-
Notifications
You must be signed in to change notification settings - Fork 25.5k
Open
Labels
module: custom-operatorscustom operators, custom ops, custom-operators, custom-opscustom operators, custom ops, custom-operators, custom-opsmodule: pt2-dispatcherPT2 dispatcher-related issues (e.g., aotdispatch, functionalization, faketensor, custom-op,PT2 dispatcher-related issues (e.g., aotdispatch, functionalization, faketensor, custom-op,oncall: pt2triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module
Description
📚 The doc issue
Hello, I was looking for more clarification about the tradeoffs for using different kinds of shape checks when creating Custom Operators. There are options at the python and C++ levels. TORCH_CHECK, torch.assert() How do these different options compose with other features like torch.compile, ATOI, and symbolic shape tracing? Which kinds of shape checks can be elided under what conditions? What are recommended best practices for shape checking with custom operators. I think just a little expansion on this topic in the custom operator section of the documentation would be very helpful.
Suggest a potential alternative/fix
No response
Metadata
Metadata
Assignees
Labels
module: custom-operatorscustom operators, custom ops, custom-operators, custom-opscustom operators, custom ops, custom-operators, custom-opsmodule: pt2-dispatcherPT2 dispatcher-related issues (e.g., aotdispatch, functionalization, faketensor, custom-op,PT2 dispatcher-related issues (e.g., aotdispatch, functionalization, faketensor, custom-op,oncall: pt2triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module