third-party-mirrors/ollama
1
0
Fork 1
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

move runner package down

Browse Source
This commit is contained in:
jmorganca 2024-05-18 23:15:51 -07:00
parent 01ccbc07fe
commit 9fe48978a8
7 changed files with 96 additions and 47 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
llama/README.md
View File

@ -58,9 +58,9 @@ go build -tags=cuda .
### ROCm
Install [ROCm 5.7.1](https://rocm.docs.amd.com/en/docs-5.7.1/) and [Strawberry Perl](https://strawberryperl.com/):
Install [ROCm 5.7.1](https://rocm.docs.amd.com/en/docs-5.7.1/) and [Strawberry Perl](https://strawberryperl.com/).
Build `ggml-hipblas.dll`:
Then, build `ggml-hipblas.dll`:
```shell
./hipblas.sh

1
llama/ggml-metal.m
View File

@ -1,3 +1,4 @@
//go:build darwin,arm64
#import "ggml-metal.h"
#import "ggml-backend-impl.h"

3
llama/llama.go
View File

@ -2,8 +2,9 @@ package llama
// #cgo darwin,arm64 CFLAGS: -std=c11 -DGGML_USE_METAL -DGGML_METAL_EMBED_LIBRARY -DGGML_USE_ACCELERATE -DACCELERATE_NEW_LAPACK -DACCELERATE_LAPACK_ILP64
// #cgo darwin,arm64 CXXFLAGS: -std=c++11 -DGGML_USE_METAL -DGGML_METAL_EMBED_LIBRARY -DGGML_USE_ACCELERATE -DACCELERATE_NEW_LAPACK -DACCELERATE_LAPACK_ILP64
// #cgo darwin,amd64 CXXFLAGS: -std=c++11
// #cgo darwin,arm64 LDFLAGS: -ld_classic ${SRCDIR}/ggml-metal.o -framework Foundation -framework Metal -framework MetalKit -framework Accelerate
// #cgo darwin,amd64 CFLAGS: -Wno-incompatible-pointer-types-discards-qualifiers
// #cgo darwin,amd64 CXXFLAGS: -std=c++11 -Wno-incompatible-pointer-types-discards-qualifiers
// #cgo darwin,amd64 LDFLAGS: -ld_classic -framework Foundation -framework Accelerate
// #cgo windows LDFLAGS: -lmsvcrt
// #cgo avx CFLAGS: -mavx

110
llama/sync.sh
View File

@ -6,51 +6,73 @@ src_dir=$1
# Set the destination directory (current directory)
dst_dir="."
# llama.cpp
cp $src_dir/unicode.cpp $dst_dir/unicode.cpp
cp $src_dir/unicode.h $dst_dir/unicode.h
cp $src_dir/unicode-data.cpp $dst_dir/unicode-data.cpp
cp $src_dir/unicode-data.h $dst_dir/unicode-data.h
cp $src_dir/llama.cpp $dst_dir/llama.cpp
cp $src_dir/llama.h $dst_dir/llama.h
cp $src_dir/sgemm.cpp $dst_dir/sgemm.cpp
cp $src_dir/sgemm.h $dst_dir/sgemm.h
# # llama.cpp
# cp $src_dir/unicode.cpp $dst_dir/unicode.cpp
# cp $src_dir/unicode.h $dst_dir/unicode.h
# cp $src_dir/unicode-data.cpp $dst_dir/unicode-data.cpp
# cp $src_dir/unicode-data.h $dst_dir/unicode-data.h
# cp $src_dir/llama.cpp $dst_dir/llama.cpp
# cp $src_dir/llama.h $dst_dir/llama.h
# cp $src_dir/sgemm.cpp $dst_dir/sgemm.cpp
# cp $src_dir/sgemm.h $dst_dir/sgemm.h
# ggml
cp $src_dir/ggml.c $dst_dir/ggml.c
