Skip to content

resize_token_embeddings does not resize lm_head #45236

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
KoichiYasuoka wants to merge 3 commits into from
Closed

resize_token_embeddings does not resize lm_head #45236

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
75008c4
deepcopy old_lm_head before changing input_embeddings
KoichiYasuoka Apr 4, 2026
cfc2062
Merge branch 'main' into patch-9
KoichiYasuoka Apr 5, 2026
8153ef2
Merge branch 'main' into patch-9
KoichiYasuoka Apr 5, 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
4 changes: 2 additions & 2 deletions src/transformers/modeling_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2662,6 +2662,7 @@ def resize_token_embeddings(

def _resize_token_embeddings(self, new_num_tokens, pad_to_multiple_of=None, mean_resizing=True):
old_embeddings = self.get_input_embeddings()
old_lm_head = copy.deepcopy(self.get_output_embeddings())
new_embeddings = self._get_resized_embeddings(
old_embeddings, new_num_tokens, pad_to_multiple_of, mean_resizing
)
Expand All @@ -2684,8 +2685,7 @@ def _resize_token_embeddings(self, new_num_tokens, pad_to_multiple_of=None, mean
new_num_tokens = new_embeddings.weight.shape[0]

# if word embeddings are not tied, make sure that lm head is resized as well
if self.get_output_embeddings() is not None:
old_lm_head = self.get_output_embeddings()
if old_lm_head is not None:
if isinstance(old_lm_head, torch.nn.Embedding):
new_lm_head = self._get_resized_embeddings(old_lm_head, new_num_tokens, mean_resizing=mean_resizing)
else:
Expand Down
Loading