All Local AI
Pick VRAM for fit estimates
Models Tools Hardware Best Of Compare Calculators Stacks Guides Blog
download tool Apache-2.0 linux macos windows

Hugging Face Hub

Official client for downloading model weights, with resumable transfers and selective file patterns.

GitHub stars

3,988

License

Apache-2.0

Install

pip

Latest release

Oct 8, 2026

Source on GitHub Documentation Official site

Best for

Pulling exactly the files you need from a repository instead of cloning tens of gigabytes you will not use.

Model formats

  • safetensors
  • GGUF

Hardware behaviour

Disk and bandwidth, not VRAM. The local cache deduplicates across repositories, so point HF_HOME at a drive with room before your first large download.

Not sure what your card can hold? Run the fit check — the entry point in our listings is the RTX 4060 8GB at 8.00 GB.

Setup

Install Hugging Face Hub

Copy-paste starting points. Always read the project's own instructions before running anything with elevated permissions.

Install

Hugging Face Hub

pip install -U huggingface_hub[cli]

Log in (for gated models)

Hugging Face Hub

huggingface-cli login

Download one quantization

Hugging Face Hub

huggingface-cli download TheBloke/Llama-2-7B-GGUF llama-2-7b.Q4_K_M.gguf --local-dir ./models

Move the cache

Hugging Face Hub

export HF_HOME=/mnt/big-disk/hf

Quick start

Getting the first model running

Most people waste bandwidth by cloning an entire repository when they need one file. huggingface-cli download REPO FILENAME fetches only what you name, resumes if the connection drops, and stores it in a shared cache.

Gated repositories — Llama and several others — require accepting the licence on the website and then huggingface-cli login with a token. Downloads fail with a 401 until both steps are done.

FAQ

Hugging Face Hub questions people actually ask

Short answers to the problems that come up most often.

How do I download a single GGUF instead of the whole repo?

Name the file: huggingface-cli download REPO FILE.gguf --local-dir ./models. You can also use --include with a glob pattern to grab a matching set.

Where are downloads cached, and how do I move them?

By default in ~/.cache/huggingface. Set HF_HOME to relocate the whole cache, which is worth doing before you fill a system drive.

Why do I get a 401 or 403 on some models?

The repository is gated. Accept its terms on the model page while signed in, then authenticate locally with huggingface-cli login using an access token.

Can downloads be resumed?

Yes. Interrupted transfers resume from the cache on the next run, which matters when a model is tens of gigabytes on an unreliable connection.

Alternatives

Compared with other download tool options

Same job, different tradeoffs.

inference engine

Ollama

One-command local model runner with an OpenAI-compatible API and automatic GPU offload.

 inference engine

llama.cpp

The C/C++ engine that made CPU and mixed CPU/GPU inference practical, and the origin of the GGUF format.

 inference server

vLLM

High-throughput serving engine with PagedAttention and continuous batching, built for concurrent requests.

Next step

A runner is only half the decision. Check which models fit your GPU with the VRAM calculator, then follow a setup guide or pick a ready-made stack.