Update index.html
Browse files- index.html +27 -59
index.html
CHANGED
|
@@ -110,7 +110,16 @@
|
|
| 110 |
color: var(--text-secondary);
|
| 111 |
margin-top: 4px;
|
| 112 |
}
|
| 113 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 114 |
.like-btn {
|
| 115 |
margin-left: auto;
|
| 116 |
display: flex;
|
|
@@ -126,42 +135,6 @@
|
|
| 126 |
flex-shrink: 0;
|
| 127 |
}
|
| 128 |
|
| 129 |
-
/* Tags */
|
| 130 |
-
.tags {
|
| 131 |
-
display: flex;
|
| 132 |
-
flex-wrap: wrap;
|
| 133 |
-
gap: 8px;
|
| 134 |
-
margin-bottom: 24px;
|
| 135 |
-
}
|
| 136 |
-
|
| 137 |
-
.tag {
|
| 138 |
-
display: inline-flex;
|
| 139 |
-
align-items: center;
|
| 140 |
-
gap: 6px;
|
| 141 |
-
padding: 5px 12px;
|
| 142 |
-
background: var(--bg-secondary);
|
| 143 |
-
border: 1px solid var(--border);
|
| 144 |
-
border-radius: 999px;
|
| 145 |
-
font-size: 12px;
|
| 146 |
-
color: var(--text-secondary);
|
| 147 |
-
}
|
| 148 |
-
|
| 149 |
-
.tag.benchmark {
|
| 150 |
-
background: rgba(251, 191, 36, 0.1);
|
| 151 |
-
border-color: rgba(251, 191, 36, 0.3);
|
| 152 |
-
color: var(--accent-gold);
|
| 153 |
-
}
|
| 154 |
-
|
| 155 |
-
.tag .dot {
|
| 156 |
-
width: 6px;
|
| 157 |
-
height: 6px;
|
| 158 |
-
border-radius: 50%;
|
| 159 |
-
background: var(--accent-blue);
|
| 160 |
-
}
|
| 161 |
-
|
| 162 |
-
.tag .dot.green { background: var(--accent-green); }
|
| 163 |
-
.tag .dot.orange { background: #f97316; }
|
| 164 |
-
|
| 165 |
/* Tabs */
|
| 166 |
.tab-bar {
|
| 167 |
display: flex;
|
|
@@ -299,10 +272,6 @@
|
|
| 299 |
gap: 4px;
|
| 300 |
}
|
| 301 |
|
| 302 |
-
.lb-rank .trophy {
|
| 303 |
-
font-size: 16px;
|
| 304 |
-
}
|
| 305 |
-
|
| 306 |
.lb-model {
|
| 307 |
display: flex;
|
| 308 |
align-items: center;
|
|
@@ -536,6 +505,16 @@
|
|
| 536 |
<div class="title-block">
|
| 537 |
<h1>Infernet Econometrics Benchmark</h1>
|
| 538 |
<p class="subtitle">CamoAiLab · HKU CAMO</p>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 539 |
</div>
|
| 540 |
<div class="like-btn">
|
| 541 |
<svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
|
|
@@ -544,27 +523,14 @@
|
|
| 544 |
Leaderboard
|
| 545 |
</div>
|
| 546 |
</div>
|
| 547 |
-
|
| 548 |
-
<div class="tags">
|
| 549 |
-
<span class="tag benchmark">🏆 Benchmark</span>
|
| 550 |
-
<span class="tag"><span class="dot"></span> Tasks: Econometrics</span>
|
| 551 |
-
<span class="tag"><span class="dot green"></span> Modalities: Code</span>
|
| 552 |
-
<span class="tag"><span class="dot orange"></span> Formats: CSV</span>
|
| 553 |
-
<span class="tag">
|
| 554 |
-
<svg width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"><rect x="3" y="11" width="18" height="11" rx="2"/><path d="M7 11V7a5 5 0 0 1 10 0v4"/></svg>
|
| 555 |
-
Open
|
| 556 |
-
</span>
|
| 557 |
-
</div>
|
| 558 |
</header>
|
| 559 |
|
| 560 |
-
<!-- Tab顺序修改:0 Partial Replication(默认active),1 Compilation Success,2 Coefficient Direction -->
|
| 561 |
<div class="tab-bar">
|
| 562 |
<button class="tab active" data-idx="0">Partial Replication</button>
|
| 563 |
<button class="tab" data-idx="1">Compilation Success</button>
|
| 564 |
<button class="tab" data-idx="2">Coefficient Direction</button>
|
| 565 |
</div>
|
| 566 |
|
| 567 |
-
<!-- panel顺序同步调换,第一个show是Partial Replication -->
|
| 568 |
<div class="panel show" id="panel-rep">
|
| 569 |
<div class="leaderboard-card">
|
| 570 |
<div class="leaderboard-header">
|
|
@@ -705,11 +671,15 @@ function buildRows(data, sortKey, containerId) {
|
|
| 705 |
const isTop = rank === 1;
|
| 706 |
const avatarColor = getOrgColor(modelId);
|
| 707 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 708 |
html += `
|
| 709 |
<div class="lb-row${isTop ? " top-1" : ""}">
|
| 710 |
-
<div class="lb-rank">
|
| 711 |
-
${isTop ? '<span class="trophy">🏆</span>' : rank}
|
| 712 |
-
</div>
|
| 713 |
<div class="lb-model">
|
| 714 |
<div class="model-avatar" style="background:${avatarColor}22;border-color:${avatarColor}44;color:${avatarColor}">
|
| 715 |
${getAvatarLetter(modelId)}
|
|
@@ -717,7 +687,6 @@ function buildRows(data, sortKey, containerId) {
|
|
| 717 |
<div class="model-info">
|
