Hey! In this tutorial you'll learn how to quickly mimic your own way of talking using AWS Bedrock. This tutorial is just the first part of a series. All suggestions for improvement or more cool ideas are welcome :)
First, go to your console and look for the AWS Bedrock service.
I'll use Meta's LLaMA 2 Chat 13B model. Make sure you have the permission to use it!
In the meantime remember to create your own S3 bucket to store the data.
As the training data I will use the messages I exported from WhatsApp. If you want, you can do the same, or export it from Facebook. For even better effect you should clean the data, and get rid of any markups. After you've done that, put it in your S3 bucket.
Now create a fine-tuning job. Go to "Custom Models" under Foundation Models" tab, scroll down, click on the "Customize Model" button, and select "Create Fine-tuning Job".
After you're done, specify the number of epochs you want it to run for (5 should be good enough for our purpose, but I will try 10 for a better effect), and hit "Fine-tune model".
In the next post I will attempt to evaluate the model. Stay tuned!
Top comments (0)