We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
Hugging Face Blog·Jun 23·5 min read
Shipping huggingface_hub every week with AI, open tools, and a human in the loop
TL;DR
Previously, releasing updates for huggingface_hub took 4 to 6 weeks and involved significant manual effort. The team implemented a streamlined process using GitHub Actions for weekly releases, integrating open-source tools and maintaining human oversight where necessary.
✦ Why It Matters
Engineers can adopt similar automation strategies to enhance their own software release processes.
Key Takeaways
- Implement a CI workflow to automate mechanical release tasks.
- Use AI to draft initial release notes, but ensure human review.
- Archive AI drafts and human edits for future reference.
- Adopt OIDC Trusted Publishing for enhanced security.
How It Works
The new release process utilizes GitHub Actions to automate tasks like version bumping and publishing to PyPI. AI models draft release notes based on PR metadata, while a deterministic script verifies the output against a ground truth list of PRs.
This ensures that the generated notes are accurate and complete before human review.