Gradient accumulation with Trainer
You're setting up Trainer for your language translation model to use gradient accumulation, so that you can effectively train on larger batches. Your model will simplify translations by training on paraphrases from the MRPC dataset. Configure the training arguments to accumulate gradients! The exercise will take some time to run with the call to trainer.train().
The model, dataset, and compute_metrics() function have been pre-defined.
This exercise is part of the course
Efficient AI Model Training with PyTorch
Exercise instructions
- Set the number of gradient accumulation steps to two.
- Pass in the training arguments to
Trainer.
Hands-on interactive exercise
Have a go at this exercise by completing this sample code.
training_args = TrainingArguments(
output_dir="./results",
evaluation_strategy="epoch",
# Set the number of gradient accumulation steps to two
____=____
)
trainer = Trainer(
model=model,
# Pass in the training arguments to Trainer
____=____,
train_dataset=dataset["train"],
eval_dataset=dataset["validation"],
compute_metrics=compute_metrics,
)
trainer.train()