Lightning Talk: TorchFix - a Linter for PyTorch-Using Code with Autofix Support
summary
A PyTorch-focused linter with autofix support analyzes code with LibCST to find and rewrite concrete issues: deprecated API moves (torch.symeig→torch.linalg.symeig), the silent typo requires_gr instead of requires_grad, missing DataLoader num_workers, and torchvision’s pretrained→weights change. It runs as autofix via LibCST, either as a flake8 plugin vs standalone --fix mode (flake8 mode lints only), and it already runs in CI on multiple Meta and open‑source repos; current state is early beta with rule opt‑ins.
AI-written summary · about
related talks
30mLint Your Code Into the Future
01h 00mLightning Talks 1
19mTorchQL: A Programming Framework for Integrity Constraints in Machine Learning
13mLightning Talk: Challenges and Standardization in PyTorch Ecosystem...
19mMake Linting Great Again
07mMake Linting Great Again
48mFireside Chat
17mLightning Thunder: Supercharged PyTorch for Modern Hardware