Modelwrap

Builds reproducible dm-verity EROFS images of Hugging Face models. Learn more about how this works on the Tinfoil blog.

Build image

docker run --rm -it \
  -v $(pwd)/cache:/cache \
  -v $(pwd)/output:/output \
  -e HF_TOKEN="${HF_TOKEN}" \
  -e MODEL=meta-llama/Llama-3.2-1B@4e20de362430cd3b72f300e6b0f18e50e7166e08 \
  ghcr.io/tinfoilsh/modelwrap

Notes:

• MODEL should include an explicit @revision (commit hash) for reproducible builds.
• If @revision is omitted, modelwrap resolves the current HEAD commit, which may change over time.

modelwrap emits two files in the output directory:

• output/meta-llama/Llama-3.2-1B/4e20de362430cd3b72f300e6b0f18e50e7166e08.mpk: dm-verity EROFS image
• output/meta-llama/Llama-3.2-1B/4e20de362430cd3b72f300e6b0f18e50e7166e08.info: metadata file in the format ROOTHASH_OFFSET_VERITYUUID