dotAI Pre-Conference Workshop: Optimize and Evaluate Your Models with Confidence

This workshop is about how to optimize and evaluate vision models using Pruna AI. The goal is to create a merch design for Pruna AI by optimizing a vision model using Pruna's "smash" functionality.

---

Checklist

• I've given a star to the Pruna repo on GitHub

• Record the original model output:

  • running time
  • prompt
  • output

• I've used Pruna to smash the vision model of my choice.

• Record the smashed model output:

  • running time
  • prompt
  • merch design

---

Submission

• Screenshot showing that you starred the Pruna repo on GitHub

• Take a screenshot of the output of the original model:

  • running time
  • prompt
  • merch design

• Take a screenshot of the output of the smashed model:

  • running time
  • prompt
  • merch design

• Upload your submission to the #chill channel on Discord with:

  • your name (or names if working in a group)
  • screenshots
  • link to the notebook

---

Run the Notebook in Modal

You can run this workshop notebook directly in Modal — no local setup needed.

Steps:

1. Log in to your Modal account.
2. Go to the Notebooks section.
3. Click Import notebook.
4. Paste the URL of this repository or the specific .ipynb file:
   dotAI-workshop-5-11-2025.ipynb
5. Once imported, you can execute and modify the notebook interactively within Modal.

---

Overview

The workshop walks through:

1. Running a vision model (FLUX.1-Kontext-dev) to generate a merch design
2. Using Pruna to optimize ("smash") the model with caching and compilation techniques
3. Comparing the performance of the original vs. optimized model