Meta Llama 3 (Deprecated)

What is the project about?

Meta Llama 3 is a collection of large language models (LLMs) released by Meta, ranging in size from 8 billion to 70 billion parameters. It includes both pre-trained and instruction-tuned models. This specific repository provides a minimal example for loading these models and performing inference. It is now deprecated, with functionality and further development moved to other repositories.

What problem does it solve?

• Provides access to powerful, state-of-the-art LLMs for a wide range of users (individuals, creators, researchers, businesses).
• Enables experimentation, innovation, and scaling of AI-powered applications.
• Democratizes access to large language model technology.
• The instruction-tuned models are specifically designed for dialogue-based applications.

What are the features of the project?

• Pre-trained Models: Foundation models suitable for various natural language processing tasks where the expected output is a continuation of the input.
• Instruction-tuned Models: Models fine-tuned for chat and Q&A, following a specific conversational format.
• Model Sizes: Offers models with 8B and 70B parameters, catering to different computational resources and performance needs.
• Sequence Length: Supports a sequence length of up to 8192 tokens.
• Hugging Face Integration: Provides downloads and integration with the Hugging Face Transformers library for easier use.
• Responsible Use Guidance: Includes a Responsible Use Guide and mechanisms for reporting issues and risky content.
• Safety Classifiers: The ability to add safety classifiers to filter inputs and outputs.

What are the technologies used in the project?

• PyTorch: The primary deep learning framework.
• CUDA: For GPU acceleration.
• Hugging Face Transformers: A popular library for working with pre-trained language models (optional, but supported).
• Python: The main programming language.
• wget and md5sum: Used for downloading the model weights.
• torchrun: Used for distributed training and inference.

What are the benefits of the project?

• Openness: The models and weights are licensed for both research and commercial use, promoting open development.
• Accessibility: Makes powerful LLMs available to a broader audience.
• Flexibility: Supports both basic text completion and interactive chat applications.
• Scalability: Offers different model sizes to suit various needs.
• Community Support: Encourages community contributions and provides channels for reporting issues.
• Responsible AI: Includes guidelines and tools for responsible use and safety.

What are the use cases of the project?

• Text Completion: Generating text continuations, writing different kinds of creative content, summarizing factual topics or creating stories.
• Chatbots and Conversational AI: Building interactive dialogue systems.
• Question Answering: Developing systems that can answer questions based on provided context or knowledge.
• Code Generation: (Potentially, with appropriate prompting and fine-tuning).
• Research: Studying and advancing the field of large language models.
• Content Creation: Assisting with writing articles, scripts, or other text-based content.
• Education: Providing tools for learning and exploration.
• Synthetic Data Generation: Creating synthetic data for training other models.

Important Note: This original repository is deprecated. Users should now refer to the following repositories for continued development and specific functionalities:

• llama-models: Foundation models, utilities, model cards, license.
• PurpleLlama: Safety risks and inference-time mitigations.
• llama-toolchain: Model development (inference, fine-tuning, safety).
• llama-agentic-system: End-to-end Llama Stack system for agentic applications.
• llama-cookbook: Community-driven scripts and integrations.

</p>