SmolVLM2-2.2B-Instruct
HuggingFaceTB/SmolVLM2-2.2B-Instruct · Apache-2.0
A vision-language model that fits on a laptop and reads video.
この解説はまだ日本語がありません。英語の原文を表示しています。
SmolVLM2 is a 2.2B VLM that handles images and short video clips with low VRAM, good for document Q&A, UI understanding and on-device perception.
スコア
導入の要点
pip install transformers 深読み
SmolVLM2 is a 2.2B vision-language model that understands both images and short video on consumer hardware — small enough for a laptop GPU, capable enough for real document and image QA.
It is trained on a compact but strong mixture (the Cauldron, Docmatix, LLaVA-OneVision data) and exposes the standard transformers path: AutoProcessor plus AutoModelForImageTextToText.
The family ships with a demo space, a SmolVLM2 highlights space and the vision docs in the smollm repository, so getting from weights to a working app is short.