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 1 Packages Projects Releases Wiki Activity

Add quantize/dequantize for mxfp8 and nvfp4 (#2688)

Browse Source 
* Add quantize/dequantize slow path for mxfp8 and nvfp4

* fast cuda kernel for mx/nv quantization

* fallback for cuda < 12.8 (#2697)

* format (#2700)

* fix (#2701)

* metal kernels

* docs

* fix jit

* add default bits and group sizes

* improve quant docs

* fix output type of mxfp4 matmuls
This commit is contained in:
Awni Hannun
2025-10-28 16:23:12 -07:00
committed by GitHub
parent 460691a0e8
commit ec72b44417
25 changed files with 1400 additions and 588 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -51,6 +51,7 @@ target_sources(
${CMAKE_CURRENT_SOURCE_DIR}/ternary.cu
${CMAKE_CURRENT_SOURCE_DIR}/utils.cpp
${CMAKE_CURRENT_SOURCE_DIR}/quantized/affine_quantize.cu
${CMAKE_CURRENT_SOURCE_DIR}/quantized/fp_quantize.cu
${CMAKE_CURRENT_SOURCE_DIR}/quantized/quantized.cpp
${CMAKE_CURRENT_SOURCE_DIR}/quantized/convert_fp8.cu
${CMAKE_CURRENT_SOURCE_DIR}/worker.cpp)
@@ -58,6 +59,11 @@ target_sources(
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/binary)
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/unary)
# fp4 is not available on < 12.8
if(CMAKE_CUDA_COMPILER_VERSION VERSION_LESS 12.8.0)
target_include_directories(mlx PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/quantized/)
endif()
if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.9.0)
target_sources(
mlx PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/gemms/cublas_gemm_batched_12_9.cu)