lewtun

Thanks a lot for re-working on this @deutschmn ❤️ ! Ping me when you'd like a review :)

> Awesome thanks ! Can you add it to https://github.com/huggingface/hub-docs/blob/main/js/src/lib/interfaces/Types.ts first please ? This is where we define the cross libraries tasks taxonomy ;) Thanks for the tip! Done in...

Thank you @severo !!

Having experienced some of these issues firsthand, I am very much in favour of this proposal as it removes potentially confusing errors for the end-user (e.g. why am I seeing...

Thanks for the tip @nateraw - it works! Nevertheless, I think it would be nice if one could still push to the Hub after calling `save_pretrained` as I gather this...

> Can you share code pointers for caching in `datasets`? Sure! You can check out the `_map_single()` method for `Dataset` objects: https://github.com/huggingface/datasets/blob/747f0c8612d4929dbdd1c72cca201815911ce660/src/datasets/arrow_dataset.py#L2561 Basically, it checks if the `Dataset` object has...

That would be wonderful - thank you @cakiki !

I like this proposal to move all `kwargs` to the `load()` function a lot! To help gauge the difference, could you share some pseudocode that demonstrates what the new API...

Hey @kaizengit, TrOCR is currently not a supported architecture in `transformers` because it involves a processor for the images. We're currently working on supporting vision models, so I'll take a...

Cool, it seems you guys went for the hardcore version :) All I had in mind were spheres and tori with gaussian noise added, but perhaps this is too limiting....