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

[Bugfix] Avoid transferring cached multi-modal items from P0 to P1 (#16273)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-04-09 15:51:27 +08:00
committed by GitHub
parent 24f6b9a713
commit e484e02857
5 changed files with 65 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/v1/engine/core.py
View File

@@ -31,7 +31,7 @@ from vllm.v1.core.sched.output import SchedulerOutput
from vllm.v1.core.sched.scheduler import Scheduler as V1Scheduler
from vllm.v1.engine import (EngineCoreOutputs, EngineCoreRequest,
EngineCoreRequestType, UtilityOutput)
from vllm.v1.engine.mm_input_cache import MMInputCacheServer
from vllm.v1.engine.mm_input_cache import MirroredProcessingCache
from vllm.v1.executor.abstract import Executor
from vllm.v1.kv_cache_interface import KVCacheConfig
from vllm.v1.outputs import ModelRunnerOutput
@@ -105,7 +105,7 @@ class EngineCore:
)
# Setup MM Input Mapper.
self.mm_input_cache_server = MMInputCacheServer(
self.mm_input_cache_server = MirroredProcessingCache(
vllm_config.model_config)
# Setup batch queue for pipeline parallelism.
@@ -173,7 +173,7 @@ class EngineCore:
# anything that has a hash must have a HIT cache entry here
# as well.
assert request.mm_inputs is not None
request.mm_inputs = self.mm_input_cache_server.get_and_update(
request.mm_inputs = self.mm_input_cache_server.get_and_update_p1(
request.mm_inputs, request.mm_hashes)
req = Request.from_engine_core_request(request)