Zephyr 7B - a HuggingFaceH4 Collection

HuggingFaceH4 's Collections

Scaling Test Time Compute with Open Models

Zephyr 7B Gemma

Papers We've Read

Awesome SFT datasets

Awesome feedback datasets

Awesome reward models

Zephyr 7B

updated Apr 12

Models, datasets, and demos associated with Zephyr 7B. For code to train the models, see: https://github.com/huggingface/alignment-handbook

Running on A100

895

🪁

Zephyr Chat

Note Chat with our Zephyr 7B models!
Zephyr: Direct Distillation of LM Alignment

Paper • 2310.16944 • Published Oct 25, 2023 • 123
HuggingFaceH4/zephyr-7b-beta

Text Generation • Updated Oct 16 • 370k • • 1.62k

Note A state-of-the-art chat model at the 7B parameter scale. Trained on synthetic data with a mix of SFT and DPO.
HuggingFaceH4/zephyr-7b-alpha

Text Generation • Updated Oct 16 • 12.7k • • 1.1k

Note The precursor to Zephyr-7B-β. Trained on synthetic data with a mix of SFT and DPO.
HuggingFaceH4/mistral-7b-sft-beta

Text Generation • Updated Sep 24 • 8.09k • 25

Note The SFT model used for the DPO training of Zephyr-7B-β
HuggingFaceH4/mistral-7b-sft-alpha

Text Generation • Updated Oct 26, 2023 • 439 • 3

Note The SFT model used for the DPO training of Zephyr-7B-α
HuggingFaceH4/ultrachat_200k

Viewer • Updated Oct 16 • 515k • 13.4k • 486

Note The SFT dataset used to train Zephyr-7B-β
HuggingFaceH4/ultrafeedback_binarized

Viewer • Updated Oct 16 • 187k • 6.58k • 254

Note The dataset of AI preferences used to train Zephyr-7B-β with DPO