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 15, 02:15 UTC
Revert "Use NVCC --compress-mode to reduce binary size by 30% #20694" (#20853)
Signed-off-by: mgoin <mgoin64@gmail.com>
Michael Goin committed
Jul 12, 2025 at 15:07 UTC
b639327ad94b3aa16022ebea49f8e525660b736b
1 file changed
-10
CMakeLists.txt
-10
@@ -171,16 +171,6 @@ if(NVCC_THREADS AND VLLM_GPU_LANG STREQUAL "CUDA")
171
list(APPEND VLLM_GPU_FLAGS "--threads=${NVCC_THREADS}")
172
endif()
173
174
-#
175
-# Set nvcc fatbin compression.
176
-#
177
-if(VLLM_GPU_LANG STREQUAL "CUDA")
178
- if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
179
- list(APPEND VLLM_GPU_FLAGS "-Xfatbin" "-compress-all" "-compress-mode=size")
180
- endif()
181
-endif()
182
-
183
-
174
#
175
# Use FetchContent for C++ dependencies that are compiled as part of vLLM's build process.
176
# setup.py will override FETCHCONTENT_BASE_DIR to play nicely with sccache.