-
### 🐛 Describe the bug
```C++
TEST_F(NVFuserTest, FusionBroadcastingIndexing_CUDA) {
Fusion fusion;
FusionGuard fg(&fusion);
auto tv0 = makeSymbolicTensor(2);
auto tv1 = makeSymbolicTe…
-
To reproduce on main, use this command with an A100 80GB PCIe:
```
NVFUSER_ENABLE=fuse_matmul NVFUSER_DISABLE=matmul_expr_eval pytest benchmarks/python/test_matmul.py -vs
```
Either with or without ou…
-
Error:
```
Traceback (most recent call last):
File "/opt/pytorch/pytorch/nvfuser/__init__.py", line 76, in execute
result = self._execute(inputs, override_user_schedule)
RuntimeError: index…
-
## 🐛 Bug
For a few models ( Platypus-30B with FSDP zero3, Gemma7b with DDP and vicuna-33b-v1.3 with FSDP zero3) we get segmentation fault error when trying to use fp8 with thunder_cudnn. When usi…
-
Recompile `building CXX object CMakeFiles/nvfuser_codegen.dir/csrc/executor_utils.cpp.o` whenever you run `python setup.py build`
-
### 🐛 Describe the bug
I have a horizontal fusion situation with `reshape` that I would like to understand if this can be fused. I think we have a knob to turn this on or a place to switch this. Ji…
-
## 🐛 Bug
When running the benchmarks for Mixtral-8x7B-v0.1 for Eager mode we get error:
> 0: [rank0]: File "/workspace/lightning-thunder/thunder/benchmarks/benchmark_litgpt.py", line 887, in…
-
## 🐛 Bug
New changes are about to be introduced in nvFuser that will break the repro script generation.
The PR is [#2831](https://github.com/NVIDIA/Fuser/pull/2831) and it will require changing…
-
When working on #582 and test softmax, I found that if FusionDefinition use `keepdim=False` then `broadcast_in_dim`, the generated code failed to explicitly use alias registers. There is no influence …
-
### 🐛 Describe the bug
I'm getting the following warning which hints at suboptimal speed, and doesn't look like it should happen at any point.
```... site-packages\torch\_functorch\vmap.py:619: Us…