-
## Issue 1: probability can't be a tensor
```py
from torch.fx.experimental.proxy_tensor import make_fx
import torch
print(make_fx(
lambda x,p: torch.nn.functional.dropout(x, p),
decomp…
-
Normally calling `model.eval()` on a model with dropout changes the behavior of the dropout ops into a noop. However, this is not true if model is already exported, since the aten dropout pattern rema…
-
### 🐛 Describe the bug
I am trying to boost model with compiling model in PyTorch2.0. However, GRU modules are not abled to be compiled.
Following is the MWE:
```python
import numpy as np
im…
-
### 🐛 Describe the bug
We are facing issues with loss curves and reproducibility when using `torch.compile()` with our models. Attached below is a graph of train loss with runs with `torch.compile(…
-
# Overview
Since there are a couple of major things that I, and I believe also other, people would like to be implemented in agbplay I want to open this issue and collect things that I do plan to i…
-
Hi!
You are mentioned here because you previously contributed to the [zeldaret/mm repo](https://github.com/zeldaret/mm), a decompilation of the game Legend of Zelda Majoras Mask.
There is a move…
-
### 🐛 Describe the bug
The `torch.nn.functional.interpolate` op in its implementation will call `torch._C.nn.upsample_bilinear2d` when `mode=bilinear` and `input.dim() == 4`. However when the decompo…
-
```
import torch
from torch._dynamo import export
from torch._subclasses import FakeTensorMode
class JankyLinear(torch.nn.Module):
def __init__(self):
super().__init__()
s…
-
Hi!
You are mentioned here because you previously contributed to the [zeldaret/oot repo](https://github.com/zeldaret/oot), a decompilation of the game Ocarina of Time 64.
There is a move to lice…
-
```
import torch
from torch.fx.experimental.proxy_tensor import make_fx
def f(x):
return torch.where(x > 0.5, x, 1.0)
g = make_fx(f)(torch.rand(2, 3))
print(g)
```
yields
```
def for…