Giovanni Puccetti

I think coca is in the training tests already, but this does not error. The tokens are not shifted by one, so it learns to copy current token instead of...

> So, been thinking about this one, I really don't like the is_training, it's not done this way elsewhere. The label shift is standard, but why do we need to...

The reason for this is that it was meant to keep it identical to how it was before (assuming I did it right) and since compared to before the tokenizer...

> @gpucce do you have any idea what might be causing it? what's the symptom and by how much is it 'off'? there are numerical changes across versions of pytorch,...

> @gpucce have you run same random inputs through the different towers, save results to verify closeness within some float eps on same env but with current main and previous...

@rwightman I think I found it, could it be the new `open_clip.tokenize` generates sequences with length 76 in some cases?

> @rwightman I think I found it, could it be the new `open_clip.tokenize` generates sequences with length 76 in some cases? specifically `open_clip.get_tokenizer("coca_ViT-B-32")("some text")` has shape `[1, 76]` in the...

Hi, I think there are two issues, the `--pretrained openai` flag makes you load models that were pretrained by openai while coca is not one of those, at the same...

Hi, @ykj467422034 can you share a snippet of the code you are actually using? From what I see the one you share is exactly the one in the readme and...

> > Hi, @ykj467422034 can you share a snippet of the code you are actually using? From what I see the one you share is exactly the one in the...