Hugging Face Hub

tl;dr: huggingface.co is the catalog of open models. huggingface_hub (Apache-2.0, 3,880 GitHub stars this fetch) is the official CLI and Python client. This page is the Hub query — not a scrape of every GGUF.

What it does

The Hub hosts model repos (weights + a model card), datasets, and Spaces. People search huggingface, huggingface models, and named weights (Qwen, DeepSeek-R1, Llama). You pick a repo_id, download a revision, then run it in Ollama, llama.cpp, or transformers.

The GitHub README calls huggingface_hub “the official CLI and Python client for the Hugging Face Hub.” Docs: huggingface.co/docs/huggingface_hub.

Install

pip install -U huggingface_hub

From 0.32.0 the extra hf_xet path ships with the client for faster Hub transfers (Hub docs: Using Xet Storage). Older 0.30–0.32 installs need hf_xet explicit.

Download a model

CLI (same helpers as the Python API; prints the local path):

hf download org/model
hf download org/model filename.gguf

Python, from the download guide this session:

from huggingface_hub import hf_hub_download, snapshot_download

hf_hub_download(repo_id="org/model", filename="config.json")
snapshot_download(repo_id="org/model")

Default is repo_type="model" and revision main. Pin revision= to a tag, PR, or commit. Cache lives under HF_HOME / HF_HUB_CACHE. HF_HUB_OFFLINE=1 skips Hub HTTP and uses cache only.

List by downloads:

hf models ls --sort downloads --limit 10

How this sits next to Ollama

We do not ship a URL per Hub file. Named-model pages are Planner cells (hundreds, not every GGUF). Promote a new model the same way: fetch the card, gold-bar page, keep it on /local-llm.

Gotchas

• Stars above are huggingface_hub, not the huggingface.co website (no public star count on the site).
• hf download still hits the network to check for a newer file unless HF_HUB_OFFLINE=1.
• Gated models need hf auth login (or a token). The docs do not make gated weights public.
• A Hub repo_id is not an Ollama library name. Check the model card for GGUF vs safetensors vs “use Ollama.”

Related: best local LLM, Ollama, llama.cpp, LM Studio, Open WebUI.

Untested here — facts from GitHub and Hugging Face docs fetched 2026-09-10.