Upload folder using huggingface_hub

README.md

@@ -0,0 +1,152 @@
+---
+license: apache-2.0
+base_model: mistral-community/Mixtral-8x22B-v0.1
+tags:
+- trl
+- orpo
+- generated_from_trainer
+datasets:
+- argilla/distilabel-capybara-dpo-7k-binarized
+model-index:
+- name: zephyr-orpo-141b-A35b-v0.1
+  results: []
+---
+
+<img src="https://huggingface.co/HuggingFaceH4/zephyr-orpo-141b-A35b-v0.1/resolve/main/logo.png" alt="Zephyr 141B Logo" width="400" style="margin-left:'auto' margin-right:'auto' display:'block'"/>
+
+
+# Model Card for Zephyr 141B-A35B
+
+Zephyr is a series of language models that are trained to act as helpful assistants. Zephyr 141B-A35B is the latest model in the series, and is a fine-tuned version of [mistral-community/Mixtral-8x22B-v0.1](https://huggingface.co/mistral-community/Mixtral-8x22B-v0.1) that was trained using a novel alignment algorithm called [Odds Ratio Preference Optimization (ORPO)](https://huggingface.co/papers/2403.07691) with **7k instances** for **1.3 hours** on 4 nodes of 8 x H100s. ORPO does not require an SFT step to achieve high performance and is thus much more computationally efficient than methods like DPO and PPO. To train Zephyr-141B-A35B, we used the [`argilla/distilabel-capybara-dpo-7k-binarized`](https://huggingface.co/datasets/argilla/distilabel-capybara-dpo-7k-binarized) preference dataset, which consists of synthetic, high-quality, multi-turn preferences that have been scored via LLMs.
+
+> [!NOTE]
+> This model was trained collaboratively between Argilla, KAIST, and Hugging Face
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+- **Model type:** A Mixture of Experts (MoE) model with 141B total parameters and 35B active parameters. Fine-tuned on a mix of publicly available, synthetic datasets.
+- **Language(s) (NLP):** Primarily English.
+- **License:** Apache 2.0
+- **Finetuned from model:** [mistral-community/Mixtral-8x22B-v0.1](https://huggingface.co/mistral-community/Mixtral-8x22B-v0.1)
+
+### Model Sources
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** https://github.com/huggingface/alignment-handbook
+- **Dataset:** https://huggingface.co/datasets/argilla/distilabel-capybara-dpo-7k-binarized
+
+## Performance
+
+Zephyr 141B-A35B was trained to test the effectiveness of ORPO at scale and the underlying dataset contains a mix of general chat capabilities. It achieves strong performance on chat benchmarks like [MT Bench](https://huggingface.co/spaces/lmsys/mt-bench) and [IFEval](https://arxiv.org/abs/2311.07911). The scores reported below were obtained using the [LightEval](https://github.com/huggingface/lighteval) evaluation suite and each prompt has been formatted with the model's corresponding chat template to simulate real-world usage. This is why some scores may differ from those reported in technical reports or on the Open LLM Leaderboard. 
+
+| Model                                                                                               | MT Bench | IFEval |   BBH | AGIEval |
+|-----------------------------------------------------------------------------------------------------|---------:|-------:|------:|--------:|
+| [zephyr-orpo-141b-A35b-v0.1](https://huggingface.co/HuggingFaceH4/zephyr-orpo-141b-A35b-v0.1)       |     8.17 |  65.06 | 58.96 |   44.16 |
+| [databricks/dbrx-instruct](https://huggingface.co/databricks/dbrx-instruct)                         |     8.26 |  52.13 | 48.50 |   41.16 |
+| [mistralai/Mixtral-8x7B-Instruct-v0.1](https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1) |     8.30 |  55.08 | 45.31 |   47.68 |
+
+
+## Intended uses & limitations
+
+The model was fine-tuned on a blend of chat, code, math, and reasoning data. Here's how you can run the model using the `pipeline()` function from 🤗 Transformers:
+
+```python
+# pip install 'transformers>=4.39.3'
+# pip install accelerate
+
+import torch
+from transformers import pipeline
+
+pipe = pipeline(
+    "text-generation",
+    model="HuggingFaceH4/zephyr-orpo-141b-A35b-v0.1",
+    device_map="auto",
+    torch_dtype=torch.bfloat16,
+)
+messages = [
+    {
+        "role": "system",
+        "content": "You are Zephyr, a helpful assistant.",
+    },
+    {"role": "user", "content": "Explain how Mixture of Experts work in language a child would understand."},
+]
+outputs = pipe(
+    messages,
+    max_new_tokens=512,
+    do_sample=True,
+    temperature=0.7,
+    top_k=50,
+    top_p=0.95,
+)
+print(outputs[0]["generated_text"][-1]["content"])
+```
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+Zephyr 141B-A35B has not been aligned to human preferences for safety within the RLHF phase or deployed with in-the-loop filtering of responses like ChatGPT, so the model can produce problematic outputs (especially when prompted to do so). 
+It is also unknown what the size and composition of the corpus was used to train the base model (`mistral-community/Mixtral-8x22B-v0.1`), however it is likely to have included a mix of Web data and technical sources like books and code. See the [Falcon 180B model card](https://huggingface.co/tiiuae/falcon-180B#training-data) for an example of this.
+
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 5e-06
+- train_batch_size: 1
+- eval_batch_size: 8
+- seed: 42
+- distributed_type: multi-GPU
+- num_devices: 32
+- total_train_batch_size: 32
+- total_eval_batch_size: 256
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: inverse_sqrt
+- lr_scheduler_warmup_steps: 100
+- num_epochs: 3
+
+### Training results
+
+
+
+### Framework versions
+
+- Transformers 4.39.3
+- Pytorch 2.1.2+cu121
+- Datasets 2.18.0
+- Tokenizers 0.15.1
+
+## Citation
+
+If you find Zephyr 141B-A35B is useful in your work, please cite the ORPO paper:
+
+```
+@misc{hong2024orpo,
+      title={ORPO: Monolithic Preference Optimization without Reference Model}, 
+      author={Jiwoo Hong and Noah Lee and James Thorne},
+      year={2024},
+      eprint={2403.07691},
+      archivePrefix={arXiv},
+      primaryClass={cs.CL}
+}
+```
+
+You may also wish to cite the creators of this model:
+
+```
+@misc{zephyr_141b,
+  author = {Alvaro Bartolome and Jiwoo Hong and Noah Lee and Kashif Rasul and Lewis Tunstall},
+  title = {Zephyr 141B A35B},
+  year = {2024},
+  publisher = {Hugging Face},
+  journal = {Hugging Face repository},
+  howpublished = {\url{https://huggingface.co/HuggingFaceH4/zephyr-orpo-141b-A35b-v0.1}}
+}
+```

