Model Zoo

Live: huggingface.co/spaces/CERN/colliderml-model-zooSource: spaces/model-zoo/

A browser for HuggingFace models tagged colliderml. Discovery is entirely tag-based: the Space calls huggingface_hub.list_models(filter="colliderml", limit=500, full=True) and lays the results out in a sortable table.

Getting your model listed

Add the colliderml tag to your model card and (optionally) an appropriate task tag:

---
tags:
- colliderml
- tracking        # or jets, anomaly, or omit for "general"
task_category: object-detection
---

Re-push the model — the Space picks up the new tag on its next refresh (tag queries go through HuggingFace's live API; no re-deploy needed).

Columns

Column	Source
model	m.modelId (repo name)
task	First of tracking / jets / anomaly in the tag list, else general
downloads	HuggingFace download count
likes	HuggingFace like count
pipeline	m.pipeline_tag
url	Direct link to the model card

The default sort is by download count, descending. The Filter by task dropdown lets you narrow to one of the three physics tasks or to general (everything else).

Authentication

Public models only — no HuggingFace sign-in needed.

Running it locally

cd spaces/model-zoo
pip install -r requirements.txt
python app.py

Limitations

• The Space caps the query at 500 models per refresh. If you have more than 500 models tagged colliderml (a nice problem to have), bump the limit= argument in fetch_models().
• Task detection is naïve — it looks for a literal tag match. If you use different tags (e.g. track-reco instead of tracking) your model shows up under general.

See also

• Benchmark Tasks guide — the tasks users train against when they publish a model
• Leaderboard — where shipped models go to battle