ros/LlamaFactory
1
0
Fork 0
mirror of https://github.com/hiyouga/LlamaFactory.git synced 2026-03-19 11:23:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix:MiniCPMVPlugin IndexError in process_messages when training with video (#10276)

Browse Source 
Co-authored-by: xxddccaa <xxddccaa@users.noreply.github.com>
This commit is contained in:
xxddccaa
2026-03-18 19:18:06 +08:00
committed by GitHub
parent 2c4f121817
commit e67ab9e2f2
1 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/llamafactory/data/mm_plugin.py
View File

@@ -1058,7 +1058,9 @@ class MiniCPMVPlugin(BasePlugin):
chunk_input=True,
sampling_rate=getattr(processor, "audio_sampling_rate", 16000),
)
audio_feature_lens = [torch.tensor(audio_feature_len) for audio_feature_len in audio_feature_lens]
audio_feature_lens = [
x.clone().detach() if isinstance(x, torch.Tensor) else torch.tensor(x) for x in audio_feature_lens
]
mm_inputs.update({"audio_features": audio_features, "audio_feature_lens": audio_feature_lens})
if kwargs.get("ret_phs", False):
mm_inputs.update({"audio_phs": audio_phs})
@@ -1098,7 +1100,7 @@ class MiniCPMVPlugin(BasePlugin):
num_image_tokens += 1
while VIDEO_PLACEHOLDER in content:
video_seqlen = len(mm_inputs["pixel_values"][num_video_tokens]) if self.expand_mm_tokens else 1
video_seqlen = len(mm_inputs["image_sizes"][num_video_tokens]) if self.expand_mm_tokens else 1
content = content.replace(VIDEO_PLACEHOLDER, "{{image}}" * video_seqlen, 1)
num_video_tokens += 1