Nous-Hermes 2 Mixtral 8x7B Surpasses Mixtral Instruct in Benchmark

Spread the love

The Large Tongue Model (LLM) known as the Nous-Hermes 2 Mixtral 8x7B was recently introduced by Nous Research when it was released. An important step forward in AI capabilities is represented by the fact that this advanced model is the first developed by the company to be enhanced using Reinforcement Learning from Human Feedback (RLHF). In addition, it is the first model that exceed the well-known Mixtral Instruct in a wide range of prominent indicators.

The Nous-Hermes 2 Mixtral 8x7B comes in two different variants: the first is simply equipped with Supervised Fine-Tuning (SFT), while the second is a more advanced combination of SFT and Decentralized Policy Optimization (DPO). An additional qlora adapter that is compatible with the DPO version is also provided by the company. Through HuggingFace, these models are available to the general public, giving people the opportunity to try and choose which choice best suits their needs.

The model’s performance is consistently good across benchmarks, with an average score of 75.70% in the ARC Challenge, AGIEval and BigBench benchmarks. In particular, it was able to achieve a high level of accuracy in tasks such as BoolQ, PIQA and Winogrande. When it comes to engaging LLMs in multi-channel chat discussions, Nous-Hermes 2 uses ChatML as a fast format that provides a more systematic way to do so. This format includes system prompts that enable management, thereby guiding the model’s rules, roles, and stylistic choices.

In order to satisfy a wide range of VRAM constraints and inference quality criteria, the Nous-Hermes 2 model offers a variety of quantization options, including 3-bit and 8-bit quantization, as well as a range of group and row sizes. action available.

Users can download and use the model using the Hugging Face Hub Python library. Additionally, the library allows pulling from multiple branches to meet different requirements. Those using text-generation-webui get an overview of a simple model download procedure, making it easy to obtain and use the model.

ALSO READ  How LLM Is Reshaping Agent-Based Modeling and Simulation

Summing it all up, the Nous-Hermes 2 Mixtral 8x7B is a big step forward in the development of open source artificial intelligence. It bridges the gap between proprietary and open source AI solutions due to its better performance and user-friendly design, making it an attractive alternative for AI applications.

Image source: Shutterstock

Leave a Comment