Setting discriminator updates to 1 causes grad_norm losses fail to update

[FirestName]

Setting discriminator updates to 1 (from v2.gin) causes grad_norm losses not updating (at least) with wasserstein config.
The training still runs so no error messages produced.

Should this be the case?

Additionally would be interesting to test even more frequent discriminator updates as recommended with wasserstein gans (and the ilk I've understood), and as already setting them discriminator updates to 2 (from 4) appears benefiting the training.