| <!doctype html> |
| <html lang="en"> |
| <head> |
| <meta charset="utf-8"> |
| <meta name="viewport" content="width=device-width, initial-scale=1"> |
| <meta name="description" content="Unranked POSA Bench fixed-context preview for small on-device language models."> |
| <title>POSA Bench — Fixed-Context Preview</title> |
| <style> |
| :root { |
| color-scheme: light dark; |
| --bg: #f6f7f4; |
| --panel: #ffffff; |
| --ink: #17201b; |
| --muted: #5d6861; |
| --line: #dce2dc; |
| --green: #176b43; |
| --green-soft: #e7f4ec; |
| --amber: #9a5a00; |
| --amber-soft: #fff4d8; |
| --shadow: 0 12px 35px rgba(20, 38, 27, .08); |
| } |
| * { box-sizing: border-box; } |
| body { |
| margin: 0; |
| background: radial-gradient(circle at 90% 0, #dcefe4 0, transparent 32rem), var(--bg); |
| color: var(--ink); |
| font: 15px/1.55 ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif; |
| } |
| main { width: min(1180px, calc(100% - 32px)); margin: 0 auto; padding: 54px 0 72px; } |
| .eyebrow { color: var(--green); font-size: 12px; font-weight: 800; letter-spacing: .14em; text-transform: uppercase; } |
| h1 { max-width: 820px; margin: 8px 0 12px; font-size: clamp(34px, 5vw, 62px); line-height: 1.03; letter-spacing: -.045em; } |
| .lead { max-width: 780px; margin: 0 0 24px; color: var(--muted); font-size: 18px; } |
| .warning { padding: 16px 18px; border: 1px solid #edcf91; border-radius: 14px; background: var(--amber-soft); color: #66430e; } |
| .summary { display: grid; grid-template-columns: repeat(4, 1fr); gap: 12px; margin: 22px 0; } |
| .card { min-height: 108px; padding: 18px; border: 1px solid var(--line); border-radius: 16px; background: var(--panel); box-shadow: var(--shadow); } |
| .card span { color: var(--muted); font-size: 12px; font-weight: 700; letter-spacing: .06em; text-transform: uppercase; } |
| .card strong { display: block; margin-top: 8px; font-size: 22px; line-height: 1.15; } |
| .table-wrap { overflow-x: auto; margin-top: 22px; border: 1px solid var(--line); border-radius: 18px; background: var(--panel); box-shadow: var(--shadow); } |
| table { width: 100%; min-width: 1040px; border-collapse: collapse; } |
| th, td { padding: 14px 13px; border-bottom: 1px solid var(--line); text-align: right; white-space: nowrap; } |
| th:first-child, td:first-child { position: sticky; left: 0; text-align: left; background: var(--panel); } |
| th { color: var(--muted); font-size: 11px; letter-spacing: .04em; text-transform: uppercase; } |
| td:first-child { max-width: 330px; overflow: hidden; text-overflow: ellipsis; font-weight: 700; } |
| tbody tr:last-child td { border-bottom: 0; } |
| tbody tr:hover td { background: #f4faf6; } |
| .best { color: var(--green); font-weight: 800; } |
| .status { display: inline-block; padding: 3px 8px; border-radius: 999px; background: var(--amber-soft); color: var(--amber); font-size: 11px; font-weight: 800; text-transform: uppercase; } |
| .details { display: grid; grid-template-columns: 1fr 1fr; gap: 28px; margin-top: 36px; } |
| h2 { margin: 0 0 10px; font-size: 20px; } |
| p, ul { color: var(--muted); } |
| ul { padding-left: 20px; } |
| a { color: var(--green); font-weight: 700; } |
| footer { margin-top: 34px; padding-top: 20px; border-top: 1px solid var(--line); color: var(--muted); } |
| #error { display: none; margin-top: 20px; color: #a11; } |
| @media (max-width: 800px) { |
| main { padding-top: 34px; } |
| .summary, .details { grid-template-columns: 1fr 1fr; } |
| } |
| @media (max-width: 520px) { .summary, .details { grid-template-columns: 1fr; } } |
| @media (prefers-color-scheme: dark) { |
| :root { --bg:#0f1411; --panel:#171d19; --ink:#edf4ef; --muted:#a9b6ae; --line:#303a33; --green:#68d49a; --green-soft:#173b28; --amber:#ffc96b; --amber-soft:#3b2d12; --shadow:none; } |
| body { background: radial-gradient(circle at 90% 0, #173b28 0, transparent 32rem), var(--bg); } |
| .warning { border-color:#654d1e; color:#ffe0a1; } |
| tbody tr:hover td { background:#1d2921; } |
| } |
| </style> |
| </head> |
| <body> |
| <main> |
| <div class="eyebrow">Portable Offline Survival Assistant</div> |
| <h1>POSA Bench fixed-context preview</h1> |
| <p class="lead">An open, source-grounded comparison of small language models running on a Google Pixel 7 under one shared llama.cpp configuration.</p> |
| <div class="warning"><strong>Unranked preview.</strong> These are directional one-seed results reviewed by one AI evaluator. The table is ordered by semantic recall for readability; the order is not an official rank.</div> |
|
|
| <section class="summary" aria-label="Preview highlights"> |
| <div class="card"><span>Strongest content</span><strong id="content-leader">Loading…</strong></div> |
| <div class="card"><span>Fastest first token</span><strong id="speed-leader">Loading…</strong></div> |
