This is a patch release containing the following changes to v3.12.1:
- Fixed correctness issue in grouped convolution on x64 processors with Intel AVX 10.1 instruction set support (680d58a, 9c3d576)
- Fixed crash in matmul with 4D shapes and trivial batch dimensions on x64 processors (9236298)
- Fixed correctness issue in convolution on x64 processors (b111531)
- Fixed crash in convolution and deconvolution with certain combinations of post-op data types on x64 processors (69c39db)