| 718 |
<div class="model-name" title="${modelId}">${modelId}</div>
|
| 719 |
<div class="model-meta">
|
| 720 |
-
<!-- 全部硬编码 verified;后续csv增加verified字段后,这里改成读取row["verified"]做判断 -->
|
| 721 |
<span class="status-badge verified">✓ verified</span>
|
| 722 |
</div>
|
| 723 |
</div>
|
|
@@ -762,7 +731,6 @@ async function loadData() {
|
|
| 762 |
throw new Error("CSV parsed zero rows");
|
| 763 |
}
|
| 764 |
|
| 765 |
-
// 对应panel‑idx 0:rep,1:compile,2:dir
|
| 766 |
buildRows(data, "Partial Replication (%)", "rows-rep");
|
| 767 |
buildRows(data, "Compilation Success (%)", "rows-compile");
|
| 768 |
buildRows(data, "Correct Coefficient Direction (%)", "rows-dir");
|
|
|
|
| 110 |
color: var(--text-secondary);
|
| 111 |
margin-top: 4px;
|
| 112 |
}
|
| 113 |
+
.intro-desc {
|
| 114 |
+
margin-top:14px;
|
| 115 |
+
font-size:14px;
|
| 116 |
+
color:var(--text-secondary);
|
| 117 |
+
line-height:1.6;
|
| 118 |
+
}
|
| 119 |
+
.header-links {
|
| 120 |
+
margin-top:12px;
|
| 121 |
+
font-size:14px;
|
| 122 |
+
}
|
| 123 |
.like-btn {
|
| 124 |
margin-left: auto;
|
| 125 |
display: flex;
|
|
|
|
| 135 |
flex-shrink: 0;
|
| 136 |
}
|
| 137 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 138 |
/* Tabs */
|
| 139 |
.tab-bar {
|
| 140 |
display: flex;
|
|
|
|
| 272 |
gap: 4px;
|
| 273 |
}
|
| 274 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 275 |
.lb-model {
|
| 276 |
display: flex;
|
| 277 |
align-items: center;
|
|
|
|
| 505 |
<div class="title-block">
|
| 506 |
<h1>Infernet Econometrics Benchmark</h1>
|
| 507 |
<p class="subtitle">CamoAiLab · HKU CAMO</p>
|
| 508 |
+
<p class="intro-desc">
|
| 509 |
+
An evaluation benchmark for large‑language‑model agents, measuring the capability to reproduce real‑world econometric empirical studies: generating executable code, replicating regression outputs, and recovering correct treatment‑effect coefficient signs.
|
| 510 |
+
</p>
|
| 511 |
+
<div class="header-links">
|
| 512 |
+
🔗 <a href="https://huggingface.co/datasets/CamoAiLab/Infernet" target="_blank">Dataset</a>
|
| 513 |
+
|
|
| 514 |
+
📄 <a href="#" target="_blank">Paper</a>
|
| 515 |
+
|
|
| 516 |
+
🐙 <a href="#" target="_blank">GitHub</a>
|
| 517 |
+
</div>
|
| 518 |
</div>
|
| 519 |
<div class="like-btn">
|
| 520 |
<svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
|
|
|
|
| 523 |
Leaderboard
|
| 524 |
</div>
|
| 525 |
</div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 526 |
</header>
|
| 527 |
|
|
|
|
| 528 |
<div class="tab-bar">
|
| 529 |
<button class="tab active" data-idx="0">Partial Replication</button>
|
| 530 |
<button class="tab" data-idx="1">Compilation Success</button>
|
| 531 |
<button class="tab" data-idx="2">Coefficient Direction</button>
|
| 532 |
</div>
|
| 533 |
|
|
|
|
| 534 |
<div class="panel show" id="panel-rep">
|
| 535 |
<div class="leaderboard-card">
|
| 536 |
<div class="leaderboard-header">
|
|
|
|
| 671 |
const isTop = rank === 1;
|
| 672 |
const avatarColor = getOrgColor(modelId);
|
| 673 |
|
| 674 |
+
// 金银铜奖牌
|
| 675 |
+
let rankDisplay = rank;
|
| 676 |
+
if(rank ===1) rankDisplay = "🥇";
|
| 677 |
+
else if(rank ===2) rankDisplay = "🥈";
|
| 678 |
+
else if(rank ===3) rankDisplay = "🥉";
|
| 679 |
+
|
| 680 |
html += `
|
| 681 |
<div class="lb-row${isTop ? " top-1" : ""}">
|
| 682 |
+
<div class="lb-rank">${rankDisplay}</div>
|
|
|
|
|
|
|
| 683 |
<div class="lb-model">
|
| 684 |
<div class="model-avatar" style="background:${avatarColor}22;border-color:${avatarColor}44;color:${avatarColor}">
|
| 685 |
${getAvatarLetter(modelId)}
|
|
|
|
| 687 |
<div class="model-info">
|
| 688 |
<div class="model-name" title="${modelId}">${modelId}</div>
|
| 689 |
<div class="model-meta">
|
|
|
|
| 690 |
<span class="status-badge verified">✓ verified</span>
|
| 691 |
</div>
|
| 692 |
</div>
|
|
|
|
| 731 |
throw new Error("CSV parsed zero rows");
|
| 732 |
}
|
| 733 |
|
|
|
|
| 734 |
buildRows(data, "Partial Replication (%)", "rows-rep");
|
| 735 |
buildRows(data, "Compilation Success (%)", "rows-compile");
|
| 736 |
buildRows(data, "Correct Coefficient Direction (%)", "rows-dir");
|