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

[V0 deprecation] Remove VLLM_USE_V1 usage in most modules (#27955)

Browse Source 
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-11-05 12:51:16 +08:00
committed by GitHub
parent 878fd5a16f
commit 428bc7bf1c
19 changed files with 107 additions and 238 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/model_executor/models/gemma3_mm.py
View File

@@ -9,7 +9,6 @@ from torch import nn
from transformers import BatchFeature, Gemma3Config, Gemma3Processor
from transformers.models.gemma3.processing_gemma3 import Gemma3ProcessorKwargs
import vllm.envs as envs
from vllm.config import VllmConfig
from vllm.config.multimodal import BaseDummyOptions
from vllm.logger import init_logger
@@ -137,11 +136,10 @@ class Gemma3ProcessingInfo(BaseProcessingInfo):
if not do_pan_and_scan:
return 0
if envs.VLLM_USE_V1:
logger.warning_once(
"`do_pan_and_scan=True` has suboptimal results on V1 "
"because of the simplified attention pattern being used."
)
logger.warning_once(
"`do_pan_and_scan=True` has suboptimal results on V1 "
"because of the simplified attention pattern being used."
)
# Based on Gemma3ImageProcessor.pan_and_scan
if image_width >= image_height: