Train OpenAI completion model using Whatsapp chat export to speak like your friend.
Install requirements in venv
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
Export creds environment variable
export OPENAI_API_KEY=<insert api key from openai here>
Use notebook create_data_from_whatsapp_export.ipynb
to parse Whatsapp chat export and save csvs with message, reply pairs.
Then use the openai command line tool to convert the CSVs to JSONL files used to train custom GPT-3 models.
openai tools fine_tunes.prepare_data -f data_alex.csv
openai tools fine_tunes.prepare_data -f data_yashodan.csv
openai api fine_tunes.create -t data_alex_prepared.jsonl -m curie
openai api fine_tunes.create -t data_yashodan_prepared.jsonl -m curie
Trained models:
- alex
curie:ft-user-3r7wl<...>e-2021-08-21-13-09-19
- yashodan
curie:ft-user-3r7w<...>e-2021-08-21-13-20-04
remember to include prompt
->
Alex
curl https://api.openai.com/v1/completions \
-H "Authorization: Bearer sk-Ge<...>QC" \
-H "Content-Type: application/json" \
-d '{"prompt": "Do you want to test it out? ->", "model": "curie:ft-user-3r7wl<...>e-2021-08-21-13-09-19"}'
Yashodan
curl https://api.openai.com/v1/completions \
-H "Authorization: Bearer sk-Ge<...>QC" \
-H "Content-Type: application/json" \
-d '{"prompt": "Do you want to test it out? ->", "model": "curie:ft-user-3r7w<...>e-2021-08-21-13-20-04"}'