VIDRAFT

VIDRAFT

The home page of VIDRAFT, a Seoul-based AI model lab, as a static site. It presents its open models and tools for training, merging, compressing, testing and speeding up language models.

Made by VIDraft 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 the home page of VIDRAFT, an AI company in Seoul. It publishes open language models and tools for checking them. The page is a static site, with pages, images, a paper PDF and a Darwin model list in the repository. The page shows four ways it makes models: pretraining from scratch, merging, cross-architecture breeding and compression.

Who it suits: Machine learning researchers and engineers who follow open model labs and want one overview of what VIDRAFT offers. People who want to run large models on a PC or phone can look at the compression line.

Industry | Who uses it | How they use it | Example
AI research | Model researchers | Read about the merging, pretraining and verification approaches | Looking at how the Darwin family is described
On-device AI | Developers running models on PCs and phones | Look at the compression line for local use | Checking which model sizes POCKET covers
Security | Testers of code and AI agents | Look at the proof-based testing tool | Reading what Chitos is meant to test
Data and databases | Analysts | Look at the natural language to SQL tool | Reading how ViSQL checks queries by running them

If you want a tool to run right now, you can skip it, because the page is an overview of a lab.

Where to use it:

• A web browser, on the Space's page at Hugging Face
• Windows, macOS or Linux, by opening a downloaded index.html in any browser

Use case | Who | What they get back
Survey a model lab | Researchers | A table of model lines and tools with what each one does
Find the named models and datasets | Engineers | Six model names and six dataset names to look up on Hugging Face

How to set it up: Online use needs nothing, since the page runs in the browser. To keep a copy, you need little.

• Python and pip, only to download the files with the Hugging Face tool
• A web browser to open the page

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

pip install -U huggingface_hub
hf download VIDraft/AI --repo-type space --local-dir AI
cd AI

How to use it: Open the page and read through the product lines. The table below lists the main ones as the page describes them.

Tool | What it does
Darwin | Merges models by evolution without any training
AETHER | A foundation model pretrained from scratch with a mix of 5, 7 and 11 attention types, with open weights, code and logs under Apache 2.0
POCKET | Runs models from 26B to 180B on a PC, laptop or phone through stock llama.cpp, with no GPU needed
ZTC | Judges how far an answer can be trusted without generating any tokens
VKAE, VKUE and VKIE | Speed up inference on a GPU, a CPU and an NPU
NPU (VKORE) | Runs inference, LoRA tuning and pretraining on NPUs made only for inference
AX-RAY | Diagnoses an LLM for safety, causality and trust before deployment
Chitos | Security tests for code and AI agents that rest on proof
ViSQL | Turns questions into SQL and checks the result by running it
CellOS | Cell simulation that can be differentiated, plus research agents that correct themselves

Example: You open the page and read the POCKET entry, which says big models can run on an ordinary PC, laptop or phone through llama.cpp, with no graphics card. You then look up the model names the Space lists on Hugging Face.

For newcomers and experts: If you are new to this, open the Space's page and read the opening paragraph and the product table. Nothing needs installing, and you get a map of what this lab works on.

If you know the field, the page reports benchmark and leaderboard results for the Darwin family and others. These are the lab's own claims, and the page also links the project to a paper PDF, so check them against the original leaderboards. The page itself is plain static files, with models and datasets named for you to fetch on Hugging Face.

Open the page in a browser
Read the lab's four ways of making models
Browse the product table
Look up the named models and datasets

Good to know: The numbers on the page are the lab's own reports, and the source gives no way to check them here.

• License: Apache 2.0, as the Space's license line says
• Cost to run: none named for the page itself. Running the models it names would depend on their own terms
• Data it reaches: it names six models and six datasets on Hugging Face, and its README mentions the Hugging Face public API for usage figures

On Hugging Face, 13 November 2026

• Likes: 24
• Built with: static
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/VIDraft/AI
• Files: https://huggingface.co/spaces/VIDraft/AI/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.