File size: 21,959 Bytes
3ed78a4 6853af9 3ed78a4 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 | <!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>RRSI — Main Results</title>
<style>
:root{
--evolve:#6366f1; --held:#0ea5e9; --ood:#10b981; --base:#94a3b8;
--ink:#0f172a; --muted:#64748b; --line:#e2e8f0; --bg:#f8fafc; --card:#ffffff;
}
*{box-sizing:border-box}
body{
margin:0; background:var(--bg); color:var(--ink);
font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,Helvetica,Arial,sans-serif;
line-height:1.55; -webkit-font-smoothing:antialiased;
}
.wrap{max-width:1080px;margin:0 auto;padding:40px 22px 70px}
header h1{font-size:26px;margin:0 0 6px;letter-spacing:-.02em}
header .sub{color:var(--muted);font-size:14px;margin:0 0 4px}
header a{color:#2563eb;text-decoration:none}
header a:hover{text-decoration:underline}
.note{
margin-top:14px;font-size:13px;color:var(--muted);
background:#fff;border:1px solid var(--line);border-radius:10px;padding:10px 14px;
}
section{margin-top:34px}
h2{font-size:17px;margin:0 0 4px;letter-spacing:-.01em}
h2 .tag{
font-size:11px;font-weight:600;text-transform:uppercase;letter-spacing:.06em;
color:#475569;background:#eef2ff;border:1px solid #e0e7ff;border-radius:999px;
padding:2px 8px;margin-right:8px;vertical-align:2px;
}
.card{background:var(--card);border:1px solid var(--line);border-radius:14px;padding:20px 22px;
box-shadow:0 1px 2px rgba(15,23,42,.04)}
.cap{font-size:13px;color:var(--muted);margin:12px 0 0}
.legend{display:flex;gap:18px;flex-wrap:wrap;font-size:13px;color:#334155;margin:6px 0 14px}
.legend i{display:inline-block;width:12px;height:12px;border-radius:3px;margin-right:6px;vertical-align:-1px}
svg{display:block;width:100%;height:auto;overflow:visible}
/* bar chart */
.domains{display:flex;gap:26px;flex-wrap:wrap;align-items:flex-end}
.dgroup{flex:1 1 300px;min-width:260px}
.dgroup h4{margin:0 0 10px;font-size:13px;letter-spacing:.04em;text-transform:uppercase;color:#475569}
.bars{display:flex;gap:14px;align-items:flex-end;border-bottom:2px solid var(--line);padding-bottom:0;min-height:230px}
.bar{flex:1 1 0;display:flex;flex-direction:column;justify-content:flex-end;align-items:center;gap:6px;min-width:48px}
.bar .val{font-size:13px;font-weight:650;font-variant-numeric:tabular-nums}
.bar .col{width:100%;border-radius:6px 6px 0 0;transition:height .3s ease}
.bar .nm{font-size:11px;color:var(--muted);text-align:center;line-height:1.3;min-height:30px}
.bar .nm b{display:block;color:#334155;font-weight:600}
/* token cost */
.tokens{display:flex;gap:22px;flex-wrap:wrap;margin-top:6px}
.tok{flex:1 1 200px;background:#fff;border:1px solid var(--line);border-radius:12px;padding:14px 16px}
.tok .lab{font-size:12px;color:var(--muted);text-transform:uppercase;letter-spacing:.05em}
.tok .big{font-size:26px;font-weight:700;font-variant-numeric:tabular-nums;margin:2px 0 0}
.tok .big span{font-size:13px;font-weight:500;color:var(--muted)}
