Hi everyone,
We have officially kicked off a coordinated initiative to refactor the PyTorch test suite. The core goal is to break the tight coupling between existing test cases and specific hardware, systematically categorizing them into:
- Accelerator-unrelated (runs on CPU)
- Accelerator-agnostic (generic across backends)
- Accelerator-specific (strictly limited to a particular hardware)
This will enable new hardware backends to seamlessly reuse PyTorch’s extensive existing test suite and ensure delivery quality more efficiently.
We have audited all the target test files and organized them into a central tracking sheet for community ownership assignment and workflow tracking.
Please check our main tracking issue for full architectural context, discussion history, and the collaboration links:
Main Tracking Issue: [Test] PyTorch Test Case Refactoring and Track · Issue #185590 · pytorch/pytorch · GitHub
We highly welcome Maintainers and Contributors to join the effort and claim ownership of specific modules. Looking forward to your feedback!
Best regards,
Jiawei Li