LLM Tinker

Basics on how to set up a local coding environment to use an open weights model for data classification and other tasks.

Work through the notebooks in order listed below if you're new to this topic.

IMPORTANT: Before you start, make sure to follow the setup instructions below.

• llm_tinker.ipynb: Basic setup/usage of Ollama + LangChain in Jupyter, and some important notes.
• classify_trump_tweets.ipynb: A Jupyter Notebook that demonstrates how to use Ollama with LangChain to classify, or label, tweets by Trump. The notebook demonstrates how to identify tweets by type (text-only, media-only, or both). Guidance on how to refine and perform additional classification tasks are provided at the end of the notebook.
• classify_trump_tweets_for_export.ipynb: Similar to classify_trump_tweets.ipynb, but illustrates how to add classification as a new DataFrame column and then export the results to a CSV file. This notebook is useful if you want to save the classification results for downstream evaluation.
• eval_tweet_classification.ipynb: A Jupyter Notebook that demonstrates how to evaluate the performance of a classification model using Precision, Recall, and F1 Score metrics. The notebook uses the results from the classify_trump_tweets_for_export.ipynb notebook to evaluate the classification model's performance on trump_tweets_sample_labeled_with_predictions.csv.

Setup

Install Ollama

Install Ollama, a tool for installing and running open weights models locally.

Install an open model using Ollama. For example, to install the multi-modal 'gemma 3' model, run the following command in the terminal:

WARNING: This will install a model that is 3.3GB in size, so make sure you have enough disk space. You also likely will need at least 8GB of free RAM -- ie amount beyond what you're already using -- to run this model. If you have less RAM or disk space, you can try a smaller model, such as gemma3:1b

# Install the gemma3 model
ollama pull gemma3

If you want to test Ollama and your model in the command line, you can run the following command:

ollama run gemma3

Then type in a question and hit enter. For example, you can ask:

What is the capital of France?

Set up a new Python project

Open a terminal shell and run the below commands, one by one:

# Navigate to your code directory, e.g.
cd ~/code

# Create a new directory for the project
mkdir llm-tinker

# Navigate into the new directory
cd llm-tinker

# Create a new virtual environment
uv init

# Install the required packages
uv add jupyterlab ipykernel pandas langchain_ollama

Working in VS Code

Below steps are adapated from the Ollama Quickstart and illustrated in llm_tinker.ipynb.

Open up the new code project in VS Code.

Then create a new Jupyter Notebook file, e.g. my-llm-classifier.ipynb.

Make sure to select the correct Python kernel for the notebook in the upper right corner of the notebook.

Add the following cell to the notebook and execute it.

from langchain_ollama import OllamaLLM

If you see a green check mark, then the package is installed correctly.

Here's a simple example of how to use the OllamaLLM class to invoke a model:

# Note we're specifying the name of the model we installed with Ollama
llm = OllamaLLM(model="gemma3")
llm.invoke("What is the capital of France?")

See label_trump_tweets.ipynb for a more advanced example of how to use Ollama with LangChain to classify tweets about Trump, with guidance on additional classification tasks you can perform.