.tok .meter{height:8px;border-radius:6px;background:#eef2f7;margin-top:10px;overflow:hidden}
.tok .meter i{display:block;height:100%;border-radius:6px}
/* tables */
table{width:100%;border-collapse:collapse;font-size:13.5px;font-variant-numeric:tabular-nums}
th,td{padding:8px 10px;text-align:right;border-bottom:1px solid var(--line)}
th:first-child,td:first-child{text-align:left}
thead th{font-size:11.5px;text-transform:uppercase;letter-spacing:.05em;color:#475569;
border-bottom:2px solid var(--line);background:#fbfcfe}
thead tr:last-child th{background:#f4f6fb}
tr.ours td{background:#ecfdf5;font-weight:650}
tr.h0 td{color:#475569}
.tblnotes{font-size:12.5px;color:var(--muted);margin-top:10px}
.win{color:#047857;font-weight:650}
.lose{color:#b91c1c;font-weight:650}
ul.mech{margin:6px 0 0;padding-left:20px;font-size:14px;color:#1e293b}
ul.mech li{margin-bottom:5px}
ul.mech b{color:#0f172a}
.grid2{display:grid;grid-template-columns:1fr 1fr;gap:18px}
@media(max-width:760px){.grid2{grid-template-columns:1fr}}
</style>
</head>
<body>
<div class="wrap">
<header>
<h1>RRSI: Regularized Recursive Self-Improvement of Agent Harnesses</h1>
<p class="sub">Peng Xia, Rujun Han, Zifeng Wang, Yanfei Chen, Yufan Zhang, Yoonho Lee, Chengsong Huang, Han Yu, Zhongying CuiZhu, Yifei Ming, Huaxiu Yao, Burak Gokturk, Tomas Pfister, Chen-Yu Lee — Google Cloud AI Research</p>
<p class="sub"><a href="https://arxiv.org/abs/2609.24972">arXiv:2609.24972</a></p>
<p class="note">Charts below re-plot numbers reported in the paper (Figures 1 & 3, Tables 1–3). Policy frozen at Claude Opus 4.8; the harness is evolved on one suite per domain and then evaluated <em>unchanged</em> on held-out and out-of-distribution (OOD) benchmarks.</p>
</header>
<section>
<h2><span class="tag">Figure 1a</span>Evolve-set gain vs. out-of-distribution gain</h2>
<p class="cap" style="margin-top:0">Agentic workspace instance. Points far to the upper-left are the goal: small evolve-set gain, large transfer. Points left of the dashed line at 0 fall <em>below</em> the unevolved harness H₀.</p>
<div class="card">
<div class="legend">
<span><i style="background:var(--evolve)"></i>Evolve-set gain (Harvey LAB)</span>
<span><i style="background:var(--ood)"></i>OOD gain (mean of JobBench, GDPval, APEX-Agents)</span>
</div>
<div id="scatter"></div>
<p class="cap">Reading it: Meta-Harness has the biggest evolve-set gain (+3.6) but keeps only +0.9 OOD. AHE and TTHE are negative OOD (−0.5 and −1.7) — their evolution actively hurt tasks they were never scored on. RRSI has the <em>smallest</em> evolve gain (+1.1) and the largest transfer (+3.9), the exact trade the regularizers are designed to make.</p>
</div>
</section>
<section>
<h2><span class="tag">Figure 3</span>Main results in all three domains (Δ vs. H₀)</h2>
<div class="card">
<div class="legend">
<span><i style="background:var(--evolve)"></i>Evolve split</span>
<span><i style="background:var(--held)"></i>In-distribution held-out</span>
