Skip to content

README.md

Latest commit

 

History

History
171 lines (110 loc) · 4.55 KB

README.md

File metadata and controls

171 lines (110 loc) · 4.55 KB

From Prompts to Programs in 120 Minutes (DSPy Workshop)

PyCon Ireland 2025

What You'll Learn

  • What DSPy is and its building blocks
  • How to structure your prompts as Python code
  • How to build multi-step pipelines
  • How to evaluate and optimize your LLM applications

Groq provides free-tier access. We will be using the Llama 3.1 8B model and the GPT OSS 20B model for the workshop.

Prerequisites

  • Python 3.12+
  • Git

Quick Start

Clone and Setup

  1. Install uv, a fast Python package and project manager.

    uv installation page

  2. Download the project repository:

    git clone https://github.com/Pencroff/dspy_workshop.git

  3. Enter the workshop directory:

    cd dspy_workshop

  4. Create a virtual environment and install dependencies:

    uv venv --python 3.12 venv

    and activate it:

    source venv/bin/activate  # On Windows: venv\Scripts\activate

  5. Install dependencies:

    uv pip install -r requirements.txt

Register on Groq

  1. Open Groq and register for a free account.

  2. Open the API Keys page.

    groq api key

  3. Click the Create API Key button and save your API key.

  4. In the pop-up, enter the key name: dspy_workshop and click the Submit button.

    groq create api key

  5. Copy your API key and save it in a .env file.

    groq api key

    The .env file should look like this:

    GROQ_API_KEY=gsk_YXEdbRF7...   # your_api_key

Run Jupyter Notebooks

From the workshop directory dspy_workshop, run:

./venv/bin/jupyter lab

This will open JupyterLab in your browser.

jupyter notebook

Validate Setup

Open the notebooks/00_setup_check.ipynb notebook and run all cells.

jupyter notebook

Run MLflow

Open a separate terminal and run the MLflow server.

Activate the virtual environment:

source venv/bin/activate

Run the MLflow server:

./venv/bin/mlflow server --backend-store-uri sqlite:///data/mlflow.db --port 5005

mlflow terminal

Check the MLflow UI at http://localhost:5005/ or http://127.0.0.1:5005/.

We will use it for the evaluation and optimization notebooks.

mlflow ui

Agenda

  1. Welcome & Framing - 15 mins (notebooks/00_setup_check.ipynb)
  2. Setup & First LLM Call - 15 mins (notebooks/01_first_call.ipynb)
  3. Core Concepts - 15 mins (notebooks/02_core_concepts.ipynb)
  4. Multi-step Pipeline - 15 mins (notebooks/03_pipeline.ipynb)
  5. Break - 5 mins
  6. Evaluation & Metrics - 20 mins (notebooks/04_evaluation.ipynb)
  7. Optimization Demo - 20 mins (notebooks/05_optimization_demo.ipynb)
  8. Wrap-up - 5 mins

Dataset

Twenty synthetic examples of Python snippets with syntax errors and possible logical issues.

Structure of each dataset element:

  id: number
  name: short description
  comment: context of the example and issue
  traceback: traceback of the error, usually syntax error
  content: snippet of the code
  test_case: assert test case, verifies the correctness of the code

API Providers

Below are a few providers that have a free tier (might not be the best for optimization):

The optimization process might face ratelimits.

References