fused_adam.py 1.0 KB

12345678910111213141516171819202122232425262728293031323334353637
  1. # Copyright (c) Microsoft Corporation.
  2. # SPDX-License-Identifier: Apache-2.0
  3. # DeepSpeed Team
  4. from .builder import CUDAOpBuilder
  5. import sys
  6. class FusedAdamBuilder(CUDAOpBuilder):
  7. BUILD_VAR = "DS_BUILD_FUSED_ADAM"
  8. NAME = "fused_adam"
  9. def __init__(self):
  10. super().__init__(name=self.NAME)
  11. def absolute_name(self):
  12. return f'deepspeed.ops.adam.{self.NAME}_op'
  13. def sources(self):
  14. return ['csrc/adam/fused_adam_frontend.cpp', 'csrc/adam/multi_tensor_adam.cu']
  15. def include_paths(self):
  16. return ['csrc/includes', 'csrc/adam']
  17. def cxx_args(self):
  18. args = super().cxx_args()
  19. return args + self.version_dependent_macros()
  20. def nvcc_args(self):
  21. nvcc_flags = ['-O3'] + self.version_dependent_macros()
  22. if not self.is_rocm_pytorch():
  23. nvcc_flags.extend(
  24. ['-allow-unsupported-compiler' if sys.platform == "win32" else '', '-lineinfo', '--use_fast_math'] +
  25. self.compute_capability_args())
  26. return nvcc_flags