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
[XPU] Fix `compile_size` is `None` case. (#25433)
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
Kunshang Ji committed
Sep 23, 2025 at 11:09 UTC
f225ea7dd98e9f29752e5c032cd4a8ee1d712f16
1 file changed
+2
vllm/platforms/xpu.py
+2
@@ -113,6 +113,8 @@ class XPUPlatform(Platform):
113
# lazy import to avoid circular import
114
from vllm.config import CompilationLevel, CUDAGraphMode
115
compilation_config = vllm_config.compilation_config
116
+ if compilation_config.compile_sizes is None:
117
+ compilation_config.compile_sizes = []
118
119
assert compilation_config.cudagraph_mode == CUDAGraphMode.NONE, \
120
"CUDA graph mode should be NONE on XPU"