Investigating Llama 3 together with Mixtral

We are building an AI Assistant, where the choice of LLM models is fundamental for the quality of the result. The models must meet several requirements. One is that they must be open-source and licensed to install on-premises. The other is that they must support on-premises fine-tuning. In this study, we have examined the performance of Llama 3 70b in comparison to Mixtral 8x7b on the Gaudi 2 server. The results are preliminary but surprising.

Introduction to LLaMA

The world of artificial intelligence (AI) is continuously evolving, with language models playing a pivotal role in this technological revolution. One of the latest advancements in this domain is Meta AI’s introduction of LLaMA (Large Language Model Meta AI). This article explains the basic concepts of LLaMA, how it works, its use cases, and its potential for the future of AI.

Key Concepts of LLaMA

LLaMA is a collection of foundation language models that were significantly impacted due to their smaller size yet exceptional performance. These models range from 7 billion to 70 billion parameters, which is notably smaller compared to other state-of-the-art models like GPT-3 (175 billion parameters) and PaLM (540 billion parameters). Despite their compact size, LLaMA models deliver impressive results on…