all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 3.0,
+    "train_loss": 0.812556631554107,
+    "train_runtime": 4771.9621,
+    "train_samples": 6932,
+    "train_samples_per_second": 4.358,
+    "train_steps_per_second": 0.136
+}

config.json

@@ -0,0 +1,42 @@
+{
+    "_name_or_path": "mistral-community/Mixtral-8x22B-v0.1",
+    "architectures": [
+        "MixtralForCausalLM"
+    ],
+    "attention_dropout": 0.0,
+    "bos_token_id": 1,
+    "eos_token_id": 2,
+    "hidden_act": "silu",
+    "hidden_size": 6144,
+    "initializer_range": 0.02,
+    "intermediate_size": 16384,
+    "max_position_embeddings": 65536,
+    "model_type": "mixtral",
+    "num_attention_heads": 48,
+    "num_experts_per_tok": 2,
+    "num_hidden_layers": 56,
+    "num_key_value_heads": 8,
+    "num_local_experts": 8,
+    "output_router_logits": false,
+    "rms_norm_eps": 1e-05,
+    "rope_theta": 1000000,
+    "router_aux_loss_coef": 0.001,
+    "router_jitter_noise": 0.0,
+    "sliding_window": null,
+    "tie_word_embeddings": false,
+    "torch_dtype": "bfloat16",
+    "transformers_version": "4.39.3",
+    "use_cache": true,
+    "vocab_size": 32000,
+    "quantization_config": {
+        "quant_method": "exl2",
+        "version": "0.0.16",
+        "bits": 4.2,
+        "head_bits": 6,
+        "calibration": {
+            "rows": 100,
+            "length": 2048,
+            "dataset": "(default)"
+        }
+    }
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "transformers_version": "4.39.3"
+}

output-00001-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3506730823900f1c947faf7f97cb94b291e9a0556f460d6fb38d7246ddbf9de
+size 8590126132

output-00002-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:966fd951f3525646de44c61a4680e1b517759b81c9fc4f55db87978846d6f926
+size 8587693564

output-00003-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0015a07664a2b8ac8f2cebbed130178dd940876a2ad25247ca6e4bec583b8808
+size 8567735480

output-00004-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ad478843d882805ef38cc1d801290b737734e4b0447bc1815d6ed7ed597736b
+size 8590034680

output-00005-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a090507e830936de2109426ccd0eafebfa17e9dc2a1129ea365a364b844bcf82
+size 8585140552

output-00006-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2806ef0dbd4d899a90b64476d3be1a367b86638c0c0c5d286256c988582904f
+size 8583037656

output-00007-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3467b4368d468cd50c4a60e0dbdbc777c702f8236f8c69cf470e9e34be887933
+size 8583086808

output-00008-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e098603b062e3631d3ffa940cacc0389eefc7b326db63315f6edc82859a7a30
+size 8584921800

output-00009-of-00009.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:32f5d44f31c4cc2a55dd272dfbb04d68bf70c1a9fce0d0f7dccd2ae65fdb1c60
+size 5503058304

special_tokens_map.json

@@ -0,0 +1,24 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "</s>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dadfd56d766715c61d2ef780a525ab43b8e6da4de6865bda3d95fdef5e134055
+size 493443

tokenizer_config.json

@@ -0,0 +1,43 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [],
+  "bos_token": "<s>",
+  "chat_template": "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ '<|user|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'system' %}\n{{ '<|system|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'assistant' %}\n{{ '<|assistant|>\n'  + message['content'] + eos_token }}\n{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ '<|assistant|>' }}\n{% endif %}\n{% endfor %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": true,
+  "model_max_length": 2048,
+  "pad_token": "</s>",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 3.0,
+    "train_loss": 0.812556631554107,
+    "train_runtime": 4771.9621,
+    "train_samples": 6932,
+    "train_samples_per_second": 4.358,
+    "train_steps_per_second": 0.136
+}

