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] refactor code by import as for torch._inductor.config (#23677)
Signed-off-by: Andy Xie <andy.xning@gmail.com>
Ning Xie committed
Sep 1, 2025 at 14:05 UTC
499b074bfd7d131dc90bba718b46ea4f35c7436e
1 file changed
+4
-4
vllm/v1/worker/cpu_model_runner.py
+4
-4
@@ -138,14 +138,14 @@ def _torch_cuda_wrapper():
138
139
@contextmanager
140
def _set_global_compilation_settings(config: VllmConfig):
141
- import torch._inductor.config
141
+ import torch._inductor.config as torch_inductor_config
142
143
inductor_config = config.compilation_config.inductor_compile_config
144
# Note: The MKLDNN and CPPGEMM backend requires freezing parameters.
145
- freezing_value = torch._inductor.config.freezing
145
+ freezing_value = torch_inductor_config.freezing
146
try:
147
if inductor_config.get("max_autotune", False):
148
- torch._inductor.config.freezing = True
148
+ torch_inductor_config.freezing = True
149
yield
150
finally:
151
- torch._inductor.config.freezing = freezing_value
151
+ torch_inductor_config.freezing = freezing_value