Anchor Points: Benchmarking Models with Much Fewer Examples

TL;DR

This paper introduces Anchor Point Selection, a method to evaluate and compare language models using a small, representative subset of data points, reducing the need for large benchmarks while maintaining accuracy.

Contribution

The paper proposes a novel technique for selecting small dataset subsets called Anchor Points, enabling effective model ranking and failure prediction with fewer examples.

Findings

Anchor points strongly correlate with model confidence across datasets.

Evaluating with 1-30 anchor points outperforms uniform sampling in ranking models.

Few anchor points can estimate model predictions with low error.

Abstract

Modern language models often exhibit powerful but brittle behavior, leading to the development of larger and more diverse benchmarks to reliably assess their behavior. Here, we suggest that model performance can be benchmarked and elucidated with much smaller evaluation sets. We first show that in six popular language classification benchmarks, model confidence in the correct class on many pairs of points is strongly correlated across models. We build upon this phenomenon to propose Anchor Point Selection, a technique to select small subsets of datasets that capture model behavior across the entire dataset. Anchor points reliably rank models: across 87 diverse language model-prompt pairs, evaluating models using 1-30 anchor points outperforms uniform sampling and other baselines at accurately ranking models. Moreover, just several anchor points can be used to estimate model per-class predictions on all other points in a dataset with low mean absolute error, sufficient for gauging where the model is likely to fail. Lastly, we present Anchor Point Maps for visualizing these insights and facilitating comparisons of the performance of different models on various regions within the dataset distribution.