remove duplicated multihead_attn install (#729)

setup.py

@@ -259,70 +259,6 @@ if "--fast_multihead_attn" in sys.argv:
                                                       '--expt-extended-lambda',
                                                       '--use_fast_math'] + version_dependent_macros}))
 
-if "--fast_multihead_attn" in sys.argv:
-    from torch.utils.cpp_extension import CUDAExtension
-    sys.argv.remove("--fast_multihead_attn")
-
-    from torch.utils.cpp_extension import BuildExtension
-    cmdclass['build_ext'] = BuildExtension
-
-    if torch.utils.cpp_extension.CUDA_HOME is None:
-        raise RuntimeError("--fast_multihead_attn was requested, but nvcc was not found.  Are you sure your environment has nvcc available?  If you're installing within a container from https://hub.docker.com/r/pytorch/pytorch, only images whose names contain 'devel' will provide nvcc.")
-    else:
-        subprocess.run(["git", "submodule", "update", "--init", "apex/contrib/csrc/multihead_attn/cutlass"])
-        ext_modules.append(
-            CUDAExtension(name='fast_self_multihead_attn',
-                          sources=['apex/contrib/csrc/multihead_attn/self_multihead_attn.cpp', 
-                                   'apex/contrib/csrc/multihead_attn/self_multihead_attn_cuda.cu'],
-                          extra_compile_args={'cxx': ['-O3',] + version_dependent_macros,
-                                              'nvcc':['-O3',
-                                                      '-gencode', 'arch=compute_70,code=sm_70',
-                                                      '-I./apex/contrib/csrc/multihead_attn/cutlass/',
-                                                      '-U__CUDA_NO_HALF_OPERATORS__',
-                                                      '-U__CUDA_NO_HALF_CONVERSIONS__',
-                                                      '--expt-relaxed-constexpr',
-                                                      '--expt-extended-lambda',
-                                                      '--use_fast_math'] + version_dependent_macros}))
-        ext_modules.append(
-            CUDAExtension(name='fast_self_multihead_attn_norm_add',
-                          sources=['apex/contrib/csrc/multihead_attn/self_multihead_attn_norm_add.cpp', 
-                                   'apex/contrib/csrc/multihead_attn/self_multihead_attn_norm_add_cuda.cu'],
-                          extra_compile_args={'cxx': ['-O3',] + version_dependent_macros,
-                                              'nvcc':['-O3',
-                                                      '-gencode', 'arch=compute_70,code=sm_70',
-                                                      '-I./apex/contrib/csrc/multihead_attn/cutlass/',
-                                                      '-U__CUDA_NO_HALF_OPERATORS__',
-                                                      '-U__CUDA_NO_HALF_CONVERSIONS__',
-                                                      '--expt-relaxed-constexpr',
-                                                      '--expt-extended-lambda',
-                                                      '--use_fast_math'] + version_dependent_macros}))
-        ext_modules.append(
-            CUDAExtension(name='fast_encdec_multihead_attn',
-                          sources=['apex/contrib/csrc/multihead_attn/encdec_multihead_attn.cpp', 
-                                   'apex/contrib/csrc/multihead_attn/encdec_multihead_attn_cuda.cu'],
-                          extra_compile_args={'cxx': ['-O3',] + version_dependent_macros,
-                                              'nvcc':['-O3',
-                                                      '-gencode', 'arch=compute_70,code=sm_70',
-                                                      '-I./apex/contrib/csrc/multihead_attn/cutlass/',
-                                                      '-U__CUDA_NO_HALF_OPERATORS__',
-                                                      '-U__CUDA_NO_HALF_CONVERSIONS__',
-                                                      '--expt-relaxed-constexpr',
-                                                      '--expt-extended-lambda',
-                                                      '--use_fast_math'] + version_dependent_macros}))
-        ext_modules.append(
-            CUDAExtension(name='fast_encdec_multihead_attn_norm_add',
-                          sources=['apex/contrib/csrc/multihead_attn/encdec_multihead_attn_norm_add.cpp', 
-                                   'apex/contrib/csrc/multihead_attn/encdec_multihead_attn_norm_add_cuda.cu'],
-                          extra_compile_args={'cxx': ['-O3',] + version_dependent_macros,
-                                              'nvcc':['-O3',
-                                                      '-gencode', 'arch=compute_70,code=sm_70',
-                                                      '-I./apex/contrib/csrc/multihead_attn/cutlass/',
-                                                      '-U__CUDA_NO_HALF_OPERATORS__',
-                                                      '-U__CUDA_NO_HALF_CONVERSIONS__',
-                                                      '--expt-relaxed-constexpr',
-                                                      '--expt-extended-lambda',
-                                                      '--use_fast_math'] + version_dependent_macros}))
-
 setup(
     name='apex',
     version='0.1',