pytorch / pytorch Public

Theoretically, there shouldn't be any error in nightly build. This Project is to track all issues and PRs related to Windows Nightly build.

Three principles

Even one error occurred only once, we should know it. If one error occurred rarely, we should keep eyes on it. If one error occurs every week, actions must be taken.

Four levels

Health No errors. subhealth The error occurs very rarely. unhealth The error occurs frequently but it can pass with rerunning. dead The error blocks all related CI workflows.

Issue related to torch.utils.data (DataLoader, Dataset, Sampler, DataPipe, etc.)

Issues related to Lazy Tensor Core.

Issues related to torch.nn

This project tracks issues with PyTorch sparse tensor support.

Issues related to torch.package

Generate fast convolutions with NNC tensor expressions.

This project tracks progress on the NNC development.

Triage performance regressions identified from torchbenchmark nightly runs

This project is used to track and triage Quantization issues.

Have type annotations for the entire PyTorch API

No description

This project is used to track and triage JIT OSS issues.

This project tracks progress on the development of the to_backend PyTorch JIT extension, a feature that allows arbitrary backends to run PyTorch JIT graphs.

No description

Used for tracking the plans and states of some topics

No description

No description

No description

No description