Spaces:

ellamind
/

base-eval

Running

App Files Files Community

maxidl commited on 25 days ago

Commit

fd32731

verified ·

1 Parent(s): cdbce30

Upload index.html with huggingface_hub

Browse files

Files changed (1) hide show

index.html +4 -134

index.html CHANGED Viewed

@@ -3,7 +3,7 @@
 <head>
   <meta charset="UTF-8">
   <meta name="viewport" content="width=device-width, initial-scale=1.0">
-  <title>Eval Suite Visualization</title>
   <script src="https://cdn.plot.ly/plotly-2.35.2.min.js" charset="utf-8"></script>
   <style>
     * { box-sizing: border-box; margin: 0; padding: 0; }
@@ -16,9 +16,7 @@
     /* ── Page header ─────────────────────────────── */
     .page-header {
-      display: flex;
-      align-items: center;
-      justify-content: space-between;
       margin-bottom: 24px;
     }
     .page-header h1 {
@@ -347,7 +345,8 @@
 </head>
 <body>
   <div class="page-header">
-    <h1>Eval Suite Visualization</h1>
   </div>
   <div id="init-loading">Initializing DuckDB...</div>
@@ -356,8 +355,6 @@
   <div class="add-panel-row" id="add-panel-row" style="display:none">
     <button class="btn btn-primary" id="btn-add-panel">+ Add Panel</button>
-    <button class="btn" id="btn-scorecard-csv">Export Scorecard CSV</button>
-    <span id="scorecard-progress" style="font-size:0.8rem;color:#6c757d;display:none"></span>
   </div>
   <script type="module">
@@ -1689,132 +1686,6 @@
       return panel;
     }
-    // ── Scorecard CSV export ────────────────────────────────────
-    async function exportScorecardCSV() {
-      const progress = document.getElementById('scorecard-progress');
-      const btn = document.getElementById('btn-scorecard-csv');
-      btn.disabled = true;
-      progress.style.display = 'inline';
-      progress.textContent = 'Querying all benchmark data...';
-      try {
-        // Get all benchmark-level tasks with checkpoint data
-        const tasks = await query(`
-          SELECT DISTINCT task, task_display_name, metric, higher_is_better
-          FROM scores
-          WHERE task_type = 'benchmark'
-            AND is_checkpoint = true
-            AND tokens_trained IS NOT NULL
-          ORDER BY task, metric
-        `);
-        // Fetch all checkpoint rows in one query
-        progress.textContent = 'Loading checkpoint scores...';
-        const allRows = await query(`
-          SELECT task, metric, model_display_name, tokens_trained, score,
-                 is_checkpoint, higher_is_better
-          FROM scores
-          WHERE task_type = 'benchmark'
-            AND is_checkpoint = true
-            AND tokens_trained IS NOT NULL
-          ORDER BY task, metric, model_display_name, tokens_trained
-        `);
-        // Group rows by task+metric
-        const grouped = {};
-        for (const r of allRows) {
-          const key = r.task + '|||' + r.metric;
-          if (!grouped[key]) grouped[key] = [];
-          grouped[key].push(r);
-        }
-        // Compute metrics for each task+metric combo
-        const stages = ['overall', 'early', 'late'];
-        const metricKeys = ['monotonicity', 'signalStrength', 'noise', 'orderingConsistency', 'discrimination'];
-        // CSV header
-        const headers = ['task', 'task_display_name', 'metric', 'higher_is_better', 'n_models', 'n_checkpoints'];
-        for (const stage of stages) {
-          for (const mk of metricKeys) {
-            headers.push(`${stage}_${mk}`);
-          }
-        }
-        const csvRows = [headers.join(',')];
-        let processed = 0;
-        const total = tasks.length;
-        for (const t of tasks) {
-          const key = t.task + '|||' + t.metric;
-          const rows = grouped[key];
-          if (!rows || rows.length === 0) continue;
-          // Build byModel structure (same as drawLineChart)
-          const byModel = {};
-          for (const r of rows) {
-            const name = r.model_display_name;
-            if (!byModel[name]) byModel[name] = { points: [], isCheckpoint: true };
-            byModel[name].points.push({ x: Number(r.tokens_trained), y: r.score });
-          }
-          for (const d of Object.values(byModel)) {
-            d.points.sort((a, b) => a.x - b.x);
-          }
-          const nModels = Object.keys(byModel).length;
-          const nCheckpoints = Math.max(...Object.values(byModel).map(d => d.points.length));
-          const metrics = computeTaskQualityMetrics(byModel, t.higher_is_better);
-          const vals = [
-            csvEsc(t.task),
-            csvEsc(t.task_display_name),
-            csvEsc(t.metric),
-            t.higher_is_better,
-            nModels,
-            nCheckpoints,
-          ];
-          for (const stage of stages) {
-            for (const mk of metricKeys) {
-              const v = metrics?.[stage]?.[mk];
-              vals.push(v != null && !isNaN(v) ? v.toFixed(6) : '');
-            }
-          }
-          csvRows.push(vals.join(','));
-          processed++;
-          if (processed % 50 === 0) {
-            progress.textContent = `Computing metrics... ${processed}/${total}`;
-            await new Promise(r => setTimeout(r, 0)); // yield to UI
-          }
-        }
-        // Download
-        const blob = new Blob([csvRows.join('\n')], { type: 'text/csv' });
-        const url = URL.createObjectURL(blob);
-        const a = document.createElement('a');
-        a.href = url;
-        a.download = 'benchmark_scorecard.csv';
-        a.click();
-        URL.revokeObjectURL(url);
-        progress.textContent = `Done! ${processed} task-metric combos exported.`;
-        setTimeout(() => { progress.style.display = 'none'; }, 3000);
-      } catch (err) {
-        progress.textContent = `Error: ${err.message}`;
-        console.error('Scorecard export failed:', err);
-      } finally {
-        btn.disabled = false;
-      }
-    }
-    function csvEsc(val) {
-      if (val == null) return '';
-      const s = String(val);
-      if (s.includes(',') || s.includes('"') || s.includes('\n')) {
-        return '"' + s.replace(/"/g, '""') + '"';
-      }
-      return s;
-    }
     // ── Init ────────────────────────────────────────────────────
     const elInitLoading = document.getElementById('init-loading');
     const elAddPanelRow = document.getElementById('add-panel-row');
@@ -1848,7 +1719,6 @@
     }
     document.getElementById('btn-add-panel').addEventListener('click', () => addPanel());
-    document.getElementById('btn-scorecard-csv').addEventListener('click', () => exportScorecardCSV());
     init();
   </script>

