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 16, 18:15 UTC
[Bugfix] Pass PYTHONPATH from setup.py to CMake (#7730)
sasha0552 committed
Aug 21, 2024 at 18:17 UTC
1b32e0264888200a0e6187496a816ef597a7f320
2 files changed
+5
-1
CMakeLists.txt
+1
-1
@@ -233,7 +233,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
233
# Generate sources:
234
execute_process(
235
COMMAND ${CMAKE_COMMAND} -E env
236
- PYTHONPATH=${CMAKE_CURRENT_SOURCE_DIR}/csrc/cutlass_extensions/:${CUTLASS_DIR}/python/:$PYTHONPATH
236
+ PYTHONPATH=${CMAKE_CURRENT_SOURCE_DIR}/csrc/cutlass_extensions/:${CUTLASS_DIR}/python/:${VLLM_PYTHON_PATH}:$PYTHONPATH
237
${Python_EXECUTABLE} ${CMAKE_CURRENT_SOURCE_DIR}/csrc/quantization/machete/generate.py
238
RESULT_VARIABLE machete_generation_result
239
OUTPUT_VARIABLE machete_generation_output
setup.py
+4
@@ -184,6 +184,10 @@ class cmake_build_ext(build_ext):
184
# match.
185
cmake_args += ['-DVLLM_PYTHON_EXECUTABLE={}'.format(sys.executable)]
186
187
+ # Pass the python path to cmake so it can reuse the build dependencies
188
+ # on subsequent calls to python.
189
+ cmake_args += ['-DVLLM_PYTHON_PATH={}'.format(":".join(sys.path))]
190
+
191
#
192
# Setup parallelism and build tool
193
#