Model routing

모델 라우팅

Select different models or processing paths according to request characteristics.

···
html
<div class="scene"><div class="heading">REQUEST ROUTER</div><div class="router"><div id="request" class="request">짧은 요약</div><div class="junction">분류기</div><div class="lanes"><div id="small">빠른 경로</div><div id="large">정밀 경로</div></div></div><div class="muted" id="route-note">빠른 경로 선택</div></div>
css
.scene{width:min(94vw,800px);height:min(88vh,326px);padding:clamp(10px,2.7vmin,19px);border:1px solid var(--line);border-radius:13px;background:var(--surface);font:500 clamp(13px,2.4vmin,17px)/1.3 var(--font-sans,sans-serif);position:relative;overflow:hidden}.scene .mono{font-family:ui-monospace,SFMono-Regular,monospace}.scene .muted{color:var(--muted)}.scene .accent{color:var(--accent)}.scene .heading{font-weight:750;color:var(--accent);margin-bottom:clamp(5px,1.6vmin,12px)}.router{height:66%;display:flex;align-items:center;justify-content:space-around;gap:5px}.request,.junction,.lanes div{border:1px solid var(--line);background:var(--bg);border-radius:7px;padding:clamp(6px,2vmin,13px);text-align:center}.junction{border-color:var(--accent)}.lanes{display:grid;gap:8px}.lanes div.active{background:color-mix(in srgb,var(--accent) 20%,var(--surface));border-color:var(--accent);color:var(--accent)}
js
let complex=false;function draw(){complex=!complex;document.getElementById('request').textContent=complex?'다단계 분석':'짧은 요약';document.getElementById('small').classList.toggle('active',!complex);document.getElementById('large').classList.toggle('active',complex);document.getElementById('route-note').textContent=complex?'정밀 경로 선택':'빠른 경로 선택'}draw();setInterval(draw,1500)

Model routing selects a path based on complexity, latency needs, or cost limits. Simple and complex requests can go to different candidate models.

The demo sends requests through a classifier into two lanes. Routing mistakes and retry cost must be measured so savings do not hide quality loss.

When to use

Use it when requests vary enough to justify balancing quality, latency, and cost.

Open as page ↗