Configuration Wizard

Adding a Local Ollama Model

To use local models with Ollama:

1. Install Ollama:

   brew install ollama  # macOS
   curl -fsSL https://ollama.com/install.sh | sh  # Linux

2. Pull a model:

   ollama pull llama3.1
   ollama pull mistral

3. Add to configuration: Use format ollama/model-name

   {
     "name": "ollama/llama3.1",
     "type": "llm",
     "prompt_manipulation": []
   }

Benefits: Zero cost, runs locally, complete privacy

Adding a Trained Transformer

To use a trained transformer model:

1. Train your model first:

   poetry run python examples/train_location_transformer.py

2. Add to configuration:

   {
     "type": "transformer",
     "label": "distilbert-classifier",
     "model_path": "./transformer_models/final_model"
   }

Benefits: Zero inference cost, fast, runs offline

Required Data Format

Your training data must be a JSON array. The content field can be a plain string or a JSON object for prompt variables:

Plain text content — use {content} in your prompt:

[
  {
    "id": "example-1",
    "content": "Your document text here...",
    "label": "yes"
  }
]

Dict content (prompt variables) — use {key} placeholders in your prompt:

[
  {
    "id": "example-1",
    "content": {
      "text": "The Amazon basin receives over 2,000 mm of rain.",
      "topic": "climate"
    },
    "label": "yes"
  }
]

With dict content, each key becomes a {key} placeholder — e.g. Text: {text}\nTopic: {topic}.

Required fields:

• content: The text to classify (string or object)
• label: The correct answer