tcsenpai/ollama
1
0
Fork 0
mirror of https://github.com/tcsenpai/ollama.git synced 2025-06-10 04:57:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

wip

Browse Source
This commit is contained in:
jmorganca 2024-05-20 15:27:10 -07:00
parent 922d0acbdb
commit 28bedcd807
2 changed files with 11 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
llama/build_hipblas.sh Normal file → Executable file
View File

@ -22,8 +22,11 @@ linux_archs=(
os="$(uname -s)" os="$(uname -s)"
additional_flags=""
if [[ "$os" == "Windows_NT" || "$os" == "MINGW64_NT"* ]]; then if [[ "$os" == "Windows_NT" || "$os" == "MINGW64_NT"* ]]; then
output="ggml-hipblas.dll" output="ggml-hipblas.dll"
additional_flags="-Xclang --dependent-lib=msvcrt -Wl,/subsystem:console"
else else
output="libggml-hipblas.so" output="libggml-hipblas.so"
archs+=("${linux_archs[@]}") archs+=("${linux_archs[@]}")
@ -37,7 +40,6 @@ done
hipcc \ hipcc \
-parallel-jobs=12 \ -parallel-jobs=12 \
-O3 \ -O3 \
$offload_arch_flags \
-DGGML_USE_CUDA \ -DGGML_USE_CUDA \
-DGGML_BUILD=1 \ -DGGML_BUILD=1 \
-DGGML_SHARED=1 \ -DGGML_SHARED=1 \
@ -51,7 +53,9 @@ hipcc \
-DK_QUANTS_PER_ITERATION=2 \ -DK_QUANTS_PER_ITERATION=2 \
-D_CRT_SECURE_NO_WARNINGS \ -D_CRT_SECURE_NO_WARNINGS \
-DCMAKE_POSITION_INDEPENDENT_CODE=on \ -DCMAKE_POSITION_INDEPENDENT_CODE=on \
-Xclang --dependent-lib=msvcrt -Wl,/subsystem:console \ -D_GNU_SOURCE \
$offload_arch_flags \
$additional_flags \
-Wno-expansion-to-defined \ -Wno-expansion-to-defined \
-Wno-invalid-noreturn \ -Wno-invalid-noreturn \
-Wno-ignored-attributes \ -Wno-ignored-attributes \

7
llama/llama.go
View File

@ -17,10 +17,13 @@ package llama
// #cgo avx2 CXXFLAGS: -mavx2 -mfma // #cgo avx2 CXXFLAGS: -mavx2 -mfma
// #cgo cuda CFLAGS: -DGGML_USE_CUDA -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 -DGGML_CUDA_MMV_Y=1 -DGGML_BUILD=1 // #cgo cuda CFLAGS: -DGGML_USE_CUDA -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 -DGGML_CUDA_MMV_Y=1 -DGGML_BUILD=1
// #cgo cuda CXXFLAGS: -DGGML_USE_CUDA -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 -DGGML_CUDA_MMV_Y=1 -DGGML_BUILD=1 // #cgo cuda CXXFLAGS: -DGGML_USE_CUDA -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 -DGGML_CUDA_MMV_Y=1 -DGGML_BUILD=1
// #cgo rocm CXXFLAGS: -DGGML_USE_CUDA -DGGML_USE_HIPBLAS -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 -DGGML_MULTIPLATFORM -DGGML_CUDA_MMV_Y=1 -DGGML_BUILD=1 // #cgo rocm CFLAGS: -DGGML_USE_CUDA -DGGML_USE_HIPBLAS -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 -DGGML_CUDA_MMV_Y=1 -DGGML_BUILD=1
// #cgo rocm CXXFLAGS: -DGGML_USE_CUDA -DGGML_USE_HIPBLAS -DGGML_CUDA_DMMV_X=32 -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 -DGGML_CUDA_MMV_Y=1 -DGGML_BUILD=1
// #cgo rocm LDFLAGS: -L${SRCDIR} -lggml-hipblas -lhipblas -lamdhip64 -lrocblas
// #cgo windows,cuda LDFLAGS: -L. -L"C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.3/lib/x64" -lggml-cuda -lcuda -lcudart -lcublas -lcublasLt // #cgo windows,cuda LDFLAGS: -L. -L"C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.3/lib/x64" -lggml-cuda -lcuda -lcudart -lcublas -lcublasLt
// #cgo windows,rocm LDFLAGS: -L. -L"C:/Program Files/AMD/ROCm/5.7/lib" -lggml-hipblas -lhipblas -lamdhip64 -lrocblas // #cgo windows,rocm LDFLAGS: -L. -L"C:/Program Files/AMD/ROCm/5.7/lib"
// #cgo linux,cuda LDFLAGS: -L${SRCDIR} -L/usr/local/cuda/lib64 -lggml-cuda -lcuda -lcudart -lcublas -lcublasLt -lpthread -ldl -lrt // #cgo linux,cuda LDFLAGS: -L${SRCDIR} -L/usr/local/cuda/lib64 -lggml-cuda -lcuda -lcudart -lcublas -lcublasLt -lpthread -ldl -lrt
// #cgo linux,rocm LDFLAGS: -L/opt/rocm/lib
// #include <stdlib.h> // #include <stdlib.h>
// #include "llama.h" // #include "llama.h"
// #include "clip.h" // #include "clip.h"