# Multi-Head Attention: N Paralel Head, Concat + Projection, Grouped-Query Attention (GQA), Multi-Query Attention (MQA)

> Source: https://sukruyusufkaya.com/learn/llm-muhendisligi/multi-head-attention-gqa-mqa-llama-3
> Updated: 2026-08-22T10:53:36.614Z
> Category: LLM Mühendisliği
> Module: Modül 8: Attention Mathematics — Transformer'ın Kalbi
**TLDR:** Tek attention'ı niye N paralel head'e bölüyoruz: her head'in farklı pattern öğrenme kapasitesi (syntactic, semantic, positional). Concat + output projection mimari, head pruning empirical bulgular, Llama-3 grouped-query attention (GQA), Mistral multi-query attention (MQA), head visualization Türkçe örneklerle.