Nous Hermes 2 Mixtral 8x7B DPO

The Nous Hermes 2 - Mixtral 8x7B (DPO) model is the latest flagship from Nous Research, trained on the Mixtral 8x7B MoE. The model was trained on 1,000,000 entries of GPT-4 generated data and other high-quality open datasets. This version introduces alignment methods like supervised finetuning (SFT) and Direct Preference Optimization (DPO) variant of Mixtral Hermes 2, providing users with options for optimal performance. The outputs showcase the model's abilities, from writing code for data visualization to creating cyberpunk psychedelic poems and performing backtranslation for prompt creation. The model surpassed the flagship Mixtral Finetune model by MistralAI across various benchmarks.

Model card on Hugging Face