gered/ollama
1
0
Fork 0
Code Issues Pull requests Actions Packages Projects Releases Wiki Activity

pack in cuda libs

Browse source
This commit is contained in:
Bruce MacDonald 2023-09-20 17:40:42 +01:00
parent e35565c567
commit 4e8be787c7
3 changed files with 10 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
docs/development.md
View file

@ -35,5 +35,5 @@ Now you can run `ollama`:
## Building on Linux with GPU support
- Install cmake and nvidia-cuda-toolkit
- run `CUDA_VERSION=$(nvcc --version | sed -n 's/^.*release \([0-9]\+\)\.\([0-9]\+\).*$/\1/p') go generate ./...`
- run `CUDA_VERSION=11 CUDA_PATH=/path/to/libcuda.so CUBLAS_PATH=/path/to/libcublas.so CUDART_PATH=/path/to/libcudart.so CUBLASLT_PATH=/path/to/libcublasLt.so go generate ./...`
- run `go build .`

8
llm/llama.cpp/generate_linux.go
View file

@ -19,3 +19,11 @@ package llm
//go:generate cmake --build ggml/build/cuda-${CUDA_VERSION} --target server --config Release
//go:generate cmake -S gguf -B gguf/build/cuda-${CUDA_VERSION} -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on
//go:generate cmake --build gguf/build/cuda-${CUDA_VERSION} --target server --config Release
//go:generate cp --dereference ${CUDA_PATH} ggml/build/cuda-${CUDA_VERSION}/bin/libcuda.so
//go:generate cp --dereference ${CUDA_PATH} gguf/build/cuda-${CUDA_VERSION}/bin/libcuda.so
//go:generate cp --dereference ${CUBLAS_PATH} ggml/build/cuda-${CUDA_VERSION}/bin/libcublas.so.11
//go:generate cp --dereference ${CUBLAS_PATH} gguf/build/cuda-${CUDA_VERSION}/bin/libcublas.so.11
//go:generate cp --dereference ${CUDART_PATH} ggml/build/cuda-${CUDA_VERSION}/bin/libcudart.so.11.0
//go:generate cp --dereference ${CUDART_PATH} gguf/build/cuda-${CUDA_VERSION}/bin/libcudart.so.11.0
//go:generate cp --dereference ${CUBLASLT_PATH} ggml/build/cuda-${CUDA_VERSION}/bin/libcublasLt.so.11
//go:generate cp --dereference ${CUBLASLT_PATH} gguf/build/cuda-${CUDA_VERSION}/bin/libcublasLt.so.11

2
llm/llama.go
View file

@ -353,7 +353,7 @@ func newLlama(model string, adapters []string, runners []ModelRunner, opts api.O
runner.Path,
append(params, "--port", strconv.Itoa(port))...,
)
cmd.Env = append(os.Environ(), fmt.Sprintf("LD_LIBRARY_PATH=%s", filepath.Dir(runner.Path)))
cmd.Stdout = os.Stderr
cmd.Stderr = os.Stderr