How to help yourself and ML community make PyTorch better.
summary
Practical steps for fixing or extending PyTorch: create a small reproducible case, run the built-in collect environment to capture Python, hardware and dependency details, and file a GitHub issue against the pytorch repo with full logs and reproduction. For features start with an RFC issue; for code submit focused PRs with tests and single-purpose changes. CI uses labels to select tests, nightly builds come from the viable strict branch, and the HUD dashboard helps trace which commit and test failed.
AI-written summary · about
chapters (14)
- 0:00Introduction to PyTorch contributions
- 0:43What is PyTorch and getting started
- 1:26Reporting issues in PyTorch
- 3:11Writing good bug reports
- 4:42Submitting feature requests
- 6:10How to contribute to PyTorch
- 7:14Submitting pull requests
- 9:19PR review process and expectations
- 10:20PyTorch PR lifecycle
- 12:47Handling CI failures and reverts
- 14:12PyTorch testing infrastructure
- 16:04Release cadence and nightly builds
- 17:08PyTorch review structure
- 18:37Q&A session
related talks
06mCommunity Updates
41mHelping the PyTorch Community Improve the Security Posture of th...
21mWhat's New for PyTorch Developer Infrastructure
30mHow to Trust Your Deep Learning Code
09mMy Journey to PyTorch
18mPyTorch Release
01h 32mIntroduction to PyTorch
25mTorchBench: Guarding the Performance of the PyTorch Ecosystem with Continuous Benchmarking