zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-12-16 01:49:05 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Fp8 conversion (#2686)

Browse Source 
* add fp8 e4m3 converters

* add cuda

* default saturate to min/max

* fix for older OS

* fix no gpu/cpu

* fix saturate

* fix compile
This commit is contained in:
Awni Hannun
2025-10-27 16:35:50 -07:00
committed by GitHub
parent d1e06117e8
commit 969924cc69
23 changed files with 363 additions and 117 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
mlx/backend/cuda/CMakeLists.txt
View File

@@ -52,6 +52,7 @@ target_sources(
${CMAKE_CURRENT_SOURCE_DIR}/utils.cpp
${CMAKE_CURRENT_SOURCE_DIR}/quantized/affine_quantize.cu
${CMAKE_CURRENT_SOURCE_DIR}/quantized/quantized.cpp
${CMAKE_CURRENT_SOURCE_DIR}/quantized/convert_fp8.cu
${CMAKE_CURRENT_SOURCE_DIR}/worker.cpp)
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/binary)
@@ -170,11 +171,6 @@ target_link_libraries(mlx PRIVATE CUDNN::cudnn_all)
# Suppress nvcc warnings on MLX headers.
target_compile_options(mlx PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:-Xcudafe
--diag_suppress=997>)
# Supress warnings: note: parameter passing for argument of type
# ‘std::pair<float, float>’ when C++17 is enabled changed to match C++14 in GCC
# 10.1
target_compile_options(mlx PRIVATE -Wno-psabi)
# Install CCCL headers for JIT.
install(DIRECTORY ${cccl_SOURCE_DIR}/include/cuda
DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/cccl)