зеркало из https://github.com/microsoft/DeepSpeed.git
38 строки
1.0 KiB
Python
38 строки
1.0 KiB
Python
# Copyright (c) Microsoft Corporation.
|
|
# SPDX-License-Identifier: Apache-2.0
|
|
|
|
# DeepSpeed Team
|
|
|
|
from .builder import CUDAOpBuilder
|
|
|
|
import sys
|
|
|
|
|
|
class FusedAdamBuilder(CUDAOpBuilder):
|
|
BUILD_VAR = "DS_BUILD_FUSED_ADAM"
|
|
NAME = "fused_adam"
|
|
|
|
def __init__(self):
|
|
super().__init__(name=self.NAME)
|
|
|
|
def absolute_name(self):
|
|
return f'deepspeed.ops.adam.{self.NAME}_op'
|
|
|
|
def sources(self):
|
|
return ['csrc/adam/fused_adam_frontend.cpp', 'csrc/adam/multi_tensor_adam.cu']
|
|
|
|
def include_paths(self):
|
|
return ['csrc/includes', 'csrc/adam']
|
|
|
|
def cxx_args(self):
|
|
args = super().cxx_args()
|
|
return args + self.version_dependent_macros()
|
|
|
|
def nvcc_args(self):
|
|
nvcc_flags = ['-O3'] + self.version_dependent_macros()
|
|
if not self.is_rocm_pytorch():
|
|
nvcc_flags.extend(
|
|
['-allow-unsupported-compiler' if sys.platform == "win32" else '', '-lineinfo', '--use_fast_math'] +
|
|
self.compute_capability_args())
|
|
return nvcc_flags
|