DataFog Instructor

Demo

datafog-instructor lets you use open-source LLMs (HF models, models ending in .gguf) and instruct them to return only specific, user-defined outputs like entity tags, JSON-only structured output, and synthetic data templates, among others.
We do this by using advanced ML methods that constrain the set of tokens a model will generate next when it's providing a response, and this allows us to achieve a far higher rate of accuracy and precision than baseline LLM output methods (for more, see the evals.ipynb file in examples).

Installation

pip install --upgrade datafog-instructor

Quick Start

To see a list of all available options, you can type

datafog-instructor --help

Which should show you a screen like this:

Initialize

Begin by entering the following into your terminal post-install:

datafog-instructor init

a default config file containing information about the model, tokenizer, regex_pattern gets loaded into your working directory.

You can see the contents of that file by typing:

datafog-instructor show-fogprint

What is a fogprint? A fogprint is a template that you can re-use, with specific configuration settings for the models, filenames, model_ids, and other important information to instruct an LLM to detect entities. This file is currently saved as fogprint.json.

Verify the installation:

datafog-instructor list-entities

You should see a list of default entity types: PERSON, COMPANY, LOCATION, and ORG.

Sample Operations

Detect Entities in Text

datafog-instructor detect-entities --prompt "Apple Inc. was founded by Steve Jobs in Cupertino, California."

This will output a table of detected entities, their positions, and types.

Display Current Configuration

datafog-instructor show-fogprint

This command will show you the current configuration stored in fogprint.json.

Reinitialize with Custom Settings

To change the default model or pattern:

1. Edit the fogprint.json file directly, or
2. Use the init command with the --force flag:

datafog-instructor init --force

Follow the prompts to update your configuration.

Advanced Usage

• Adjust the maximum number of tokens generated:

datafog-instructor detect-entities --prompt "Your text here" --max-new-tokens 100

• For batch processing or integration into your Python projects, import the EntityDetector class from models.py.

Development and Testing

For development purposes, you can install additional dependencies:

python -m venv venv && source venv/bin/activate && pip install requirements-dev.txt

## Documentation

To build the documentation locally:

pip install datafog-instructor[docs] cd docs sphinx

The documentation will be available in the `docs/_build/html` directory.

## Contributing

Contributions to the DataFog Instructor SDK are welcome! Please feel free to submit a Pull Request.

## License

This project is licensed under the MIT License.

## Support

If you encounter any problems or have any questions, please open an issue on the GitHub repository or join our Discord community at https://discord.gg/bzDth394R4.

## Links

- Homepage: https://datafog.ai
- Documentation: https://docs.datafog.ai
- Twitter: https://twitter.com/datafoginc
- GitHub: https://github.com/datafog/datafog-instructor