-
### 🐛 Describe the bug
When you have a function that uses torch.sparse, torch.compile can compile it, but not with "fullgraph=True"
### Error logs
(e3nn_env) aglover@login36:~/e3nn_benchmarking> p…
-
# 🚀 Feature
I think that it would be very helpful to allow sparse tensors to be used as masks (`attn_bias`) in the `memory_efficient_attention` function.
## Motivation
As the documentation no…
-
Dear all:
the sparse retain operation can be implemented in tf by `tf.sparse.retain` as follows:
```
def sparse_dropout(x, rate, noise_shape):
"""
Dropout for sparse tensors.
"""
…
-
I'm a bit lost in this awesome library, and could not find a suitable example, hence not sure if and how to use tensorly for a "simple" SVD decomposition for PyTorch tensors (if possible).
I.e
G…
-
Working with block-sparse reduction, my code runs properly with 2D tensors. However it fails to compile if I add an additional dimension (batch), leading to this error:
`[KeOps] The 'ranges' argume…
tkon3 updated
3 years ago
-
### 🐛 Describe the bug
Hi, I hope you don't mind me raising so many issues 😄
`torch.sparse.sampled_addmm` works fine in the forward phase for both 2D and 3D tensors. However, in the backward pas…
-
https://pytorch.org/docs/master/tensors.html#torch.Tensor.index_put_
`
Puts values from the tensor value into the tensor self using the indices specified in indices (which is a tuple of Tensors). …
-
I am getting the following error when running a custom sparse model: RuntimeError: [Minuet] [.../csrc/minuet/cuda/helpers.cu (77)] CUBLAS Error: CUBLAS_STATUS_NOT_INITIALIZED
I think it might be du…
-
this is the code:
```
import torch
from pytorch_block_sparse import BlockSparseLinear
x = torch.randn(32, 128).to('cuda')
y = torch.randn(32, 64).to('cuda')
model = torch.nn.Sequential(
…
-
I wonder if PyTorch hybrid sparse/dense tensors can be used to represent outputs of dimension-wise topk / min / max. Unfortunately I couldn't completely understand from docs if they can be used for th…