-
Hi folks, I'd like to get your feedback on my proposed design of selective decomposition of operations in Torch-MLIR.
I am building mainly on the notion of "backend-legal Ops" (merged in [this PR](…
Svoch updated
2 years ago
-
one for the moment tensor decomposition and such, one for plotting, and one for the command line interface as suggested from @emolch
-
### Bug description
```python
"""
Test pytorch linalg module with pylint
"""
import torch
def my_qr_decomposition(tensor: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
"""
M…
-
# Adding an optimization module
For now, Tensorly (TL) ships with one API for each particular tensor decomposition model. While this has the advantage of simplicity for the end-user, this limits the …
-
BLAS installation has been a recurrent problem for users and CI:
- IRC: https://irclogs.nim-lang.org/03-03-2020.html#17:24:53
- In Nim CI due to Linux distros using either blas or cblas: https://g…
-
As outlined in Meta Ticket #31991, a tensor can admit different types of decompositions / tensor networks. We propose to add visualizations to the tensors (depending on decomposition used), possibly…
-
Came up in discussion with @samdow and @Chillee.
Given an aten operator that has some schema, its decomposition should match the schema's alias information. For example, an out-of-place operation l…
-
Hi, thanks for the great implementation of tensor decomposition. However, when I apply your code to my resnet model, the decomposed model actually has more parameters than the original one. Do you hav…
-
### 🐛 Describe the bug
When running a simple model including torch.nn.LayerNorm using deepspeed zero3 with torch.compile and [compiled_autograd](https://github.com/pytorch/tutorials/blob/main/interme…
-
### Proposal
Allow ActivationCache.get_full_resid_decomposition to receive a `project_output_onto` tensor that is either a [d_model] tensor or [d_model, num_outputs] tensor, such we multiply the o…