Reference Any Hugging Face Model or Dataset in One Step
For peopleBy glitchdata Team 2 min read
Much of the world's open AI work already lives on Hugging Face. Rather than asking you to re-describe it by hand, glitchdata can now reference it directly.
How It Works
On the New model or New dataset page, paste a Hugging Face ID such as sentence-transformers/all-MiniLM-L6-v2 — or copy the address straight from your browser — and choose Import. glitchdata reads the repository through Hugging Face's public API and creates a draft under your name, or under one of your organisations.
What Comes Across
- The card. The repository's README becomes the card, with its metadata header removed and a note crediting the source at the top.
- The facts. For models: the task, the library it's built with, and the parameter count. For datasets: the file format and the number of rows across its splits. Both bring the licence, a short summary, useful tags, and the exact revision imported.
- The files. Up to 100 files, each with its size, as links to
huggingface.co/…/resolve/main/…. Larger repositories get a note linking to the full list.
What Doesn't
Nothing is downloaded. The weights and data stay on Hugging Face, and when someone downloads from glitchdata they are sent there. Download counts on Hugging Face and likes are not copied either — those belong to the original.
Safe by Default
Imports arrive as drafts so you can read the card before anyone else sees it. From there it follows the usual path: members submit it for review and an admin approves it; admins can publish it directly. Importing the same source twice into the same namespace simply opens the existing entry, and gated repositories — where Hugging Face requires you to sign in and accept terms — carry a clear note so nobody is surprised at download time. Private or missing repositories are refused with a plain explanation.
Why It Matters
A reference that keeps its source is better than a copy that drifts. Importing by ID means the licence, the card and the files all point back to the original maintainers — while glitchdata adds the things a catalogue is for: review, search, previews, and a place alongside the rest of your data.