cp $src_dir/ggml.h $dst_dir/ggml.h
cp $src_dir/ggml-quants.c $dst_dir/ggml-quants.c
cp $src_dir/ggml-quants.h $dst_dir/ggml-quants.h
cp $src_dir/ggml-metal.metal $dst_dir/ggml-metal.metal
cp $src_dir/ggml-metal.h $dst_dir/ggml-metal.h
cp $src_dir/ggml-metal.m $dst_dir/ggml-metal.m
cp $src_dir/ggml-impl.h $dst_dir/ggml-impl.h
cp $src_dir/ggml-cuda.h $dst_dir/ggml-cuda.h
cp $src_dir/ggml-cuda.cu $dst_dir/ggml-cuda.cu
cp $src_dir/ggml-common.h $dst_dir/ggml-common.h
cp $src_dir/ggml-backend.h $dst_dir/ggml-backend.h
cp $src_dir/ggml-backend.c $dst_dir/ggml-backend.c
cp $src_dir/ggml-backend-impl.h $dst_dir/ggml-backend-impl.h
cp $src_dir/ggml-alloc.h $dst_dir/ggml-alloc.h
cp $src_dir/ggml-alloc.c $dst_dir/ggml-alloc.c
# # ggml
# cp $src_dir/ggml.c $dst_dir/ggml.c
# cp $src_dir/ggml.h $dst_dir/ggml.h
# cp $src_dir/ggml-quants.c $dst_dir/ggml-quants.c
# cp $src_dir/ggml-quants.h $dst_dir/ggml-quants.h
# cp $src_dir/ggml-metal.metal $dst_dir/ggml-metal.metal
# cp $src_dir/ggml-metal.h $dst_dir/ggml-metal.h
# cp $src_dir/ggml-metal.m $dst_dir/ggml-metal.m
# cp $src_dir/ggml-impl.h $dst_dir/ggml-impl.h
# cp $src_dir/ggml-cuda.h $dst_dir/ggml-cuda.h
# cp $src_dir/ggml-cuda.cu $dst_dir/ggml-cuda.cu
# cp $src_dir/ggml-common.h $dst_dir/ggml-common.h
# cp $src_dir/ggml-backend.h $dst_dir/ggml-backend.h
# cp $src_dir/ggml-backend.c $dst_dir/ggml-backend.c
# cp $src_dir/ggml-backend-impl.h $dst_dir/ggml-backend-impl.h
# cp $src_dir/ggml-alloc.h $dst_dir/ggml-alloc.h
# cp $src_dir/ggml-alloc.c $dst_dir/ggml-alloc.c
# ggml-cuda
mkdir -p $dst_dir/ggml-cuda
cp $src_dir/ggml-cuda/*.cu $dst_dir/ggml-cuda/
cp $src_dir/ggml-cuda/*.cuh $dst_dir/ggml-cuda/
# # ggml-cuda
# mkdir -p $dst_dir/ggml-cuda
# cp $src_dir/ggml-cuda/*.cu $dst_dir/ggml-cuda/
# cp $src_dir/ggml-cuda/*.cuh $dst_dir/ggml-cuda/
sed -i 's/extern "C" GGML_CALL int ggml_backend_cuda_reg_devices();/\/\/ extern "C" GGML_CALL int ggml_backend_cuda_reg_devices();/' ggml-cuda.cu
sed -i '34iGGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_cuda_host_buffer_type(void);' ggml-cuda.h
# sed -i 's/extern "C" GGML_CALL int ggml_backend_cuda_reg_devices();/\/\/ extern "C" GGML_CALL int ggml_backend_cuda_reg_devices();/' ggml-cuda.cu
# sed -i '34iGGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_cuda_host_buffer_type(void);' ggml-cuda.h
# ggml-metal
sed -e '/#include "ggml-common.h"/r ggml-common.h' -e '/#include "ggml-common.h"/d' < ggml-metal.metal > temp.metal
TEMP_ASSEMBLY=$(mktemp)
echo ".section __DATA, __ggml_metallib" > $TEMP_ASSEMBLY
echo ".globl _ggml_metallib_start" >> $TEMP_ASSEMBLY
