Diplomovka/perun/ministral_3b_aya/sk_bench/index.html
2026-11-12 18:27:37 +01:00

409 lines
32 KiB
HTML
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>SK-Bench Leaderboard</title>
<link rel="preconnect" href="https://fonts.googleapis.com">
<link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&family=JetBrains+Mono:wght@400;500&display=swap" rel="stylesheet">
<style>
:root {
--bg: #0f1117;
--surface: #1a1d27;
--surface2: #22263a;
--border: #2e3350;
--accent: #5b7fff;
--accent2: #a78bfa;
--avg-accent: #f59e0b;
--avg-accent-dim: rgba(245,158,11,.15);
--text: #e2e8f0;
--text-muted: #8892b0;
--header-bg: #1e2235;
--sticky-bg: #1a1d27;
--row-hover: #1e2337;
--font-mono: "JetBrains Mono", "Fira Code", "Consolas", monospace;
--font-sans: "Inter", system-ui, -apple-system, sans-serif;
}
*, *::before, *::after { box-sizing: border-box; margin: 0; padding: 0; }
body {
background: var(--bg);
color: var(--text);
font-family: var(--font-sans);
min-height: 100vh;
}
/* ── Topbar ── */
.topbar {
position: sticky;
top: 0;
z-index: 100;
background: var(--surface);
border-bottom: 1px solid var(--border);
padding: 0 24px;
display: flex;
align-items: center;
gap: 24px;
height: 56px;
backdrop-filter: blur(8px);
}
.topbar h1 {
font-size: 1.1rem;
font-weight: 700;
letter-spacing: 0.03em;
background: linear-gradient(135deg, var(--accent), var(--accent2));
-webkit-background-clip: text;
-webkit-text-fill-color: transparent;
white-space: nowrap;
}
.topbar .meta {
font-size: 0.75rem;
color: var(--text-muted);
white-space: nowrap;
}
.topbar .spacer { flex: 1; }
#filter-input {
background: var(--surface2);
border: 1px solid var(--border);
border-radius: 6px;
color: var(--text);
font-size: 0.8rem;
padding: 5px 10px;
width: 200px;
outline: none;
transition: border-color .2s;
}
#filter-input:focus { border-color: var(--accent); }
#filter-input::placeholder { color: var(--text-muted); }
/* ── Wrap ── */
.table-wrap {
overflow: auto;
max-height: calc(100vh - 56px);
padding: 16px;
}
/* ── Table ── */
table {
border-collapse: separate;
border-spacing: 0;
font-size: 0.78rem;
white-space: nowrap;
min-width: max-content;
}
/* Sticky model column — use .th-model instead of th:first-child because
the Model cell has rowspan=3 and is absent from thead rows 2 and 3,
so th:first-child there would incorrectly pin the first AVG header. */
.th-model, td:first-child {
position: sticky;
left: 0;
z-index: 3;
background: var(--sticky-bg) !important;
min-width: 180px;
max-width: 280px;
overflow: hidden;
text-overflow: ellipsis;
}
.th-model { z-index: 5; }
tbody tr:hover td:first-child { background: var(--row-hover) !important; }
tbody tr:nth-child(even) td:first-child { background: #1c1f2e !important; }
tbody tr:nth-child(even):hover td:first-child { background: var(--row-hover) !important; }
thead th {
background: var(--header-bg);
color: var(--text-muted);
font-weight: 600;
text-align: center;
padding: 6px 8px;
border-bottom: 1px solid var(--border);
border-right: 1px solid var(--border);
position: sticky;
}
/* Sticky header rows */
thead tr:nth-child(1) th { top: 0; z-index: 4; }
thead tr:nth-child(2) th { top: 30px; z-index: 4; }
thead tr:nth-child(3) th { top: 61px; z-index: 4; }
thead tr:nth-child(1) th:first-child { z-index: 6; }
/* Task-level header */
.th-task {
font-size: 0.82rem;
font-weight: 700;
color: var(--accent);
letter-spacing: 0.04em;
background: var(--surface2) !important;
border-top: 2px solid var(--accent) !important;
padding: 8px 10px !important;
}
/* AVG column group gets a gold accent */
.th-task.th-avg {
color: var(--avg-accent) !important;
border-top-color: var(--avg-accent) !important;
background: rgba(245,158,11,.08) !important;
}
.th-ls.th-avg-ls {
color: var(--avg-accent) !important;
background: rgba(245,158,11,.06) !important;
}
.th-metric.th-avg-metric {
color: var(--avg-accent) !important;
background: rgba(245,158,11,.06) !important;
}
.th-metric.th-avg-metric:hover {
background: rgba(245,158,11,.15) !important;
}
.th-metric.th-avg-metric.sort-asc,
.th-metric.th-avg-metric.sort-desc {
color: var(--avg-accent) !important;
background: rgba(245,158,11,.2) !important;
}
td.avg-cell {
background: rgba(245,158,11,.04) !important;
font-weight: 500;
}
tbody tr:hover td.avg-cell { background: rgba(245,158,11,.10) !important; }
/* Language AVG column group styling (purple accent) */
.th-task.th-avg-lang {
color: var(--accent2) !important;
border-top-color: var(--accent2) !important;
background: rgba(167,139,250,.08) !important;
}
.th-ls.th-avg-lang-ls {
color: var(--accent2) !important;
background: rgba(167,139,250,.06) !important;
}
.th-metric.th-avg-lang-metric {
color: var(--accent2) !important;
background: rgba(167,139,250,.06) !important;
}
.th-metric.th-avg-lang-metric:hover {
background: rgba(167,139,250,.15) !important;
}
.th-metric.th-avg-lang-metric.sort-asc,
.th-metric.th-avg-lang-metric.sort-desc {
color: var(--accent2) !important;
background: rgba(167,139,250,.2) !important;
}
td.avg-lang-cell {
background: rgba(167,139,250,.04) !important;
font-weight: 500;
}
tbody tr:hover td.avg-lang-cell { background: rgba(167,139,250,.10) !important; }
/* Lang|Shot header */
.th-ls {
font-size: 0.72rem;
font-weight: 600;
color: var(--accent2);
background: var(--header-bg) !important;
border-bottom: 1px solid var(--border) !important;
}
/* Metric header */
.th-metric {
font-size: 0.68rem;
font-weight: 500;
color: var(--text-muted);
font-family: var(--font-mono);
border-right: 1px solid var(--border) !important;
cursor: pointer;
user-select: none;
transition: color .15s, background .15s;
}
.th-metric:hover { color: var(--text); background: var(--surface2) !important; }
