torch.nn.utils.prune.l1_unstructured#

torch.nn.utils.prune.l1_unstructured(module,name,amount,importance_scores=None)[source]#

Prune tensor by removing units with the lowest L1-norm.

Prunes tensor corresponding to parameter calledname inmoduleby removing the specifiedamount of (currently unpruned) units with thelowest L1-norm.Modifies module in place (and also return the modified module)by:

1. adding a named buffer calledname+'_mask' corresponding to thebinary mask applied to the parametername by the pruning method.

2. replacing the parametername by its pruned version, while theoriginal (unpruned) parameter is stored in a new parameter namedname+'_orig'.

Parameters

• module (nn.Module) – module containing the tensor to prune

• name (str) – parameter name withinmodule on which pruningwill act.

• amount (int orfloat) – quantity of parameters to prune.Iffloat, should be between 0.0 and 1.0 and represent thefraction of parameters to prune. Ifint, it represents theabsolute number of parameters to prune.

• importance_scores (torch.Tensor) – tensor of importance scores (of sameshape as module parameter) used to compute mask for pruning.The values in this tensor indicate the importance of the correspondingelements in the parameter being pruned.If unspecified or None, the module parameter will be used in its place.

Returns

modified (i.e. pruned) version of the input module

Return type

module (nn.Module)

Examples

>>>m=prune.l1_unstructured(nn.Linear(2,3),"weight",amount=0.2)>>>m.state_dict().keys()odict_keys(['bias', 'weight_orig', 'weight_mask'])