 <head>
   <meta charset="UTF-8">
   <meta name="viewport" content="width=device-width, initial-scale=1.0">
+  <title>ellamind base-eval</title>
   <script src="https://cdn.plot.ly/plotly-2.35.2.min.js" charset="utf-8"></script>
   <style>
     * { box-sizing: border-box; margin: 0; padding: 0; }
     /* ── Page header ─────────────────────────────── */
     .page-header {
+      text-align: center;
       margin-bottom: 24px;
     }
     .page-header h1 {
 </head>
 <body>
   <div class="page-header">
+    <h1>ellamind base-eval</h1>
+    <p style="margin:4px 0 0;font-size:13px;color:#6c757d;"><a href="https://github.com/ellamind/base-eval" target="_blank" rel="noopener" style="color:#4361ee;">ellamind/base-eval</a> · Data: <a href="https://huggingface.co/datasets/ellamind/eval-scores-ref" target="_blank" rel="noopener" style="color:#4361ee;">ellamind/eval-scores-ref</a></p>
   </div>
   <div id="init-loading">Initializing DuckDB...</div>
   <div class="add-panel-row" id="add-panel-row" style="display:none">
     <button class="btn btn-primary" id="btn-add-panel">+ Add Panel</button>
   </div>
   <script type="module">
       return panel;
     }
     // ── Init ────────────────────────────────────────────────────
     const elInitLoading = document.getElementById('init-loading');
     const elAddPanelRow = document.getElementById('add-panel-row');
     }
     document.getElementById('btn-add-panel').addEventListener('click', () => addPanel());
     init();
   </script>