ros/LlamaFactory
1
0
Fork 0
mirror of https://github.com/hiyouga/LlamaFactory.git synced 2026-03-16 22:33:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

[fix] make position_id_per_seconds configurable for Qwen2OmniPlugin (#10281)

Browse Source 
Co-authored-by: LincolnBurrows2017 <lincoln@example.com>
This commit is contained in:
LincolnBurrows2017
2026-03-16 19:42:38 +08:00
committed by GitHub
parent 246192abd2
commit 70653026f5
1 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/llamafactory/data/mm_plugin.py
View File

@@ -1987,6 +1987,7 @@ class Qwen2OmniPlugin(Qwen2VLPlugin):
f"Each {VIDEO_PLACEHOLDER} must be followed by an {AUDIO_PLACEHOLDER} when using audio in video." f"Each {VIDEO_PLACEHOLDER} must be followed by an {AUDIO_PLACEHOLDER} when using audio in video."
) )
position_id_per_seconds: int = getattr(processor, "position_id_per_seconds", 25)
audio_t_index = torch.arange(audio_lengths[num_audio_tokens]) audio_t_index = torch.arange(audio_lengths[num_audio_tokens])
video_t_index = ( video_t_index = (
torch.arange(video_grid_thw[num_video_tokens][0]) torch.arange(video_grid_thw[num_video_tokens][0])
@@ -1998,9 +1999,9 @@ class Qwen2OmniPlugin(Qwen2VLPlugin):
) )
.flatten() .flatten()
* mm_inputs["video_second_per_grid"][num_video_tokens] * mm_inputs["video_second_per_grid"][num_video_tokens]
* 25 # FIXME hardcode of position_id_per_seconds=25 * position_id_per_seconds
).long() ).long()
t_ntoken_per_chunk = 50 # FIXME hardcode: [25 * 2] t_ntoken_per_chunk = position_id_per_seconds * 2
video_chunk_indices = processor.get_chunked_index(video_t_index, t_ntoken_per_chunk) video_chunk_indices = processor.get_chunked_index(video_t_index, t_ntoken_per_chunk)
audio_chunk_indices = processor.get_chunked_index(audio_t_index, t_ntoken_per_chunk) audio_chunk_indices = processor.get_chunked_index(audio_t_index, t_ntoken_per_chunk)
placeholder_string = "" placeholder_string = ""