You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

param_pack.py 1.2 kB

12345678910111213141516171819202122232425262728293031323334
  1. # -*- coding: utf-8 -*-
  2. # MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
  3. #
  4. # Copyright (c) 2014-2020 Megvii Inc. All rights reserved.
  5. #
  6. # Unless required by applicable law or agreed to in writing,
  7. # software distributed under the License is distributed on an
  8. # "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  9. import numpy as np
  10. from ..tensor import Tensor
  11. from .distributed import all_reduce_sum
  12. from .tensor import param_pack_concat, param_pack_split
  13. def get_offsets(shapes):
  14. offsets = []
  15. offset = 0
  16. for shape in shapes:
  17. offsets.append(offset)
  18. offset += int(np.prod(shape))
  19. offsets.append(offset)
  20. return offsets
  21. def pack_allreduce_split(pack_list, shapes, group, reduce_method):
  22. offsets_val = get_offsets(shapes)
  23. offsets = Tensor(offsets_val)
  24. packed_grads = param_pack_concat(pack_list, offsets, offsets_val)
  25. packed_grads = all_reduce_sum(packed_grads, group, group.comp_node)
  26. if reduce_method == "mean":
  27. packed_grads /= group.size
  28. grads = param_pack_split(packed_grads, offsets_val, shapes)
  29. return grads

MegEngine 安装包中集成了使用 GPU 运行代码所需的 CUDA 环境,不用区分 CPU 和 GPU 版。 如果想要运行 GPU 程序,请确保机器本身配有 GPU 硬件设备并安装好驱动。 如果你想体验在云端 GPU 算力平台进行深度学习开发的感觉,欢迎访问 MegStudio 平台