extern/whisper.cpp
1
0
Fork 1
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2024-11-07 08:34:37 +01:00
Code Issues Packages Projects Releases Wiki Activity

ggml : disable CUDA graphs for non-llama.cpp projects

Browse Source
This commit is contained in:
Georgi Gerganov 2024-06-26 20:14:22 +03:00
parent dc8cc2dd6f
commit ceb77363cd
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 10 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
ggml/src/CMakeLists.txt
View File

@ -295,7 +295,16 @@ if (GGML_CUDA)
list(APPEND GGML_CDEF_PUBLIC GGML_USE_CUDA)
add_compile_definitions(GGML_CUDA_USE_GRAPHS)
# TODO: for now CUDA graphs should be used only with llama.cpp
# https://github.com/ggerganov/whisper.cpp/issues/2258
message(STATUS "CMAKE_PROJECT_NAME: ${CMAKE_PROJECT_NAME}")
if (CMAKE_PROJECT_NAME STREQUAL "llama.cpp")
add_compile_definitions(GGML_CUDA_USE_GRAPHS)
message(STATUS "GGML_CUDA_USE_GRAPHS enabled")
else()
message(STATUS "GGML_CUDA_USE_GRAPHS disabled")
endif()
add_compile_definitions(GGML_CUDA_DMMV_X=${GGML_CUDA_DMMV_X})
add_compile_definitions(GGML_CUDA_MMV_Y=${GGML_CUDA_MMV_Y})
add_compile_definitions(K_QUANTS_PER_ITERATION=${GGML_CUDA_KQUANTS_ITER})