Skip to content

Difficulty Generating Answers in the Desired Format for Low-Shot Scenario with GPT-J #11

@1836897243

Description

@1836897243

Hi,

Thank you for sharing your code. I'm trying to run it in a low-shot scenario using the GPT-J model, where the number of training samples is typically small. However, I'm facing difficulty generating answers in the expected format (question###answer@@@).

For instance, in the car dataset (data_id=40975), when the number of training samples is 64, the prompt and generated response appear as follows:
'When buying = high, maint = high, doors = 4, persons = 2, lug boot = med, safety = high, How would you rate the decision to buy this car?###@ unacceptable@@### acceptable good good@@@@ very'
As you can see, the output doesn't match the expected format.

Have you encountered this issue as well? If so, could you share any insights or potential solutions to resolve it?

Thank you for your assistance!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions