Carlos Mocholí

> @carmocca do we really want to append "-true" to all non-mixed precision stuff? intuitively, when I see precision="16", I would assume that this is always the case and there...

@awaelchli It's not clear to me if your argument also applies to the linked issue and whether it should be closed too as non-supported usage. Closing is your call, as...

Can you also update the [CHANGELOG](https://github.com/PyTorchLightning/pytorch-lightning/blob/master/CHANGELOG.md) and add a test for this?

@kaushikb11 I suggest copying the [patch](https://patch-diff.githubusercontent.com/raw/Lightning-AI/lightning/pull/12125.patch) into a branch of your own and opening a new PR that closes this one

> but it seemed to me that the models don't inherit from AutoModel IIRC `AutoModel` is the API that defines `.from_pretrained()`: https://github.com/Lightning-AI/lightning-transformers/blob/master/lightning_transformers/core/model.py#L77. But not a HF expert.

> The type hint would need to be Type[_BaseAutoModelClass] instead of Type[AutoModel]. Fine, this can be changed. > why does it have to be a forward reference? It doesn't have...

Any update here? Still unmaintained?

Can you try upgrading dependencies? I believe we test with `torch_xla==1.12`. The stacktrace also shows you are using `python==3.6` but `3.7` is our minimum supported version. Finally, have you tried...

Unfortunately, we cannot backport a bugfix for an old release. It would need to be reproducible in the latest version so that we can look into it.

Sorry, I don't. What I would do in your situation is to sequentially upgrade each patch version to update the code while minimizing breaks (look at the warning messages!). `1.2.*`...