How to train the lab to match the loaded model

Course Q&A Generative AI with Large Language Models
1

I’m on Week 2 lab, Fine-Tune a Generative AI Model for Dialogue Summarization.

The lab actually only train for 1 weights update with max_steps=1. I’m trying to get the training result to match that of the loaded model. I’ve updated the max_steps to 10, but the ROUGE metric doesn’t come close to that of the loaded model.

Does anyone have any luck in terms of training the model to get to the performance of the loaded model?

if so, would you mind sharing the parameter you have ?

2

The loaded model maybe trained on much larger dataset!

3

thank you!

1 Like