A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[nit]: Fix import for the lmcache integration (#27600)
Signed-off-by: Samuel Shen <slshen@uchicago.edu> Co-authored-by: Samuel Shen <slshen@uchicago.edu>
Samuel Shen committed
Oct 28, 2025 at 07:40 UTC
05e034f085b0255b171ff494938a775f2fe80ee1
3 files changed
+11
-4
vllm/distributed/kv_transfer/kv_connector/v1/lmcache_connector.py
+5
-3
@@ -13,9 +13,6 @@ from vllm.distributed.kv_transfer.kv_connector.v1.base import (
13
KVConnectorMetadata,
14
KVConnectorRole,
15
)
16
-from vllm.distributed.kv_transfer.kv_connector.v1.lmcache_integration import (
17
- vllm_v1_adapter as _adapter,
18
-)
16
from vllm.logger import init_logger
17
from vllm.v1.core.sched.output import SchedulerOutput
18
@@ -37,6 +34,11 @@ class LMCacheConnectorV1(KVConnectorBase_V1):
34
)
35
if use_native:
36
logger.info("Initializing native LMCache connector")
37
+ # lazy import
38
+ from vllm.distributed.kv_transfer.kv_connector.v1 import lmcache_integration
39
+
40
+ _adapter = lmcache_integration.vllm_v1_adapter
41
+
42
cls = _adapter.LMCacheConnectorV1Impl
43
else:
44
logger.info("Initializing latest dev LMCache connector")
vllm/distributed/kv_transfer/kv_connector/v1/lmcache_integration/__init__.py
+5
@@ -1,2 +1,7 @@
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
+
4
+
5
+from . import vllm_v1_adapter
6
+
7
+__all__ = ["vllm_v1_adapter"]
vllm/distributed/kv_transfer/kv_connector/v1/lmcache_integration/utils.py
+1
-1
@@ -131,7 +131,7 @@ def create_lmcache_metadata(
131
# First Party
132
from lmcache.config import LMCacheEngineMetadata
133
134
- from vllm.utils import get_kv_cache_torch_dtype
134
+ from vllm.utils.torch_utils import get_kv_cache_torch_dtype
135
136
config = lmcache_get_or_create_config()
137
# Support both vllm_config object and individual config parameters