Kyle Richardson

Just an update: in terms of nltk's wordnet mapping using `synset_from_sense_key`, something seems to be wrong. Your gloss/id pair is consistent with wordnet when I searched here: http://wordnetweb.princeton.edu/perl/webwn?s=climate&sub=Search+WordNet&o2=&o0=1&o8=1&o1=1&o7=&o5=&o9=&o6=1&o3=&o4=&h=0000 . This...

Like @pacman100 I'm getting good performance with Stage3 training (using a similar setup), but also getting these really concerning warnings: ``` [WARNING] [parameter_offload.py:55:_apply_to_tensors_only] A module has unknown inputs or outputs...

Thanks for the response. Ah, yes, I redacted the type because it wouldn't make sense, it's a custom dataclass that holds a bunch of tensors (it's the output of some...

Hi again @stas00, perhaps before I make an issue we can quickly check if this is an issue to be discussed in the context of pytorch-lightning (which is what I...

For this case involving batches with additional information, yes, it's never the case that I have objects with tensors inside or anything else that involve backprop, but I wouldn't want...

Thanks. These are tensors that receive grads, yes, this has been checked. (Like I mentioned, this is a terrible design that was fixed before and that doesn't make a lot...

Aren't these usually reported by default? What's the issue exactly?

Oh okay.