Sequence architectures and attention · MLD-0777
Multi-query attention
Attention with many query heads sharing a single set of key and value heads.
Original reference page · detailed expansion pending. Original definitions, examples, notes and citations are preserved below. New examples are illustrative, not benchmark results.
Original example
Reduce decoding KV-cache size.
Original caveat
Sharing can trade representation flexibility for efficiency.
Context and aliases
Level: Intermediate. Stages: Training · Inference · Theory.
Tags: DL · SLM · LLM · NLP.
Aliases: MQA.
Related concepts
Concepts linking here
Original references
- GQA: Training Generalized Multi-Query Transformer Models ↗ · Reference link
- Transformers glossary ↗ · Consulted
- Original reference URL ↗