---
title: Agent
emoji: 🌖
colorFrom: red
colorTo: gray
sdk: docker
pinned: false
---

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

SOTA open VLM is [InternVL-1.5](https://huggingface.co/spaces/opencompass/open_vlm_leaderboard), which is *22B*, for practical deployment I choose moondream which is a model can answer real-world questions about images (378x378). It's tiny by today's models, with only *1.6B* parameters. That enables it to run on a variety of devices, including mobile phones and edge devices.