.th-metric.sort-asc,
.th-metric.sort-desc { color: var(--accent) !important; background: rgba(91,127,255,.12) !important; }
.th-metric .sort-arrow { margin-left: 3px; opacity: 0.5; font-size: 0.6rem; }
.th-metric.sort-asc .sort-arrow,
.th-metric.sort-desc .sort-arrow { opacity: 1; }
/* Task group separators */
.task-start th { border-left: 2px solid var(--border) !important; }
/* Model column header */
.th-model {
color: var(--text) !important;
font-size: 0.8rem;
text-align: left !important;
padding-left: 12px !important;
background: var(--surface2) !important;
z-index: 6 !important;
}
/* Body */
tbody tr { transition: background .15s; }
tbody tr:hover td { background: var(--row-hover) !important; }
tbody tr:nth-child(even) td { background: rgba(255,255,255,0.02); }
td {
padding: 5px 7px;
text-align: right;
border-right: 1px solid var(--border);
border-bottom: 1px solid rgba(46,51,80,.4);
font-family: var(--font-mono);
font-size: 0.75rem;
color: var(--text);
}
td:first-child {
text-align: left;
font-family: var(--font-sans);
font-size: 0.8rem;
padding-left: 12px;
border-right: 2px solid var(--border);
}
td.missing { color: var(--text-muted); font-size: 0.65rem; }
.task-start-cell { border-left: 2px solid var(--border) !important; }
/* Heat-map cells */
td.val-cell { position: relative; }
td.val-cell::before {
content: "";
position: absolute;
inset: 0;
opacity: var(--heat-opacity, 0);
background: var(--heat-color, transparent);
pointer-events: none;
z-index: 0;
}
td.val-cell span { position: relative; z-index: 1; }
/* Legend */
.legend {
display: flex;
gap: 16px;
align-items: center;
font-size: 0.72rem;
color: var(--text-muted);
padding: 0 16px 12px;
}
.legend-swatch {
display: inline-block;
width: 40px; height: 10px;
border-radius: 3px;
}
.legend-label { margin-left: 4px; }
</style>
</head>
<body>
<div class="topbar">
<h1>⚡ SK-Bench Leaderboard</h1>
<span class="meta">2 models &nbsp;·&nbsp; 62 metric columns &nbsp;·&nbsp; generated 2026-11-12 15:24 UTC</span>
<div class="spacer"></div>
<input id="filter-input" type="search" placeholder="Filter models…">
</div>
<div class="legend">
<span>Heat-map:</span>
<span class="legend-swatch" style="background:linear-gradient(to right,rgb(248,113,113),rgb(248,211,113),rgb(52,211,153))"></span>
<span class="legend-label">low → high within each metric column</span>
</div>
<div class="table-wrap">
<table>
<thead>
<tr><th rowspan="3" class="th-model">Model (2)</th><th colspan="10" class="th-task th-avg">⌀ AVG</th><th colspan="10" class="th-task task-start th-avg-lang">lang AVG</th><th colspan="1" class="th-task task-start">arc</th><th colspan="1" class="th-task task-start">belebele</th><th colspan="1" class="th-task task-start">chiby</th><th colspan="1" class="th-task task-start">driving_tests</th><th colspan="1" class="th-task task-start">financial_exams</th><th colspan="4" class="th-task task-start">ifeval_sk</th><th colspan="1" class="th-task task-start">globalmmlu_lite</th><th colspan="1" class="th-task task-start">mmlu</th><th colspan="1" class="th-task task-start">movie_reviews</th><th colspan="2" class="th-task task-start">multi_wiki_qa</th><th colspan="1" class="th-task task-start">senti_sk:sentiment_3_class</th><th colspan="1" class="th-task task-start">senti_sk:sentiment_binary</th><th colspan="1" class="th-task task-start">sklep:hate-speech</th><th colspan="1" class="th-task task-start">sklep:nli</th><th colspan="2" class="th-task task-start">sklep:question-answering</th><th colspan="1" class="th-task task-start">sklep:rte</th><th colspan="1" class="th-task task-start">sklep:sentiment-analysis</th><th colspan="1" class="th-task task-start">skj1</th><th colspan="1" class="th-task task-start">slovakcopa</th><th colspan="1" class="th-task task-start">slovak_trivia</th><th colspan="1" class="th-task task-start">toxic_sk</th><th colspan="1" class="th-task task-start">truthfulqa</th><th colspan="1" class="th-task task-start">analogies_sk</th><th colspan="1" class="th-task task-start">slohry</th><th colspan="1" class="th-task task-start">scio</th><th colspan="1" class="th-task task-start">e_tests_mqa</th><th colspan="5" class="th-task task-start">e_tests_qa</th><th colspan="1" class="th-task task-start">slovak_tests_mqa</th><th colspan="5" class="th-task task-start">slovak_tests_qa</th></tr>
<tr><th colspan="10" class="th-ls th-avg-ls">ALL|0</th><th colspan="10" class="th-ls task-start th-avg-lang-ls">SK_AVERAGE|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="4" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="2" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="2" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="5" class="th-ls task-start">SK|0</th><th colspan="1" class="th-ls task-start">SK|0</th><th colspan="5" class="th-ls task-start">SK|0</th></tr>
<tr><th class="th-metric th-avg-metric">acc</th><th class="th-metric th-avg-metric">inst​_level​_loose​_acc</th><th class="th-metric th-avg-metric">inst​_level​_strict​_acc</th><th class="th-metric th-avg-metric">prompt​_level​_loose​_acc</th><th class="th-metric th-avg-metric">prompt​_level​_strict​_acc</th><th class="th-metric th-avg-metric">qem</th><th class="th-metric th-avg-metric">qf1</th><th class="th-metric th-avg-metric">rouge1</th><th class="th-metric th-avg-metric">rouge2</th><th class="th-metric th-avg-metric">rougeL</th><th class="th-metric task-start">acc</th><th class="th-metric">inst​_level​_loose​_acc</th><th class="th-metric">inst​_level​_strict​_acc</th><th class="th-metric">prompt​_level​_loose​_acc</th><th class="th-metric">prompt​_level​_strict​_acc</th><th class="th-metric">qem</th><th class="th-metric">qf1</th><th