-
```py
import torch
import thunder
@thunder.jit
def f(x):
x.add_(1)
return x.copy_(x.sin())
f(torch.tensor(0.0, device='cuda'))
```
The above results in the following error from nv…
-
Molmo7BDbnb
all() received an invalid combination of arguments - got (Tensor, dim=tuple, keepdim=bool), but expected one of: * (Tensor input, *, Tensor out) didn't match because some of the keywords …
-
The `*` op is supposed to concatenate a shape tensor to itself. However, if we use the resulting shape tensor with a fill ops, we get a failure:
```py
>>> a = tp.Tensor([1, 2, 3])
>>> tp.ones(a.shape[…
-
## 🚀 Feature
It would be useful if we would model constants and use that to optimize away `getitem`.
### Motivation
The ThunderFX path sometimes ends up giving us some rather silly graphs:
…
-
### 🐛 Describe the bug
When I try multi-gpu on torch with `backend = custom_backend` it leads to the error-
`aot_export is not currently supported with traceable tensor subclass`
The following …
-
## 🚀 Feature Request
Currently we have computation traces with the generated tensor shapes as part of comments next to the computation like
```
t908 = torch.nn.functional.linear(t907, t19, t17) …
-
## Bug Description
acc tracer doesn't handle torch.max(tensor).values correctly
## To Reproduce
```
import torch
import torch.nn as nn
import torch_tensorrt.fx.tracer.acc_tracer.acc_trace…
-
### What happened?
There are 14,620 child zones that are intializer dispatches of slow_memcpy's. Is there a way to decrease those somehow? The tracy file is provided in Azure [here](https://sharkpu…
-
**Describe the bug**
New conv api i.e., ttnn.conv2d returns list of int and tensor ([output_tensor, _out_height, _out_width, self.weights, self.bias]]. As type output visualizer is not enabled for …
-
### Description
I am having issues initializing a Flax.linen neural network when running with GPU support. I have narrowed it down to the flax.linen.initializers.orthogonal. Running the below cod…