Using & Downloading Models

Every model has a detail page at /models/<slug> with three things you'll use as a consumer: the model card, the file browser, and the clone panel.

Read the model card

The card (rendered from the repo's README.md) is where you decide if a model fits: intended use, training data, benchmarks, limitations and license. Always check the license before downloading — it governs how you may use the weights.

Browse files

The file browser on the model page lists everything in the model's object storage — checkpoints, tokenizers, configs. You can preview small text files in the in-browser code editor and download any file directly.

Download

In the browser

Open the model page, find the file in the browser, and click download.

With curl

Each file resolves through a stable URL that 302-redirects to a presigned / CDN link:

# $API = https://api.chaimind.in (or your local backend)
curl -L "$API/api/v1/models/<slug>/resolve/<path>" -o model.safetensors

-L is required so curl follows the redirect to the actual bytes. Public models need no auth; private ones require -H "Authorization: Bearer <token>".

List files via the API

curl "$API/api/v1/models/<slug>/files"

Returns the file list (paths + sizes) backing the browser.

Clone the code repo

Code and docs (README, config.json, tokenizer configs) live in an auto-provisioned git repo, separate from the large weights. Copy the URL from the clone panel on the model page:

git clone <clone-url>

Two stores, by design. Text/code → git (clone it). Large binaries → object storage (download via resolve). See Publishing Models for the full split.

Versions & history

The Versions panel on the model page lists the repo's commit history — each entry shows the author, commit message and how long ago it landed, and links to the full diff on the git server. Use it to see what changed between checkpoints or to pin to a specific commit when cloning.

Run it

To actually load the model — with 🤗 Transformers, an inference server (vLLM / Ollama) or the Docker Model Runner — see Running & Integrating Models. The model page surfaces copy-paste snippets for each.

Fine-tune / fork

To build on a model, clone its repo, download the weights you need, and publish your result as a new model (crediting the original in your card). See Publishing Models.

Next: Browsing & Discovery · Publishing Models · Import from GitHub