|
--- |
|
title: VLM Demo |
|
sdk: docker |
|
license: mit |
|
--- |
|
|
|
This demo illustrates the work published in the paper ["Prismatic VLMs: Investigating the Design Space of Visually-Conditioned Language Models"](https://arxiv.org/pdf/2402.07865.pdf) |
|
|
|
|
|
# Source code |
|
|
|
For more information, please refer to this repository: |
|
|
|
> *VLM Demo*: Lightweight repo for chatting with VLMs supported by our |
|
[VLM Evaluation Suite](https://github.com/TRI-ML/vlm-evaluation/tree/main). |
|
|
|
# Huffing Face Space architecture |
|
|
|
Hugging Face Space build a container image based on the `Dockerfile`. In this file, we use the base Nvidia base image and install additional packages and external repositories. |
|
|
|
The Hugging Face Space start the container and execute `startup.sh`. The script loads each model on a separate GPU of the 4xA10G. Then it launches several processes: one for each model, the Gradio API controller and frontend. |
