Hugging Face CLI

Use the current hf CLI for Hub operations. Do not use the deprecated huggingface-cli command.

Workflow

1. Run hf version and hf --help before relying on a command that may vary by installed version.
2. Use public read-only operations without authentication when possible.
3. For private or write operations, run hf auth whoami without printing stored tokens.
4. Prefer structured output such as --format json when another tool will consume the result.
5. Use --dry-run when supported before downloads, cache cleanup, synchronization, or deletion.
6. Confirm the repository, revision, visibility, and affected paths before uploads or destructive changes.

Common operations

hf models info namespace/model --format json
hf models list --search "query" --limit 20 --format json
hf datasets info namespace/dataset --format json
hf datasets list --search "query" --limit 20 --format json
hf download namespace/model --include "*.json" --dry-run
hf cache list --limit 30
hf papers list --limit 20 --format json

Use hf upload for a bounded file or directory and hf upload-large-folder for resumable large uploads. Never place a token in command arguments, source files, logs, or chat output. Use the authenticated credential store or HF_TOKEN environment variable.

If hf is unavailable, report that prerequisite clearly. Do not install software unless the user requested installation.