github/gpt4all
1
0
Fork 0
mirror of https://github.com/nomic-ai/gpt4all.git synced 2025-09-06 19:10:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

python: reduce size of wheels built by CI, other build tweaks (#2802)

Browse Source 
* Read CMAKE_CUDA_ARCHITECTURES directly
* Disable CUBINs for python build in CI
* Search for CUDA 11 as well as CUDA 12

Signed-off-by: Jared Van Bortel <jared@nomic.ai>
This commit is contained in:
Jared Van Bortel
2024-08-07 11:27:50 -04:00
committed by GitHub
parent be66ec8ab5
commit de7cb36fcc
6 changed files with 42 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
gpt4all-backend/CMakeLists.txt
View File

@@ -63,10 +63,6 @@ if (LLMODEL_VULKAN)
list(APPEND BUILD_VARIANTS vulkan vulkan-avxonly)
endif()
if (LLMODEL_CUDA)
if (DEFINED CMAKE_CUDA_ARCHITECTURES)
set(GGML_CUDA_ARCHITECTURES "${CMAKE_CUDA_ARCHITECTURES}")
endif()
include(CheckLanguage)
check_language(CUDA)
if (NOT CMAKE_CUDA_COMPILER)

13
gpt4all-backend/llama.cpp.cmake
View File

@@ -378,19 +378,19 @@ function(include_ggml SUFFIX)
find_package(CUDAToolkit REQUIRED)
set(CUDAToolkit_BIN_DIR ${CUDAToolkit_BIN_DIR} PARENT_SCOPE)
if (NOT DEFINED GGML_CUDA_ARCHITECTURES)
if (NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
# 52 == lowest CUDA 12 standard
# 60 == f16 CUDA intrinsics
# 61 == integer CUDA intrinsics
# 70 == compute capability at which unrolling a loop in mul_mat_q kernels is faster
if (GGML_CUDA_F16 OR GGML_CUDA_DMMV_F16)
set(GGML_CUDA_ARCHITECTURES "60;61;70;75") # needed for f16 CUDA intrinsics
set(CMAKE_CUDA_ARCHITECTURES "60;61;70;75") # needed for f16 CUDA intrinsics
else()
set(GGML_CUDA_ARCHITECTURES "52;61;70;75") # lowest CUDA 12 standard + lowest for integer intrinsics
#set(GGML_CUDA_ARCHITECTURES "OFF") # use this to compile much faster, but only F16 models work
set(CMAKE_CUDA_ARCHITECTURES "52;61;70;75") # lowest CUDA 12 standard + lowest for integer intrinsics
#set(CMAKE_CUDA_ARCHITECTURES "OFF") # use this to compile much faster, but only F16 models work
endif()
endif()
message(STATUS "Using CUDA architectures: ${GGML_CUDA_ARCHITECTURES}")
message(STATUS "Using CUDA architectures: ${CMAKE_CUDA_ARCHITECTURES}")
set(GGML_HEADERS_CUDA ${DIRECTORY}/ggml/include/ggml-cuda.h)
file(GLOB GGML_HEADERS_CUDA "${DIRECTORY}/ggml/src/ggml-cuda/*.cuh")
@@ -1018,9 +1018,6 @@ function(include_ggml SUFFIX)
C_STANDARD 11
C_STANDARD_REQUIRED true
)
if (GGML_CUDA_ARCHITECTURES)
set_property(TARGET ggml${SUFFIX} llama${SUFFIX} PROPERTY CUDA_ARCHITECTURES "${GGML_CUDA_ARCHITECTURES}")
endif()
target_compile_options(ggml${SUFFIX} PRIVATE "${GGML_COMPILE_OPTS}")
target_compile_options(llama${SUFFIX} PRIVATE "${GGML_COMPILE_OPTS}")