README.md

---
pipeline_tag: text-generation
inference: false
tags:
- zephyr
- mlx
language:
- en
license: mit
library_name: mlx
---

# Zephyr 7B β (✨ 4-bit)

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. You can find more details in the [technical report](https://arxiv.org/abs/2310.16944).

This repository contains the `zephyr-7b-beta` weights in `npz` format in 4-bit suitable for use with Apple's MLX framework (from 0.6.0 onwards).

## Use with MLX

```bash
pip install mlx
pip install huggingface_hub hf_transfer
git clone https://github.com/ml-explore/mlx-examples.git
cd mlx-examples

# Download model
export HF_HUB_ENABLE_HF_TRANSFER=1
huggingface-cli download --local-dir-use-symlinks False --local-dir zephyr-7b-beta-4bit mlx-community/zephyr-7b-beta-4bit

# Run example
python llms/mistral/mistral.py --model-path zephyr-7b-beta-4bit --prompt "My name is"
```

Please, refer to the [original model card](https://huggingface.co/HuggingFaceH4/zephyr-7b-beta) for more details on Zephyr 7B β.

## Prompt Format

Please note that this model expects a specific prompt structure. Here is an example:

```
<|system|>
You are a pirate chatbot who always responds with Arr!</s>
<|user|>
There's a llama on my lawn, how can I get rid of him?</s>
<|assistant|>
```