trainer_state.json

@@ -0,0 +1,1200 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "eval_steps": 500,
+  "global_step": 651,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.05,
+      "grad_norm": 117760.0,
+      "learning_rate": 5.000000000000001e-07,
+      "log_odds_chosen": 0.36438828706741333,
+      "log_odds_ratio": -0.6397662162780762,
+      "logits/chosen": 3.8861491680145264,
+      "logits/rejected": 5.231001853942871,
+      "logps/chosen": -0.9861465692520142,
+      "logps/rejected": -1.2529093027114868,
+      "loss": 1.953,
+      "nll_loss": 3.2415008544921875,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.04930732399225235,
+      "rewards/margins": 0.013338141143321991,
+      "rewards/rejected": -0.06264545768499374,
+      "step": 10
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 29184.0,
+      "learning_rate": 1.0000000000000002e-06,
+      "log_odds_chosen": 0.17107267677783966,
+      "log_odds_ratio": -0.6301043033599854,
+      "logits/chosen": 4.779696464538574,
+      "logits/rejected": 5.251872539520264,
+      "logps/chosen": -1.1045284271240234,
+      "logps/rejected": -1.2445374727249146,
+      "loss": 1.7108,
+      "nll_loss": 1.8614288568496704,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.05522642284631729,
+      "rewards/margins": 0.007000453770160675,
+      "rewards/rejected": -0.06222687289118767,
+      "step": 20
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 3932160.0,
+      "learning_rate": 1.5e-06,
+      "log_odds_chosen": 0.478428453207016,
+      "log_odds_ratio": -0.5682710409164429,
+      "logits/chosen": 4.58956241607666,
+      "logits/rejected": 5.215265274047852,
+      "logps/chosen": -0.9884525537490845,
+      "logps/rejected": -1.2604442834854126,
+      "loss": 2.1071,
+      "nll_loss": 1.525723934173584,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.04942262917757034,
+      "rewards/margins": 0.013599586673080921,
+      "rewards/rejected": -0.06302221864461899,
+      "step": 30
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 63232.0,
+      "learning_rate": 2.0000000000000003e-06,
+      "log_odds_chosen": 0.2763148248195648,
+      "log_odds_ratio": -0.6428317427635193,
+      "logits/chosen": 5.248695373535156,
+      "logits/rejected": 5.335747718811035,
+      "logps/chosen": -0.9019734263420105,
+      "logps/rejected": -1.058569312095642,
+      "loss": 1.6367,
+      "nll_loss": 1.17539644241333,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.045098677277565,
+      "rewards/margins": 0.007829795591533184,
+      "rewards/rejected": -0.052928466349840164,
+      "step": 40
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1802240.0,
+      "learning_rate": 2.5e-06,
+      "log_odds_chosen": -0.07109338045120239,
+      "log_odds_ratio": -0.9068069458007812,
+      "logits/chosen": 4.34699821472168,
+      "logits/rejected": 5.148941993713379,
+      "logps/chosen": -1.0307292938232422,
+      "logps/rejected": -1.0001896619796753,
+      "loss": 2.0499,
+      "nll_loss": 2.4581282138824463,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.05153647065162659,
+      "rewards/margins": -0.0015269846189767122,
+      "rewards/rejected": -0.05000948905944824,
+      "step": 50
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 473088.0,
+      "learning_rate": 3e-06,
+      "log_odds_chosen": 0.7022095918655396,
+      "log_odds_ratio": -0.47877854108810425,
+      "logits/chosen": 5.137725830078125,
+      "logits/rejected": 5.073107719421387,
+      "logps/chosen": -0.7480964660644531,
+      "logps/rejected": -1.172572374343872,
+      "loss": 1.9116,
+      "nll_loss": 1.2398216724395752,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.037404827773571014,
+      "rewards/margins": 0.021223794668912888,
+      "rewards/rejected": -0.058628618717193604,
+      "step": 60
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 4161536.0,
+      "learning_rate": 3.5e-06,
+      "log_odds_chosen": -0.30822521448135376,
+      "log_odds_ratio": -1.0616459846496582,
+      "logits/chosen": 4.378929615020752,
+      "logits/rejected": 5.239219665527344,
+      "logps/chosen": -1.115562081336975,
+      "logps/rejected": -0.8684147596359253,
+      "loss": 2.0166,
+      "nll_loss": 2.142368793487549,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.05577809736132622,
+      "rewards/margins": -0.012357364408671856,
+      "rewards/rejected": -0.043420739471912384,
+      "step": 70
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 211968.0,
+      "learning_rate": 4.000000000000001e-06,
+      "log_odds_chosen": 0.38707518577575684,
+      "log_odds_ratio": -0.5776039361953735,
+      "logits/chosen": 5.019408226013184,
+      "logits/rejected": 5.3371453285217285,
+      "logps/chosen": -0.9375723004341125,
+      "logps/rejected": -1.1710981130599976,
+      "loss": 1.8918,
+      "nll_loss": 1.6274166107177734,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.046878617256879807,
+      "rewards/margins": 0.011676294729113579,
+      "rewards/rejected": -0.058554910123348236,
+      "step": 80
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1704.0,
+      "learning_rate": 4.5e-06,
+      "log_odds_chosen": 0.5051761865615845,
+      "log_odds_ratio": -0.5127500295639038,
+      "logits/chosen": 4.478859901428223,
+      "logits/rejected": 4.748915672302246,
+      "logps/chosen": -0.8121053576469421,
+      "logps/rejected": -1.1007237434387207,
+      "loss": 1.8015,
+      "nll_loss": 1.57842218875885,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -0.04060526192188263,
+      "rewards/margins": 0.014430919662117958,
+      "rewards/rejected": -0.055036187171936035,
+      "step": 90
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 8.9375,
+      "learning_rate": 5e-06,
+      "log_odds_chosen": 1.037414789199829,
+      "log_odds_ratio": -0.3248421549797058,
+      "logits/chosen": 4.653676509857178,
+      "logits/rejected": 5.350204944610596,
+      "logps/chosen": -0.6695261001586914,
+      "logps/rejected": -1.300445795059204,
+      "loss": 0.9356,
+      "nll_loss": 0.7057152986526489,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.03347630053758621,
+      "rewards/margins": 0.031545985490083694,
+      "rewards/rejected": -0.0650222972035408,
+      "step": 100
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 2.8125,
+      "learning_rate": 4.767312946227961e-06,
+      "log_odds_chosen": 0.6677854061126709,
+      "log_odds_ratio": -0.5610898733139038,
+      "logits/chosen": 4.671368598937988,
+      "logits/rejected": 5.119524002075195,
+      "logps/chosen": -0.8684868812561035,
+      "logps/rejected": -1.2020485401153564,
+      "loss": 0.8288,
+      "nll_loss": 0.9947482347488403,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.043424345552921295,
