ks-llm-101

Knowledge sharing on fundamentals of LLMs

In this repo we go over the basics of working with LLMs.
Each topic is covered by a notebook:

• Chat Templates - What do LLMs actually receive when we send them a prompt? We discuss the concept of chat templates and how to are used to format conversations and tool calls in such the format expected by the LLM.
• Tool Calling - How to allow LLMs to make tool calls. This is a crucial ingredient in creating Agentic applications.
• Structured Output - How to get LLMs to return structured output. This is a crucial ingredient in creating robust LLM applications.

We provide hands-on code examples with both OpenAI, Ollama and LangChain/LangGraph.

Getting started

To run these examples, you will need an OpenAI API key and put it in a .env file in the root of the repo. For part of the examples, you may also need an HF_TOKEN.
Get it from Hugging Face and put it in the .env file as well. Finally, you will need Ollama installed and running.
For example on Linux:

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

Once Ollama is installed, pull the mistral-small model as follows:

ollama pull mistral-small

To run the python examples, we setup an environment with uv. Install it as follows:

curl -LsSf https://astral.sh/uv/install.sh | sh

Create and activate a virtual environment with the dependencies installed

uv sync --frozen
source .venv/bin/activate

You now have a virtual environment with the dependencies installed. To run the notebooks, we recommend using VSCode with the Jupyter extension. Alternatively, spin up a Jupyter server and open the notebooks in your browser.