paddleslim.nas.one_shot package¶

class paddleslim.nas.one_shot.OneShotSuperNet(name_scope)¶

Bases: paddle.fluid.dygraph.layers.Layer

The base class of super net used in one-shot searching strategy. A super net is a dygraph layer.

Parameters:	name_scope (str) – The name scope of super net.

forward(input, tokens=None)¶

Defines the computation performed at every call.

Parameters:	input (variable) – The input of super net. tokens (list) – The tokens used to generate a sub-network. None means computing in super net training mode. Otherwise, it will execute the sub-network generated by tokens. The tokens should be set in searching stage and final training stage. Default: None.
Returns:	The output of super net.
Return type:	Varaible

init_tokens()¶

Get init tokens in search space.

Returns:	The init tokens which is a list of integer.
Return type:	lis<int>t

range_table()¶

Get range table of current search space.

Returns:	The maximum value and minimum value in each position of tokens with format (min_values, max_values). The min_values is a list of integers indicating the minimum values while max_values indicating the maximum values.
Return type:	range_table(tuple)

paddleslim.nas.one_shot.OneShotSearch(model, eval_func, strategy='sa', search_steps=100)¶

Search a best tokens which represents a sub-network.

Parameters:	model (fluid.dygraph.Layer) – A dynamic graph module whose sub-modules should contain one instance of OneShotSuperNet at least. eval_func (function) – A callback function which accept model and tokens as arguments. strategy (str) – The name of strategy used to search. Default: ‘sa’. search_steps (int) – The total steps for searching.
Returns:	The best tokens searched.
Return type:	list<int>

class paddleslim.nas.one_shot.SuperMnasnet(name_scope, input_channels=3, out_channels=1280, repeat_times=[6, 6, 6, 6, 6, 6], stride=[1, 1, 1, 1, 2, 1], channels=[16, 24, 40, 80, 96, 192, 320], use_auxhead=False)¶

Bases: paddleslim.nas.one_shot.one_shot_nas.OneShotSuperNet

get_flops(input, output, op)¶

init_tokens()¶

range_table()¶

Submodules¶

paddleslim.nas.one_shot.one_shot_nas module¶

class paddleslim.nas.one_shot.one_shot_nas.OneShotSuperNet(name_scope)¶