+      "rewards/margins": 0.01667807623744011,
+      "rewards/rejected": -0.06010241433978081,
+      "step": 110
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.9921875,
+      "learning_rate": 4.564354645876385e-06,
+      "log_odds_chosen": 0.5597886443138123,
+      "log_odds_ratio": -0.5193617343902588,
+      "logits/chosen": 5.5248026847839355,
+      "logits/rejected": 6.067958354949951,
+      "logps/chosen": -0.9015194773674011,
+      "logps/rejected": -1.2406196594238281,
+      "loss": 0.7451,
+      "nll_loss": 0.8342186212539673,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.045075975358486176,
+      "rewards/margins": 0.01695500686764717,
+      "rewards/rejected": -0.06203098222613335,
+      "step": 120
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 2.875,
+      "learning_rate": 4.385290096535147e-06,
+      "log_odds_chosen": 0.26957136392593384,
+      "log_odds_ratio": -0.7732787728309631,
+      "logits/chosen": 4.8973588943481445,
+      "logits/rejected": 5.552582263946533,
+      "logps/chosen": -0.877202033996582,
+      "logps/rejected": -0.9518612623214722,
+      "loss": 0.7319,
+      "nll_loss": 0.6940464377403259,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.04386010393500328,
+      "rewards/margins": 0.0037329583428800106,
+      "rewards/rejected": -0.04759306460618973,
+      "step": 130
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 2.5,
+      "learning_rate": 4.2257712736425835e-06,
+      "log_odds_chosen": 0.7680839896202087,
+      "log_odds_ratio": -0.5321913957595825,
+      "logits/chosen": 5.471996307373047,
+      "logits/rejected": 5.644137382507324,
+      "logps/chosen": -0.6714180111885071,
+      "logps/rejected": -0.9587985277175903,
+      "loss": 0.732,
+      "nll_loss": 0.6440631151199341,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.033570900559425354,
+      "rewards/margins": 0.014369020238518715,
+      "rewards/rejected": -0.047939930111169815,
+      "step": 140
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 2.375,
+      "learning_rate": 4.082482904638631e-06,
+      "log_odds_chosen": 0.5068908929824829,
+      "log_odds_ratio": -0.604145884513855,
+      "logits/chosen": 5.474297523498535,
+      "logits/rejected": 5.376832485198975,
+      "logps/chosen": -0.8357957005500793,
+      "logps/rejected": -1.0136160850524902,
+      "loss": 0.706,
+      "nll_loss": 0.6737378835678101,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.04178978502750397,
+      "rewards/margins": 0.00889101903885603,
+      "rewards/rejected": -0.05068080872297287,
+      "step": 150
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 2.015625,
+      "learning_rate": 3.952847075210474e-06,
+      "log_odds_chosen": 0.615983784198761,
+      "log_odds_ratio": -0.4876289963722229,
+      "logits/chosen": 5.473410129547119,
+      "logits/rejected": 6.06318998336792,
+      "logps/chosen": -0.9676389694213867,
+      "logps/rejected": -1.349385142326355,
+      "loss": 0.6996,
+      "nll_loss": 0.6852242350578308,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -0.048381954431533813,
+      "rewards/margins": 0.019087309017777443,
+      "rewards/rejected": -0.06746925413608551,
+      "step": 160
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 2.171875,
+      "learning_rate": 3.834824944236852e-06,
+      "log_odds_chosen": 0.4551977515220642,
+      "log_odds_ratio": -0.5428072214126587,
+      "logits/chosen": 4.785284042358398,
+      "logits/rejected": 6.005092620849609,
+      "logps/chosen": -0.7350739240646362,
+      "logps/rejected": -1.0496256351470947,
+      "loss": 0.6959,
+      "nll_loss": 0.5339438319206238,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.03675369173288345,
+      "rewards/margins": 0.015727588906884193,
+      "rewards/rejected": -0.052481282502412796,
+      "step": 170
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 2.015625,
+      "learning_rate": 3.72677996249965e-06,
+      "log_odds_chosen": 0.5587902665138245,
+      "log_odds_ratio": -0.6063727140426636,
+      "logits/chosen": 4.6595892906188965,
+      "logits/rejected": 5.4700422286987305,
+      "logps/chosen": -0.7482207417488098,
+      "logps/rejected": -0.9887701272964478,
+      "loss": 0.7233,
+      "nll_loss": 0.5874465703964233,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.03741103783249855,
+      "rewards/margins": 0.012027469463646412,
+      "rewards/rejected": -0.04943850636482239,
+      "step": 180
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 2.0625,
+      "learning_rate": 3.6273812505500587e-06,
+      "log_odds_chosen": 0.9965683817863464,
+      "log_odds_ratio": -0.4162277281284332,
+      "logits/chosen": 5.304540157318115,
+      "logits/rejected": 5.486930847167969,
+      "logps/chosen": -0.7579169869422913,
+      "logps/rejected": -1.1843591928482056,
+      "loss": 0.7298,
+      "nll_loss": 0.6787526607513428,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -0.03789585083723068,
+      "rewards/margins": 0.021322116255760193,
+      "rewards/rejected": -0.059217967092990875,
+      "step": 190
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.9609375,
+      "learning_rate": 3.5355339059327378e-06,
+      "log_odds_chosen": 0.2911016047000885,
+      "log_odds_ratio": -0.6208275556564331,
+      "logits/chosen": 5.865508556365967,
+      "logits/rejected": 5.9140448570251465,
+      "logps/chosen": -1.0318800210952759,
+      "logps/rejected": -1.2233208417892456,
+      "loss": 0.6888,
+      "nll_loss": 0.8277570009231567,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.051594000309705734,
+      "rewards/margins": 0.009572046808898449,
+      "rewards/rejected": -0.061166055500507355,
+      "step": 200
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 2.546875,
+      "learning_rate": 3.450327796711771e-06,
+      "log_odds_chosen": 0.3929597735404968,
+      "log_odds_ratio": -0.6252869367599487,
+      "logits/chosen": 5.480368137359619,
+      "logits/rejected": 5.818605899810791,
+      "logps/chosen": -0.8382253646850586,
+      "logps/rejected": -1.1194109916687012,
+      "loss": 0.703,
+      "nll_loss": 0.7914389967918396,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.04191126674413681,
+      "rewards/margins": 0.014059278182685375,
+      "rewards/rejected": -0.05597054958343506,
+      "step": 210
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 2.234375,
+      "learning_rate": 3.3709993123162106e-06,
+      "log_odds_chosen": 1.1686198711395264,
+      "log_odds_ratio": -0.39844751358032227,
+      "logits/chosen": 4.818378448486328,
+      "logits/rejected": 5.660789966583252,
+      "logps/chosen": -0.5040851831436157,
+      "logps/rejected": -0.9685913324356079,
+      "loss": 0.6554,
+      "nll_loss": 0.49605101346969604,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -0.025204259902238846,
