Few-shot in this model

#7
by LGFU - opened

In the scorecard, there is a performance increase for few-shot prompts. Do I integrate the few-shot prompts into the hypothesis?

Thank you

in this case, few-shot means that the model has seen some example texts for the same task during training. There is nothing you need to change during inference. It's not like with a generative LLM, where you need to add few-shot examples in a prompt. The model will have this performance on the respective datasets by default, because it has seen some training data from the task.

Got it, thank you!

Sign up or log in to comment