Upload a model

With the hf CLI

HF_TOKEN=<your-key> HF_ENDPOINT=https://siahub.app \
  hf upload <owner>/<repo> ./path/to/files

This works for single files (./model.safetensors) or directories (./my-model/). Recursion is on by default.

Filtering

Upload only the weights:

HF_TOKEN=<your-key> HF_ENDPOINT=https://siahub.app \
  hf upload <owner>/<repo> . --include="*.safetensors"

Multiple patterns are supported (--include="*.json" --include="*.md").

What gets stored where

• Small text files (README, config.json, tokenizer.json) land as inline LFS in Postgres — shown once on the model page, cheap to fetch.
• Large binaries go through the Xet chunk pipeline: hf_xet chunks and compresses them, SiaHub receives xorbs, caches them locally, and queues a Sia pin. The Sia upload runs asynchronously; the hf upload command returns as soon as SiaHub has durable bytes.

Authentication

HF_TOKEN is your SiaHub API key, not your huggingface.co token. They can coexist — your hf auth login for huggingface.co keeps working. The per-command prefix (HF_TOKEN=... HF_ENDPOINT=...) only applies to that invocation.