+      "rewards/margins": 0.02322530373930931,
+      "rewards/rejected": -0.04842956364154816,
+      "step": 220
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 2.046875,
+      "learning_rate": 3.296902366978936e-06,
+      "log_odds_chosen": 0.7159255743026733,
+      "log_odds_ratio": -0.5276229977607727,
+      "logits/chosen": 4.3275017738342285,
+      "logits/rejected": 5.1829423904418945,
+      "logps/chosen": -0.7593253254890442,
+      "logps/rejected": -1.0148638486862183,
+      "loss": 0.6289,
+      "nll_loss": 0.609928548336029,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.03796626627445221,
+      "rewards/margins": 0.012776928022503853,
+      "rewards/rejected": -0.05074319988489151,
+      "step": 230
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 2.375,
+      "learning_rate": 3.2274861218395142e-06,
+      "log_odds_chosen": 0.7326894998550415,
+      "log_odds_ratio": -0.5214331150054932,
+      "logits/chosen": 4.783654689788818,
+      "logits/rejected": 5.283537864685059,
+      "logps/chosen": -0.7465990781784058,
+      "logps/rejected": -0.9910147786140442,
+      "loss": 0.6382,
+      "nll_loss": 0.7347540855407715,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.03732995316386223,
+      "rewards/margins": 0.012220785021781921,
+      "rewards/rejected": -0.04955074191093445,
+      "step": 240
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 2.0625,
+      "learning_rate": 3.1622776601683796e-06,
+      "log_odds_chosen": 0.040362291038036346,
+      "log_odds_ratio": -0.7654204964637756,
+      "logits/chosen": 4.929324150085449,
+      "logits/rejected": 4.940483570098877,
+      "logps/chosen": -0.939703106880188,
+      "logps/rejected": -0.9395262598991394,
+      "loss": 0.6626,
+      "nll_loss": 0.7169132232666016,
+      "rewards/accuracies": 0.4000000059604645,
+      "rewards/chosen": -0.04698516055941582,
+      "rewards/margins": -8.843839168548584e-06,
+      "rewards/rejected": -0.04697632044553757,
+      "step": 250
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 2.59375,
+      "learning_rate": 3.1008683647302113e-06,
+      "log_odds_chosen": 0.8304751515388489,
+      "log_odds_ratio": -0.4627406597137451,
+      "logits/chosen": 4.34907341003418,
+      "logits/rejected": 4.541801929473877,
+      "logps/chosen": -0.7797168493270874,
+      "logps/rejected": -1.0878037214279175,
+      "loss": 0.6408,
+      "nll_loss": 0.6424815058708191,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.03898584097623825,
+      "rewards/margins": 0.015404346399009228,
+      "rewards/rejected": -0.05439019203186035,
+      "step": 260
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 2.328125,
+      "learning_rate": 3.0429030972509227e-06,
+      "log_odds_chosen": 0.2547241747379303,
+      "log_odds_ratio": -0.7041358351707458,
+      "logits/chosen": 4.1212077140808105,
+      "logits/rejected": 5.139257431030273,
+      "logps/chosen": -0.5988011360168457,
+      "logps/rejected": -0.7647382020950317,
+      "loss": 0.6441,
+      "nll_loss": 0.4384763836860657,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.029940057545900345,
+      "rewards/margins": 0.008296851068735123,
+      "rewards/rejected": -0.03823690861463547,
+      "step": 270
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2.988071523335984e-06,
+      "log_odds_chosen": 0.7432643175125122,
+      "log_odds_ratio": -0.4928904175758362,
+      "logits/chosen": 4.240169525146484,
+      "logits/rejected": 4.746310234069824,
+      "logps/chosen": -0.7583116292953491,
+      "logps/rejected": -1.0217373371124268,
+      "loss": 0.6349,
+      "nll_loss": 0.5912537574768066,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.03791557624936104,
+      "rewards/margins": 0.013171288184821606,
+      "rewards/rejected": -0.05108686536550522,
+      "step": 280
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 2.5,
+      "learning_rate": 2.9361010975735177e-06,
+      "log_odds_chosen": 0.6404408812522888,
+      "log_odds_ratio": -0.5461726784706116,
+      "logits/chosen": 4.347890377044678,
+      "logits/rejected": 5.2955708503723145,
+      "logps/chosen": -0.8145158886909485,
+      "logps/rejected": -1.124975323677063,
+      "loss": 0.6204,
+      "nll_loss": 0.5651360154151917,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.040725789964199066,
+      "rewards/margins": 0.015522971749305725,
+      "rewards/rejected": -0.05624876171350479,
+      "step": 290
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 1.96875,
+      "learning_rate": 2.8867513459481293e-06,
+      "log_odds_chosen": 0.4704459607601166,
+      "log_odds_ratio": -0.6623938083648682,
+      "logits/chosen": 4.255876064300537,
+      "logits/rejected": 5.063040733337402,
+      "logps/chosen": -0.7718355059623718,
+      "logps/rejected": -1.144460916519165,
+      "loss": 0.6404,
+      "nll_loss": 0.6724303364753723,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.03859177231788635,
+      "rewards/margins": 0.01863126829266548,
+      "rewards/rejected": -0.05722304433584213,
+      "step": 300
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 2.453125,
+      "learning_rate": 2.839809171235324e-06,
+      "log_odds_chosen": 1.5952459573745728,
+      "log_odds_ratio": -0.2707791328430176,
+      "logits/chosen": 2.7694969177246094,
+      "logits/rejected": 5.479510307312012,
+      "logps/chosen": -0.4962679445743561,
+      "logps/rejected": -1.2316776514053345,
+      "loss": 0.6428,
+      "nll_loss": 0.3623020648956299,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.024813394993543625,
+      "rewards/margins": 0.03677048534154892,
+      "rewards/rejected": -0.06158388406038284,
+      "step": 310
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 2.078125,
+      "learning_rate": 2.7950849718747376e-06,
+      "log_odds_chosen": 0.4402007460594177,
+      "log_odds_ratio": -0.5388344526290894,
+      "logits/chosen": 4.8701372146606445,
+      "logits/rejected": 4.049181938171387,
+      "logps/chosen": -0.8427563905715942,
+      "logps/rejected": -1.1280080080032349,
+      "loss": 0.6661,
+      "nll_loss": 0.6774541735649109,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.04213782027363777,
+      "rewards/margins": 0.014262576587498188,
+      "rewards/rejected": -0.056400395929813385,
+      "step": 320
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 1.9765625,
+      "learning_rate": 2.752409412815902e-06,
+      "log_odds_chosen": 1.4536019563674927,
+      "log_odds_ratio": -0.3178521990776062,
+      "logits/chosen": 4.046222686767578,
+      "logits/rejected": 4.855486869812012,
+      "logps/chosen": -0.4614998400211334,
+      "logps/rejected": -1.0025476217269897,
+      "loss": 0.6396,
+      "nll_loss": 0.46759381890296936,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -0.02307499200105667,
