ChatHFD: model and dataset downloads
Download public Hugging Face repositories with one command. HF-Mirror is the default endpoint; aria2 provides split and multi-file concurrency. Preview the selected files and size before transferring them.
-
Start here
Install the package, check aria2, and download a model config and a small math dataset.
-
Find a command
Real CLI trees, every option, defaults, exit behavior and installation confirmation.
-
Choose endpoint and output
Endpoint precedence, file filters, concurrency, pinned revisions and resume identity.
-
Call from Python
Reuse metadata and download functions with their exact arguments, output and errors.
-
Understand the boundary
Public versus authenticated repositories, compatible mirrors and the separate ModelScope platform.
-
Resolve a failure
Missing dependencies, destination conflicts, network errors and legacy HFD flags.
Choose by task
| Task | Starting point | File effects |
|---|---|---|
| Inspect repository files and size | download --dry-run |
No download files created |
| Download model config or weights | Model example | Writes selected files |
| Download a math dataset | Dataset example | Writes repository files |
| Check or install aria2 | setup |
Read-only unless installation is confirmed |
| Understand a rejected directory | Output and resume | Existing files are not deleted |
| Map commands to implementation | Interface Tree | Reference only |
Minimal working path
pip install ChatHFD==0.1.6
chathfd setup
chathfd download Qwen/Qwen3.8-Flash-Next --include config.json --dry-run
This only queries metadata for a config file, not model weights. Continue with Quickstart.
Current boundary
Public repositories only; no account token input or forwarding. Completion checks file sizes, not upstream cryptographic hashes. The 10 GiB gate is metadata preflight on the file selection, not a hard network-byte quota.