423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-08-23 06:12:50 +08:00
Code Issues Packages Projects Releases Wiki Activity

[data] fix mllama collator (#6874)

Browse Source 
Former-commit-id: b68199db274a53d5916179e1aaf9722fd94fa2dc
This commit is contained in:
hoshi-hiyouga 2025-02-09 22:42:25 +08:00 committed by GitHub
parent b93333685b
commit 1bb3d17d9e
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/llamafactory/data/collator.py
View File

@ -121,7 +121,6 @@ class MultiModalDataCollatorForSeq2Seq(DataCollatorForSeq2Seq):
)
batch_images = fake_images
batch_imglens[0] = 1
batch_input_ids[0] = features[0]["input_ids"]
if (
self.template.mm_plugin.audio_token is not None and sum(batch_audlens) == 0
@ -137,7 +136,6 @@ class MultiModalDataCollatorForSeq2Seq(DataCollatorForSeq2Seq):
)
batch_audios = fake_audios
batch_audlens[0] = 1
batch_input_ids[0] = features[0]["input_ids"]
if fake_input_ids is not None:
if self.tokenizer.padding_side == "right":
@ -149,6 +147,8 @@ class MultiModalDataCollatorForSeq2Seq(DataCollatorForSeq2Seq):
features[0]["attention_mask"] = [0] * len(fake_input_ids) + features[0]["attention_mask"]
features[0]["labels"] = [IGNORE_INDEX] * len(fake_input_ids) + features[0]["labels"]
batch_input_ids[0] = features[0]["input_ids"]
mm_inputs = self.template.mm_plugin.get_mm_inputs(
batch_images,
batch_videos,