W++ / JSON "persona"?
Hi,
Loving PygmalionAI. Super fun to play around with in Python using HuggingFace transformers. :)
I've noticed with some of the notebook implementations allow for more character persona options, rather than just describing "persona: xyz" and passing that as the first part of the input to the HF transformer model. Is there a way to use W++ or even a JSON as the "persona" when using PygmalionAI / python / HF transformers / AutoModelforCausalLM?
Thanks!
At the end of the day, every single notebook/UI that you use eventually just generates a glob of text (usually called the "prompt") to send off to the model. That means that if you want to use W++ or JSON for the persona, all you have to do is dump that into the prompt text. For example, if you have a character called Emma you could try a really basic prompt like:
Emma's Persona: (your JSON or W++ text here)
<START>
(possibly some example dialogue/interactions here in the same format as the 2 messages below)
You: (your message here)
Emma:
Theoretically, due to the way the model was trained, you should get better results by following the outline described above, but you're free to experiment and try to see if you can find something that works better.
Awesome, thank you so much 11b. Based on your example, I'm pretty sure I've got it formatted the proper way. Theoretically, of course. :)
Can you be more spesific please. I keep getting same results due to not being able to create characters.