This repo shows how to finetune GPT-3.5-Turbo on tweets. Specifically, Elon Musk's tweets.
In this example, data is first exported from Twitter using Apify. A copy of this data can be found in dataset_twitter-scraper_2023-08-23_22-13-19-740.json.
This data is then loaded to a format with which it can be used to finetune a GPT-3.5-Turbo model, and is then used to do exactly that. This can be done by running python ingest.py.
A Streamlit app is this created to compare this finetuned model to a prompted GPT-3.5-Turbo model.
This can be run with streamlit run app.py.
Access the final app hosted on Streamlit here.