-
### š Describe the bug
```
>>> import torch
>>> torch.randn(46000, 46000, device='cuda:3').to_sparse()
Traceback (most recent call last):
File "", line 1, in
RuntimeError: CUDA error: an ilā¦
-
### š Describe the bug
see the log outputs here: https://gist.github.com/alita-moore/a23f57d2adfaa885185308e922aef44a
To reproduce this run the following code:
```
import os
import torch
impā¦
-
I think about the learning design that is implemented here, and I just can't resolve to myself two questions. The core function for the learning is [the environment step function](https://github.com/gā¦
-
### š Describe the bug
So I've been trying to compile some auto-regressive transformer models and they consist of for loops within the model architecture. I'm posting a simplified version of the archā¦
-
### š Describe the bug
I may have discovered a bug when I registered the corresponding ProcessGroup development.
I found that the destructor of the ProcessGroup subclass object was not called whenā¦
-
### š Describe the bug
As Pytorch does not (yet?) support broadcasting on sparse matrices, I implemented a simple autograd class. Forward propagation works fine, but backprop fails with a cryptic:
ā¦
-
### š Describe the bug
Bit of a weird one, not sure if this is something interesting but just in case:
```python
import torch
torch.tensor([torch.tensor(0)]) # works fine
torch.Tensor.__getitā¦
-
I'm looking at Metalhead integration in MLJFlux. To do this well, I'm looking for some uniformity in the Metalhead.jl API that seems to be lacking. In particular, it would help if `nclasses` and `inchā¦
-
190126~
-
### š Describe the bug
Hello,
I am trying to combine both pytorch 2.0 compile + fsdp on TPU but it doesn't work.
What does work on TPU:
base training.
base training + PyTorch compile.
base tā¦