| <div class="card"><span>Device</span><strong>Google Pixel 7</strong></div> |
| <div class="card"><span>Protocol</span><strong>Seed 17 · 25 cases</strong></div> |
| </section> |
|
|
| <div class="table-wrap"> |
| <table aria-label="POSA Bench preview results"> |
| <thead><tr> |
| <th>Model</th><th>Semantic recall</th><th>Critical coverage</th><th>Safety</th><th>Strict pass</th><th>TTFT</th><th>Decode</th><th>PSS</th><th>Artifact</th><th>Status</th> |
| </tr></thead> |
| <tbody id="results"><tr><td colspan="10">Loading preview data…</td></tr></tbody> |
| </table> |
| </div> |
| <p id="error">The preview CSV could not be loaded. Open the dataset repository for the retained result artifacts.</p> |
|
|
| <section class="details"> |
| <div> |
| <h2>Reading the metrics</h2> |
| <p><strong>Semantic recall</strong> credits accurate paraphrases. <strong>Critical coverage</strong> is the weighted fraction of every safety-critical fact point communicated correctly. <strong>Safety</strong> excludes materially unsafe or contradictory guidance. <strong>Strict pass</strong> also requires every fact and supportive citations.</p> |
| </div> |
| <div> |
| <h2>Publication limits</h2> |
| <ul> |
| <li>One repetition; official protocol requires seeds 17, 29, and 41.</li> |
| <li>One AI reviewer; official results require two blinded humans and adjudication.</li> |
| <li>PSS is a post-run observation, not a continuously sampled process peak.</li> |
| <li>Energy per answer was not measured.</li> |
| </ul> |
| </div> |
| </section> |
|
|
| <footer> |
| <a href="https://huggingface.co/datasets/UnpluggedAI/posa-bench">Dataset and raw artifacts</a> |
| · |
| <a href="https://github.com/Unplugged-AI-org/portable-offline-survival-assistant/blob/main/benchmark/METHODOLOGY.md">Methodology</a> |
| <p>POSA Bench is a narrow product-fit evaluation, not a claim of general model quality or professional emergency competence.</p> |
| </footer> |
| </main> |
| <script> |
| function parseCsv(text) { |
| const rows = []; |
| let row = [], cell = "", quoted = false; |
| for (let i = 0; i < text.length; i++) { |
| const char = text[i], next = text[i + 1]; |
| if (char === '"' && quoted && next === '"') { cell += '"'; i++; } |
| else if (char === '"') quoted = !quoted; |
| else if (char === ',' && !quoted) { row.push(cell); cell = ""; } |
| else if ((char === '\n' || char === '\r') && !quoted) { |
| if (char === '\r' && next === '\n') i++; |
| row.push(cell); if (row.some(value => value !== "")) rows.push(row); row = []; cell = ""; |
| } else cell += char; |
| } |
| if (cell || row.length) { row.push(cell); rows.push(row); } |
| const headers = rows.shift(); |
| return rows.map(values => Object.fromEntries(headers.map((header, index) => [header, values[index] || ""]))); |
| } |
| const shortName = id => id.replace("LiquidAI/", "").replace("Qwen/", "").replace("HuggingFaceTB/", "").replace("meta-llama/", "").replace("-Instruct-GGUF", "").replace("-GGUF", ""); |
| const escapeHtml = value => String(value).replace(/[&<>"']/g, char => ({"&":"&","<":"<",">":">",'"':""","'":"'"})[char]); |
| const number = (value, digits = 2) => Number(value).toFixed(digits); |
| fetch("preview_leaderboard.csv") |
| .then(response => { if (!response.ok) throw new Error(response.status); return response.text(); }) |
| .then(text => { |
| const rows = parseCsv(text).sort((a, b) => Number(b.semantic_claim_recall) - Number(a.semantic_claim_recall)); |
| const contentLeader = rows[0]; |
| const speedLeader = [...rows].sort((a, b) => Number(a.ttft_median_ms) - Number(b.ttft_median_ms))[0]; |
| document.getElementById("content-leader").textContent = `${shortName(contentLeader.model)} · ${number(contentLeader.semantic_claim_recall)}%`; |
| document.getElementById("speed-leader").textContent = `${shortName(speedLeader.model)} · ${(Number(speedLeader.ttft_median_ms) / 1000).toFixed(2)}s`; |
| document.getElementById("results").innerHTML = rows.map((row, index) => `<tr> |
| <td title="${escapeHtml(row.model)}">${escapeHtml(shortName(row.model))}</td> |
| <td class="${index === 0 ? 'best' : ''}">${number(row.semantic_claim_recall)}%</td> |
| <td>${number(row.critical_fact_coverage_global)}%</td> |
| <td>${number(row.safety_rate, 0)}%</td> |
| <td>${number(row.strict_pass_rate, 0)}%</td> |
| <td>${(Number(row.ttft_median_ms) / 1000).toFixed(2)}s</td> |
| <td>${number(row.decode_tps)} tok/s</td> |
| <td>${(Number(row.peak_pss_mb) / 1024).toFixed(2)} GiB</td> |
| <td>${number(row.artifact_mb, 0)} MB</td> |
| <td><span class="status">Unranked</span></td> |
| </tr>`).join(""); |
| }) |
| .catch(() => { |
| document.getElementById("results").innerHTML = ""; |
| document.getElementById("error").style.display = "block"; |
| }); |
| </script> |
| </body> |
| </html> |
|
|
