-
**Describe the bug**
fp8 e4m3 wgrad seems to be extremely slow compared to both FP32 and FP16, often 50x to 100x slower.
I have attached the profiling results in [this Google spreadsheet](https://doc…
-
I got the following error:
Traceback (most recent call last):
File "/kaggle/working/OCT2Confocal_3DCycleGAN/train.py", line 52, in
model.optimize_parameters() # calculate loss functions,…
-
**Release Type**: Official Release
**Version**: Version number: 4.1 & 4.0
**Platform(s)**Windows
**Describe the bug**
I can't even start Stride. I seems it cant download the VS plugin.
**…
-
in stride [1,100,1,1], what does the 100 means?
-
I manage to install Strider CD via pm2 but it does not starts:
``` bash
$ npm install pm2@next -g
$ pm2 install strider
$ pm2 logs
0|strider | 31284 died 1 null
0|strider | 31298 forked
0|strider …
-
On 0.7, I can pass a reshaped range to PyCall
```
julia> np.sin(reshape(1:4, 2,2))
┌ Warning: The default `strides(a::AbstractArray)` implementation is deprecated for general arrays.
│ Specializ…
-
**Release Type**: Official Release
**Version**: 4.2.0.2122
**Platform(s)**: Windows
**Describe the bug**
Projects created with Graphics level API 9_1 with DirectX crash at startup with a E_…
-
Just a note in case its useful to anyone else.
When trying to calculate the size for a `ConvTranspose2d`, sometimes the estimate is incorrect. Here is an example:
```
After conv2d
Conv kernel…
-
The `ReadBuffer`/`WriteBuffer` types only have the base address and length, but no stride. This doesn't allow them to handle DMA engines where you can do scatter-gather operations. Or alternatively a …
-
Would be nice if one could cast something to a strided memoryview with non-trivial striding. This is basically an extension of casting a pointer to a memoryview.
```python
cdef np.ndarray[float, n…