Hi,
I’m looking for a developer who can help fine-tune the model with furniture related terms. Our end goal is to write 500+ furniture product descriptions with the help of GPT3. I don’t know if it’s the right way to solve this task, so any advice is appreciated.
Thanks
1 Like
What sort of input data do you have?
I would start using Divinci instruct with a prompt similar to:
“Describe a chair with wooden arms in great detail:
Description:”
This can get you pretty far changing the initial phrase “chair with wooden arms” and leaving the temperature at a high setting. I would collect each generation for each initial phrase in a spreadsheet of some type. If you wish to fine-tune, place the initial phrase in a prompt column and the generation in the completion column. Try to have at least 50 examples. After training, this would give you a fine-tuned model that takes an input of a short description of a furniture item (chair with wooden arms) and generates a longer description. Hope this helps your project!
1 Like