package#

Module Contents#

Package

A basic network package data structure used in FedLab. Everything is Tensor in FedLab.

supported_torch_dtypes

supported_torch_dtypes#
class Package(message_code=None, content=None)#

Bases: object

A basic network package data structure used in FedLab. Everything is Tensor in FedLab.

Note

slice_size_i = tensor_i.shape[0], that is, every element in slices indicates the size of a sub-Tensor in content.

Package maintains 3 variables:

  • header : torch.Tensor([sender_rank, recv_rank, content_size, message_code, data_type])

  • slices : list[slice_size_1, slice_size_2]

  • content : torch.Tensor([tensor_1, tensor_2, ...])

Parameters

  • message_code (MessageCode) – Message code

  • content (torch.Tensor, optional) – Tensors contained in this package.

append_tensor(self, tensor)#

Append new tensor to Package.content

Parameters

tensor (torch.Tensor) – Tensor to append in content.

append_tensor_list(self, tensor_list)#

Append a list of tensors to Package.content.

Parameters

tensor_list (list[torch.Tensor]) – A list of tensors to append to Package.content.

to(self, dtype)#
static parse_content(slices, content)#

Parse package content into a list of tensors

Parameters

  • slices (list[int]) – A list containing number of elements of each tensor. Each number is used as offset in parsing process.

  • content (torch.Tensor) – Package.content, a 1-D tensor composed of several 1-D tensors and their corresponding offsets. For more details about Package.

Returns

A list of 1-D tensors parsed from content

Return type

list[torch.Tensor]

static parse_header(header)#

Parse header to get information of current package.

Parameters

header (torch.Tensor) – Package.header, a 1-D tensor composed of 4 elements: torch.Tensor([sender_rank, recv_rank, slice_size, message_code, data_type]).

:param For more details about Package.:

Returns

A tuple containing 5 elements: (sender_rank, recv_rank, slice_size, message_code, data_type).

Return type

tuple