Wrap fallback submodules for Inductor in a new nn.Module
#2551
+59
−1
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
shino16
requested review from
mruberry,
lantiga,
t-vi and
KaelanDt
as code owners
shino16
force-pushed
the
wrap-inductor-submodule
branch
from
f3311ea to
ec67c18
Compare
|
The test failures are coming from this: import torch
class GraphModule(torch.nn.Module):
def forward(self, y: "f32[2, 2]"):
# No stacktrace found for following nodes
_enter_autocast = torch.amp.autocast_mode._enter_autocast('cpu', None, True, None)
# File: /opt/pytorch/lightning-thunder/thunder/tests/test_dynamo.py:280 in func, code: y = torch.sinc(y)
y_1: "f32[2, 2]" = torch.sinc(y); y = None
# No stacktrace found for following nodes
_exit_autocast = torch.amp.autocast_mode._exit_autocast(_enter_autocast); _enter_autocast = _exit_autocast = None
return y_1
model = GraphModule()
y = torch.randn(2, 2)
torch.compile(model)(y) |
|
PyTorch checks that the global state is kept unchanged while tracing (ref). It may be better to rely on |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #2539, which in turn fixes #2527 and #2501. This makes PR #2538 obsolete.
As in #2527 (comment), when we pass a GraphModule to
torch.compileit doesn't get lowered properly. As a workaround, this PR wraps the fallback GraphModule in a newnn.Module. This workaround was found in #2527 (comment) (credit @mattteochen).Alternatively we could call
torch._inductor.compile_fx.compile_fxdirectly to compile the GraphModule, but as it returns a bareforwardfunction instead of ann.Moduleinstance, it has troubles with registering as a submodule to the outer GraphModule.