kavin-sai-krishna commented on PR #18040:
URL: https://github.com/apache/tvm/pull/18040#issuecomment-2944134843
@tqchen Thank you. I understood the high-level idea you suggested, but I
have a few specific questions regarding the design choices:
* Q1: What’s the difference between decomposing an op using Relax ops vs.
TOPI ops vs. TIR?
How does the abstraction level impact performance or correctness?
* Q2: If the op compiles on CUDA, is numerical verification still required?
* Q3: Are there nightly tests that check numerical correctness?
I ask because I found a case (`fmod`) where the op ran but didn't match
PyTorch output.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
