huggingface-cli

Login to Hugging Face Hub

$ huggingface-cli login

$ huggingface-cli whoami

$ huggingface-cli logout

$ huggingface-cli env

$ huggingface-cli download --repo-type [repo_type] [repo_id] [filename1 filename2 ...]

$ huggingface-cli upload --repo-type [repo_type] [repo_id] [path/to/local_file_or_directory] [path/to/repo_file_or_directory]

$ huggingface-cli scan-cache

$ huggingface-cli delete-cache

huggingface-cli <subcommand> [ <options> ] [ <arguments> ]

Common subcommands include login, logout, whoami, env, repo, download, and cache.

login
    Log in to the Hugging Face Hub. Prompts for a token or uses an environment variable.

logout
    Log out from the Hugging Face Hub. Clears the local token.

whoami
    Display the currently authenticated user information.

env
    Show environment details relevant to Hugging Face Hub operations, including cache paths and library versions.

repo
    Manage repositories on the Hugging Face Hub. Used for creating, deleting, or cloning repositories.

download
    Download files from a model, dataset, or Space repository on the Hub. Supports specific patterns or full downloads.

upload
    Upload files to a model, dataset, or Space repository on the Hub. Used for pushing local changes.

cache
    Manage the local cache used by Hugging Face libraries. Allows clearing or listing cached files.

refresh-token
    Refresh your Hugging Face authentication token.

lfs-enable-largefiles
    Enable Git LFS for handling large files in local repositories, essential for models and datasets.

scan-for-patterns
    Scan local repositories for common security issues like hardcoded tokens.

convert
    Convert models to different formats (e.g., from `tensorflow` to `pytorch`).

The huggingface-cli is the command-line interface tool for interacting with the Hugging Face Hub. It provides a convenient way for machine learning engineers and developers to manage models, datasets, and Spaces directly from their terminal. This includes authentication, uploading and downloading files, managing repositories, viewing environment information, and configuring local cache settings. It simplifies workflows involving large models and datasets by integrating with the underlying huggingface_hub Python library, making it an essential tool for anyone deeply involved with the Hugging Face ecosystem.

Requires the Python huggingface_hub library to be installed.
Network connectivity is essential for most operations, especially login, download, and upload.
Authentication tokens should be managed securely to prevent unauthorized access to your Hub resources.
Rate limiting may apply for extensive download/upload operations.

The CLI often uses configuration stored in ~/.cache/huggingface/token or environment variables like HF_HOME and HF_TOKEN for authentication and cache location.

HF_HOME: Specifies the directory for caching models, datasets, and other Hugging Face assets.
HF_TOKEN: Directly provides the authentication token, useful for CI/CD environments.

The huggingface-cli login command provides an interactive prompt for entering your Hugging Face token, which is then securely stored locally.

The huggingface-cli tool emerged as a natural extension of the Hugging Face ecosystem, growing alongside the Hugging Face Hub itself. It provides a convenient command-line interface to the functionalities offered by the huggingface_hub Python library. Its development has focused on streamlining common developer workflows, such as authentication, model/dataset sharing, and cache management, making the Hub more accessible and integrated into developer tooling.

git(1), python(1), curl(1)