Skip to content

Skip _check_forward_args validation during make_fx tracing #454

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
IvanKobzarev wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Skip _check_forward_args validation during make_fx tracing #454

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 9 additions & 0 deletions autoparallel/input_validation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,6 +71,15 @@ def _check_forward_args(args, expected_inputs, dynamic_dims=frozenset()):
expected_inputs: Expected shapes from _compute_expected_inputs.
dynamic_dims: Set of (arg_index, dim) pairs for dynamic dimensions.
"""
# Skip validation during make_fx tracing -- FakeTensors have different

@fmassa fmassa May 11, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Normally this check only runs after AutoParallel has finished, as is meant as a safety for users calling the parallelized model in their code.

When we do torch.compile(parallel_mod), this check should normally be properly captured.

So I'm not sure if this skipping is actually doing what we want globally, but it would be good to discuss with others from the GraphTrainer if they should just completely skip this in their implementation of AutoParallelGraph

# device/repr than the meta tensors used during AutoParallel tracing.
if any(
a.is_meta or getattr(a, "fake_mode", None) is not None
for a in args
if isinstance(a, torch.Tensor)
):
return

if len(args) != len(expected_inputs):
raise ValueError(
f"AutoParallel: expected {len(expected_inputs)} arguments "
Expand Down
Loading