Compiler and Systems on Jaehun's Bloghttps://jaehun.me/categories/compiler-and-systems/Recent content in Compiler and Systems on Jaehun's BlogHugoko-krTue, 08 Sep 2026 13:42:29 +0000[논문 리뷰] Hardware-Efficient Attention for Fast Decodinghttps://jaehun.me/posts/%EB%85%BC%EB%AC%B8-%EB%A6%AC%EB%B7%B0-hardware-efficient-attention-for-fast-decoding/Thu, 23 Oct 2025 00:00:00 +0900https://jaehun.me/posts/%EB%85%BC%EB%AC%B8-%EB%A6%AC%EB%B7%B0-hardware-efficient-attention-for-fast-decoding/<p><a href="https://arxiv.org/abs/2505.21487v1"target="_blank" class="inline-flex items-center gap-1" >논문 링크<svg class="h-3 w-3 flex-shrink-0" id="external-link" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M15 3h6v6m-11 5L21 3m-3 10v6a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2V8a2 2 0 0 1 2-2h6"/></svg> </a></p> <h2 id="gta--gla-디코딩의-메모리-지배를-깨는-하드웨어-효율-어텐션">GTA &amp; GLA: 디코딩의 ‘메모리-지배’를 깨는 하드웨어 효율 어텐션<a href="#gta--gla-%eb%94%94%ec%bd%94%eb%94%a9%ec%9d%98-%eb%a9%94%eb%aa%a8%eb%a6%ac-%ec%a7%80%eb%b0%b0%eb%a5%bc-%ea%b9%a8%eb%8a%94-%ed%95%98%eb%93%9c%ec%9b%a8%ec%96%b4-%ed%9a%a8%ec%9c%a8-%ec%96%b4%ed%85%90%ec%85%98" class="heading-anchor" aria-label="이 섹션에 대한 링크"><svg class="h-4 w-4" aria-hidden="true" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><g fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2"><path d="M10 13a5 5 0 0 0 7.54.54l3-3a5 5 0 0 0-7.07-7.07l-1.72 1.71"/><path d="M14 11a5 5 0 0 0-7.54-.54l-3 3a5 5 0 0 0 7.07 7.07l1.71-1.71"/></g></svg></a></h2><h2 id="tldr">TL;DR<a href="#tldr" class="heading-anchor" aria-label="이 섹션에 대한 링크"><svg class="h-4 w-4" aria-hidden="true" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><g fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2"><path d="M10 13a5 5 0 0 0 7.54.54l3-3a5 5 0 0 0-7.07-7.07l-1.72 1.71"/><path d="M14 11a5 5 0 0 0-7.54-.54l-3 3a5 5 0 0 0 7.07 7.07l1.71-1.71"/></g></svg></a></h2><p><strong>GTA(키·값 타이잉)와 GLA(잠재 헤드 샤딩)</strong> 으로 <strong>FLOPs/byte</strong>를 끌어올리고 <strong>per-GPU KV 캐시</strong>를 줄여, <strong>품질 손실 없이</strong> <strong>커널 1.2–2× 가속</strong>, <strong>서빙 지연 대폭↓</strong>, <strong>처리량 최대 2×↑</strong> 를 달성한다. (근거: §Abstract, Fig.4–5, Fig.15, Tab.38–41)</p>