fixed to work transformers after v4.45.2

[DemirTonchev]

Pull Request overview

fixes #564

Details

Fixed by setting:

tokenizer=self.processing_class,

here:

setfit/src/setfit/trainer.py

Line 75 in 146c7c9

tokenizer=self.tokenizer,

and adding

    def log(self, logs: Dict[str, float], start_time: Optional[float] = None) -> None:
        logs = {f"{self.logs_prefix}_{k}" if k == "loss" else k: v for k, v in logs.items()}
        return super().log(logs)

here:

setfit/src/setfit/trainer.py

Line 159 in 146c7c9

def log(self, logs: Dict[str, float]) -> None:

[coolhannes]

Seems reasonable to me but I've got no merging auth here

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[DemirTonchev]

@coolhannes can we merge this one? Who has rights? Can you tag them?

[tomaarsen]

Big apologies for the delay from this. It's always a shame when SetFit doesn't work with the latest versions of transformers and Sentence Transformers anymore.

There should be 2 issues that remain that I can't resolve in setfit:

• AttributeError: 'float' object has no attribute 'size' - related to Fix size attribute error for precision/recall/f1  evaluate#656: the f1 Space doesn't work with the latest scikit-learn anymore
• FileNotFoundError: [Errno 2] No such file or directory: '/tmp/pytest-of-runner/pytest-0/test_model_card0/checkpoint-1' - caused by Trainer sets state.best_model_checkpoint even when it doesn't save there; leads to training crash transformers#35609, will need to await a fix in transformers

But that does mean that this should be good to go!

• Tom Aarsen