-
Currently, 40 models are passing `torchbench.py -d cuda --inductor --training --float16` mode, this issue tracks the ones that don't work. Note that due to `dropout()` RNG differences we don't expect …
-
### 🐛 Describe the bug
When using HuggingFace's Trainer API I noticed that PyTorch eager mode succeeds as expected but inductor fails with a shape mismatch error:
```
ValueError: Cannot view a …
-
### 🐛 Describe the bug
```python
import torch
@torch.compile(dynamic=True, backend='eager')
def f(x):
return x >> 0
f(torch.tensor(0))
```
The operations `>>` and ` 0
You can…
-
Hi,
I am trying run for inference another model on an android device using executorch. Execution fails with a
```torch._dynamo.exc.Unsupported: call_function args: UserDefinedObjectVariable(EasyDict…
-
### 🐛 Describe the bug
Recently I tried tracing a inplace op with a hand written [decomp](https://github.com/pytorch/pytorch/blob/cdab6c8df9ff9331126f69ea59c23f06109f03d7/torch/distributed/_spmd/ap…
-
### 🐛 Describe the bug
~~~
File "/scratch/anijain/work/pytorch/torch/autograd/__init__.py", line 319, in grad
result = Variable._execution_engine.run_backward( # Calls into the C++ engine to…
-
### 🐛 Describe the bug
Not a bug per se, just a (perhaps unnecessary) inconsistency with `max_pool2d` when using `make_fx(functionalize(...))` instead of `make_fx(...)`. To illustrate this, conside…
-
### 🐛 Describe the bug
This is a WIP version of https://github.com/lucidrains/x-transformers that has zero graph breaks. It converges as expected when not compiled, but when compiled the loss curve…
-
After watching chaomix's video, I decided to go ahead and put these decomps on my steam deck. All of a sudden, the game audio has completely disappeared. tested on both modded sonic 1 and vanilla soni…
-
### 🐛 Describe the bug
For lowering graphs using compile_fx_inner, we are hitting a cyclical error re: decompositions for 20 operators. These operators are all present in the torch decomp table, but …