423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-11-28 11:14:18 +08:00
Code Issues Packages Projects Releases Wiki Activity

[data] fix qwen3omni audio length calculation (#9467)

Browse Source
This commit is contained in:
Kingsley 2025-11-12 10:37:15 +08:00 committed by GitHub
parent 5afa851f71
commit fef86fa7fe
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 8 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
src/llamafactory/data/mm_plugin.py
View File

@ -1885,8 +1885,14 @@ class Qwen2OmniPlugin(Qwen2VLPlugin):
image_grid_thw = mm_inputs.get("image_grid_thw", [])
video_grid_thw = mm_inputs.get("video_grid_thw", [])
if "feature_attention_mask" in mm_inputs:
input_lengths = (mm_inputs["feature_attention_mask"].sum(-1).numpy() - 1) // 2 + 1
audio_lengths = (input_lengths - 2) // 2 + 1
if processor.__class__.__name__ == "Qwen3OmniMoeProcessor": # for qwen3omni
input_lengths = mm_inputs["feature_attention_mask"].sum(-1)
input_lengths_leave = input_lengths % 100
feature_lengths = (input_lengths_leave - 1) // 2 + 1
audio_lengths = ((feature_lengths - 1) // 2 + 1 - 1) // 2 + 1 + (input_lengths // 100) * 13
else:
input_lengths = (mm_inputs["feature_attention_mask"].sum(-1).numpy() - 1) // 2 + 1
audio_lengths = (input_lengths - 2) // 2 + 1
else:
mm_inputs = {}
image_grid_thw = [None] * len(images)