[Build] Avoid building too many extensions (#1624)
This commit is contained in:
@@ -142,87 +142,25 @@ if nvcc_cuda_version >= Version("11.2"):
|
||||
NVCC_FLAGS += ["--threads", str(num_threads)]
|
||||
|
||||
ext_modules = []
|
||||
|
||||
# Cache operations.
|
||||
cache_extension = CUDAExtension(
|
||||
name="vllm.cache_ops",
|
||||
sources=["csrc/cache.cpp", "csrc/cache_kernels.cu"],
|
||||
extra_compile_args={
|
||||
"cxx": CXX_FLAGS,
|
||||
"nvcc": NVCC_FLAGS,
|
||||
},
|
||||
)
|
||||
ext_modules.append(cache_extension)
|
||||
|
||||
# Attention kernels.
|
||||
attention_extension = CUDAExtension(
|
||||
name="vllm.attention_ops",
|
||||
sources=["csrc/attention.cpp", "csrc/attention/attention_kernels.cu"],
|
||||
extra_compile_args={
|
||||
"cxx": CXX_FLAGS,
|
||||
"nvcc": NVCC_FLAGS,
|
||||
},
|
||||
)
|
||||
ext_modules.append(attention_extension)
|
||||
|
||||
# Positional encoding kernels.
|
||||
positional_encoding_extension = CUDAExtension(
|
||||
name="vllm.pos_encoding_ops",
|
||||
sources=["csrc/pos_encoding.cpp", "csrc/pos_encoding_kernels.cu"],
|
||||
extra_compile_args={
|
||||
"cxx": CXX_FLAGS,
|
||||
"nvcc": NVCC_FLAGS,
|
||||
},
|
||||
)
|
||||
ext_modules.append(positional_encoding_extension)
|
||||
|
||||
# Layer normalization kernels.
|
||||
layernorm_extension = CUDAExtension(
|
||||
name="vllm.layernorm_ops",
|
||||
sources=["csrc/layernorm.cpp", "csrc/layernorm_kernels.cu"],
|
||||
extra_compile_args={
|
||||
"cxx": CXX_FLAGS,
|
||||
"nvcc": NVCC_FLAGS,
|
||||
},
|
||||
)
|
||||
ext_modules.append(layernorm_extension)
|
||||
|
||||
# Activation kernels.
|
||||
activation_extension = CUDAExtension(
|
||||
name="vllm.activation_ops",
|
||||
sources=["csrc/activation.cpp", "csrc/activation_kernels.cu"],
|
||||
extra_compile_args={
|
||||
"cxx": CXX_FLAGS,
|
||||
"nvcc": NVCC_FLAGS,
|
||||
},
|
||||
)
|
||||
ext_modules.append(activation_extension)
|
||||
|
||||
# Quantization kernels.
|
||||
quantization_extension = CUDAExtension(
|
||||
name="vllm.quantization_ops",
|
||||
vllm_extension = CUDAExtension(
|
||||
name="vllm._C",
|
||||
sources=[
|
||||
"csrc/quantization.cpp",
|
||||
"csrc/cache_kernels.cu",
|
||||
"csrc/attention/attention_kernels.cu",
|
||||
"csrc/pos_encoding_kernels.cu",
|
||||
"csrc/activation_kernels.cu",
|
||||
"csrc/layernorm_kernels.cu",
|
||||
"csrc/quantization/awq/gemm_kernels.cu",
|
||||
"csrc/quantization/squeezellm/quant_cuda_kernel.cu",
|
||||
"csrc/cuda_utils_kernels.cu",
|
||||
"csrc/pybind.cpp",
|
||||
],
|
||||
extra_compile_args={
|
||||
"cxx": CXX_FLAGS,
|
||||
"nvcc": NVCC_FLAGS,
|
||||
},
|
||||
)
|
||||
ext_modules.append(quantization_extension)
|
||||
|
||||
# Misc. CUDA utils.
|
||||
cuda_utils_extension = CUDAExtension(
|
||||
name="vllm.cuda_utils",
|
||||
sources=["csrc/cuda_utils.cpp", "csrc/cuda_utils_kernels.cu"],
|
||||
extra_compile_args={
|
||||
"cxx": CXX_FLAGS,
|
||||
"nvcc": NVCC_FLAGS,
|
||||
},
|
||||
)
|
||||
ext_modules.append(cuda_utils_extension)
|
||||
ext_modules.append(vllm_extension)
|
||||
|
||||
|
||||
def get_path(*filepath) -> str:
|
||||
|
||||
Reference in New Issue
Block a user