<span><i style="background:var(--ood)"></i>OOD held-out (never scored during search)</span>
</div>
<div class="domains" id="domains"></div>
<p class="cap"><b>Every</b> held-out split improves and none regresses. Frontier-Eng is measured in medal points: +4.3 is a 24.3% relative gain. The engineering-design domain matters because its tasks are graded by frozen deterministic simulators, not a judge model — so the transfer can't be explained by a harness learning to write the way a judge rewards.</p>
</div>
</section>
<section>
<h2><span class="tag">Efficiency</span>Policy tokens per trial</h2>
<div class="card">
<div class="tokens" id="tokens"></div>
<p class="cap">Regularization is <em>cheaper</em>, not more expensive: RRSI spends 2.42M tokens/trial against 3.80M for unregularized evolution (≈36% less), and only 1.6× the unevolved baseline. Unregularized selection spends most of its accepted edits on noise and on context rather than on mechanism — it buys score with compute, and the compute doesn't transfer.</p>
</div>
</section>
<section>
<h2><span class="tag">Table 1</span>Comparison with prior harness evolution methods</h2>
<div class="card">
<table>
<thead>
<tr>
<th rowspan="2">Method</th>
<th colspan="2">In-distribution</th>
<th colspan="3">Out-of-distribution</th>
<th rowspan="2">OOD avg.</th>
</tr>
<tr>
<th>Harvey LAB (evolve)</th>
<th>Harvey LAB (ID held-out)</th>
<th>JobBench</th>
<th>GDPval</th>
<th>APEX-Agents</th>
</tr>
</thead>
<tbody id="t1"></tbody>
</table>
<p class="tblnotes">OOD avg. is computed as the mean of JobBench, GDPval and APEX-Agents (the paper reports only the H₀ and RRSI averages explicitly, 39.7 and 43.6; the baseline averages shown for the others are recomputed from the table's own columns). Same H₀, same evolve split, same candidate budget for every row.</p>
</div>
</section>
<section>
<h2><span class="tag">Table 2</span>Ablation: is it the regularizers or the score?</h2>
<div class="card">
<table>
<thead>
<tr>
<th>Variant</th>
<th>Harvey LAB (evolve)</th>
<th>Harvey LAB (ID held-out)</th>
<th>OOD avg.</th>
<th>Tokens/trial (M) ↓</th>
</tr>
</thead>
<tbody id="t2"></tbody>
</table>
<p class="tblnotes">The strongest single signal in the paper: evolve-set score and transfer move in <em>opposite</em> directions. Dropping both regularizer groups yields the highest evolve score of any arm (92.8) and an OOD average within a point of the unevolved harness. Dropping only the acceptance constraints raises the evolve score by 1.0 pt, costs 2.6 pts OOD, and inflates token cost by half.</p>
</div>
</section>
<section>
<h2><span class="tag">Table 3</span>Robustness to the backbone policy (coding)</h2>
<div class="card">
<table>
<thead>
<tr><th>Policy</th><th>Benchmark</th><th>H₀</th><th>RRSI</th><th>Δ</th></tr>
</thead>
<tbody id="t3"></tbody>
</table>
