-
Hi,
First of all, EIN looks awesome!
I was wondering about one of the design decisions on this project. It appears that this notebook frontend connects to the IPython notebook server, and I'm wonder…
tompx updated
10 years ago
-
```
Our current kernel configuration supports only 4G memory. Systems with
large amount of memory can't see the extra memory with zumastor kernel
installed. We may want to build separate kernels with …
-
# Problem Statement
Currently in MXNet we implement operator kernels in C++. Developers need to specify the detail logic of each computation, which slows down the development process. Given the fac…
-
Hi!
I'm using xESMF in my Python package to take care of regridding. A certain input dataset constantly raises `UserWarning: Input array is not C_CONTIGUOUS. Will affect performance.`. The dataset …
-
This is open-ended. The problem is that many key use cases, such as matrix multiplication kernels, need to know a number of SIMD vector registers that they can count on using. In practice, the number …
-
Many scalar string kernels only use ASCII characters in their tests. This should be extended to other UTF-8 codepoints for move coverage.
This will most likely require changing the **TYPED_TEST_SUITE…
-
The static buffer result_buffer is declared to have size 256, but the unsafe functions sprintf() and strcpy() are used instead of snprintf() and strncpy(). For long results (such as on my Dell L702X …
-
```
Our current kernel configuration supports only 4G memory. Systems with
large amount of memory can't see the extra memory with zumastor kernel
installed. We may want to build separate kernels with …
-
### Fast Pytorch dequantize() + matmul
I would like to open the discussion about faster inference with quantized models using pure Pytorch calls.
As you know, quantization is extremely importan…
-
many gpu algorithms require global data structure as a "working space".
In some cases the data structures are used to communicate between the various kernels that compose a more complex algo (encapsu…