Errors are Useful Prompts: Instruction Guided Task Programming with Verifier-Assisted Iterative Prompting
- Project website: https://ac-rad.github.io/clairify/
This repository contains
- The source code for CLAIRify
- Dataset (Chem-RnD and Chem-EDU)
- CLAIRify web interface
A tutorial on CLAIRify is provided as Jupyter Notebook.
- OpenAI Python Library
You need to set your OpenAI API key in OPENAI_API_KEY environment variable.
To generate a XDL protocol from a natural language description of an experiment, run the following:
python3 xdlgenerator/nlp2xdl.py --input_dir /path/to/experiment/dir
where /path/to/experiment/dir is a directory containing natural language experiments. Each experiment is assumed to be its own file in the dictory (e.g. expertiment1.txt, experiment2.txt). Running the script will automatically generate an output directory /path/to/experiment/dir_output. Each file in the new directory contains a XDL description.