576171875,
-      "logps/chosen": -32.27037048339844,
-      "logps/rejected": -33.92435836791992,
-      "loss": 0.5157,
-      "rewards/accuracies": 0.75,
-      "rewards/chosen": 0.4708884358406067,
-      "rewards/margins": 0.7487217783927917,
-      "rewards/rejected": -0.27783337235450745,
+      "logits/chosen": 80.8781509399414,
+      "logits/rejected": 80.92503356933594,
+      "logps/chosen": -32.53740310668945,
+      "logps/rejected": -34.52736282348633,
+      "loss": 0.5705,
+      "rewards/accuracies": 0.7124999761581421,
+      "rewards/chosen": 0.13494810461997986,
+      "rewards/margins": 0.3666823208332062,
+      "rewards/rejected": -0.23173420131206512,
       "step": 340
     },
     {
       "epoch": 0.91,
-      "grad_norm": 13.0,
+      "grad_norm": 9.375,
       "learning_rate": 1.2518018074041684e-07,
-      "logits/chosen": 80.89490509033203,
-      "logits/rejected": 80.90391540527344,
-      "logps/chosen": -32.372520446777344,
-      "logps/rejected": -33.53419876098633,
-      "loss": 0.4983,
-      "rewards/accuracies": 0.762499988079071,
-      "rewards/chosen": 0.6356090307235718,
-      "rewards/margins": 0.8568087816238403,
-      "rewards/rejected": -0.22119958698749542,
+      "logits/chosen": 79.81520080566406,
+      "logits/rejected": 79.83287048339844,
+      "logps/chosen": -32.53202819824219,
+      "logps/rejected": -33.971153259277344,
+      "loss": 0.561,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.22234222292900085,
+      "rewards/margins": 0.39821282029151917,
+      "rewards/rejected": -0.1758706122636795,
       "step": 350
     },
     {
       "epoch": 0.94,
-      "grad_norm": 20.125,
+      "grad_norm": 9.25,
       "learning_rate": 6.41315865106129e-08,
-      "logits/chosen": 82.41497802734375,
-      "logits/rejected": 82.44314575195312,
-      "logps/chosen": -28.537517547607422,
-      "logps/rejected": -31.826416015625,
-      "loss": 0.5313,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.5073467493057251,
-      "rewards/margins": 0.5956294536590576,
-      "rewards/rejected": -0.08828280866146088,
+      "logits/chosen": 81.41835021972656,
+      "logits/rejected": 81.45225524902344,
+      "logps/chosen": -28.548419952392578,
+      "logps/rejected": -32.2237434387207,
+      "loss": 0.5819,
+      "rewards/accuracies": 0.675000011920929,
+      "rewards/chosen": 0.20075836777687073,
+      "rewards/margins": 0.3155379891395569,
+      "rewards/rejected": -0.11477959156036377,
       "step": 360
     },
     {
       "epoch": 0.96,
-      "grad_norm": 21.75,
+      "grad_norm": 12.4375,
       "learning_rate": 2.3150941078050325e-08,
-      "logits/chosen": 81.85018920898438,
-      "logits/rejected": 81.87150573730469,
-      "logps/chosen": -31.847997665405273,
-      "logps/rejected": -35.508277893066406,
-      "loss": 0.5486,
-      "rewards/accuracies": 0.6875,
-      "rewards/chosen": 0.5140814781188965,
-      "rewards/margins": 0.7135335206985474,
-      "rewards/rejected": -0.19945205748081207,
+      "logits/chosen": 80.87324523925781,
+      "logits/rejected": 80.88945007324219,
+      "logps/chosen": -32.29994583129883,
+      "logps/rejected": -36.0130615234375,
+      "loss": 0.6003,
+      "rewards/accuracies": 0.6625000238418579,
+      "rewards/chosen": 0.1152433529496193,
+      "rewards/margins": 0.295980840921402,
+      "rewards/rejected": -0.1807374656200409,
       "step": 370
     },
     {
       "epoch": 0.99,
-      "grad_norm": 25.5,
+      "grad_norm": 11.1875,
       "learning_rate": 2.575864278703266e-09,
-      "logits/chosen": 75.74685668945312,
-      "logits/rejected": 75.61808013916016,
-      "logps/chosen": -29.746694564819336,
-      "logps/rejected": -28.432201385498047,
-      "loss": 0.5715,
-      "rewards/accuracies": 0.699999988079071,
-      "rewards/chosen": 0.4147160053253174,
-      "rewards/margins": 0.5505291223526001,
-      "rewards/rejected": -0.1358131319284439,
+      "logits/chosen": 74.64203643798828,
+      "logits/rejected": 74.50636291503906,
+      "logps/chosen": -29.965755462646484,
+      "logps/rejected": -28.94148826599121,
+      "loss": 0.5896,
+      "rewards/accuracies": 0.737500011920929,
+      "rewards/chosen": 0.12207402288913727,
+      "rewards/margins": 0.2782566249370575,
+      "rewards/rejected": -0.15618260204792023,
       "step": 380
     },
     {
       "epoch": 1.0,
       "step": 385,
       "total_flos": 0.0,
-      "train_loss": 0.5860598966672823,
-      "train_runtime": 2558.949,
+      "train_loss": 0.6168378699909557,
+      "train_runtime": 2560.1818,
       "train_samples_per_second": 1.203,
       "train_steps_per_second": 0.15
     }