Lesson 4 - IMDB - Training results differs that video



This is what I got when training on Google Colab:

This is the accuracy in the YouTube lecture:

I haven’t changed any code in the notebook. Just ran it on colab.
How can the accuracy difference be explained / resolved please?

Thank you.


(Bobak Farzin) #2

Hard to tell from your code, but your learner appears to be the language model (data_lm) and what is in the video is the IMDB sentiment analysis classifier. Can you share more of what you are doing here? 30-50% in an LM model is about right for IMDB fine-tuning.