meta-llama/llama3

This is the official but now-deprecated GitHub repository for Meta's Llama 3 large language models. A large language model (LLM) is an AI system trained on vast amounts of text that can generate, summarize, translate, and answer questions in natural language. The repository provided model weights, the trained numerical parameters, and minimal starter code for running those models locally. Llama 3 was released in sizes of 8 billion and 70 billion parameters. Larger models generally produce more capable outputs but require more memory and computing power. To run the 70-billion-parameter version, for example, you needed 8 GPUs working in parallel. The models came in two forms: pretrained versions that continue text naturally, and instruction-tuned versions fine-tuned to respond to conversational prompts. This repository has since been superseded. Meta split its model infrastructure across several dedicated repositories, one for the core model files, one for safety tools, one for fine-tuning and inference tooling, and one for agent-based applications. The README directs users to those newer repos instead. You would have used this repository if you wanted to run a Llama 3 model on your own hardware using Python, or if you wanted to download the raw model weights for research. Today, the successor repositories serve that purpose. The tech stack is Python, with PyTorch required for running the models.