d3-turbo: beyond one question at a time

d3-turbo: beyond one question at a time

A web page that shows d3-turbo, a model that reads one picture, clip or text and answers many questions about it together, with scenes and five live demos. For engineers and researchers.

Made by vllm-sr on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: This Space is a showcase page for d3-turbo, a Decision 3 release from the vLLM Semantic Router project. The model takes one picture, one clip or one text as input, then answers a whole set of questions about it in a single pass. The page opens with real requests to the free Decision API, then runs five live demos.

Who it suits: Engineers and researchers who work with vision and language models, and app builders who want to see what a single pass over an input can answer.

Industry | Who uses it | How they use it | Example
Computer vision | ML engineers | See boxes with a confidence for things you name | Open the demo with four photos
Research | Model researchers | Compare d3-turbo with d3-flash on timing | Run the demo with 32 questions on one photo
Product development | App builders | See what one pass over an input can answer | Click a scene to read its questions and answers
Games | Game developers | Look at scenes drawn from open-source games | Open one of the game scenes

People who want a model they can download and run offline can skip it, since the page talks to a hosted API.

Where to use it:

• A browser: the Space runs on Hugging Face with nothing to install
• Windows, macOS or Linux: open the downloaded index.html in a web browser

Use case | Who | What they get back
Read the scenes | Engineers | Real questions and answers from the API for each scene
Race two models | Researchers | Times for d3-turbo and d3-flash, measured in your browser
See tracking | App builders | Objects followed through short clips

How to set it up:

• Python with pip on Windows, macOS or Linux, only to download the files
• A web browser
• Nothing at all if you only use the Space in a browser

1. Install the Hugging Face hub package.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Open index.html in a web browser.

pip install -U huggingface_hub
hf download vllm-sr/d3-turbo --repo-type space --local-dir d3-turbo
cd d3-turbo

How to use it: Open the page and click any scene to see the questions asked and the answers returned. Then try the five live demos.

• Many questions in one message: 64 questions about one short text, with d3-turbo timed against d3-flash
• Many questions in one look: 32 questions about one photo, timed the same way
• Pictures as options: three puzzles where you choose among images
• Boxes and scores: four photos where any object you name gets a box and a score
• Tracks through time: birds and a kitten's ball are tracked as the clips play

Example: You open the page, click a scene to read its questions and answers, then run the demo with 32 questions on one photo. The page shows how long d3-turbo and d3-flash each took, measured in your browser.

For newcomers and experts: If you are new, open the Space and click through the scenes first. Each one shows a real request and what the model returned, so you see the idea before you run anything.

For experts, the page sets up its own no-cost key inside your browser, then sends d3-turbo a single request as proof the service is up. If the API is out of reach, it replays answers recorded earlier. Boxes, tracks, choices and yes or no answers all come from the same one-pass reading.

You open the page
The page sets up its own free key
It asks d3-turbo about the input once
All the answers come back together

Good to know:

• Live results need the Decision API: when it is out of reach, the page replays recorded answers
• License: apache-2
• Cost to run: none named, the API key is free
• Data it reaches: the free Decision API, plus photos and clips from Wikimedia Commons and open-source games, credited on the page
• Not for: offline use of the model, since the page calls an API

On Hugging Face, 13 November 2026

• Likes: 9
• Built with: static
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/vllm-sr/d3-turbo
• Files: https://huggingface.co/spaces/vllm-sr/d3-turbo/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.