Skip to content

fix: gemma4 mm_token_type_ids padding #10359

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

Merged
Kuangdd01 merged 1 commit into from
Apr 6, 2026
Merged

fix: gemma4 mm_token_type_ids padding #10359

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
13 changes: 13 additions & 0 deletions src/llamafactory/data/collator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -380,6 +380,19 @@ def __call__(self, features: list[dict[str, Any]]) -> dict[str, "torch.Tensor"]:
for i, feature in enumerate(features):
feature["token_type_ids"] = token_type_ids[i]

if "mm_token_type_ids" in mm_inputs: # need tensor-like for gemma4
mm_token_type_ids = mm_inputs.pop("mm_token_type_ids")
max_len = max(len(ids) for ids in mm_token_type_ids)
padded = []
for ids in mm_token_type_ids:
pad_len = max_len - len(ids)
if self.tokenizer.padding_side == "right":
padded.append(ids + [0] * pad_len)
else:
padded.append([0] * pad_len + ids)
Comment thread
Kuangdd01 marked this conversation as resolved.

mm_inputs["mm_token_type_ids"] = torch.tensor(padded, dtype=torch.long)

features: dict[str, torch.Tensor] = super().__call__(features)

bsz, seq_len = features["input_ids"].shape[:2]
Expand Down
Loading