+      "rewards/margins": 0.027052391320466995,
+      "rewards/rejected": -0.05012737959623337,
+      "step": 330
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 2.65625,
+      "learning_rate": 2.711630722733202e-06,
+      "log_odds_chosen": 0.4552677273750305,
+      "log_odds_ratio": -0.5441101789474487,
+      "logits/chosen": 4.233187198638916,
+      "logits/rejected": 4.776756286621094,
+      "logps/chosen": -0.9984881281852722,
+      "logps/rejected": -1.3039405345916748,
+      "loss": 0.6326,
+      "nll_loss": 0.7266319990158081,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.04992440715432167,
+      "rewards/margins": 0.01527262944728136,
+      "rewards/rejected": -0.06519703567028046,
+      "step": 340
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2.6726124191242444e-06,
+      "log_odds_chosen": 0.3951299488544464,
+      "log_odds_ratio": -0.6442996263504028,
+      "logits/chosen": 4.592418193817139,
+      "logits/rejected": 4.885247707366943,
+      "logps/chosen": -0.9690208435058594,
+      "logps/rejected": -1.1191128492355347,
+      "loss": 0.6271,
+      "nll_loss": 0.7028160095214844,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.04845104366540909,
+      "rewards/margins": 0.007504602428525686,
+      "rewards/rejected": -0.055955640971660614,
+      "step": 350
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 2.109375,
+      "learning_rate": 2.6352313834736496e-06,
+      "log_odds_chosen": 0.6397253274917603,
+      "log_odds_ratio": -0.4948647916316986,
+      "logits/chosen": 3.1035220623016357,
+      "logits/rejected": 4.4074320793151855,
+      "logps/chosen": -0.7063679695129395,
+      "logps/rejected": -1.086042881011963,
+      "loss": 0.6133,
+      "nll_loss": 0.5209956765174866,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.03531839698553085,
+      "rewards/margins": 0.0189837496727705,
+      "rewards/rejected": -0.054302144795656204,
+      "step": 360
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2.599376224550182e-06,
+      "log_odds_chosen": 0.5072129368782043,
+      "log_odds_ratio": -0.5375211834907532,
+      "logits/chosen": 4.4618144035339355,
+      "logits/rejected": 4.897726535797119,
+      "logps/chosen": -0.8658114671707153,
+      "logps/rejected": -1.161678433418274,
+      "loss": 0.625,
+      "nll_loss": 0.7147814035415649,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.043290577828884125,
+      "rewards/margins": 0.014793348498642445,
+      "rewards/rejected": -0.058083921670913696,
+      "step": 370
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 2.28125,
+      "learning_rate": 2.564945880212886e-06,
+      "log_odds_chosen": 0.5736058950424194,
+      "log_odds_ratio": -0.4948197305202484,
+      "logits/chosen": 4.31764554977417,
+      "logits/rejected": 4.153486251831055,
+      "logps/chosen": -0.8540223836898804,
+      "logps/rejected": -1.1471771001815796,
+      "loss": 0.6393,
+      "nll_loss": 0.6763076186180115,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.0427011176943779,
+      "rewards/margins": 0.014657735824584961,
+      "rewards/rejected": -0.05735884979367256,
+      "step": 380
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 3.640625,
+      "learning_rate": 2.5318484177091667e-06,
+      "log_odds_chosen": 0.8381564021110535,
+      "log_odds_ratio": -0.5308811068534851,
+      "logits/chosen": 4.037534236907959,
+      "logits/rejected": 5.888669013977051,
+      "logps/chosen": -0.700161337852478,
+      "logps/rejected": -1.2042081356048584,
+      "loss": 0.6318,
+      "nll_loss": 0.5512461066246033,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.03500806540250778,
+      "rewards/margins": 0.025202345103025436,
+      "rewards/rejected": -0.060210417956113815,
+      "step": 390
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 2.171875,
+      "learning_rate": 2.5e-06,
+      "log_odds_chosen": 0.7038768529891968,
+      "log_odds_ratio": -0.43052348494529724,
+      "logits/chosen": 3.822885036468506,
+      "logits/rejected": 4.210227012634277,
+      "logps/chosen": -0.6150542497634888,
+      "logps/rejected": -0.9889954328536987,
+      "loss": 0.6218,
+      "nll_loss": 0.5013046264648438,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -0.030752714723348618,
+      "rewards/margins": 0.01869705691933632,
+      "rewards/rejected": -0.04944976791739464,
+      "step": 400
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 2.03125,
+      "learning_rate": 2.4693239916239746e-06,
+      "log_odds_chosen": 0.49417972564697266,
+      "log_odds_ratio": -0.5454962253570557,
+      "logits/chosen": 3.7158710956573486,
+      "logits/rejected": 4.625822067260742,
+      "logps/chosen": -0.7136448621749878,
+      "logps/rejected": -0.9806584119796753,
+      "loss": 0.6163,
+      "nll_loss": 0.5766875147819519,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.03568224236369133,
+      "rewards/margins": 0.013350683264434338,
+      "rewards/rejected": -0.04903292655944824,
+      "step": 410
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 1.96875,
+      "learning_rate": 2.4397501823713327e-06,
+      "log_odds_chosen": 1.2905668020248413,
+      "log_odds_ratio": -0.3054632544517517,
+      "logits/chosen": 4.375031471252441,
+      "logits/rejected": 5.165828704833984,
+      "logps/chosen": -0.6634560823440552,
+      "logps/rejected": -1.2297804355621338,
+      "loss": 0.6299,
+      "nll_loss": 0.5654190182685852,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -0.03317280486226082,
+      "rewards/margins": 0.02831621840596199,
+      "rewards/rejected": -0.06148902326822281,
+      "step": 420
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 2.234375,
+      "learning_rate": 2.411214110852061e-06,
+      "log_odds_chosen": 0.4614163041114807,
+      "log_odds_ratio": -0.5477044582366943,
+      "logits/chosen": 3.945091724395752,
+      "logits/rejected": 4.783943176269531,
+      "logps/chosen": -0.670985758304596,
+      "logps/rejected": -0.8528381586074829,
+      "loss": 0.6328,
+      "nll_loss": 0.5353778004646301,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.03354928642511368,
+      "rewards/margins": 0.009092616848647594,
+      "rewards/rejected": -0.04264190047979355,
+      "step": 430
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 2.140625,
+      "learning_rate": 2.3836564731139807e-06,
+      "log_odds_chosen": 0.519318699836731,
+      "log_odds_ratio": -0.5034213066101074,
+      "logits/chosen": 3.990828037261963,
+      "logits/rejected": 4.283727645874023,
+      "logps/chosen": -0.7843809723854065,
+      "logps/rejected": -1.1084554195404053,
+      "loss": 0.598,
+      "nll_loss": 0.6064985394477844,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.03921904414892197,
