Week 1

You will practice multiple prompting techniques by crafting prompts to complete specific tasks. Each task’s instructions are at the top of its corresponding source file.

Installation

Make sure you have first done the installation described in the top-level README.md.

Ollama installation

We will be using a tool to run different state-of-the-art LLMs locally on your machine called Ollama. Use one of the following methods:

• macOS (Homebrew):

  brew install --cask ollama 
  ollama serve
  

• Linux (recommended):

  curl -fsSL https://ollama.com/install.sh | sh
  

• Windows: Download and run the installer from ollama.com/download.

Verify installation:

ollama -v

Before running the test scripts, make sure you have the following models pulled. You only need to do this once (unless you remove the models later):

ollama run mistral-nemo:12b
ollama run llama3.1:8b

Techniques and source files

• K-shot prompting — week1/k_shot_prompting.py
• Chain-of-thought — week1/chain_of_thought.py
• Tool calling — week1/tool_calling.py
• Self-consistency prompting — week1/self_consistency_prompting.py
• RAG (Retrieval-Augmented Generation) — week1/rag.py
• Reflexion — week1/reflexion.py

Deliverables

• Read the task description in each file.
• Design and run prompts (look for all the places labeled TODO in the code). That should be the only thing you have to change (i.e. don't tinker with the model).
• Iterate to improve results until the test script passes.
• Save your final prompt(s) and output for each technique.
• Make sure to include in your submission the completed code for each prompting technique file. Double check that all TODOs have been resolved.

Evaluation rubric (60 pts total)

• 10 for each completed prompt across the 6 different prompting techniques