samuel-vitorino/lm.rs: Minimal LLM inference in Rust

A Rust-based tool, lm.rs, enables local CPU operation of language models without ML libraries. It supports Gemma 2, Llama 3.2, and PHI-3.5 with vision capabilities, offering various quantized sizes. Speeds range from 7 to 35 tokens per second on a 16-core AMD Epyc processor. Users can download pre-prepared models or convert their own. The project aims to increase AI accessibility, potentially running advanced models on basic hardware in the future.