Skip to content

context compression #432

@klsr27

Description

@klsr27

Question

how to compress chat history when training for long horizons tasks

Context

I'm looking for examples or pointers on how to correctly implement context compression through summarization of N-1 turns, and how how change he chat history in the middle of an ongoing episode.

Relevant Code / Config

Environment

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    questionFurther information is requested

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions