Introduction to the Pi SDK
The Pi SDK empowers software engineers to harness the power of machine learning without specialized ML expertise. Our approach allows you to embed your application’s definition of “goodness” into a Pi Scoring System which makes it easy for you to try out and assess a number of model optimization techniques.
Getting started
Our SDK supports various model optimization techniques. If this is your first time using Pi, we recommend starting with the below steps for trying Prompt Optimization, but you can use this same workflow for most of our techniques.
Check out a code notebook demonstrating the full workflow below. Find the full repo of notebooks here
Grab an API key
You can get one for free from the Techniques Catalog.
Create your Scoring System
Pi helps you create a customized scoring system that reflects your application’s specific success metrics.
Start optimizing!
-
Optimize your prompt: Have Pi use your scoring system to write you a better prompt for your application.
-
Optimize your inference: Have Pi use your scoring system to optimally route between a smaller and larger model
-
Tune your model: Use your scoring system to filter training data and run Supervised Fine Tuning
Once you’ve defined “goodness” in your scoring system, you can reuse the above workflow for almost all of the techniques in the API Reference tab (with a few exceptions).