support for packaging in multiple cuda runners (#509)

* enable packaging multiple cuda versions
* use nvcc cuda version if available

---------

Co-authored-by: Michael Yang <mxyng@pm.me>
This commit is contained in:
Bruce MacDonald
2023-09-14 15:08:13 -04:00
committed by GitHub
parent 83ffb154bc
commit 2540c9181c
5 changed files with 96 additions and 38 deletions

View File

@@ -7,9 +7,15 @@ package llm
//go:generate git-apply ../ggml_patch/0001-add-detokenize-endpoint.patch
//go:generate git-apply ../ggml_patch/0002-34B-model-support.patch
//go:generate git-apply ../ggml_patch/0005-ggml-support-CUDA-s-half-type-for-aarch64-1455-2670.patch
//go:generate cmake -S ggml -B ggml/build/gpu -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on
//go:generate cmake --build ggml/build/gpu --target server --config Release
//go:generate cmake -S ggml -B ggml/build/cpu -DLLAMA_K_QUANTS=on
//go:generate cmake --build ggml/build/cpu --target server --config Release
//go:generate git submodule update --force gguf
//go:generate cmake -S gguf -B gguf/build/gpu -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on
//go:generate cmake --build gguf/build/gpu --target server --config Release
//go:generate cmake -S gguf -B gguf/build/cpu -DLLAMA_K_QUANTS=on
//go:generate cmake --build gguf/build/cpu --target server --config Release
//go:generate cmake -S ggml -B ggml/build/cuda-${CUDA_VERSION} -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on
//go:generate cmake --build ggml/build/cuda-${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