ByteResearch/Llama-3-8B-Instruct

The ByteResearch/Llama-3-8B-Instruct is an 8 billion parameter instruction-tuned generative text model developed by Meta, part of the Llama 3 family. Optimized for dialogue use cases, it utilizes an optimized transformer architecture with Grouped-Query Attention (GQA) and was trained on over 15 trillion tokens. This model excels in assistant-like chat applications and demonstrates strong performance across various benchmarks, including MMLU and HumanEval, surpassing its predecessor, Llama 2.