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

8
tests/multimodal/test_cache.py
View File

@@ -10,8 +10,8 @@ from vllm.config import ModelConfig, ParallelConfig, VllmConfig
from vllm.multimodal.cache import (MultiModalCache,
MultiModalProcessorCacheItem,
MultiModalProcessorCacheItemMetadata,
processor_cache_from_config,
receiver_cache_from_config)
engine_receiver_cache_from_config,
processor_cache_from_config)
from vllm.multimodal.hasher import MultiModalHasher
from vllm.multimodal.inputs import (MultiModalFieldElem, MultiModalKwargsItem,
MultiModalKwargsItems,
@@ -115,9 +115,9 @@ def _compare_caches(
):
mm_registry = MultiModalRegistry()
cache_0_p0 = processor_cache_from_config(config_0, mm_registry)
cache_0_p1 = receiver_cache_from_config(config_0, mm_registry)
cache_0_p1 = engine_receiver_cache_from_config(config_0, mm_registry)
cache_1_p0 = processor_cache_from_config(config_1, mm_registry)
cache_1_p1 = receiver_cache_from_config(config_1, mm_registry)
cache_1_p1 = engine_receiver_cache_from_config(config_1, mm_registry)
cache_size_gb = max(
config_0.model_config.mm_processor_cache_gb,