This repo will help you fine-tune GPT-3 with a Google Chat conversation history. The trained model will be able to converse as one or both sides of the conversation in the participants' style.
- Download your Chat archive from Google Takeout.
- Locate the
messages.jsonfile of the conversation you would like to use as a training set. - Use the script to prepare data for training:
python preparer.py --messages <MESSAGES_FILE>- Test your training data with OpenAI's tool:
openai tools fine_tunes.prepare_data -f <TRAINING_FILE>You should see: No remediations found.
- Fine-tune GPT-3 with your training data:
openai api fine_tunes.create -t <TRAINING_FILE>You should see: Job complete! Status: succeeded 🎉. Don't forget to note the name of the model.
- Try out your model in the Playground or with the CLI:
openai api completions.create -m <MODEL_NAME>