-
Let's start with saying that based on my reading of various papers Model Parallelism (MP) is a very inconsistent term. One can slice vertically or horizontally. One can implement a naive slow version …
-
I want to do matrix multiplication of 2D tensors where I only care about a few diagonals of the resulting matrix, and I want to run this on GPU on PyTorch. Is this something that can be done with `keo…
-
`CSD` stands for Compressed Sparse Dimensions. It is a multidimensional generalization of `CSR`, `CSC` and `COO`.
`CSD` has a few "compressed" dimensions and a few "uncompressed" dimensions. The "c…
-
The recommended `WSP` builder in the raster section of `weights` seems to return `WSP` objects that can only have `int8` values. This makes if fail, for example, on computations that require products …
-
Since this package saves data as column-order, the order field should be `F` per https://zarr.readthedocs.io/en/stable/spec/v1.html?highlight=row%20major#metadata, but currently this is set to `C`
…
-
It's a well-known fact that many convolutions can be thought of as a direct matrix multiplication(Im2Col and more subtle ideas). cuDNN white-paper directly states that NVIDIA developers use precisely …
-
### What happened?
When compiling for cuda with `--iree-codegen-llvmgpu-enable-transform-dialect-matmul-tensorcore-strategy` and `--td-matmul-strategy-use-mma-sync` I run into miscompiles.
### Steps…
-
@steo85it
What has been done with this so far? I would like to push this along
-
Relu is a quite simple function and I would expect that different definitions would all be optimized to the fastest calculation and thus that all different definitions would have the same speed. Howev…
-
Dear @xianyi and everyone!
Is there a way to forbid denormals in the OpenBLAS?
I tried to execute the following code (MSVC2015) before calling any OpenBLAS routines:
unsigned int current_wo…
Arech updated
7 years ago