-
### 🐛 Describe the bug
Here, a heavily used pattern lifted from NVIDIA NeMo / Riva TTS use case. Not exportable with neither method:
```python
import torch
def batch_from_ragged(text: torch.Tensor,…
-
This may be a better sample.
Please provide feedback on the image.
This also addresses Issue #15.
Please rename if the file if desired.
Please check if they meet the requirement, or even re-cr…
-
We should audit existing decompositions and replace them with refs/decomps where appropriate, e.g. `clamp` decomposition currently incorrectly type promotes and doesn't handle non-number boundaries (s…
-
### 🚀 The feature, motivation and pitch
Take code like
```
assert x.numel() != 0
```
vs.
```
assert all([i > 0 for i in x.shape])
```
Both of these are mathematically equivalent. But, th…
-
**Describe the bug**
pennylane circuits are un-parametrized if there the qnode has `custom_decomps` keyword argument.
**To reproduce**
```
import pennylane.numpy as np
import pennylane as qml
…
-
With the latest version of PyTorch, I've been migrating TorchInductor to prims (see https://github.com/pytorch/torchdynamo/pull/431) and I've noticed many cases where PrimTorch burns in shapes in case…
-
### 🐛 Describe the bug
```python
import torch
from torch import nn
from torchvision.models import resnet18
from torch._dynamo import allow_in_graph
from functools import wraps
from functorch im…
-
### 🐛 Describe the bug
Repro - `python benchmarks/dynamo/huggingface.py --backend=aot_eager_decomp_partition --amp --training --device cuda --accuracy --only=AlbertForQuestionAnswering`
Setup - …
-
Thank you for providing a great tool, PyGCGOpt.
I want to solve a MILP with Dantzig-Wolfe reformulation using PyCGCOpt in my research.
In the following code, I try to compute a solution of a MIP '…
-
The following ops are using `ir.FallbackKernel` via `make_fallback()` in [lowering.py](https://github.com/pytorch/torchdynamo/blob/main/torchinductor/lowering.py#L894) and appear in benchmarks. We sh…