class="th-metric">rouge1</th><th class="th-metric">rouge2</th><th class="th-metric">rougeL</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">inst​_level​_loose​_acc</th><th class="th-metric">inst​_level​_strict​_acc</th><th class="th-metric">prompt​_level​_loose​_acc</th><th class="th-metric">prompt​_level​_strict​_acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">qem</th><th class="th-metric">qf1</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">qem</th><th class="th-metric">qf1</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">qem</th><th class="th-metric">qf1</th><th class="th-metric">rouge1</th><th class="th-metric">rouge2</th><th class="th-metric">rougeL</th><th class="th-metric task-start">acc</th><th class="th-metric task-start">qem</th><th class="th-metric">qf1</th><th class="th-metric">rouge1</th><th class="th-metric">rouge2</th><th class="th-metric">rougeL</th></tr>
</thead>
<tbody>
<tr><td title="models/ministral-base">models/ministral-base</td><td class="val-cell avg-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0003</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.1731</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.1731</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0782</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0782</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0039</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0045</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0007</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0007</span></td><td class="val-cell task-start-cell avg-lang-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0002</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.1731</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.1731</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0782</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0782</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0039</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0045</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0007</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0007</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2449</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.2537</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.1374</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0887</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.1731</span></td><td class="val-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.1731</span></td><td class="val-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0782</span></td><td class="val-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0782</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.2108</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.2225</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0005</span></td><td class="val-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0053</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0003</span></td><td class="val-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0070</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.5967</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0533</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.1992</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0647</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0023</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.2385</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.1501</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0023</span></td><td class="val-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0026</span></td><td class="val-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0003</span></td><td class="val-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0003</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.1739</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0126</span></td><td class="val-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0030</span></td><td class="val-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0012</span></td><td class="val-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0011</span></td></tr>
<tr><td title="models/ministral-last-109064">models/ministral-last-109064</td><td class="val-cell avg-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2603</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2131</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2131</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1024</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1024</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0382</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0708</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0149</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0032</span></td><td class="val-cell avg-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0149</span></td><td class="val-cell task-start-cell avg-lang-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2405</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2131</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2131</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1024</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1024</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0382</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0708</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0149</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0032</span></td><td class="val-cell