-
### 🚀 The feature, motivation and pitch
Hi,
both `torch.softmax` and `torch.sparse.softmax` don't support the CSR format.
Namely, when I try to apply any softmax to `torch.sparse_csr tensor`, I rec…
-
### 🐛 Describe the bug
[bug.pt.gz](https://github.com/pytorch/pytorch/files/8147613/bug.pt.gz)
1. I don't understand why `a.indices()` gives an error (maybe related https://github.com/pytorch/py…
-
Hello, thanks for the nice work; well done.
Can the SAM take multiple Bounding boxes as prompts for segmentation? For example, if I draw boxes around two objects, say a building and a dog,
![image](…
-
- [ ] Sparse Tensor Methods
- [x] ndims
- [x] norm
- [x] shape slice
- [ ] mttkrp (partial implementation that only takes in factors, but not full ktensor)
- [ ] ttv (needed for…
-
### Describe the feature request
The current C-API only offers the following functions for loading attributes:
- `KernelInfoGetAttribute_float`
- `KernelInfoGetAttribute_int64`
- `KernelInfoGetAtt…
-
## 🐛 Bug
## To Reproduce
```python
@dtypes(torch.double)
def test_matmul_device_mismatch(self, device, dtype):
cpu = torch.rand((10, 10))
cuda = cpu.cuda()
for s, m1, m2 in i…
aocsa updated
3 years ago
-
Models trained by this pipeline perform great. But how to host them using Tesorflow Model Serving? Checkpoint needs to be converted into SavedModel (.pb) format.
What I've done so far is:
1. I'…
-
### 📚 The doc issue
Hey,
it is a bit confusing for me and my group for which cases [SparseAdam](https://pytorch.org/docs/stable/generated/torch.optim.SparseAdam.html) can be used:
There is a …
-
the MPS classes rely on initializing new tensors from shapes of existing tensors, using `Node.shape`. This is not working for symmetric tensors. We need to add a `sparse_shape` attribute to `Node`, wh…
-
Hi,
I am trying to implement a self-attention block for sparse tensors. For doing that, I need to multiply the features coming from two sparse tensors but cause I can't multiply features from two …