`super().init()` call should use keyword arguments #5

bauwenst · 2024-12-01T21:25:03Z

The library is broken in the most recent version of transformers because the HfMultiTaskTrainer class doesn't use keyword arguments to call super().__init__(), which means that when HuggingFace adds a constructor argument (which they have done), the values get passed to the wrong argument.

hf-multitask-trainer/hf_mtask_trainer/trainer.py

Lines 79 to 83 in 8c40c26

    
           super().__init__( 
        
               model, args, data_collator, train_dataset, eval_dataset, tokenizer, 
        
               model_init, compute_metrics, callbacks, optimizers, 
        
               preprocess_logits_for_metrics 
        
           )

It should be something like

super().__init__( 
    model=model, 
    args=args, 
    data_collator=data_collator, 
    train_dataset=train_dataset, 
    eval_dataset=eval_dataset, 
    tokenizer=tokenizer, 
    model_init=model_init, 
    compute_metrics=compute_metrics, 
    callbacks=callbacks, 
    optimizers=optimizers, 
    preprocess_logits_for_metrics=preprocess_logits_for_metrics  
)

The text was updated successfully, but these errors were encountered:

bauwenst linked a pull request Dec 9, 2024 that will close this issue

Fix misaligned arguments in super() constructor call #6

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

`super().init()` call should use keyword arguments #5

`super().init()` call should use keyword arguments #5

bauwenst commented Dec 1, 2024 •

edited

Loading

super().__init__() call should use keyword arguments #5

super().__init__() call should use keyword arguments #5

Comments

bauwenst commented Dec 1, 2024 • edited Loading

`super().init()` call should use keyword arguments #5

`super().init()` call should use keyword arguments #5

bauwenst commented Dec 1, 2024 •

edited

Loading