The Introduction of Nous-Hermes 2 Mixtral 8x7B
Nous Research has introduced the Large Language Model (LLM) called Nous-Hermes 2 Mixtral 8x7B, which represents a significant advancement in artificial intelligence. This model is the first developed by the firm to utilize Reinforcement Learning from Human Feedback (RLHF), marking a milestone in AI capabilities. Additionally, it surpasses the well-known Mixtral Instruct across various benchmarks.
Two Variants with Different Features
The Nous-Hermes 2 Mixtral 8x7B is available in two variants: one with Supervised Fine-Tuning (SFT) and the other with a combination of SFT and Decentralised Policy Optimisation (DPO). The firm has also provided an qlora adaptor specifically designed for the DPO version. These models are accessible to the public through HuggingFace, allowing users to choose the most suitable option for their needs.
Impressive Performance and Prompt Format
The model consistently performs well across various benchmarks, achieving an average score of 75.70% in benchmarks such as ARC Challenge, AGIEval, and BigBench. It excels in tasks like BoolQ, PIQA, and Winogrande. In multi-turn chat discussions, Nous-Hermes 2 utilizes ChatML as the prompt format, enabling steerability by incorporating system prompts to guide the model’s rules, roles, and stylistic choices.
Flexible Quantization Choices
The Nous-Hermes 2 model offers a range of quantization options, including 3-bit and 8-bit quantization, as well as different group sizes and act orders. This flexibility allows users to meet VRAM constraints and inference quality criteria.
Easy Model Download and Usage
Users can download and utilize the model through the Hugging Face Hub Python library. The library supports downloading from multiple branches to accommodate different requirements. The text-generation-webui provides a straightforward overview of the model download process, making it convenient for users to obtain and utilize the model.
Hot Take: Advancing Open-Source AI
The Nous-Hermes 2 Mixtral 8x7B represents a significant advancement in open-source artificial intelligence. With its superior performance and user-friendly design, it bridges the gap between proprietary and open-source AI solutions, making it an appealing choice for various artificial intelligence applications.
By
By
By
By