EmbedLLM: Learning Compact Representations of Large Language Models

TL;DR

EmbedLLM introduces a novel framework for learning compact vector representations of large language models, enabling efficient model routing and performance prediction across multiple benchmarks without extra inference costs.

Contribution

The paper presents EmbedLLM, a new encoder-decoder approach for creating embeddings of LLMs, improving model routing accuracy and latency, and enabling performance forecasting without additional inference.

Findings

EmbedLLM outperforms prior methods in model routing accuracy and latency.

The embeddings can predict model performance on benchmarks without extra inference.

Probing shows embeddings capture key model characteristics, like coding specialization.

Abstract

With hundreds of thousands of language models available on Huggingface today, efficiently evaluating and utilizing these models across various downstream, tasks has become increasingly critical. Many existing methods repeatedly learn task-specific representations of Large Language Models (LLMs), which leads to inefficiencies in both time and computational resources. To address this, we propose EmbedLLM, a framework designed to learn compact vector representations, of LLMs that facilitate downstream applications involving many models, such as model routing. We introduce an encoder-decoder approach for learning such embeddings, along with a systematic framework to evaluate their effectiveness. Empirical results show that EmbedLLM outperforms prior methods in model routing both in accuracy and latency. Additionally, we demonstrate that our method can forecast a model's performance on multiple benchmarks, without incurring additional inference cost. Extensive probing experiments validate that the learned embeddings capture key model characteristics, e.g. whether the model is specialized for coding tasks, even without being explicitly trained on them. We open source our dataset, code and embedder to facilitate further research and application.