Show HN: HuggingFaceModelDownloader v2.0–resumable, filterable HF Downloads

1 points by bodaay 7 hours ago

I just released v2.0 of HuggingFaceModelDownloader, a Go-based CLI for downloading models and datasets from the Hugging Face Hub.

Key features in this release:

Filters for selective downloads → fetch only what you need (e.g. specific GGUF quant variants like q4_0, q5_0), with optional auto-subdir layout.

Resumable by default → SHA-256 verified for LFS files, size checks for others, multipart range resume per part.

Terminal UX → live TUI with per-file progress, speeds, ETA; plain text fallback in constrained environments.

CI-friendly → structured --json progress events, concurrency limits, retries with backoff, dry-run planning.

No meta files → resuming is purely based on what’s on disk.

Compared to huggingface-cli/snapshot_download, this emphasizes resiliency, filtering, and machine-readable outputs for automation.

Example with filters:

hfdownloader download TheBloke/Mistral-7B-Instruct-v0.2-GGUF:q4_0,q5_0 \ --append-filter-subdir -o ./Models -c 8 --max-active 3