Skip to content

Inquiry about generation length control in xsum, squad, and writingprompt datasets #21

Description

@bigbrother001

While processing the xsum, squad, and writingprompt datasets, I was quite surprised to find that the average length of the human-written "original" texts is remarkably close to that of the LLM-generated "sampled" texts.
Could you please share what strategies you used to control the length of the generated text? As far as I know, earlier LLMs tend to generate significantly longer text compared to human-written content.

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