<p class="tblnotes">Evolution is run independently per policy on Terminal-Bench 2.1 only; the resulting harness is then evaluated unchanged on SWE-bench Verified. Gemini 3.5 Flash gains more on the evolve split (+14.1 — the paper's headline number) while the stronger Claude Opus 4.8 sits closer to both ceilings and gains less.</p>
</div>
</section>
<section>
<h2><span class="tag">Figure 2</span>What is actually regularized</h2>
<div class="card grid2">
<div>
<h4 style="margin:0 0 6px;font-size:13px;letter-spacing:.04em;text-transform:uppercase;color:#475569">Proposal side — how search capacity is spent</h4>
<ul class="mech">
<li><b>L₀-style annealed edit budget:</b> b_t = ⌈b_min + (b_max−b_min)·½(1+cos(πt/T))⌉. Early rounds bundle coordinated edits to find mechanisms; late rounds become sparse and attributable.</li>
<li><b>Evidence-aware credit assignment:</b> every candidate's component, hypothesis, diff, score/cost delta and accept/reject outcome are recorded; falsified hypotheses stay negative evidence.</li>
<li><b>Structured exploration:</b> when progress over w rounds stays inside the noise band δ, part of the budget is reserved for components never yet exercised.</li>
</ul>
</div>
<div>
<h4 style="margin:0 0 6px;font-size:13px;letter-spacing:.04em;text-transform:uppercase;color:#475569">Selection side — what becomes permanent state</h4>
<ul class="mech">
<li><b>Leakage screening:</b> a critic rejects diffs encoding task/entity names, task-specific values, answers, or inert machinery — <em>before</em> evaluation, so a leaking candidate never banks an inflated score.</li>
<li><b>Stability-aware acceptance:</b> Ŝ(H′) ≥ S⋆ − δ, with δ estimated from repeated runs of the unchanged base harness. Blocks slow downhill walks of individually noise-sized regressions.</li>
<li><b>L₂-style complexity-aware acceptance:</b> for ΔS > δ, require ΔC ≤ β₀ + β₁ΔS — extra tokens must be justified by measured improvement.</li>
<li><b>L₁-style structural pruning:</b> components with no strictly positive measured gain over a window are handed back to the proposer as deletion targets.</li>
</ul>
</div>
</div>
</section>
</div>
<script>
const C = {evolve:'#6366f1', held:'#0ea5e9', ood:'#10b981', base:'#94a3b8'};
/* ---------- Figure 1a scatter ---------- */
(function(){
const W=660,H=430,m={l:66,r:30,t:26,b:56};
const xMin=-0.5,xMax=4.3,yMin=-2.5,yMax=4.7;
const X=v=>m.l+(v-xMin)/(xMax-xMin)*(W-m.l-m.r);
const Y=v=>m.t+(yMax-v)/(yMax-yMin)*(H-m.t-m.b);
const pts=[
{n:'H₀ (base)',x:0,y:0,c:C.base,r:6,ax:-6,ay:-13,anchor:'end'},
{n:'Meta-Harness',x:3.6,y:0.93,c:C.evolve,r:7,ax:-10,ay:-12,anchor:'end'},
{n:'HarnessX',x:2.4,y:0.00,c:C.evolve,r:7,ax:10,ay:5,anchor:'start'},
{n:'AHE',x:1.3,y:-0.53,c:C.evolve,r:7,ax:11,ay:4,anchor:'start'},
{n:'TTHE',x:1.7,y:-1.73,c:C.evolve,r:7,ax:11,ay:4,anchor:'start'},
{n:'RRSI (ours)',x:1.1,y:3.90,c:C.ood,r:9,ax:0,ay:-16,anchor:'middle',bold:true}
];
let s=`<svg viewBox="0 0 ${W} ${H}" role="img" aria-label="Evolve-set gain versus out-of-distribution gain">`;
// horizontal gridlines
for(let gy=-2;gy<=4;gy++){
