[CUDA fuser] fails to run basic InceptionV3 #64062
Labels
high priority
oncall: jit
Add this issue/PR to JIT oncall triage queue
triaged
This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Projects
Milestone
馃悰 Bug
In PyTorch 1.9, the CUDA fuser addition makes it impossible to run (part of) the NVIDIA's InceptionV3 TorchScript model (ckpt url). After loading, the model works fine when running directly on an image (calling
model(x)
), but using a submodule (callingmodel.layers(x)
failsRuntimeError: MALFORMED INPUT: lanes don't match
.1.10.0.dev20210826
) at the time of writing, with either the aboveRuntimeError
orsegfault
.A couple notes:
model.code
andmodel.layers.code
, I think they are very standard code without anything too fancy. Is the fuser trying to fuse across the boundary (ofself.layers
call innetwork.forward
)?To Reproduce
PyTorch 1.8 output
PyTorch 1.9 output
PyTorch nightly
1.10.0.dev20210826
Output 1 (
RuntimeError
)Output 2 (
segfault
)Expected behavior
I expect the code to work as it did in PyTorch 1.8.1.
Environment
cc @ezyang @gchanan @zou3519 @bdhirsh @jbschlosser @anjali411 @ngimel
The text was updated successfully, but these errors were encountered: