Summary and Takeaways#

Summary

  • Setting up an OpenAI account

  • Multimodal OpenAI use cases

  • Advanced topic: Function calling and RAG

Development Workflow

  • Identify a dataset and specify precisely what you want to do with it

  • Engineer a prompt (use OpenAI Playground)

  • Evaluate performance on a sample

    • If performance is unacceptable, try further prompt engineering, functions, etc.

    • If performance is still not good enough, try fine-tuning

  • Deploy at scale

_images/gpt-dev-cycle.png
Structuring and Testing Code

  • Structure your code into source code (functions) along with unit tests

  • Functions are composed in scripts, which should have runtime tests and generate logs

  • All code (source, scripts, tests) need to be in source control, typically git

  • Save logs and output files in a secure location. Do not modify them.

_images/testing.png

Moving forward

  • This is a living document, we will be updating it with new topics regularly

  • We will also be adding code to the attached openai-helper package

  • Come to us for help during office hours at 5424, or send us an email us at

    rs@kellogg.northwestern.edu