PR #1085 (Benjamin Demaille's native Apple Silicon port) notes that Homebrew tap packaging is pending (Phase 6).
I just archived my competing fork after benchmarking showed the native port is 2.8× faster (14s vs 61s on M1 Max, chr20 5Mb) with identical accuracy (F1=0.9965).
Before archiving, I had a working Homebrew formula with:
- Pre-built ARM64 binary download (no build-from-source)
- Wrapper scripts:
run_deepvariant, deepvariant-download-model, deepvariant-quicktest, deepvariant-doctor
- Auto-download + CoreML auto-conversion of WGS model
brew tap + brew install one-command install
Formula source: https://github.com/antomicblitz/homebrew-deepvariant/blob/main/Formula/deepvariant.rb
It would need adaptation for the single-binary architecture (no Python venv), but the wrapper patterns and model download infrastructure are reusable. Posting here in case it helps the native port reach a brew install experience.
PR #1085 (Benjamin Demaille's native Apple Silicon port) notes that Homebrew tap packaging is pending (Phase 6).
I just archived my competing fork after benchmarking showed the native port is 2.8× faster (14s vs 61s on M1 Max, chr20 5Mb) with identical accuracy (F1=0.9965).
Before archiving, I had a working Homebrew formula with:
run_deepvariant,deepvariant-download-model,deepvariant-quicktest,deepvariant-doctorbrew tap+brew installone-command installFormula source: https://github.com/antomicblitz/homebrew-deepvariant/blob/main/Formula/deepvariant.rb
It would need adaptation for the single-binary architecture (no Python venv), but the wrapper patterns and model download infrastructure are reusable. Posting here in case it helps the native port reach a
brew installexperience.