mirror of
https://github.com/dogkeeper886/ollama37.git
synced 2025-12-10 15:57:04 +00:00
pack in cuda libs
This commit is contained in:
@@ -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
|
||||
|
||||
@@ -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
|
||||
|
||||
|
||||
Reference in New Issue
Block a user