-
On the latest Thunder (dafc79d21c04769c5e9d1fb737b8cd21d0841e69) I tried the snippet from
https://github.com/Lightning-AI/lightning-thunder/issues/1174#issuecomment-2383823134
and got the following…
-
```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…
-
### Your question
Not sure why but I updated comfy and suddenly started getting errors loading torch. The file IS there in that location, so I'm not sure what the problem might be. Originally it ha…
-
Hi @chuanqi129,
Currently guilty commit inductor single run cannot support other backend except inductor.
https://github.com/chuanqi129/inductor-tools/blob/main/scripts/modelbench/inductor_single…
-
For a fair comparison with the baseline in test_transformer_engine.py.
I tried the following briefly and hit big number mismatches.
```patch
diff --git a/tests/cpp/test_multidevice_transformer.cpp…
-
Tracking the perf of a specific shape in this specific unit test using `nsys nvprof`.
TOT main branch (as measured [here](https://github.com/NVIDIA/Fuser/issues/3137#issuecomment-2406541381)):
```C++…
-
## 🚀 Feature
Today, nvfuser requirements a manual install. It is NOT part of requirements.txt.
Ideally users should be able to get all default executors without having to go through the README.MD an…
-
I believe the issue is somewhere inside some DynamicType that is not handling the evaluation properly. We might be able to work around it via change something in expression evaluator, but I wasn't abl…
-
🚀 The feature, motivation and pitch
# RFC: Multi-Gpu Python Frontend API
This RFC compares and contrasts some ideas for exposing multi-gpu support in the python frontend.
1. The current `multigpu_sc…
-
The re-written rope example has quite different indexing when input `q` / `cos` / `sin` is defined with static or dynamic shapes.
I think this is coming from the inconsistent fusion definition. i.e…