Skip to content

Pull your first model

A Pull fetches a Repo, or some of its files, from an Upstream such as huggingface.co into the Arsenal. It keeps the upstream ID and Commit SHAs, so code that names org/model works against Hangar unchanged.

From the web UI

  1. Open Discover and search by name, or enter a full Repo ID such as hf-internal-testing/tiny-random-bert.
  2. Open a result, or for a Repo ID select Open hf-internal-testing/tiny-random-bert. huggingface.co search does not list every Repo (it leaves out hf-internal-testing, for example), so a Repo ID opens the Repo directly. Hangar shows the files, their sizes, any GGUF quants and the Dependencies it would pull.
  3. Select Pull into arsenal. The Pull runs as a Job; follow it under Jobs.

For a GGUF Repo, the quant picker lets you pull one quantisation, such as Q4_K_M, instead of every file. The Repo then becomes a Partial mirror: the full history is there, and files you did not pull show as Not in arsenal with a Pull action.

From the CLI

hangar pull hf-internal-testing/tiny-random-bert
hangar pull bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M

hangar pull asks the Instance to Pull what is missing, shows the progress, then downloads the files into your local Hugging Face cache, where Python libraries find them. Add --server-only to fill the Arsenal without downloading to your machine.

Gated and private Repos

Pulls of gated or private Repos need your huggingface.co token. Save it under Settings, Upstream tokens; Pulls you start use it. See Gated models.

Next: Point HF_ENDPOINT at Hangar.