cpu_model_runner.py: skip the warm up if CompilationMode.NONE (#50547)

Signed-off-by: YAMAMOTO Takashi <yamamoto@midokura.com>
This commit is contained in:
YAMAMOTO Takashi
2026-08-03 04:34:29 +00:00
committed by GitHub
parent 5c4fe4b17e
commit 5e35a6f4f9
+6 -1
View File
@@ -8,7 +8,10 @@ import torch
import torch.nn as nn
import vllm.utils.cpu_triton_utils as cpu_tl
from vllm.config import VllmConfig
from vllm.config import (
CompilationMode,
VllmConfig,
)
from vllm.logger import init_logger
from vllm.model_executor.model_loader import get_model
from vllm.tracing import instrument
@@ -142,6 +145,8 @@ class CPUModelRunner(GPUModelRunner):
@instrument(span_name="Warmup (CPU)")
def warming_up_model(self) -> None:
if self.vllm_config.compilation_config.mode == CompilationMode.NONE:
return
logger.info("Warming up model for the compilation...")
# Only generate graph for the generic shape
with _set_global_compilation_settings(self.vllm_config):