Llama 3 — открытые LLM-модели от Meta

★ 27k

Llama 3 is Meta's official repository for the open-weight Llama 3 model family: weights, minimal inference code, and a license for local running and fine-tuning. Reach for it when you need a free LLM to deploy on your own hardware, fine-tune for a domain, and embed in a product without depending on a closed API. It provides access to models of various sizes (for example 8B and 70B), instructions to download the weights, and examples for running chat/generation. Its focus is a reference entry point to the Llama 3 models and running them, not a finished app or a high-throughput server: for speed and batching in production use vLLM, for a local chat with no code use Ollama or LM Studio (which also offer quantized builds for weaker hardware), and for fine-tuning on your data, unsloth and text-generation-webui are convenient.