Skip to content

[PyTorch] weight tensor dimension assumption #7865

@ssnl

Description

@ssnl

Now in both inits and spectral norm, we assume that dim 0 of weight is the output feature dim. However, this is incorrect for ConvTranspose.

We need to figure out how to properly handle those.

cc @albanD @mruberry

Metadata

Metadata

Assignees

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