[BUG][FRONTEND] Pytorch Importer mm

[tqchen]

Caught this when reading through the code. Dense requires transposition.

https://github.com/apache/tvm/blob/main/python/tvm/relay/frontend/pytorch.py#L1880

[tqchen]

cc @masahi

[tqchen]

tracking an related error here https://discuss.tvm.apache.org/t/pytorch-unable-to-import-a-simple-torch-nn-linear-to-relay/10383

[tqchen]

cc @jcf94 @comaniac

[masahi]

Yeah torch.mm expects non-transposed input, so our converter doesn't seem correct. https://pytorch.org/docs/stable/generated/torch.mm.html

I wonder why the test cases didn't catch this...

[masahi]

Sorry for the confusion, this converter is not used anywhere. Removed in #8397

[jcf94]

Sorry for my late reply, what's the latest update to this bug?
The nn.matmul should currently only be enabled in TensorFlow frontend with extra parameters.

Ok, seems it has been fixed in #8397 ? Thanks @masahi .