SmolLM is a series of efficient, on-device capable AI models from Hugging Face spanning text and vision modalities. SmolLM3 is a 3B parameter language model trained on 11T tokens that achieves state-of-the-art results at its scale while supporting multilingual input across six languages and extended 128k context windows. SmolVLM is a compact multimodal model capable of processing images and text for tasks like visual QA and image description. The repository provides model weights, training configurations, and inference tools.