Question answering AI training with MRQA datasets using Autotrain taking longer than expected

I have been using MRQA datasets to train an AI model using Autotrain for question answering for about 2 days straight. I’m concerned whether it’s normal for the process to take this long, or if there might be errors occurring.