# Latent Multi-Head Attention for Small Language Models

arXiv: https://arxiv.org/abs/2506.09342 · HTML: https://mlsanity.com/paper/2506.09342/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 19 of its 37 references that are in the map.
- First posted on arXiv 11 Jun 2025.

## Builds on (most cited first, 19 of 19)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://mlsanity.com/paper/1910.01108/) (2019)
- [On the Opportunities and Risks of Foundation Models](https://mlsanity.com/paper/2108.07258/) (2021)
- [FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness](https://mlsanity.com/paper/2205.14135/) (2022)
- [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://mlsanity.com/paper/1909.11942/) (2019)
- [G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment](https://mlsanity.com/paper/2303.16634/) (2023)
- [Reformer: The Efficient Transformer](https://mlsanity.com/paper/2001.04451/) (2020)
- [Rethinking Attention with Performers](https://mlsanity.com/paper/2009.14794/) (2020)
- [Linformer: Self-Attention with Linear Complexity](https://mlsanity.com/paper/2006.04768/) (2020)
- [GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints](https://mlsanity.com/paper/2305.13245/) (2023)
- [DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model](https://mlsanity.com/paper/2405.04434/) (2024)
- [Self-Attention with Relative Position Representations](https://mlsanity.com/paper/1803.02155/) (2018)
- [Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation](https://mlsanity.com/paper/2108.12409/) (2021)
- [MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies](https://mlsanity.com/paper/2404.06395/) (2024)
- [Fast Transformer Decoding: One Write-Head is All You Need](https://mlsanity.com/paper/1911.02150/) (2019)
- [Contextual Position Encoding: Learning to Count What's Important](https://mlsanity.com/paper/2405.18719/) (2024)
- [TransMLA: Multi-Head Latent Attention Is All You Need](https://mlsanity.com/paper/2502.07864/) (2025)

## Built on by (most cited first, 2 of 2)

- [A Survey of Context Engineering for Large Language Models](https://mlsanity.com/paper/2507.13334/) (2025)
- [Muon: Training and Trade-offs with Latent Attention and MoE](https://mlsanity.com/paper/2509.24406/) (2025)

## Abstract (arXiv)

We present the first comprehensive study of latent multi-head attention (MLA) for small language models, revealing interesting efficiency-quality trade-offs. Training 30M-parameter GPT models on 100,000 synthetic stories, we benchmark three architectural variants: standard multi-head attention (MHA), MLA, and MLA with rotary positional embeddings (MLA+RoPE). Our key finding is that MLA+RoPE with half-rank latent dimensions (r = d/2) achieves a 45% KV-cache memory reduction while incurring only a 0.3% increase in validation loss (essentially matching MHA quality)- a Pareto improvement for memory constrained deployment. We further show that RoPE is crucial for MLA in small models: without it, MLA underperforms vanilla attention by 3-5%, but with RoPE, it surpasses vanilla by 2%. Inference benchmarks on NVIDIA A100 GPUs reveal that MLA with r=d/2 achieves a 1.4 times speedup over full-rank MLA while maintaining the memory savings. GPT-4 evaluations corroborate perplexity results, with ours achieving the highest quality scores (7.4/10) across grammar, creativity, and consistency metrics. Code and models will be released upon acceptance.