Use Stanford Alpaca's Instruction generating script instead, this is just a demo
This repo is a demo of how to create new instruction prompts using ChatGPT API and using ray as workers to distribute the work across up to 1000 workers.
I wanted to see in 5 minutes, how many features I can write from scratch for an instruction generating script.
The variants is the generated instructions is pretty low. Probably using more seed instructions and a similiarity filter like in Alpaca would help.
Optional: Set up your ray cluster
## set up OPENAI_API_KEY
export OPENAI_API_KEY=YOUR_KEY
# Install dependencies
pip install -r requirements.txt
# Run the script
python main.py