Skip to content

The repository for papaer "Distance between Relevant Information Pieces Causes Bias in Long-Context LLMs"

License

Notifications You must be signed in to change notification settings

Rachum-thu/LongPiBench

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

21 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

LongPiBench

LongPiBench is a benchmark designed to assess the positional biases in large language models (LLMs) when processing long contexts, specifically targeting biases arising from the spacing of multiple pieces of relevant information.

Overview

LongPiBench evaluates how well long-context LLMs handle tasks that involve multiple relevant information pieces distributed throughout the input context. The benchmark focuses on two types of positional biases:

Illustration of Absolute and Relative Positions

  • Absolute Positional Bias: The position of relevant information in the input (beginning, middle, end).
  • Relative Positional Bias: The spacing between relevant information pieces and how densely they are distributed across the context.

The benchmark includes tasks of varying complexity and spans four different input lengths: 32K, 64K, 128K, and 256K tokens.

Notice

  • This repo will be have a major updated soon.

About

The repository for papaer "Distance between Relevant Information Pieces Causes Bias in Long-Context LLMs"

Resources

License

Stars

Watchers

Forks

Packages

No packages published