НачатьНачать бесплатно

CV fine-tuning: trainer configuration

Now that you have prepared the dataset and adapted a pretrained model to the new classes, it is time to configure your trainer.

The TrainingArguments and Trainer have been loaded from the transformers library. The model (model) and dataset (dataset) have been loaded as you previously configured them.

Это упражнение является частью курса

Multi-Modal Models with Hugging Face

Посмотреть курс

Инструкции к упражнению

  • Adjust the learning rate to 6e-5.
  • Provide the model, training data, and test data to the Trainer instance.

Интерактивное практическое упражнение

Попробуйте выполнить это упражнение, дополнив этот пример кода.

training_args = TrainingArguments(
    output_dir="dataset_finetune",
    # Adjust the learning rate
    ____,
    gradient_accumulation_steps=4,
    num_train_epochs=3,
    push_to_hub=False
)

trainer = Trainer(
    # Provide the model and datasets
    model=____,
    args=training_args,
    data_collator=data_collator,
    train_dataset=____,
    eval_dataset=____,
    processing_class=image_processor,
    compute_metrics=compute_metrics,
)
Редактировать и запускать код