xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-07-09 10:17:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Model] Deprecate merge_by_field_config=False (#27551)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung 2025-10-27 16:43:00 +08:00 committed by GitHub
parent a806c14cc7
commit b368382964
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 14 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
vllm/multimodal/utils.py
View File

@ -18,6 +18,7 @@ from PIL import Image, UnidentifiedImageError
import vllm.envs as envs import vllm.envs as envs
from vllm.connections import HTTPConnection, global_http_connection from vllm.connections import HTTPConnection, global_http_connection
from vllm.logger import init_logger
from vllm.utils.jsontree import json_map_leaves from vllm.utils.jsontree import json_map_leaves
from .audio import AudioMediaIO from .audio import AudioMediaIO
@ -25,8 +26,6 @@ from .base import MediaIO
from .image import ImageEmbeddingMediaIO, ImageMediaIO from .image import ImageEmbeddingMediaIO, ImageMediaIO
from .video import VideoMediaIO from .video import VideoMediaIO
_M = TypeVar("_M")
if TYPE_CHECKING: if TYPE_CHECKING:
from .inputs import ( from .inputs import (
BatchedTensorInputs, BatchedTensorInputs,
@ -38,11 +37,15 @@ else:
MultiModalKwargsItem = Any MultiModalKwargsItem = Any
MultiModalPlaceholderDict = Any MultiModalPlaceholderDict = Any
logger = init_logger(__name__)
global_thread_pool = ThreadPoolExecutor( global_thread_pool = ThreadPoolExecutor(
max_workers=envs.VLLM_MEDIA_LOADING_THREAD_COUNT max_workers=envs.VLLM_MEDIA_LOADING_THREAD_COUNT
) )
atexit.register(global_thread_pool.shutdown) atexit.register(global_thread_pool.shutdown)
_M = TypeVar("_M")
class MediaConnector: class MediaConnector:
def __init__( def __init__(
@ -413,14 +416,21 @@ def group_mm_kwargs_by_modality(
"`merge_by_field_config` arg, please update your model runner " "`merge_by_field_config` arg, please update your model runner "
"according to https://github.com/vllm-project/vllm/pull/25676." "according to https://github.com/vllm-project/vllm/pull/25676."
) )
if merge_by_field_config is False:
logger.warning_once(
"The legacy code for batching multi-modal kwargs is deprecated and "
"will be removed in v0.12. Please update your model with "
"`merge_by_field_config=True` to use the new code defined by "
"`MultiModalFieldConfig`. You can refer to "
"https://github.com/vllm-project/vllm/issues/26149 "
"for some examples on how to do this."
)
from vllm.multimodal.inputs import MultiModalKwargs, MultiModalKwargsItems from vllm.multimodal.inputs import MultiModalKwargs, MultiModalKwargsItems
for modality, items in groupby(mm_kwargs, key=lambda item: item.modality): for modality, items in groupby(mm_kwargs, key=lambda item: item.modality):
items_lst = list(items) items_lst = list(items)
# TODO: Deprecate `merge_by_field_config` once
# we have migrated all in-tree models
if merge_by_field_config: if merge_by_field_config:
mm_kwargs_group: BatchedTensorInputs = dict( mm_kwargs_group: BatchedTensorInputs = dict(
MultiModalKwargsItems.from_seq(items_lst).get_data( MultiModalKwargsItems.from_seq(items_lst).get_data(