Add activation functions (ReLU and SiLU for now) for structured spars…

…e linear operator (#101339) Differential Revision: [D46453476](https://our.internmc.facebook.com/intern/diff/D46453476) Pull Request resolved: #101339 Approved by: https://github.com/cpuhrsch
pytorch · Jun 16, 2023 · 8fc687f · 8fc687f
1 parent 0da3840
commit 8fc687f
Show file tree

Hide file tree

Showing 3 changed files with 370 additions and 118 deletions.
diff --git a/aten/src/ATen/native/native_functions.yaml b/aten/src/ATen/native/native_functions.yaml
@@ -3221,7 +3221,7 @@
     MkldnnCPU: mkldnn_linear_backward
   autogen: mkldnn_linear_backward.out
 
-- func: _structured_sparse_linear(Tensor input, Tensor weight, Tensor mask_or_meta, *, Tensor? bias=None) -> (Tensor, Tensor)
+- func: _structured_sparse_linear(Tensor input, Tensor weight, Tensor mask_or_meta, *, Tensor? bias=None, str? activation=None) -> (Tensor, Tensor)
   dispatch:
     CUDA: _structured_sparse_linear