Discovering Hidden Gems in Model Repositories

TL;DR

This paper uncovers 'hidden gems' in model repositories that outperform popular models, introduces a bandit-based method for efficient discovery, and demonstrates significant acceleration in finding top models.

Contribution

It presents a novel approach to identify high-performing but overlooked models using a Multi-Armed Bandit framework and optimized search algorithms.

Findings

Rare models improve math performance from 83.2% to 96.0%.

Proposed method accelerates model discovery by over 50 times.

Efficient search requires as few as 50 queries per candidate.

Abstract

Public repositories host millions of fine-tuned models, yet community usage remains disproportionately concentrated on a small number of foundation checkpoints. We investigate whether this concentration reflects efficient market selection or if superior models are systematically overlooked. Through an extensive evaluation of over 2,000 models, we show the prevalence of "hidden gems", unpopular fine-tunes that significantly outperform their popular counterparts. Notably, within the Llama-3.1-8B family, we find rarely downloaded checkpoints that improve math performance from 83.2% to 96.0% without increasing inference costs. However, discovering these models through exhaustive evaluation of every uploaded model is computationally infeasible. We therefore formulate model discovery as a Multi-Armed Bandit problem and accelerate the Sequential Halving search algorithm by using shared query sets and aggressive elimination schedules. Our method retrieves top models with as few as 50 queries per candidate, accelerating discovery by over 50x.