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

[Core] Shared memory based object store for Multimodal data caching and IPC (#20452)

Browse Source 
Signed-off-by: donglu <donglu@cohere.com>
This commit is contained in:
dongluw
2025-09-12 10:54:17 -04:00
committed by GitHub
parent 9f04d9d55f
commit a5b84f1cbf
17 changed files with 1487 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/v1/engine/core.py
View File

@@ -23,7 +23,7 @@ from vllm.logger import init_logger
from vllm.logging_utils.dump_input import dump_engine_exception
from vllm.lora.request import LoRARequest
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.multimodal.cache import receiver_cache_from_config
from vllm.multimodal.cache import engine_receiver_cache_from_config
from vllm.tasks import POOLING_TASKS, SupportedTask
from vllm.transformers_utils.config import (
maybe_register_config_serialize_by_value)
@@ -131,7 +131,7 @@ class EngineCore:
self.use_spec_decode = vllm_config.speculative_config is not None
self.mm_registry = mm_registry = MULTIMODAL_REGISTRY
self.mm_receiver_cache = receiver_cache_from_config(
self.mm_receiver_cache = engine_receiver_cache_from_config(
vllm_config, mm_registry)
# Setup batch queue for pipeline parallelism.