I am trying to train a T5 (t5_large) transformer model on some data. Since it's out of cuda memory, I was forced to set batch_size to 1 so that I can run the model on my computer.

Now, my question is what other consideration I must take into account. Should I check the model convergence? if yes how? Then, should I increase the epochs? It's currently 1 epoch.

  • $\begingroup$ If your dataset is large, why would the batch size affect the convergence? $\endgroup$ Apr 3 at 15:06
  • $\begingroup$ @AryaMcCarthy Yes, the dataset is large. I don't know how or when or why it affects or not, that's actually my question. $\endgroup$
    – Ahmad
    Apr 3 at 16:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.