GHPT-experiments

This is inspired by GHPT: https://github.com/enmerk4r/GHPT

The aim is to "flow engineer" a Large Language Model (LLM) powered program to generate Grasshopper3d scripts based on a user prompt as input.

This repo goes as far as generating JSON formatted string which can be ingested by a custom Grasshopper component. The JSON string structure is aligned with the structure in the original GHPT repo (see above).

A modified GHPT version (saved in a separate branch) can be found here:

https://github.com/samgregson/GHPT/tree/llm-result-as-input

This allows JSON input to be used directly to allow for testing the output of the colab notebook to aid development and test validity of generated JSON.

Get Started (in VS Code)

• clone repo
• run pip install -r requirements.txt to install dependancies
• run pip install jupyter to install dependancies
• run pip install -e . to install the project module
• set up OPENAI_API_KEY in .env file (assuming you have set one up)

Built With

• Instructor
• OpenAI

Contributing

There are two ways to contribute to the notebook files, via Google Colab or locally in and IDE such as VS Code

1. clone the repository
2. create a new branch with suitable name to reflect the changes you will make

Colab

3. open the colab notebook in google colab
4. modify the notebook as required
5. "save a copy in github"

6. select the correct branch and write a suitable commit message (not as below)

7. make a pull request

Local

3. modify any file
4. push changes to your new branch
5. make a pull request

Acknowledgements

• This is inspired by GHPT: https://github.com/enmerk4r/GHPT