Training#
Here we describe the available trainers in Argilla:
Base Trainer: Internal mechanism to handle Trainers
SetFit Trainer: Internal mechanism for handling training logic of SetFit models
OpenAI Trainer: Internal mechanism for handling training logic of OpenAI models
PEFT (LoRA) Trainer: Internal mechanism for handling training logic of PEFT (LoRA) models
AutoTrain Trainer: Internal mechanism for handling training logic of AutoTrain models
spaCy Trainer: Internal mechanism for handling training logic of spaCy models
Transformers Trainer: Internal mechanism for handling training logic of Transformers models
SpanMarker Trainer: Internal mechanism for handling training logic of SpanMarker models
Base Trainer#
- class argilla.training.base.ArgillaTrainerSkeleton(name, dataset, record_class, workspace=None, multi_label=False, settings=None, model=None, seed=None, *arg, **kwargs)#
- Parameters:
name (str) –
record_class (Union[TokenClassificationRecord, Text2TextRecord, TextClassificationRecord]) –
workspace (Optional[str]) –
multi_label (bool) –
settings (Union[TextClassificationSettings, TokenClassificationSettings]) –
model (str) –
seed (int) –
- abstract init_model()#
Initializes a model.
- abstract init_training_args()#
Initializes the training arguments.
- abstract predict(text, as_argilla_records=True, **kwargs)#
Predicts the label of the text.
- Parameters:
text (Union[List[str], str]) –
as_argilla_records (bool) –
- abstract save(output_dir)#
Saves the model to the specified path.
- Parameters:
output_dir (str) –
- abstract train(output_dir=None)#
Trains the model.
- Parameters:
output_dir (Optional[str]) –
- abstract update_config(*args, **kwargs)#
Updates the configuration of the trainer, but the parameters depend on the trainer.subclass.