Richard Löwenström

I've tested this a bit when I was implementing it. I got the best results with `spherical_distance_itc` (but cosine similarity itc was very similar). I kept the softmax itm loss...

Very interesting indeed, thanks for letting us try out your work! :) Yes, I'm also hoping it will sometimes improve things but it's of course very hard to evaluate if...

It seems you can patch the issue easily by wrapping SWR and replacing the local mutate ``` const { mutate } = useSWRConfig(); const backend = useSWR( key, ... );...

I see. I think it makes sense that the bound `mutate` should be specified to `useEffect` like you wrote above and anything else would be wrong but I also only...

On a very related note: it would also be nice if `viewcode` worked for the class methods (especially at the moment since the documentation is sparse)

I started looking into this and I found out that the reason is that everything is actually a `@property` so naturally they would not have arguments. However, the code style...

This is just my opinion but I don't think `__call__` should be implemented like this since it will cause confusion about what it's meant to do. I think it would...

> ... even other API Wrapper don't do filter, they do it with either __init__ or __call__ Ok I didn't know about this convention, if this is the convention then...

Probably not the exact versions they used but try and see if this works for you ``` pip install -r requirements-lock.txt ``` save as `requirements-lock.txt` ``` absl-py==0.10.0 astunparse==1.6.3 attrs==21.2.0 cachetools==4.2.2...

Have you tried making the mask a little larger?