Create Taxonomies for Red Hat AI Instructlab with a Jupyter Notebook

By Ken Ocheltree

Overview

This Jupyter notebook provides a template for constructing data taxonomies for the Red Hat AI Instructlab service. The InstructLab method consists of three major components:

• Taxonomy-driven data curation: The taxonomy is a set of training data curated by humans as examples of new knowledge and skills for the model.
• Large-scale synthetic data generation: A teacher model is used to generate new examples based on the seed training data.
• Iterative model alignment tuning: The model is trained based on the synthetic data. The InstructLab method includes two tuning phases: knowledge tuning, followed by skill tuning.

The Jupyter notebook assists in the first component of constructing taxonomy-based skill and knowledge representation.

Steps

Step 1. Open the Jupyter Notebook in Colab

1. Open Google Colab.

2. If you already have a Google account, select it. If you do not have a Google account, click Sign in > Create account. Proceed though the steps to create a Google Account and then come back to the Google Colab and sign in with your Google account.

3. If you have used Google Colab before, click File > Open. If you have never used Google Colab before, the Open Notebook window opens automatically.

4. Enter the details for this repo to open the notebook.

   a. Select the Github tab.

   b. For the organization, enter IBM-Cloud.

   c. Select the IBM-Cloud/redhat-ai-instructlab-jupyter-notebooks repository.

   

   d. Select the creating_instructlab_taxonomies.jpynb notebook.

   

Step 2. Run the Notebook

You can decide how to run this notebook.

• Run the code cell by cell by clicking the Run cell button next to each cell.

• Run all the following cells at once, select the second code cell, shown with the arrow below, without running it, and click Runtime->Run cell and below.

The run does the following for you:

• Completes the environment setup by installing InstructLab and other required packages
• Configures the InstructLab installation
• Creates a taxonomy with the new data
• Uploads the taxonomy to the InstructLab service

Summary and next steps

Click the Folder icon to explore the files in the ilab folder. Preloaded QNA files and synthetically generated questions and answers can be found in this directory tree.

After the run completes, you can make another run if desired. Try running the notebook with the other example data sets or create your own QNA files.

1. Either choose a different data set or place your own data in the /content/ilab/data/your_content_1 or the /content/ilab/data/your_content_2 folder.

2. Select the second code cell without running it and click Runtime -> Run cell and below to run the rest of the notebook. This step creates and uploads a new taxonomy with the newly provided data.

This tutorial demonstrated utilizing InstructLab to introduce AI data sets.