# Create a new Trainer with optimal parameters
optimal_trainer = DistillationTrainer(
    student_model,
    training_args,
    teacher_model=teacher_model,
    train_dataset=tokenized_datasets["train"],
    eval_dataset=tokenized_datasets["validation"],
    data_collator=data_collator,
    tokenizer=tokenizer,
    compute_metrics=compute_metrics,
)
 
optimal_trainer.train()
 
 
# save best model, metrics and create model card
trainer.create_model_card(model_name=training_args.hub_model_id)
trainer.push_to_hub()