README.md

---
license: other
library_name: transformers
datasets:
- HuggingFaceH4/ultrafeedback_binarized
base_model: wandb/gemma-2b-zephyr-sft
license_name: gemma-terms-of-use
license_link: https://ai.google.dev/gemma/terms
---

[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/llm_surgery/gemma-zephyr)

# Gemma 2B Zephyr DPO

The [Zephyr](https://huggingface.co/HuggingFaceH4/zephyr-7b-beta) DPO recipe applied on top of SFT finetuned Gemma 2B

## Model description

- **Model type:** A 8.5B parameter GPT-like model fine-tuned on a mix of publicly available, synthetic datasets.
- **Language(s) (NLP):** Primarily English
- **Finetuned from model:** [wandb/gemma-2b-zephyr-sft](https://huggingface.co/wandb/gemma-2b-zephyr-sft/)

## Recipe

We trained using the DPO script in [alignment handbook recipe](https://github.com/huggingface/alignment-handbook/blob/main/scripts/run_dpo.py) and logging to W&B

Visit the [W&B workspace here](https://wandb.ai/llm_surgery/gemma-zephyr?nw=nwusercapecape)


## License
This model has the same license as the [original Gemma model collection](https://ai.google.dev/gemma/terms)

## Compute provided by [Lambda Labs](https://lambdalabs.com/) - 8xA100 80GB node

around 13 hours of training