hf_text-generation-inference/server/exllamav2_kernels/setup.py

29 lines
703 B
Python

from setuptools import setup
from torch.utils.cpp_extension import BuildExtension, CUDAExtension
import torch
extra_cuda_cflags = ["-lineinfo", "-O3"]
if torch.version.hip:
extra_cuda_cflags += ["-DHIPBLAS_USE_HIP_HALF"]
extra_compile_args = {
"nvcc": extra_cuda_cflags,
}
setup(
name="exllamav2_kernels",
ext_modules=[
CUDAExtension(
name="exllamav2_kernels",
sources=[
"exllamav2_kernels/ext.cpp",
"exllamav2_kernels/cuda/q_matrix.cu",
"exllamav2_kernels/cuda/q_gemm.cu",
],
extra_compile_args=extra_compile_args,
)
],
cmdclass={"build_ext": BuildExtension},
)