Skip to content

Parametrization cannot be parametrized #65163

@albanD

Description

@albanD

The Parametrization that contains the original Tensor is not traversing itself properly during the forward.
In particular here:

for module in list(self._modules.values())[1:]:

We use all the _modules even the ones that are not indexed by integers.
This prevent parametrization of parametrization because it would add a parametrizations submodule to this and the forward would try to traverse it.

cc @albanD @mruberry @jbschlosser @walterddr

Metadata

Metadata

Assignees

No one assigned

    Labels

    module: nnRelated to torch.nntriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions