Training compiler for all NLP models?

Hi @philschmid,

I liked very much your post “Hugging Face Transformers BERT fine-tuning using Amazon SageMaker and Training Compiler” (and its promises of saving time and money when training NLP models :slight_smile: ) and I did try it with a T5 model.

However, I got the following error: The training task fails due to a missing XLA configuration

I found an article from AWS on “Training Job Fails Due to Missing XLA Configuration” but it did not fix the problem.

Do you have any suggestion? Thank you.

Hey @pierreguillou,

could you share how you created your Training Job (python estiomator + hyperparameters) and the training script you use?

Iin addition to this, I am not sure if T5 is well supported yet. As mentioned in my blog

The Amazon Training Compiler works best with Encoder Type models, like BERT , RoBERTa , ALBERT , DistilBERT .

The Training Compiler currently works best with encoder-type models. In the test I ran a couple of weeks back T5 performed worse than the default training.

Thanks for your answer @philschmid. Do you plan to work on it?