Llama can now see and run on your device - welcome Llama 3.2

Llama 3.2 is out! Today, we welcome the next iteration of the Llama collection to Hugging Face. This time, we’re excited to collaborate with Meta on the release of multimodal and small models. Ten open-weight models (5 multimodal models and 5 text-only ones) are available on the Hub.

Llama 3.2 Vision comes in two sizes: 11B for efficient deployment and development on consumer-size GPU, and 90B for large-scale applications. Both versions come in base and instruction-tuned variants. In addition to the four multimodal models, Meta released a new version of Llama Guard with vision support. Llama Guard 3 is a safeguard model that can classify model inputs and generations, including detecting harmful multimodal prompts or assistant responses.

Llama 3.2 also includes small text-only language models that can run on-device. They come in two new sizes (1B and 3B) with base and instruct variants, and they have strong capabilities for their sizes. There’s also a small 1B version of Llama Guard that can be deployed alongside these or the larger text models in production use cases.

Among the features and integrations being released, we have:

• Model checkpoints on the Hub
• Hugging Face Transformers and TGI integration for the Vision models
• Inference & Deployment Integration with Inference Endpoints, Google Cloud, Amazon SageMaker & DELL Enterprise Hub
• Fine-tuning Llama 3.2 11B Vision on a single GPU with transformers🤗 and TRL

Table of contents

• What is Llama 3.2 Vision?
• Llama 3.2 license changes. Sorry, EU :(
• What is special about Llama 3.2 1B and 3B?
• Demo
• Using Hugging Face Transformers
• Llama 3.2 1B & 3B Language Models
• Llama 3.2 Vision
• On-device
• Llama.cpp & Llama-cpp-python
• Transformers.js
• Fine-tuning Llama 3.2
• Hugging Face Partner Integrations
• Additional Resources
• Acknowledgements

What is Llama 3.2 Vision?

The full list of languages in text-only mode is:

• English
• German
• French
• Italian
• Portuguese
• Hindi
• Spanish
• Thai

Example output