Warning: cudnn.convert does not work with nngraph yet. Ignoring nn.gModule

When I try to run on new image, it throws this warning. And then I found the generated captions are all like that:
"a large brown and white and white and white and white and white."repeated many times the same caption.

How to fix this problem? Thank you.

Start by reproducing the warning involving cudnn.convert, nngraph, and nn.gModule with the new image described in the issue. Compare the generated captions with the reported repeated output; done means the warning is addressed and captions no longer repeat the same phrase.

Written by the indexing model from the issue text.