Summary Snack on 4k Context

[fayvor]

#146:

• Ensure model receives inputs less than 4k tokens
• Add language about hierarchical abstractive summarization

PR Checklist

• Commits signed: All commits must be GPG or SSH signed.
• DCO Compliance: Developer Certificate of Origin (DCO) applies to the code, documentation, and any example data provided. Ensure commits are signed off.
• Notebook outputs cleared: Ensure all notebook outputs are cleared.
• Automated testing: Add the recipe to the automated tests.
• Test in Google Colab:
  • Test that it works in Google Colab (Python 3.10.12).
  • Colab has its own package set and Python version, so ensure compatibility.
• Test locally:
  • Ensure the code works in a fresh Python virtual environment (venv).
• Flexible LLM platform support:
  • The platform should be easily switchable. Use LangChain for now.
  • Include !pip install git+https://github.com/ibm-granite-community/granite-kitchen in the instructions.
• Example data: Follow the example data guidance.
• README.md updates:
  • Add a link to the recipe in the Table of Contents (ToC).
  • Include a Colab button after that link.

[rawkintrevo]

The changes look good. How is the output?

[fayvor]

The changes look good. How is the output?

The output is pretty good with the language model. It sometimes emphasizes things in the summary that should be left out. That would be improved by creating a hierarchy better that better isolates passages by topic, but that's out of scope for this snack..