A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 13:30 UTC
[Model] Qwen2.5-Omni Cleanup (#16872)
Roger Wang committed
Apr 19, 2025 at 02:37 UTC
5124f5bf51b83e6f344c1bc6652e8c4d81313b34
2 files changed
+2
-5
docs/source/models/supported_models.md
+2
-2
@@ -1117,8 +1117,8 @@ Our PaliGemma implementations have the same problem as Gemma 3 (see above) for b
1117
:::
1118
1119
:::{note}
1120
-To use Qwen2.5-Omni, you have to install a fork of Hugging Face Transformers library from source via
1121
-`pip install git+https://github.com/BakerBunker/transformers.git@qwen25omni`.
1120
+To use Qwen2.5-Omni, you have to install Hugging Face Transformers library from source via
1121
+`pip install git+https://github.com/huggingface/transformers.git`.
1122
1123
Read audio from video pre-processing is currently supported on V0 (but not V1), because overlapping modalities is not yet supported in V1.
1124
`--mm-processor-kwargs '{"use_audio_in_video": True}'`.
vllm/model_executor/models/qwen2_5_omni_thinker.py
-3
@@ -518,9 +518,6 @@ class Qwen2_5OmniThinkerMultiModalProcessor(
518
"""
519
Qwen2.5-Omni reimplements this function to handle text only.
520
"""
521
- print(prompt)
522
- print(hf_processor_mm_kwargs)
523
- print(mm_items)
521
if isinstance(prompt, str):
522
if enable_hf_prompt_update:
523
return self._apply_hf_processor_text_mm(