Recommendations on NLP models for ad copy

I’m looking for the best model to help generate ideas for ad copy - can anyone give me some advice? What are the differences between the NLP models that you host? Thanks!

Welcome! The GPT models available on the Model Hub are all ‘generic’ models trained on a substantial amount of text, which enables them to generalise to specific downstream tasks such as ad copy, without the need for fine-tuning.

So my recommendation would be to use GPT-J (the largest / most powerful model we serve, at 6B parameters) and experiment with different prompts to get some useful results. Then once you have a prompt that you’re happy with, you can try out the different models and see if you prefer the speed of a smaller model or if you’d prefer to stick with the capable GPT-J.

I like to supply at least one ‘example’ of the desired output to help guide the model towards a useful generation. Here’s a prompt template you can copy:

Write 10 catchy taglines for a new pair of rainbow socks. The socks are soft and environmentally-friendly. The socks are one-size-fits-all.

1: There's socks at the end of this rainbow.

I also suggest you use Playgrounds while experimenting because it’s easier to test out different models quickly there, and then you can deploy them in Pipeline over API.

1 Like

Thanks for the advice - I’m making progress and playgrounds is extremely helpful.