+      "rewards/margins": 0.01620371639728546,
+      "rewards/rejected": -0.055422764271497726,
+      "step": 440
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 2.015625,
+      "learning_rate": 2.357022603955159e-06,
+      "log_odds_chosen": 1.2161670923233032,
+      "log_odds_ratio": -0.5558447241783142,
+      "logits/chosen": 2.7631869316101074,
+      "logits/rejected": 4.014997959136963,
+      "logps/chosen": -0.4891352653503418,
+      "logps/rejected": -1.057556390762329,
+      "loss": 0.6063,
+      "nll_loss": 0.5005042552947998,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.02445676364004612,
+      "rewards/margins": 0.028421055525541306,
+      "rewards/rejected": -0.052877821028232574,
+      "step": 450
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 2.0625,
+      "learning_rate": 2.3312620206007847e-06,
+      "log_odds_chosen": 0.8278636932373047,
+      "log_odds_ratio": -0.43884754180908203,
+      "logits/chosen": 4.009448051452637,
+      "logits/rejected": 4.671367645263672,
+      "logps/chosen": -0.7134698629379272,
+      "logps/rejected": -1.146784782409668,
+      "loss": 0.5862,
+      "nll_loss": 0.5619599223136902,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.03567349165678024,
+      "rewards/margins": 0.021665748208761215,
+      "rewards/rejected": -0.05733924359083176,
+      "step": 460
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 2.609375,
+      "learning_rate": 2.3063280200722128e-06,
+      "log_odds_chosen": 1.677671194076538,
+      "log_odds_ratio": -0.2895694375038147,
+      "logits/chosen": 2.985790491104126,
+      "logits/rejected": 4.190914630889893,
+      "logps/chosen": -0.5018793344497681,
+      "logps/rejected": -1.0572091341018677,
+      "loss": 0.5765,
+      "nll_loss": 0.5001329183578491,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -0.025093963369727135,
+      "rewards/margins": 0.02776649035513401,
+      "rewards/rejected": -0.05286044999957085,
+      "step": 470
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 2.0,
+      "learning_rate": 2.2821773229381924e-06,
+      "log_odds_chosen": 1.0791471004486084,
+      "log_odds_ratio": -0.37350553274154663,
+      "logits/chosen": 3.676426649093628,
+      "logits/rejected": 3.8374907970428467,
+      "logps/chosen": -0.7438164353370667,
+      "logps/rejected": -1.29355788230896,
+      "loss": 0.5652,
+      "nll_loss": 0.6555451154708862,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.037190817296504974,
+      "rewards/margins": 0.027487074956297874,
+      "rewards/rejected": -0.0646779015660286,
+      "step": 480
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 2.21875,
+      "learning_rate": 2.2587697572631284e-06,
+      "log_odds_chosen": 0.275502473115921,
+      "log_odds_ratio": -0.7135687470436096,
+      "logits/chosen": 4.321534156799316,
+      "logits/rejected": 4.41732120513916,
+      "logps/chosen": -0.9727070927619934,
+      "logps/rejected": -1.0810346603393555,
+      "loss": 0.5952,
+      "nll_loss": 0.7110171914100647,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.04863535612821579,
+      "rewards/margins": 0.005416377447545528,
+      "rewards/rejected": -0.05405173450708389,
+      "step": 490
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 2.15625,
+      "learning_rate": 2.23606797749979e-06,
+      "log_odds_chosen": 0.34863442182540894,
+      "log_odds_ratio": -0.6463712453842163,
+      "logits/chosen": 4.6876606941223145,
+      "logits/rejected": 5.054124355316162,
+      "logps/chosen": -0.9338000416755676,
+      "logps/rejected": -1.1037800312042236,
+      "loss": 0.5953,
+      "nll_loss": 0.8528131246566772,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.04669000208377838,
+      "rewards/margins": 0.008499005809426308,
+      "rewards/rejected": -0.05518900603055954,
+      "step": 500
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 2.171875,
+      "learning_rate": 2.2140372138502386e-06,
+      "log_odds_chosen": 0.9548345804214478,
+      "log_odds_ratio": -0.39882007241249084,
+      "logits/chosen": 3.5289406776428223,
+      "logits/rejected": 3.8287463188171387,
+      "logps/chosen": -0.6570809483528137,
+      "logps/rejected": -1.1388274431228638,
+      "loss": 0.609,
+      "nll_loss": 0.5968061685562134,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -0.032854050397872925,
+      "rewards/margins": 0.024087321013212204,
+      "rewards/rejected": -0.05694136768579483,
+      "step": 510
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2.1926450482675734e-06,
+      "log_odds_chosen": 0.4539831280708313,
+      "log_odds_ratio": -0.5872747302055359,
+      "logits/chosen": 3.2061939239501953,
+      "logits/rejected": 4.589787006378174,
+      "logps/chosen": -0.7979894280433655,
+      "logps/rejected": -1.0285401344299316,
+      "loss": 0.5827,
+      "nll_loss": 0.6084668636322021,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.039899468421936035,
+      "rewards/margins": 0.011527536436915398,
+      "rewards/rejected": -0.051426999270915985,
+      "step": 520
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 2.484375,
+      "learning_rate": 2.1718612138153473e-06,
+      "log_odds_chosen": 0.8493059277534485,
+      "log_odds_ratio": -0.6372500658035278,
+      "logits/chosen": 3.078615665435791,
+      "logits/rejected": 4.099945068359375,
+      "logps/chosen": -0.6704202890396118,
+      "logps/rejected": -0.7899671792984009,
+      "loss": 0.5788,
+      "nll_loss": 0.5733928084373474,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.03352101519703865,
+      "rewards/margins": 0.005977341439574957,
+      "rewards/rejected": -0.039498358964920044,
+      "step": 530
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 1.859375,
+      "learning_rate": 2.151657414559676e-06,
+      "log_odds_chosen": 0.6374627351760864,
+      "log_odds_ratio": -0.5592355728149414,
+      "logits/chosen": 3.680483341217041,
+      "logits/rejected": 3.9816291332244873,
+      "logps/chosen": -0.8559755086898804,
+      "logps/rejected": -1.1612054109573364,
+      "loss": 0.6003,
+      "nll_loss": 0.6403124928474426,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.04279877990484238,
+      "rewards/margins": 0.015261486172676086,
+      "rewards/rejected": -0.05806026607751846,
+      "step": 540
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2.132007163556104e-06,
+      "log_odds_chosen": 1.399209976196289,
+      "log_odds_ratio": -0.5735031366348267,
+      "logits/chosen": 3.132289171218872,
+      "logits/rejected": 3.5427193641662598,
+      "logps/chosen": -0.5963010191917419,
+      "logps/rejected": -0.9639393091201782,
+      "loss": 0.5984,
+      "nll_loss": 0.5058175325393677,
+      "rewards/accuracies": 0.6000000238418579,
+      "rewards/chosen": -0.029815051704645157,
