Function foobar_unstructured

intermediate_source/pruning_tutorial.py:371–396

Prunes tensor corresponding to parameter called `name` in `module` by removing every other entry in the tensors. Modifies module in place (and also return the modified module) by: 1) adding a named buffer called `name+'_mask'` corresponding to the binary mask applied to the par

(module, name)

# also provide a simple function that instantiates the method and
# applies it.
def foobar_unstructured(module, name):
    """Prunes tensor corresponding to parameter called `name` in `module`
    by removing every other entry in the tensors.
    Modifies module in place (and also return the modified module) 
    by:
    1) adding a named buffer called `name+'_mask'` corresponding to the 
    binary mask applied to the parameter `name` by the pruning method.
    The parameter `name` is replaced by its pruned version, while the 
    original (unpruned) parameter is stored in a new parameter named 
    `name+'_orig'`.

    Args:
        module (nn.Module): module containing the tensor to prune
        name (string): parameter name within `module` on which pruning
                will act.

    Returns:
        module (nn.Module): modified (i.e. pruned) version of the input
            module
    
    Examples:
        >>> m = nn.Linear(3, 4)
        >>> foobar_unstructured(m, name='bias')
    """
    FooBarPruningMethod.apply(module, name)
    return module

######################################################################
# Let's try it out!