Skip to main content
PaperLLMsv1.0

Fast Transformer Decoding: One Write-Head is All You Need (Multi-Query Attention)

by Google Brain · free · Last verified 2026-03-17

Introduces Multi-Query Attention (MQA), sharing a single set of key-value heads across all query heads during decoding. Achieves significant memory bandwidth savings for autoregressive generation with minimal quality degradation, pioneering efficient LLM inference.

https://arxiv.org/abs/1911.02150
B
BAbove Average
Adoption: B+Quality: AFreshness: CCitations: BEngagement: F

Specifications

License
Open Access
Pricing
free
Capabilities
efficient-decoding, kv-cache-optimization, autoregressive-generation
Integrations
Use Cases
inference-optimization, memory-efficient-serving
API Available
No
Tags
multi-query-attention, mqa, inference-speed, kv-cache, decoding
Added
2026-03-17
Completeness
100%

Index Score

63.2
Adoption
75
Quality
85
Freshness
45
Citations
65
Engagement
0

Put AI to work for your business

Deploy this paper alongside autonomous AaaS agents that handle tasks end-to-end — no babysitting required.

Explore the full AI ecosystem on Agents as a Service