[0.11.0][Perf] Add padding vision tower for Qwen2_5_Omni (#4041)

### What this PR does / why we need it?
This PR repalce the vision tower in Qwen2.5-Omni-Thinker model,
Qwen2_5_VisionTransformer, with AscendQwen2_5_VisionTransformer, which
use QKV padding for padding performance.

- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

Signed-off-by: Ting FU <futing10@huawei.com>
This commit is contained in:
tingfu
2025-11-08 13:56:05 +08:00
committed by GitHub
parent d4e2a44307
commit f9842560cb
3 changed files with 66 additions and 0 deletions

View File

@@ -23,6 +23,10 @@ def register_model():
"Qwen2_5_VLForConditionalGeneration",
"vllm_ascend.models.qwen2_5_vl:AscendQwen2_5_VLForConditionalGeneration"
)
ModelRegistry.register_model(
"Qwen2_5OmniModel",
"vllm_ascend.models.qwen2_5_omni_thinker:AscendQwen2_5OmniThinkerForConditionalGeneration"
)
else:
ModelRegistry.register_model(
"Qwen2_5_VLForConditionalGeneration",