-
### 🚀 The feature, motivation and pitch
It would be good to make sure we have strong exposure of our expression evaluator. Going to note some interfaces we would likely want to expose:
Input bindi…
-
https://github.com/csarofeen/pytorch/blob/8ed9540b84b976fa5cd373e2d5a55a253223196b/third_party/nvfuser/csrc/ir_utils.cpp#L435
I don't have a failing case with devel, but encountered a problem with …
-
**Describe the bug**
o3.spherical_harmonics raises RuntimeError due to TorchScript. This only happens with PyTorch>=1.12. I guess it is because of a recently introduced feature, [nvfuser](https://pyt…
-
This can lead to false negatives because the threshold is overly relaxed.
```diff
diff --git a/tests/cpp/test_gpu_fused_reduction.cpp b/tests/cpp/test_gpu_fused_reduction.cpp
index e67875f4..b3923d6…
-
### 🐛 Describe the bug
For the issues I'm seeing in our benchmark, I think we can work around it by making batch_norm promotion explicit in the graph. So I'll try to work around it in nvprim/primto…
-
Because we do not expect users to deal with flatbuffers directly, should we restrict flatbuffer includes to cpp-only? This way, users linking with nvFuser will not have to deal with flatbuffers.
An…
-
When running a python script that includes `import torch`, it works fine as follows.
```
$ python3 -m pip install torch
$ cat import-torch.py
#!/usr/bin/env python3
import torch
$ echo +x i…
-
我早上安装好的,中午关机,晚上打开,回答的就特别慢。。。是什么原因?
有QQ群可以请教吗?
-
### 🐛 Describe the bug
# there is output diff between origin and script
``` python
import torch
import os
os.environ['TORCH_JIT_DISABLE_NEW_EXECUTOR'] = '1'
torch._C._jit_set_nvfuser_enabled…
-
When running Gemma-2-27b with Thunder we get an error:
> Networkx.exception.NetworkXUnbounded: Infinite capacity path, flow unbounded above.
The error can be recreated for 1 node with 8 gpus as…