SmolVLM

Tiny 2B vision-language model from HuggingFace — runs on 8GB VRAM, Apache 2.0.