xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-01-27 14:46:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Core][MultiModalHasher] Hash images without converting image mode (#24969)

Browse Source 
Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com>
This commit is contained in:
Lukas Geiger 2025-09-17 08:57:34 +01:00 committed by GitHub
parent b77bf34e53
commit 03191cd8f0
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 6 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/multimodal/hasher.py
View File

@ -12,7 +12,6 @@ from blake3 import blake3
from PIL import Image
from vllm.logger import init_logger
from vllm.multimodal.image import convert_image_mode
logger = init_logger(__name__)
@ -35,8 +34,12 @@ class MultiModalHasher:
exif[Image.ExifTags.Base.ImageID], uuid.UUID):
# If the image has exif ImageID tag, use that
return (exif[Image.ExifTags.Base.ImageID].bytes, )
return cls.iter_item_to_bytes(
"image", np.asarray(convert_image_mode(obj, "RGBA")))
data = {"mode": obj.mode, "data": np.asarray(obj)}
if obj.palette is not None:
data["palette"] = obj.palette.palette
if obj.palette.rawmode is not None:
data["palette_rawmode"] = obj.palette.rawmode
return cls.iter_item_to_bytes("image", data)
if isinstance(obj, torch.Tensor):
tensor_obj: torch.Tensor = obj.cpu()
tensor_dtype = tensor_obj.dtype