---
tags:
- generated_from_trainer
model-index:
- name: tulu-2-dpo-70b
  results: []
# license: mit
datasets:
# - HuggingFaceH4/ultrachat_200k
- HuggingFaceH4/ultrafeedback_binarized
language:
- en
base_model: meta-llama/Llama-2-70b-hf
---



<img src="https://huggingface.co/datasets/allenai/blog-images/resolve/main/tulu-v2/Tulu%20V2%20banner.png" alt="TuluV2 banner" width="800" style="margin-left:'auto' margin-right:'auto' display:'block'"/>


# Model Card for Tulu V2 DPO 70B

Zephyr is a series of language models that are trained to act as helpful assistants. Zephyr-7B-β is the second model in the series, and is a fine-tuned version of [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) that was trained on on a mix of publicly available, synthetic datasets using [Direct Preference Optimization (DPO)](https://arxiv.org/abs/2305.18290). We found that removing the in-built alignment of these datasets boosted performance on [MT Bench](https://huggingface.co/spaces/lmsys/mt-bench) and made the model more helpful. However, this means that model is likely to generate problematic text when prompted to do so and should only be used for educational and research purposes. You can find more details in the [technical report](https://arxiv.org/abs/2310.16944).


## Model description

- **Model type:** The flagship model of a suite of instruction and RLHF tuned chat models on a mix of publicly available, synthetic and human-created datasets.
- **Language(s) (NLP):** Primarily English
- **License:** MIT
- **Finetuned from model:** [meta-llama/Llama-2-70b-hf](https://huggingface.co/ meta-llama/Llama-2-70b-hf)

### Model Sources

- **Repository:** https://github.com/allenai/https://github.com/allenai/open-instruct
- TODO add collection

## Performance

At the time of release, the Tulu-v2-dpo-70b model is approximately equal to GPT4 on AlpacaEval, and has a score of TODO on MT-Bench.
All smaller DPO'd models have strong performance per model size in the category and with lower verbosity (average completion length).
| Model | Size | Alignment | MT-Bench (score) | AlpacaEval (win rate %) |
|-------------|-----|----|---------------|--------------|
| **Tulu-v2-7b** 🪁 | **7B** | **dDPO** | **TODO** | **TODO** |
| **Tulu-v2-13b** 🪁 | **13B** | **dDPO** | **TODO** | **TODO** |
| **Tulu-v2-70b** 🪁 | **70B** | **dDPO** | **TODO** | **TODO** |
| **Tulu-v2-dpo-7b** 🪁 | **7B** | **dDPO** | **TODO** | **TODO** |
| **Tulu-v2-dpo-13b** 🪁 | **13B** | **dDPO** | **TODO** | **TODO** |
| **Tulu-v2-dpo-70b** 🪁 | **70B** | **dDPO** | **TODO** | **TODO** |
| StableLM-Tuned-α | 7B| dSFT |2.75| -|
| MPT-Chat |  7B |dSFT |5.42| -|
| Xwin-LMv0.1 | 7B| dPPO| 6.19| 87.83|
| Mistral-Instructv0.1 | 7B|  - | 6.84 |-|
| Zephyr-7b-α |7B|  dDPO| 6.88| -|
| Zephyr-7b-β 🪁 | 7B | dDPO | 7.34 | 90.60 |
| Falcon-Instruct |  40B |dSFT |5.17 |45.71|
| Guanaco | 65B |  SFT |6.41| 71.80|
| Llama2-Chat |  70B |RLHF |6.86| 92.66|
| Vicuna v1.3 |  33B |dSFT |7.12 |88.99|
| WizardLM v1.0 |  70B |dSFT |7.71 |-|
| Xwin-LM v0.1 |   70B |dPPO |- |95.57|
| GPT-3.5-turbo | - |RLHF |7.94 |89.37|
| Claude 2 |  - |RLHF |8.06| 91.36|
| GPT-4 |  -| RLHF |8.99| 95.28|


## Intended uses & limitations

The model was initially fine-tuned on a filtered and preprocessed of the Tulu V2 mix dataset (TODO add link), which contains a diverse range of human created instructions and synthetic dialogues generated primarily by other LLMs. 
We then further aligned the model with a [Jax DPO trainer](https://github.com/hamishivi/EasyLM/blob/main/EasyLM/models/llama/llama_train_dpo.py) built on [EasyLM](https://github.com/young-geng/EasyLM) on the [openbmb/UltraFeedback](https://huggingface.co/datasets/openbmb/UltraFeedback) dataset, which contains 64k prompts and model completions that are ranked by GPT-4. As a result, the model can be used for chat and you can check out our [demo](https://huggingface.co/spaces/HuggingFaceH4/zephyr-chat) to test its capabilities. 

<!--  You can find the datasets used for training Tulu V2 [here]() -->

Here's how you can run the model using the `pipeline()` function from 🤗 Transformers:

```python
# Install transformers from source - only needed for versions <= v4.34
# pip install git+https://github.com/huggingface/transformers.git
# pip install accelerate

import torch
from transformers import pipeline

pipe = pipeline("text-generation", model="HuggingFaceH4/tulu-2-dpo-70b", torch_dtype=torch.bfloat16, device_map="auto")

# We use the tokenizer's chat template to format each message - see https://huggingface.co/docs/transformers/main/en/chat_templating
messages = [
    {
        "role": "system",
        "content": "You are a friendly chatbot who always responds in the style of a pirate",
    },
    {"role": "user", "content": "How many helicopters can a human eat in one sitting?"},
]
prompt = pipe.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = pipe(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])
# <|system|>
# You are a friendly chatbot who always responds in the style of a pirate.</s>
# <|user|>
# How many helicopters can a human eat in one sitting?</s>
# <|assistant|>
# Ah, me hearty matey! But yer question be a puzzler! A human cannot eat a helicopter in one sitting, as helicopters are not edible. They be made of metal, plastic, and other materials, not food!
```

## Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->

The Tulu models have not been aligned to generate safe completions 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 Llama 2 models, 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 hyperparameters

The following hyperparameters were used during training:
- learning_rate: 5e-07
- total_train_batch_size: 32
- total_eval_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 3.0


## Citation

If you find Tulu V2 is useful in your work, please cite it with:

```
TODO
```

*Model card adapted from [Zephyr Beta](https://huggingface.co/HuggingFaceH4/zephyr-7b-beta/blob/main/README.md)*