biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[CustomOp][MM] Extract MMEncoderAttention as CustomOp and replace the backend of QwenVisionAttention with it. (#30125)

Browse Source 
Signed-off-by: shen-shanshan <467638484@qq.com>
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Co-authored-by: tjtanaa <tunjian.tan@embeddedllm.com>
This commit is contained in:
Shanshan Shen
2025-12-15 11:13:32 +08:00
committed by GitHub
parent 84e23d103d
commit 87b4d1557d
24 changed files with 1262 additions and 851 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/model_executor/models/opencua.py
View File

@@ -240,18 +240,12 @@ class OpenCUAForConditionalGeneration(Qwen2_5_VLForConditionalGeneration):
)
if multimodal_config.get_limit_per_prompt("image"):
attn_backend_override = (
multimodal_config.mm_encoder_attn_backend
if multimodal_config is not None
else None
)
self.visual = OpenCUAVisionTransformer(
vision_config=config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=self.quant_config,
multimodal_config=self.multimodal_config,
prefix=maybe_prefix(prefix, "visual"),
use_data_parallel=self.use_data_parallel,
attn_backend_override=attn_backend_override,
)
else:
self.visual = None