tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-26 19:55:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: quantized KV support for FA vec (#7527)

Browse Source 
* CUDA: quantized KV support for FA vec

* try CI fix

* fix commented-out kernel variants

* add q8_0 q4_0 tests

* fix nwarps > batch size

* split fattn compile via extern templates

* fix flake8

* fix metal tests

* fix cmake

* make generate_cu_files.py executable

* add autogenerated .cu files

* fix AMD

* error if type_v != FP16 and not flash_attn

* remove obsolete code
This commit is contained in:
Johannes Gäßler
2024-06-01 08:44:14 +02:00
committed by GitHub
parent a323ec60af
commit 9b596417af
110 changed files with 2697 additions and 1200 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
CMakeLists.txt
View File

@ -106,6 +106,7 @@ set(LLAMA_CUDA_PEER_MAX_BATCH_SIZE "128" CACHE STRING
"llama: max. batch size for using peer access")
option(LLAMA_CUDA_NO_PEER_COPY "llama: do not use peer to peer copies" OFF)
option(LLAMA_CUDA_NO_VMM "llama: do not try to use CUDA VMM" OFF)
option(LLAMA_CUDA_FA_ALL_QUANTS "llama: compile all quants for FlashAttention" OFF)
option(LLAMA_CURL "llama: use libcurl to download model from an URL" OFF)
option(LLAMA_HIPBLAS "llama: use hipBLAS" OFF)
@ -402,6 +403,8 @@ if (LLAMA_CUDA)
file(GLOB GGML_SOURCES_CUDA "ggml-cuda/*.cu")
list(APPEND GGML_SOURCES_CUDA "ggml-cuda.cu")
file(GLOB SRCS "ggml-cuda/template-instances/fattn-wmma*.cu")
list(APPEND GGML_SOURCES_CUDA ${SRCS})
add_compile_definitions(GGML_USE_CUDA)
add_compile_definitions(GGML_CUDA_USE_GRAPHS)
@ -427,6 +430,18 @@ if (LLAMA_CUDA)
if (LLAMA_CUDA_NO_PEER_COPY)
add_compile_definitions(GGML_CUDA_NO_PEER_COPY)
endif()
if (LLAMA_CUDA_FA_ALL_QUANTS)
file(GLOB SRCS "ggml-cuda/template-instances/fattn-vec*.cu")
list(APPEND GGML_SOURCES_CUDA ${SRCS})
add_compile_definitions(GGML_CUDA_FA_ALL_QUANTS)
else()
file(GLOB SRCS "ggml-cuda/template-instances/fattn-vec*q4_0-q4_0.cu")
list(APPEND GGML_SOURCES_CUDA ${SRCS})
file(GLOB SRCS "ggml-cuda/template-instances/fattn-vec*q8_0-q8_0.cu")
list(APPEND GGML_SOURCES_CUDA ${SRCS})
file(GLOB SRCS "ggml-cuda/template-instances/fattn-vec*f16-f16.cu")
list(APPEND GGML_SOURCES_CUDA ${SRCS})
endif()
if (LLAMA_STATIC)
if (WIN32)
@ -571,6 +586,8 @@ if (LLAMA_HIPBLAS)
file(GLOB GGML_SOURCES_ROCM "ggml-cuda/*.cu")
list(APPEND GGML_SOURCES_ROCM "ggml-cuda.cu")
file(GLOB SRCS "ggml-cuda/template-instances/fattn-wmma*.cu")
list(APPEND GGML_SOURCES_ROCM ${SRCS})
add_compile_definitions(GGML_USE_HIPBLAS GGML_USE_CUDA)
@ -590,6 +607,19 @@ if (LLAMA_HIPBLAS)
add_compile_definitions(GGML_CUDA_NO_PEER_COPY)
endif()
if (LLAMA_CUDA_FA_ALL_QUANTS)
file(GLOB SRCS "ggml-cuda/template-instances/fattn-vec*.cu")
list(APPEND GGML_SOURCES_ROCM ${SRCS})
add_compile_definitions(GGML_CUDA_FA_ALL_QUANTS)
else()
file(GLOB SRCS "ggml-cuda/template-instances/fattn-vec*q4_0-q4_0.cu")
list(APPEND GGML_SOURCES_ROCM ${SRCS})
file(GLOB SRCS "ggml-cuda/template-instances/fattn-vec*q8_0-q8_0.cu")
list(APPEND GGML_SOURCES_ROCM ${SRCS})
file(GLOB SRCS "ggml-cuda/template-instances/fattn-vec*f16-f16.cu")
list(APPEND GGML_SOURCES_ROCM ${SRCS})
endif()
add_compile_definitions(GGML_CUDA_DMMV_X=${LLAMA_CUDA_DMMV_X})
add_compile_definitions(GGML_CUDA_MMV_Y=${LLAMA_CUDA_MMV_Y})
add_compile_definitions(K_QUANTS_PER_ITERATION=${LLAMA_CUDA_KQUANTS_ITER})