Llama3-S

Llama 3-S: The optimized, high-performance large language model (LLM) variant designed for speed and efficient deployment on consumer-grade hardware.

This is the Llama 3-Small class: a family of highly efficient, openly available LLMs from Meta, engineered for superior performance on resource-constrained systems. Specifically, models like the 8B parameter version and the ultra-compact Llama 3.2 (1B and 3B parameters) deliver state-of-the-art results while maintaining a small footprint. They feature an 8K context window and leverage grouped-query attention (GQA) for faster inference, making them ideal for local deployment, edge computing, and rapid application development without sacrificing critical reasoning or code generation capabilities. We’ve seen them outperform larger models on key industry benchmarks (e.g., MMLU, ARC), proving that efficiency does not mean compromise.

https://llama.meta.com/