English

Hymba: A Hybrid-head Architecture for Small Language Models

Computation and Language 2024-11-22 v1 Artificial Intelligence Machine Learning

Abstract

We propose Hymba, a family of small language models featuring a hybrid-head parallel architecture that integrates transformer attention mechanisms with state space models (SSMs) for enhanced efficiency. Attention heads provide high-resolution recall, while SSM heads enable efficient context summarization. Additionally, we introduce learnable meta tokens that are prepended to prompts, storing critical information and alleviating the "forced-to-attend" burden associated with attention mechanisms. This model is further optimized by incorporating cross-layer key-value (KV) sharing and partial sliding window attention, resulting in a compact cache size. During development, we conducted a controlled study comparing various architectures under identical settings and observed significant advantages of our proposed architecture. Notably, Hymba achieves state-of-the-art results for small LMs: Our Hymba-1.5B-Base model surpasses all sub-2B public models in performance and even outperforms Llama-3.2-3B with 1.32% higher average accuracy, an 11.67x cache size reduction, and 3.49x throughput.

Keywords

Cite

@article{arxiv.2411.13676,
  title  = {Hymba: A Hybrid-head Architecture for Small Language Models},
  author = {Xin Dong and Yonggan Fu and Shizhe Diao and Wonmin Byeon and Zijia Chen and Ameya Sunil Mahabaleshwarkar and Shih-Yang Liu and Matthijs Van Keirsbilck and Min-Hung Chen and Yoshi Suhara and Yingyan Lin and Jan Kautz and Pavlo Molchanov},
  journal= {arXiv preprint arXiv:2411.13676},
  year   = {2024}
}

Comments

20 pages, models are available on huggingface

R2 v1 2026-06-28T20:07:05.478Z