Tapered Language Models
2026-06-22 • Machine Learning
Machine LearningArtificial IntelligenceComputation and Language
AI summaryⓘ
The authors studied how layers in language models share computing resources and found that not all layers need the same amount of capacity. They showed that giving more resources to early layers and less to later ones improves performance compared to using the same amount throughout. Based on this, they created Tapered Language Models (TLMs), which gradually reduce the size of certain parts of the model across its depth. Testing this idea on multiple model types and sizes, they consistently saw better results without extra cost. Their work suggests spreading resources unevenly across layers is a simple way to make language models work better.
language modeltransformerperplexityMLPmodel capacitylayerparameter allocationcosine scheduleresidual streambenchmark performance
Authors
Reza Bayat, Ali Behrouz, Aaron Courville
Abstract
Modern language models, including transformer, recurrent, and memory-based variants, share a common chassis: a stack of identical layers in which parameters are allocated uniformly across depth. This is a default inherited from the original transformer and largely unchanged since, yet a growing body of evidence suggests that layers contribute non-uniformly to the final output, with later layers refining the residual stream rather than transforming it. We ask whether parameter capacity should reflect this asymmetry. Our controlled experiment shows that, under a fixed budget, allocating more capacity to earlier layers and less to later layers improves perplexity over a uniform-width baseline, while the reverse allocation hurts. Building on this result, we introduce Tapered Language Models (TLMs), an architectural principle in which a parameter-bearing component is monotonically tapered across depth under a fixed total budget. MLPs are the natural site for this instantiation: they dominate parameter count across all modern LM families and expose width as a single, clean axis of variation. Across three model scales and four architectures (Transformer, Gated Attention, Hope-attention, and Titans), tapering MLP width via a smooth cosine schedule consistently improves perplexity and downstream benchmark performance over uniform baselines, at no additional parameter or compute cost. These findings establish depth-aware capacity allocation as a simple, architecture-agnostic axis of language model design, a free lever hidden in plain sight.