OpenHermes-2-Mistral-7B is a state of the art Mistral Fine-tune.

OpenHermes was trained on 900,000 entries of primarily GPT-4 generated data, from open datasets across the AI landscape. 

Filtering was extensive of these public datasets, as well as conversion of all formats to ShareGPT, which was then further transformed by axolotl to use ChatML.   Get more info  on huggingface

Zephyr is a series of language models that are trained to act as helpful assistants. Zephyr-7B-α is the first model in the series, and is a fine-tuned version of mistralai/Mistral-7B-v0.1 that was trained on on a mix of publicly available, synthetic datasets using Direct Preference Optimization (DPO). We found that removing the in-built alignment of these datasets boosted performance on 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.

Chat with zephyr