Mapping 1,000+ Language Models via the Log-Likelihood Vector

TL;DR

This paper introduces a scalable method using log-likelihood vectors to compare over 1,000 language models, enabling efficient large-scale analysis based on theoretical foundations related to model divergence.

Contribution

The paper presents a novel, scalable approach for comparing language models using log-likelihood vectors, grounded in a theoretical approximation of divergence.

Findings

Constructed a comprehensive model map of 1,000+ language models

Demonstrated the method's linear scalability with models and data

Provided new insights into large-scale model relationships

Abstract

To compare autoregressive language models at scale, we propose using log-likelihood vectors computed on a predefined text set as model features. This approach has a solid theoretical basis: when treated as model coordinates, their squared Euclidean distance approximates the Kullback-Leibler divergence of text-generation probabilities. Our method is highly scalable, with computational cost growing linearly in both the number of models and text samples, and is easy to implement as the required features are derived from cross-entropy loss. Applying this method to over 1,000 language models, we constructed a "model map," providing a new perspective on large-scale model analysis.