echo "_ggml_metallib_start:" >> $TEMP_ASSEMBLY
echo ".incbin \"temp.metal\"" >> $TEMP_ASSEMBLY
echo ".globl _ggml_metallib_end" >> $TEMP_ASSEMBLY
echo "_ggml_metallib_end:" >> $TEMP_ASSEMBLY
as -mmacosx-version-min=11.3 $TEMP_ASSEMBLY -o ggml-metal.o
rm -f $TEMP_ASSEMBLY
rm -rf temp.metal
# # ggml-metal
# sed -i '' '1s;^;//go:build darwin,arm64\n;' ggml-metal.m
# sed -e '/#include "ggml-common.h"/r ggml-common.h' -e '/#include "ggml-common.h"/d' < ggml-metal.metal > temp.metal
# TEMP_ASSEMBLY=$(mktemp)
# echo ".section __DATA, __ggml_metallib" > $TEMP_ASSEMBLY
# echo ".globl _ggml_metallib_start" >> $TEMP_ASSEMBLY
# echo "_ggml_metallib_start:" >> $TEMP_ASSEMBLY
# echo ".incbin \"temp.metal\"" >> $TEMP_ASSEMBLY
# echo ".globl _ggml_metallib_end" >> $TEMP_ASSEMBLY
# echo "_ggml_metallib_end:" >> $TEMP_ASSEMBLY
# as -mmacosx-version-min=11.3 $TEMP_ASSEMBLY -o ggml-metal.o
# rm -f $TEMP_ASSEMBLY
# rm -rf temp.metal
# add license info
LICENSE=$(mktemp)
cleanup() {
rm -f $LICENSE
}
trap cleanup 0
cat <<EOF | sed 's/ *$//' >$LICENSE
/**
* llama.cpp - git $SHA1
*
$(sed 's/^/ * /' <$1/$src_dir/LICENSE)
*/
for IN in $OUT/*.{c,h,cpp,m,metal,cu}; do
TMP=$(mktemp)
status "updating license $IN"
cat $LICENSE $IN >$TMP
mv $TMP $IN
done

0
llama/runner/README.md → runner/README.md
View File

0
llama/runner/runner.go → runner/runner.go
View File

25
scripts/build_llama_darwin.sh Executable file
View File

@ -0,0 +1,25 @@
#!/bin/sh
set -e
export VERSION=${VERSION:-$(git describe --tags --first-parent --abbrev=7 --long --dirty --always | sed -e "s/^v//g")}
export GOFLAGS="'-ldflags=-w -s \"-X=github.com/ollama/ollama/version.Version=$VERSION\" \"-X=github.com/ollama/ollama/server.mode=release\"'"
mkdir -p dist
# amd64 runners
export CGO_CFLAGS_ALLOW=-mfma
export CGO_CXXFLAGS_ALLOW=-mfma
CGO_ENABLED=1 GOOS=darwin GOARCH=amd64 go build -x -ldflags="-s -w" -trimpath -o dist/ollama_llama_runner_darwin_amd64 ./runner &
CGO_ENABLED=1 GOOS=darwin GOARCH=amd64 go build -x -ldflags="-s -w" -tags avx -trimpath -o dist/ollama_llama_runner_darwin_amd64_avx ./runner &
CGO_ENABLED=1 GOOS=darwin GOARCH=amd64 go build -x -ldflags="-s -w" -tags avx,avx2 -trimpath -o dist/ollama_llama_runner_darwin_amd64_avx2 ./runner &
wait
# amd64
CGO_ENABLED=1 GOOS=darwin GOARCH=amd64 go build -x -ldflags="-s -w" -o dist/ollama_darwin_amd64 .
# arm64 runner
CGO_ENABLED=1 GOOS=darwin GOARCH=arm64 go build -ldflags="-s -w" -trimpath -o dist/ollama_llama_runner_darwin_arm64 ./runner
# arm64
CGO_ENABLED=1 GOOS=darwin GOARCH=arm64 go build -ldflags="-s -w" -trimpath -o dist/ollama_darwin_arm64 .