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 2, 08:30 UTC
[ROCm] cmake: support PYTORCH_FOUND_HIP for torch 2.13 native HIP language support (#43881)
Signed-off-by: nemanjaudovic <nudovic@amd.com> Co-authored-by: Shengqi Chen <harry-chen@outlook.com>
nemanjaudovic committed
May 30, 2026 at 07:16 UTC
c0056b19bf4930ae7830e753b048b3daca0fbfee
1 file changed
+2
-2
CMakeLists.txt
+2
-2
@@ -144,14 +144,14 @@ endif()
144
# Set up GPU language and check the torch version and warn if it isn't
145
# what is expected.
146
#
147
-if (NOT HIP_FOUND AND CUDA_FOUND)
147
+if (NOT HIP_FOUND AND NOT PYTORCH_FOUND_HIP AND CUDA_FOUND)
148
set(VLLM_GPU_LANG "CUDA")
149
150
if (NOT Torch_VERSION VERSION_EQUAL ${TORCH_SUPPORTED_VERSION_CUDA})
151
message(WARNING "Pytorch version ${TORCH_SUPPORTED_VERSION_CUDA} "
152
"expected for CUDA build, saw ${Torch_VERSION} instead.")
153
endif()
154
-elseif(HIP_FOUND)
154
+elseif(HIP_FOUND OR PYTORCH_FOUND_HIP)
155
set(VLLM_GPU_LANG "HIP")
156
157
# Importing torch recognizes and sets up some HIP/ROCm configuration but does