s+=`<line x1="${m.l}" y1="${Y(gy)}" x2="${W-m.r}" y2="${Y(gy)}" stroke="${gy===0?'#cbd5e1':'#eef2f7'}" stroke-width="${gy===0?1.5:1}" ${gy===0?'stroke-dasharray="5 4"':''}/>`;
s+=`<text x="${m.l-10}" y="${Y(gy)+4}" font-size="11.5" fill="#94a3b8" text-anchor="end">${gy>0?'+':''}${gy}</text>`;
}
for(let gx=0;gx<=4;gx++){
s+=`<line x1="${X(gx)}" y1="${m.t}" x2="${X(gx)}" y2="${H-m.b}" stroke="#eef2f7" stroke-width="1"/>`;
s+=`<text x="${X(gx)}" y="${H-m.b+20}" font-size="11.5" fill="#94a3b8" text-anchor="middle">${gx===0?'0':'+'+gx}</text>`;
}
// axes
s+=`<line x1="${m.l}" y1="${H-m.b}" x2="${W-m.r}" y2="${H-m.b}" stroke="#cbd5e1" stroke-width="1.5"/>`;
s+=`<line x1="${m.l}" y1="${m.t}" x2="${m.l}" y2="${H-m.b}" stroke="#cbd5e1" stroke-width="1.5"/>`;
s+=`<text x="${(m.l+W-m.r)/2}" y="${H-14}" font-size="12.5" fill="#475569" text-anchor="middle">Evolve-set gain (points vs H₀)</text>`;
s+=`<text transform="translate(18 ${(m.t+H-m.b)/2}) rotate(-90)" font-size="12.5" fill="#475569" text-anchor="middle">Out-of-distribution gain (points vs H₀)</text>`;
// y = x reference
s+=`<line x1="${X(0)}" y1="${Y(0)}" x2="${X(xMax)}" y2="${Y(xMax)}" stroke="#94a3b8" stroke-width="1.2" stroke-dasharray="3 4"/>`;
s+=`<text x="${X(3.55)}" y="${Y(3.35)}" font-size="11" fill="#94a3b8" text-anchor="end">y = x (gain fully retained)</text>`;
s+=`<text x="${X(0)+6}" y="${Y(0)-7}" font-size="11" fill="#94a3b8">H₀ line — below here, evolution made things worse</text>`;
// points
pts.forEach(p=>{
s+=`<circle cx="${X(p.x)}" cy="${Y(p.y)}" r="${p.r}" fill="${p.c}" stroke="#fff" stroke-width="2"/>`;
s+=`<text x="${X(p.x)+p.ax}" y="${Y(p.y)+p.ay}" font-size="12.5" fill="#1e293b" text-anchor="${p.anchor}" ${p.bold?'font-weight="700"':''}>${p.n}</text>`;
});
s+=`</svg>`;
document.getElementById('scatter').innerHTML=s;
})();
/* ---------- Figure 3 bars ---------- */
(function(){
const groups=[
{label:'Coding',bars:[
{b:'Terminal-Bench 2.1',k:'evolve',v:6.0,t:'evolve'},
{b:'SWE-bench Verified',k:'ood',v:1.8,t:'OOD'}
]},
{label:'Agentic workspace',bars:[
{b:'Harvey LAB',k:'evolve',v:1.1,t:'evolve'},
{b:'Harvey LAB held-out',k:'held',v:2.3,t:'ID held-out'},
{b:'JobBench',k:'ood',v:4.7,t:'OOD'},
{b:'GDPval',k:'ood',v:3.5,t:'OOD'},
{b:'APEX-Agents',k:'ood',v:3.7,t:'OOD'}
]},
{label:'Engineering design',bars:[
{b:'EngDesign',k:'evolve',v:4.9,t:'evolve'},
{b:'Frontier-Eng',k:'ood',v:4.3,t:'OOD'}
]}
];
const MAX=6.0, PX=200;
let html='';
groups.forEach(g=>{
html+=`<div class="dgroup"><h4>${g.label}</h4><div class="bars">`;
g.bars.forEach(x=>{
const h=Math.max(6,Math.round(x.v/MAX*PX));
html+=`<div class="bar">
<div class="val">+${x.v.toFixed(1)}</div>
<div class="col" style="height:${h}px;background:${C[x.k]}"></div>
<div class="nm"><b>${x.b}</b>${x.t}</div>
</div>`;
});
html+=`</div></div>`;
});
document.getElementById('domains').innerHTML=html;
})();
/* ---------- efficiency ---------- */
(function(){
const rows=[
{lab:'H₀ — no evolution',v:1.56,c:C.base},
