Do Language Models Use Their Depth Efficiently?

TL;DR

This paper investigates whether deeper language models utilize their additional layers for new computations or merely distribute existing ones, finding evidence that they mainly spread the same computations across more layers, leading to diminishing returns.

Contribution

The study provides empirical analysis of residual streams in large language models, revealing that increased depth mainly results in finer adjustments rather than new types of computation.

Findings

Deeper layers contribute less to the residual stream after a phase transition.

Skipping later layers has minimal impact on outputs.

Models do not show evidence of using depth for multi-hop reasoning or new computations.

Abstract

Modern LLMs are increasingly deep, and depth correlates with performance, albeit with diminishing returns. However, do these models use their depth efficiently? Do they compose more features to create higher-order computations that are impossible in shallow models, or do they merely spread the same kinds of computation out over more layers? To address these questions, we analyze the residual stream of the Llama 3.1, Qwen 3, and OLMo 2 family of models. We find: First, comparing the output of the sublayers to the residual stream reveals that layers in the second half contribute much less than those in the first half, with a clear phase transition between the two halves. Second, skipping layers in the second half has a much smaller effect on future computations and output predictions. Third, for multihop tasks, we are unable to find evidence that models are using increased depth to compose subresults in examples involving many hops. Fourth, we seek to directly address whether deeper models are using their additional layers to perform new kinds of computation. To do this, we train linear maps from the residual stream of a shallow model to a deeper one. We find that layers with the same relative depth map best to each other, suggesting that the larger model simply spreads the same computations out over its many layers. All this evidence suggests that deeper models are not using their depth to learn new kinds of computation, but only using the greater depth to perform more fine-grained adjustments to the residual. This may help explain why increasing scale leads to diminishing returns for stacked Transformer architectures.