Skip to main content

hugging-face-quickstart

Last updated on

Prerequisites​

  • Install the Hugging Face CLI (pip install -U huggingface_hub gives you the hf command).
additional information

Depending on your environment, you may need to run addition command, go to Hugging Face installation for more information.

  • Access to a Harness account with permissions to create registries and generate identity tokens.

Create a Hugging Face artifact registry​

Start by creating a new Hugging Face registry in your Harness project.

  1. Under Registries, click New Artifact Registry.
  2. Select Hugging Face as the registry type.
  3. Provide a Registry Name.
registry name criteria

Your registry name must start with a letter and can include lowercase alphanumerics, _, . and -.

  1. (Optional) Add a Description and Labels.
  2. Choose visibility between Public and Private.
  3. Click Create Registry.
private registry

This registry will serve as your private Hugging Face registry within Harness.

No upstream proxy

Hugging Face registries do not proxy to the public Hub. If content is not present in your Harness registry, it will not automatically resolve from huggingface.co. Teams can switch endpoints when needed or pre-populate the registry with required artifacts. Client-side caching still applies.

Configure client​

Configure environment variables​

  1. Set up environment variables to connect to Harness Artifact Registry.
# Recommended timeouts for model/dataset resolution.
export HF_HUB_ETAG_TIMEOUT=86400
export HF_HUB_DOWNLOAD_TIMEOUT=86400

# Point the client at your Harness registry endpoint (example shown).
export HF_ENDPOINT="https://pkg.<env>.harness.io/pkg/<HARNESS_ACCOUNT_ID>/<REGISTRY_NAME>/huggingface"
Client version

For Hugging Face client version 0.19.0 and above, this timeout setting enables model resolution via pipelines and tokenizer libraries.

Configure authentication​

  1. Click Generate token to generate an identity token.
  2. Copy the token.
  3. Set the token as an environment variable with the following command:
export HF_TOKEN="<token from step 2>"

Deploy models and datasets​

Upload a model to Artifact Registry using the huggingface_hub library:

from huggingface_hub import HfApi
api = HfApi()
api.upload_folder(
    folder_path="<folder_name>", # local folder containing model files
    repo_id="<ARTIFACT_NAME>", # name for the model in the registry
    revision="<VERSION>", # model version (defaults to 'main')
    repo_type="model"
)

Resolve models and datasets​

Download a model from Artifact Registry:

from huggingface_hub import snapshot_download
snapshot_download(
    repo_id="<ARTIFACT_NAME>", revision="<VERSION>", etag_timeout=86400
)
model resolution

With Hugging Face client version 0.19.0+ and HF_HUB_ETAG_TIMEOUT set, you can resolve models with transformers, diffusers, and other libraries.

Troubleshooting​

  • Slow downloads or timeouts. Increase HF_HUB_ETAG_TIMEOUT and HF_HUB_DOWNLOAD_TIMEOUT. Verify network egress to the Harness endpoint.
  • File too large. Ensure Git LFS is used automatically by the CLI; confirm the LFS filter in your repo and re-try the upload.