sentence-transformers_all-MiniLM-L6-v2/fixed_chunks/_accelerate.txt_chunk_3.txt

Backward
The last addition is to replace the typical loss.backward() in your training loop with 🤗 Accelerate's [~accelerate.Accelerator.backward]method:

for epoch in range(num_epochs):
     for batch in train_dataloader:
         outputs = model(**batch)
         loss = outputs.loss
         accelerator.backward(loss)

         optimizer.step()
         lr_scheduler.step()
         optimizer.zero_grad()
         progress_bar.update(1)