Llama 3 8B

Meta's flagship 8-billion parameter model delivers state-of-the-art performance for local inference and edge deployment.

Llama 3 8B redefines the efficiency frontier for small-form factor LLMs. Built on a 128k token vocabulary and trained on over 15 trillion tokens (a 7x increase over Llama 2), this model excels at reasoning, code generation, and instruction following. It utilizes Grouped Query Attention (GQA) for optimized inference speeds and fits comfortably on consumer-grade hardware like an NVIDIA RTX 4090. Developers leverage this 8B variant for low-latency applications where data privacy and local execution are non-negotiable.

https://llama.meta.com/llama3/