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
[Misc] local import code clean (#23420)
Signed-off-by: Andy Xie <andy.xning@gmail.com>
Ning Xie committed
Aug 22, 2025 at 22:01 UTC
325aa3dee922b344a26b9e74d9ae3c769828e70e
2 files changed
+1
-3
vllm/v1/worker/gpu_worker.py
-1
@@ -292,7 +292,6 @@ class Worker(WorkerBase):
292
allocator = CuMemAllocator.get_instance()
293
context = allocator.use_memory_pool(tag="kv_cache")
294
else:
295
- from contextlib import nullcontext
295
context = nullcontext()
296
with context:
297
self.model_runner.initialize_kv_cache(kv_cache_config)
vllm/worker/worker.py
+1
-2
@@ -3,6 +3,7 @@
3
"""A GPU worker class."""
4
import gc
5
import os
6
+from contextlib import nullcontext
7
from typing import Dict, List, Optional, Set, Tuple, Type, Union
8
9
import torch
@@ -206,7 +207,6 @@ class Worker(LocalOrDistributedWorkerBase):
207
"used for one instance per process.")
208
context = allocator.use_memory_pool(tag="weights")
209
else:
209
- from contextlib import nullcontext
210
context = nullcontext()
211
with context:
212
self.model_runner.load_model()
@@ -330,7 +330,6 @@ class Worker(LocalOrDistributedWorkerBase):
330
allocator = CuMemAllocator.get_instance()
331
context = allocator.use_memory_pool(tag="kv_cache")
332
else:
333
- from contextlib import nullcontext
333
context = nullcontext()
334
with context:
335
self._init_cache_engine()