avg-lang-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0149</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.2447</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2667</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1424</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2710</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1862</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2131</span></td><td class="val-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2131</span></td><td class="val-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1024</span></td><td class="val-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1024</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2758</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2415</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1489</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0788</span></td><td class="val-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1235</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.3862</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.4936</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2143</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0240</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0520</span></td><td class="val-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1428</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.3151</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1139</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0000</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2700</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2786</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1998</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.3188</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0006</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0170</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2500</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.1815</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0120</span></td><td class="val-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0079</span></td><td class="val-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0148</span></td><td class="val-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0032</span></td><td class="val-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0148</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.2601</span></td><td class="val-cell task-start-cell" style="--heat-color:rgb(248,113,113);--heat-opacity:0.120;"><span>0.0098</span></td><td class="val-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0089</span></td><td class="val-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0149</span></td><td class="val-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0031</span></td><td class="val-cell" style="--heat-color:rgb(52,211,153);--heat-opacity:0.470;"><span>0.0149</span></td></tr>
</tbody>
</table>
</div>
<script>
(function () {
// ── Filter ────────────────────────────────────────────────────────────────
const input = document.getElementById('filter-input');
// Helper function to apply filter
function applyFilter(q) {
document.querySelectorAll('tbody tr').forEach(tr => {
const name = tr.querySelector('td')?.textContent.toLowerCase() ?? '';
tr.style.display = name.includes(q) ? '' : 'none';
});
}
// Parse URL parameters on page load
const urlParams = new URLSearchParams(window.location.search);
const filterParam = urlParams.get('filter');
if (filterParam) {
input.value = filterParam;
applyFilter(filterParam.toLowerCase());
}
// Listen to input changes and update URL
input.addEventListener('input', function () {
const q = this.value.toLowerCase();
applyFilter(q);
// Update URL without page reload
const newParams = new URLSearchParams(window.location.search);
if (q) {
newParams.set('filter', this.value);
} else {
newParams.delete('filter');
}
const newUrl = newParams.toString() ? `?${newParams.toString()}` : window.location.pathname;
window.history.replaceState({}, '', newUrl);
});
// ── Sort ──────────────────────────────────────────────────────────────────
// Metric <th> cells live in header row 3. Because the Model <th> spans 3
// rows it is absent from rows 2–3, so in row 3 cellIndex 0 maps to tbody
// column 1. Therefore: td_column_index = th.cellIndex + 1.
let activeTh = null;
let sortDesc = true;
function getSortValue(tr, colIdx) {
const td = tr.cells[colIdx];
if (!td) return null;
const span = td.querySelector('span');
const text = span ? span.textContent : td.textContent;
const v = parseFloat(text);
return isNaN(v) ? null : v;
}
function applySort(th) {
const colIdx = th.cellIndex + 1;
if (th === activeTh) {
sortDesc = !sortDesc;
} else {
if (activeTh) {
activeTh.classList.remove('sort-asc', 'sort-desc');
const prev = activeTh.querySelector('.sort-arrow');
if (prev) prev.textContent = ' ⇅';
}
activeTh = th;
sortDesc = true;
}
th.classList.toggle('sort-desc', sortDesc);
th.classList.toggle('sort-asc', !sortDesc);
const arrow = th.querySelector('.sort-arrow');
if (arrow) arrow.textContent = sortDesc ? ' ▼' : ' ▲';
const tbody = document.querySelector('tbody');
const rows = Array.from(tbody.querySelectorAll('tr'));
rows.sort((a, b) => {
const va = getSortValue(a, colIdx);
const vb = getSortValue(b, colIdx);
if (va === null && vb === null) return 0;
if (va === null) return 1; // nulls always last
if (vb === null) return -1;
return sortDesc ? vb - va : va - vb;
});
rows.forEach(r => tbody.appendChild(r));
}
document.querySelectorAll('.th-metric').forEach(th => {
th.insertAdjacentHTML('beforeend', '<span class="sort-arrow"> ⇅</span>');
th.addEventListener('click', () => applySort(th));
});
})();
</script>
</body>
</html>