Matt

In general, I find `pip install transformers[dev]` isn't really necessary! `pip install transformers[quality]` should be sufficient for most of what you need for a PR.

Hi @a8nova, when you say you're not implemeting caching, does that mean `past_key_values` just isn't implemented at all, or we're not implementing the `PyTorch` `StaticCache`? Not implementing `StaticCache` is totally...

Yeah - rather than implementing `StaticCache`, maybe we can just return tensors with variable shapes, like the other TF models do? You can probably copy the relevant code from another...

Hm, this is challenging! I'm not sure how the model could autodetect which files should be pushed. I guess it would need to 1) Inspect the Python code being pushed...

Some backstory: a lot of the dynamic module loading code was written by Sylvain Gugger, who has since left HF, and so no-one really owns it right now! I'm probably...

Cool, I'll see if I can make it work! And don't stress about the extra work - we need to have someone take ownership of it again, so this is...

Hi @ariG23498, the cause there is most likely that Numpy doesn't support `bfloat16` dtypes, and so the code fails because there is no direct conversion from Torch -> TF, it...

PR is open at #29755

PR merged! If you rebase, those loads should now work.

@ariG23498 yes, errors like these almost always indicate that weights haven't been built! My guess is that since a lot of weights in each layer are missing, the problem is...