Charlie Lin

Results 35 comments of Charlie Lin

How exactly would this be better than the current method?

Closing, we do pass verify accuracy with MLIR's update and testing with the program mentioned in #2949 and a couple of different random seeds I tried.

> Looks like CI is failing, didn't see anything sticking out here. > > Is there any more changes you're going to push up? I see the issue. Changed the...

I had a similar idea to this where we have a `lazy_add_instruction()` or `lazy_replace_instruction()` that doesn't propagate the `op.compute_shape()` calls until the end of the matcher. Another way to do...

> Unit tests need to be added for fusing the pointwise modules. You will need to add a test suite for the `fuse_pointwise_reduce` pass and add a test there. Running...