Step-by-step guide to deploying Llama 3.2 Vision (11B and 90B) on Amazon SageMaker using the Hugging Face LLM Deep Learning Container powered by Text Generation Inference (TGI). Covers environment setup, retrieving the DLC container URI, hardware requirements for each model size (g5/g6/p4d instance types), deploying the 11B model with the SageMaker Python SDK, and running multimodal inference using the OpenAI-compatible messages API with image URLs or base64-encoded images. Also notes the EU licensing restriction on Llama 3.2 multimodal models.
