Finn Dayton comments

Repositories
Issues
Comments

Results 1 comments of


                                            Finn Dayton

`bfloat16` and FP16 support for custom kernels

So adding on to @sidnb13 's comments here, it looks like `segment_matmul` just takes in two `Tensor` types [here](https://github.com/pyg-team/pyg-lib/blob/2a0d5586d5f8b4ec9288018b2b071e01eff6f180/pyg_lib/csrc/ops/matmul.cpp#L41C37-L41C43) which are simply `torch.Tensors`, right? And `torch.Tensor` does have native support...