Max Ren
Max Ren
@digantdesai hmm I'm wondering if this suggests that XNNPACK doesn't offer a default operator for looking up weights then?
@kirklandsign
It looks like the decomposition for matrix multiply introduces an expand copy. However expand copy doesn't look it can be traced with dynamic shapes?
@JacobSzwejbka @SS-JIA @angelayi. Dynamic shapes seems to be failing with because of some issue with expand copy when decomposing torch.matmul. This seems like a bug. Do you have any idea...
cc: @digantdesai @fbarchard @alankelly @gonnet for reviews
cc. @digantdesai @alankelly @fbarchard @gonnet for reviews
Can't add reviewers, so ccing: @alankelly @fbarchard @digantdesai
@gonnet as well
hi @alankelly, I noticed that these changes were rolled back. Is that the intention for the future going forward? Just want to know now to understand whether or not the...
Link to landed trunk PR: * https://github.com/pytorch/executorch/pull/3739 Link to release branch PR: * https://github.com/pytorch/executorch/pull/3778 Criteria category and reasons: * Replace buck with cmake for all docs ------ @mcr229 merged