Prompt-as-Method Example

A simple example to showcase using a prompt as a method.

The prompt is defined in the prompt.json.

{
    "task": "https://raw.githubusercontent.com/GESIS-Methods-Hub/prompt-as-method/refs/heads/main/examples/sentiment/task.json",
    "model": "llama3.1",
    "messages": [
        {
            "content": "Judge whether the following text has 'positive', 'neutral' or 'negative' sentiment. Answer with a JSON object with the sentiment and your confidence (between 0 and 1).\n\n{{{text}}}",
            "role": "user"
        }
    ]
}

In the prompt, {{{text}}} is replaced by the values in the text-column of data.csv:

text,sentiment
The food was terrible,negative
This phone is awesome!,positive

To run it, make sure you have an Ollama (or another LLM server) running on the default port on your local machine:

# ensure LLM is downloaded
ollama pull llama3.1

# initialize virtual environemnt
python3 -m venv .venv && source .venv/bin/activate

# install prompt-as-method
pip install prompt-as-method

# run the prompt
python -m prompt_as_method --prompt prompt.json --data data.csv

Output (indentation added, confidence might change):

{
    "input":{"text":"The food was terrible","sentiment":"negative"},
    "output":{"sentiment":"negative","confidence":0.9}
}
{
    "input":{"text":"This phone is awesome!","sentiment":"positive"},
    "output":{"sentiment":"positive","confidence":1}
}