{lab:'RRSI (ours)',v:2.42,c:C.ood},
{lab:'Unregularized evolution',v:3.80,c:C.evolve}
];
const MAX=4.0;
let html='';
rows.forEach(r=>{
html+=`<div class="tok">
<div class="lab">${r.lab}</div>
<div class="big">${r.v.toFixed(2)}<span> M policy tokens / trial</span></div>
<div class="meter"><i style="width:${r.v/MAX*100}%;background:${r.c}"></i></div>
</div>`;
});
document.getElementById('tokens').innerHTML=html;
})();
/* ---------- tables ---------- */
(function(){
const t1=[
{m:'H₀ (no evolution)',e:89.4,h:86.9,j:36.0,g:48.8,a:34.2,cls:'h0'},
{m:'Meta-Harness',e:93.0,h:89.2,j:37.1,g:49.1,a:35.7},
{m:'AHE',e:90.7,h:88.7,j:37.2,g:47.2,a:33.1},
{m:'TTHE',e:91.1,h:88.5,j:35.2,g:47.0,a:31.7},
{m:'HarnessX',e:91.8,h:89.1,j:36.3,g:48.5,a:34.3},
{m:'RRSI (ours)',e:90.5,h:89.2,j:40.7,g:52.3,a:37.9,cls:'ours'}
];
document.getElementById('t1').innerHTML=t1.map(r=>{
const avg=(r.j+r.g+r.a)/3;
const cell=(v,i)=>`<td>${i===1&&v<40?'<span class="lose">'+v.toFixed(1)+'</span>':v.toFixed(1)}</td>`;
return `<tr class="${r.cls||''}"><td>${r.m}</td>`+
[r.e,r.h,r.j,r.g,r.a].map(cell).join('')+
`<td>${avg.toFixed(1)}</td></tr>`;
}).join('');
const t2=[
{m:'H₀ (no evolution)',e:89.4,h:86.9,o:39.7,k:1.56,cls:'h0'},
{m:'Unregularized evolution',e:92.8,h:88.9,o:40.3,k:3.80},
{m:'w/o proposal regularizers',e:90.7,h:88.8,o:41.9,k:2.69},
{m:'w/o acceptance regularizers',e:91.5,h:88.7,o:41.0,k:3.59},
{m:'RRSI (ours)',e:90.5,h:89.2,o:43.6,k:2.42,cls:'ours'}
];
document.getElementById('t2').innerHTML=t2.map(r=>
`<tr class="${r.cls||''}"><td>${r.m}</td><td>${r.e.toFixed(1)}</td><td>${r.h.toFixed(1)}</td><td>${r.o.toFixed(1)}</td><td>${r.k.toFixed(2)}</td></tr>`
).join('');
const t3=[
{p:'Claude Opus 4.8',b:'Terminal-Bench 2.1 (evolve)',h0:74.2,r:80.2,d:6.0},
{p:'',b:'SWE-bench Verified (OOD)',h0:82.0,r:83.8,d:1.8},
{p:'Gemini 3.5 Flash',b:'Terminal-Bench 2.1 (evolve)',h0:64.6,r:78.7,d:14.1},
{p:'',b:'SWE-bench Verified (OOD)',h0:76.8,r:79.0,d:2.2}
];
document.getElementById('t3').innerHTML=t3.map(r=>
`<tr><td>${r.p?'<b>'+r.p+'</b>':''}</td><td>${r.b}</td><td>${r.h0.toFixed(1)}</td><td>${r.r.toFixed(1)}</td><td class="win">+${r.d.toFixed(1)}</td></tr>`
).join('');
})();
</script>
</body>
</html>
<script data-huggingchat-badge="script">
(function () {
try {
if (document.querySelector("[data-huggingchat-badge]:not(script)")) return;
var host = document.createElement("div");
host.setAttribute("data-huggingchat-badge", "");
var styles = [["all","initial"],["position","fixed"],["right","12px"],["bottom","12px"],["z-index","2147483647"],["display","block"],["visibility","visible"],["opacity","1"],["pointer-events","auto"],["width","auto"],["height","auto"],["margin","0"],["padding","0"],["max-width","none"],["max-height","none"],["transform","none"],["filter","none"],["clip-path","none"],["color-scheme","light"]];
for (var i = 0; i < styles.length; i++) {
host.style.setProperty(styles[i][0], styles[i][1], "important");
}