+      "rewards/margins": 0.018381912261247635,
+      "rewards/rejected": -0.04819696769118309,
+      "step": 550
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 1.859375,
+      "learning_rate": 2.1128856368212917e-06,
+      "log_odds_chosen": 0.688880443572998,
+      "log_odds_ratio": -0.4902462959289551,
+      "logits/chosen": 2.6950721740722656,
+      "logits/rejected": 3.1528286933898926,
+      "logps/chosen": -0.6383022665977478,
+      "logps/rejected": -0.9691828489303589,
+      "loss": 0.5718,
+      "nll_loss": 0.4289799630641937,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.03191510960459709,
+      "rewards/margins": 0.016544032841920853,
+      "rewards/rejected": -0.048459142446517944,
+      "step": 560
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 2.421875,
+      "learning_rate": 2.0942695414584777e-06,
+      "log_odds_chosen": 1.3283271789550781,
+      "log_odds_ratio": -0.3012233078479767,
+      "logits/chosen": 3.4564871788024902,
+      "logits/rejected": 4.7043867111206055,
+      "logps/chosen": -0.6779360771179199,
+      "logps/rejected": -1.523970365524292,
+      "loss": 0.6138,
+      "nll_loss": 0.5768535137176514,
+      "rewards/accuracies": 1.0,
+      "rewards/chosen": -0.033896803855895996,
+      "rewards/margins": 0.042301714420318604,
+      "rewards/rejected": -0.0761985182762146,
+      "step": 570
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 1.953125,
+      "learning_rate": 2.0761369963434992e-06,
+      "log_odds_chosen": 1.4566174745559692,
+      "log_odds_ratio": -0.32581037282943726,
+      "logits/chosen": 2.691676616668701,
+      "logits/rejected": 4.661564826965332,
+      "logps/chosen": -0.4493564963340759,
+      "logps/rejected": -1.0139671564102173,
+      "loss": 0.5782,
+      "nll_loss": 0.37120580673217773,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.022467825561761856,
+      "rewards/margins": 0.028230536729097366,
+      "rewards/rejected": -0.05069836229085922,
+      "step": 580
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 2.0625,
+      "learning_rate": 2.058467423981546e-06,
+      "log_odds_chosen": 1.0190517902374268,
+      "log_odds_ratio": -0.5730624198913574,
+      "logits/chosen": 3.407086133956909,
+      "logits/rejected": 4.482596397399902,
+      "logps/chosen": -0.7345553040504456,
+      "logps/rejected": -0.9309635162353516,
+      "loss": 0.5723,
+      "nll_loss": 0.5519307851791382,
+      "rewards/accuracies": 0.5,
+      "rewards/chosen": -0.03672776371240616,
+      "rewards/margins": 0.009820410050451756,
+      "rewards/rejected": -0.04654817283153534,
+      "step": 590
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 2.375,
+      "learning_rate": 2.0412414523193154e-06,
+      "log_odds_chosen": 1.107779860496521,
+      "log_odds_ratio": -0.40593117475509644,
+      "logits/chosen": 3.215078830718994,
+      "logits/rejected": 4.503358840942383,
+      "logps/chosen": -0.663019597530365,
+      "logps/rejected": -1.2786920070648193,
+      "loss": 0.5815,
+      "nll_loss": 0.5633824467658997,
+      "rewards/accuracies": 0.800000011920929,
+      "rewards/chosen": -0.03315097838640213,
+      "rewards/margins": 0.030783619731664658,
+      "rewards/rejected": -0.06393460184335709,
+      "step": 600
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 2.09375,
+      "learning_rate": 2.0244408254472904e-06,
+      "log_odds_chosen": 0.7602224349975586,
+      "log_odds_ratio": -0.5018362998962402,
+      "logits/chosen": 3.604353666305542,
+      "logits/rejected": 4.481316089630127,
+      "logps/chosen": -0.7105517387390137,
+      "logps/rejected": -1.0740478038787842,
+      "loss": 0.5873,
+      "nll_loss": 0.5312780737876892,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.035527586936950684,
+      "rewards/margins": 0.018174810335040092,
+      "rewards/rejected": -0.05370239168405533,
+      "step": 610
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 1.90625,
+      "learning_rate": 2.0080483222562476e-06,
+      "log_odds_chosen": 1.3286904096603394,
+      "log_odds_ratio": -0.36574870347976685,
+      "logits/chosen": 3.620469331741333,
+      "logits/rejected": 4.373411655426025,
+      "logps/chosen": -0.4990506172180176,
+      "logps/rejected": -0.953050971031189,
+      "loss": 0.5716,
+      "nll_loss": 0.5527733564376831,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -0.024952532723546028,
+      "rewards/margins": 0.022700021043419838,
+      "rewards/rejected": -0.047652553766965866,
+      "step": 620
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 2.359375,
+      "learning_rate": 1.9920476822239895e-06,
+      "log_odds_chosen": 0.4847317636013031,
+      "log_odds_ratio": -0.5640643835067749,
+      "logits/chosen": 3.125113010406494,
+      "logits/rejected": 3.340205669403076,
+      "logps/chosen": -0.8360971212387085,
+      "logps/rejected": -1.0480194091796875,
+      "loss": 0.5738,
+      "nll_loss": 0.6136351823806763,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.041804857552051544,
+      "rewards/margins": 0.010596117004752159,
+      "rewards/rejected": -0.05240097641944885,
+      "step": 630
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.976423537605237e-06,
+      "log_odds_chosen": 0.8931509256362915,
+      "log_odds_ratio": -0.40087467432022095,
+      "logits/chosen": 3.574153423309326,
+      "logits/rejected": 4.537802219390869,
+      "logps/chosen": -0.6440940499305725,
+      "logps/rejected": -1.088226556777954,
+      "loss": 0.5846,
+      "nll_loss": 0.5598152875900269,
+      "rewards/accuracies": 0.8999999761581421,
+      "rewards/chosen": -0.032204702496528625,
+      "rewards/margins": 0.0222066268324852,
+      "rewards/rejected": -0.054411329329013824,
+      "step": 640
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.5625,
+      "learning_rate": 1.961161351381841e-06,
+      "log_odds_chosen": 1.2053475379943848,
+      "log_odds_ratio": -0.430248886346817,
+      "logits/chosen": 2.245370388031006,
+      "logits/rejected": 3.5309462547302246,
+      "logps/chosen": -0.5642444491386414,
+      "logps/rejected": -0.9910544157028198,
+      "loss": 0.5605,
+      "nll_loss": 0.45886915922164917,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.028212225064635277,
+      "rewards/margins": 0.021340493112802505,
+      "rewards/rejected": -0.04955272004008293,
+      "step": 650
+    },
+    {
+      "epoch": 3.0,
+      "step": 651,
+      "total_flos": 0.0,
+      "train_loss": 0.812556631554107,
+      "train_runtime": 4771.9621,
+      "train_samples_per_second": 4.358,
+      "train_steps_per_second": 0.136
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 651,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500,
+  "total_flos": 0.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}