Return of the Encoder: Maximizing Parameter Efficiency for SLMs

TL;DR

This paper demonstrates that encoder-decoder architectures are more efficient than decoder-only models for small language models, offering significant latency and throughput advantages on edge devices, and introduces a knowledge distillation method to enhance their performance.

Contribution

The paper provides a systematic analysis of encoder-decoder versus decoder-only models for small language models and introduces a knowledge distillation framework to improve encoder-decoder capabilities.

Findings

Encoder-decoder models achieve 47% lower first-token latency.

Encoder-decoder models have 4.7x higher throughput on edge devices.

Knowledge distillation improves encoder-decoder performance by up to 6 points.

Abstract

The dominance of large decoder-only language models has overshadowed encoder-decoder architectures, despite their fundamental efficiency advantages in sequence processing. For small language models (SLMs) - those with 1 billion parameters or fewer - our systematic analysis across GPU, CPU, and NPU platforms reveals that encoder-decoder architectures achieve 47% lower first-token latency and 4.7x higher throughput compared to decoder-only models on edge devices. These gains may be attributed to encoder-decoder's one-time input processing and efficient separation of understanding and generation phases. We introduce a novel knowledge distillation framework that enables encoder-decoder models to leverage capabilities from large scalable decoder-only teachers while preserving their architectural advantages, achieving up to 6 average performance points improvement across diverse tasks, with significant gains in asymmetric sequence tasks where input and output distributions can benefit from different processing approaches. When combined with modern advances like Rotary Positional Embeddings (RoPE) and Vision encoders, our systematic investigation demonstrates that encoder-decoder architectures provide a more practical path toward deploying capable language models in resource-constrained environments. Our findings challenge the prevailing trend toward decoder-only scaling, showing that architectural choices become increasingly crucial as parameter budgets decrease, particularly for on-device and edge deployments where computational efficiency is paramount.