Get help

Downloading, verifying, and picking the right file — everything in one place. No account, no token, no throttle.

Download a model

Start from the .torrent file on each model page — hand it to your client and the swarm does the rest. Each model page also carries a magnet link that opens the swarm directly. The full three-command walkthrough (download → fetch the signed manifest set → verify the minisign signature) lives on /verify.

Pick a client

• Transmission (desktop) and aria2 (command line) are solid, dependable picks.
• qBittorrent works too — any client that speaks BitTorrent can fetch and seed these torrents. If an initial fetch ever stalls, aria2c is the dependable second choice.
• Leave the client running afterward — a finished torrent seeds automatically. Share ratio 2.0+ is the goal.

Which file do I need?

safetensors vs GGUF: the archive mirrors the upstream .safetensors files verbatim — those are the canonical weights of record. GGUF files are third-party quantized repacks for local inference tools (llama.cpp family); unless a GGUF is listed in the signed manifest, it did not come from this archive.

Picking a quant: if you run a GGUF locally, a Q4 quant needs roughly 0.6 GB of RAM per billion parameters plus overhead; Q8 needs about 1 GB per billion. When in doubt, start one size smaller than you think fits.

Client RAM: the 0.6B model in the current catalog runs comfortably in 1–2 GB at Q4; 7B-class models want ~5 GB at Q4.

Guides

Longer walkthroughs live in the guides section:

• Checking the files you downloaded: verify a model download
• What the link hands your client: magnet links and infohashes
• Torrents as the distribution method: AI model torrents
• A download from huggingface.co that runs but crawls: Hugging Face download slow
• Getting weights somewhere else entirely: download models without Hugging Face
• Refused rather than slow: Hugging Face 429 and 403 errors
• The ollama client path: ollama pull slow
• Whether a model file can hurt you: can model files be malicious
• Trusting the hub as a whole: are Hugging Face models safe
• What the licenses permit: open-weight licenses, explained
• Picking a Llama size: Llama models, every size compared