# FlashAttention: IO-Aware Attention — Dao 2022 Algoritması ve Modern Implementations

> Source: https://sukruyusufkaya.com/learn/llm-muhendisligi/flashattention-dao-2022-io-aware-attention
> Updated: 2026-08-23T19:51:12.917Z
> Category: LLM Mühendisliği
> Module: Modül 8: Attention Mathematics — Transformer'ın Kalbi
**TLDR:** FlashAttention'ın matematiksel ve sistemsel anatomi: niye standard attention memory-bound, GPU memory hierarchy (HBM vs SRAM), tile-based computation, online softmax, recomputation backward. FlashAttention-1 (Dao 2022), FlashAttention-2, FlashAttention-3 evrimi. PyTorch flash_attn library, performance benchmarks, long context enablement.