host.attachShadow({ mode: "closed" }).innerHTML = "<style>\n:host { all: initial; }\na {\n display: flex;\n align-items: center;\n gap: 5px;\n box-sizing: border-box;\n height: 26px;\n padding: 0 9px 0 7px;\n border-radius: 999px;\n border: 1px solid rgba(0, 0, 0, 0.07);\n background: rgba(255, 255, 255, 0.92);\n -webkit-backdrop-filter: saturate(180%) blur(8px);\n backdrop-filter: saturate(180%) blur(8px);\n box-shadow: 0 1px 2px rgba(0, 0, 0, 0.08), 0 6px 16px rgba(0, 0, 0, 0.06);\n color: #111827;\n font: 500 11px/1 ui-sans-serif, system-ui, -apple-system, \"Segoe UI\", sans-serif;\n text-decoration: none;\n white-space: nowrap;\n opacity: 0.72;\n transition: opacity 0.15s ease, transform 0.15s ease;\n}\na:hover, a:focus-visible { opacity: 1; transform: translateY(-1px); }\nsvg { width: 14px; height: 14px; flex: none; }\n@media print { a { display: none; } }\n<\/style><a href=\"https://huggingface.co/chat\" target=\"_blank\" rel=\"noopener noreferrer nofollow\"><svg viewBox=\"0 0 32 32\" fill=\"none\" aria-hidden=\"true\"><path d=\"M16.0006 25.9992C13.8266 25.999 11.7118 25.2901 9.97686 23.9799C8.2419 22.6698 6.98127 20.8298 6.38599 18.7388C5.79071 16.6478 5.89323 14.4198 6.678 12.3923C7.46278 10.3648 8.88705 8.64837 10.735 7.50308C12.5829 6.35779 14.7538 5.84606 16.9187 6.04544C19.0837 6.24481 21.1246 7.14442 22.7323 8.60795C24.34 10.0715 25.4268 12.0192 25.8281 14.1559C26.2293 16.2926 25.9232 18.5019 24.9561 20.449C24.7703 20.8042 24.7223 21.2155 24.8211 21.604L25.4211 23.8316C25.4803 24.0518 25.4805 24.2837 25.4216 24.5039C25.3627 24.7242 25.2468 24.925 25.0856 25.0862C24.9244 25.2474 24.7235 25.3633 24.5033 25.4222C24.283 25.4811 24.0512 25.4809 23.831 25.4217L21.6034 24.8217C21.2172 24.7248 20.809 24.7729 20.4558 24.9567C19.0683 25.6467 17.5457 26.0068 16.0006 26.0068V25.9992Z\" fill=\"currentColor\"/><path d=\"M9.62598 16.0013C9.62598 15.3799 10.1294 14.8765 10.7508 14.8765C11.3721 14.8765 11.8756 15.3799 11.8756 16.0013C11.8756 17.0953 12.3102 18.1448 13.0838 18.9184C13.8574 19.692 14.9069 20.1266 16.001 20.1267C17.095 20.1267 18.1445 19.692 18.9181 18.9184C19.6918 18.1448 20.1264 17.0953 20.1264 16.0013C20.1264 15.3799 20.6299 14.8765 21.2512 14.8765C21.8725 14.8765 22.3759 15.3799 22.3759 16.0013C22.3759 17.6921 21.7046 19.3137 20.509 20.5093C19.3134 21.7049 17.6918 22.3762 16.001 22.3762C14.3102 22.3762 12.6885 21.7049 11.4929 20.5093C10.2974 19.3137 9.62598 17.6921 9.62598 16.0013Z\" fill=\"#fff\"/><\/svg><span>Made with HuggingChat<\/span><\/a>";
function attach() {
var parent = document.body || document.documentElement;
if (parent && host.parentNode !== parent) parent.appendChild(host);
}
attach();
// Artifacts that rewrite document.body wholesale would drop the badge.
if (document.body && typeof MutationObserver === "function") {
new MutationObserver(attach).observe(document.body, { childList: true });
} else {
document.addEventListener("DOMContentLoaded", attach);
}
} catch (e) {}
})();
</script>
|