Skip to content

Add skip_names and LoRALinear guard to replace_linear_with_split_linear #18346

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

Closed
lucylq wants to merge 6 commits into from
+13 −1
Closed

Add skip_names and LoRALinear guard to replace_linear_with_split_linear #18346

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
c470898
Update
lucylq Mar 19, 2026
23f921a
Update
lucylq Mar 19, 2026
ae03b26
Update
lucylq Mar 19, 2026
182abb6
Update
lucylq Mar 19, 2026
51f5322
Update
lucylq Mar 19, 2026
d58628d
Update
lucylq Mar 20, 2026
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
14 changes: 13 additions & 1 deletion examples/apple/coreml/llama/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -91,9 +91,20 @@ def forward(self, x):


def replace_linear_with_split_linear(
model, out_target_split_size, out_max_splits, in_target_split_size, in_max_splits=1
model,
out_target_split_size,
out_max_splits,
in_target_split_size,
in_max_splits=1,
skip_names=None,
):
from executorch.examples.models.llama.lora import LoRALinear

for name, module in model.named_children():
if skip_names and name in skip_names:
Copy link
Contributor Author

@lucylq lucylq Mar 20, 2026 •
edited
Loading

Choose a reason for hiding this comment

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

in the next PR, this ends up being 'wqs', 'wks', 'wvs', 'wos'. Regular linear and LoraLinear both have to be skipped.

continue
if isinstance(module, LoRALinear):
continue
if isinstance(module, torch.nn.Linear):
assert module.bias is None, "SplitLinearModule does not support bias"
new_module = SplitLinearModule(
Expand All @@ -113,4 +124,5 @@ def replace_linear_with_split_linear(
out_max_splits,
in_target_split_size,
in_max_splits,
skip_names=skip_names,
)
Loading