Add custom kernel for RMS normalization (#16)

This commit is contained in:
Woosuk Kwon
2023-03-31 09:51:22 -07:00
committed by GitHub
parent c45f3c3ab6
commit 09e9245478
9 changed files with 243 additions and 58 deletions

View File

@ -31,6 +31,14 @@ positional_encoding_extension = cpp_extension.CUDAExtension(
)
ext_modules.append(positional_encoding_extension)
# Layer normalization kernels.
layernorm_extension = cpp_extension.CUDAExtension(
name='cacheflow.layernorm_ops',
sources=['csrc/layernorm.cpp', 'csrc/layernorm_kernels.cu'],
extra_compile_args={'cxx': CXX_FLAGS, 'nvcc': NVCC_FLAGS},
)
ext_modules.append(layernorm_extension)
setuptools.setup(
name='cacheflow',
ext_modules=ext_modules,