add v2

analysis/data/os_models/v2/final_rankings.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc908af138b83e2b2760975929742886c2fc8b5095ff87ae95a2a10a7b9626b3
+size 872

analysis/data/os_models/v2/results_ar.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40fb9d04723ff5ffd5189222d0496a0bb1b4b5d62fcb43f0b3bd1f52fee8ad43
+size 12363

analysis/data/os_models/v2/results_fr.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86837a26cd20191cfb6e321a47ecb6bc50cf541a6313cbd7a69e8bb7778c32da
+size 7235

analysis/data/os_models/v2/results_hi.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2f6d09787fa06e86d7407e418d15cd83d5a32a5ac947257fbf7bff251e93fda
+size 7433

analysis/data/os_models/v2/results_ru.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ab46c8317084966648116ac11e3d739dbd492013fcce48d2ec0283b1b50a74f
+size 8586

analysis/data/os_models/v2/results_sw.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:caec2333bd61ec53164555ff8629b11d79fb1d4a1272e3d31b0731a380d2748f
+size 7352

analysis/data/os_models/v2/results_te.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db36b5f9a229368749d0ed9196155eb7a2bd2ffe5604e73912e72d98533429c1
+size 6394

analysis/data/os_models/v2/results_th.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:808f72a53b1de3ec5a8839db3cc4393e0e4cfcf5e52f13a11ee8988fdeb03432
+size 7582

analysis/data/os_models/v2/results_tr.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ffd3d95b5a8d9d4c4db2706cc254c098ad017d98a81b8288ac64b8c2c716038
+size 7202

analysis/data/os_models/v2/results_zh.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86c959789f6425ec5e4e5dc4d05140e15502a7c37c2de65493ea18bb474599c4
+size 10203

app/src/leaderboard_results.js

@@ -14,6 +14,20 @@ const languageMap = {
   'Hindi': 'results_hi.csv',
 };
 
+const versionMap = {
+  'v1': 'v1',
+  'v2': 'v2'
+};
+
+const versionChangelog = {
+  'v1': 'Initial release of FineTasks Leaderboard',
+  'v2': 'Changes in v2:\n' +
+        '• Fixed a bug in the rescaling of scores\n' +
+        '• Switched to using Native choice prefixes for Thai/Telugu/Hindi/Arabics\n' +
+        '• Added Options: anchors before showing options for continuation tasks (e.g Hellawag) - consistent improvement in scores\n' +
+        '• Removed openai/gpt-4o-mini'
+};
+
 const columnNameMap = {
   'runname': 'Model',
   'agg_score_macro': 'Score',
@@ -25,12 +39,15 @@ const columnNameMap = {
   'rank': 'Rank'
 };
 
-function createDropdown(options, onChange) {
+function createDropdown(options, onChange, initialValue = null) {
   const select = document.createElement('select');
   options.forEach(option => {
     const optionElement = document.createElement('option');
     optionElement.value = option;
     optionElement.textContent = option;
+    if (initialValue && option === initialValue) {
+      optionElement.selected = true;
+    }
     select.appendChild(optionElement);
   });
   select.addEventListener('change', onChange);
@@ -72,11 +89,35 @@ function createResultsTable(data, extraColumn) {
   data.forEach((row, index) => {
     if (!row.runname) return; // Skip rows without a model name
     const tr = body.insertRow();
+    
+    // Add gradient background for top 3 positions
+    if (index < 3) {
+      const opacity = 1 - (index * 0.25); // Creates a fading effect: 1, 0.75, 0.5
+      tr.style.backgroundColor = `rgba(255, 165, 0, ${opacity * 0.2})`; // Light orange with fading opacity
+      tr.style.fontWeight = 600; // Make text slightly bolder for top 3
+    }
+
     columns.forEach(column => {
       const td = tr.insertCell();
-      td.className = `column-${sanitizeColumnName(column)}`; // Sanitize the column name
+      td.className = `column-${sanitizeColumnName(column)}`;
+      
       if (column === 'rank') {
         td.textContent = index + 1;
+        // Add special styling for top 3 ranks
+        if (index < 3) {
+          td.style.fontWeight = 'bold';
+          switch(index) {
+            case 0:
+              td.style.color = '#FFB800'; // Gold
+              break;
+            case 1:
+              td.style.color = '#C0C0C0'; // Silver
+              break;
+            case 2:
+              td.style.color = '#CD7F32'; // Bronze
+              break;
+          }
+        }
       } else if (column === 'runname') {
         const modelName = row[column];
         let displayName;
@@ -112,6 +153,40 @@ function createResultsTable(data, extraColumn) {
   return tableWrapper;
 }
 
+function createChangelog() {
+  const changelogContainer = document.createElement('div');
+  changelogContainer.className = 'changelog-container';
+
+  const changelogHeader = document.createElement('div');
+  changelogHeader.className = 'changelog-header';
+  
+  const arrow = document.createElement('span');
+  arrow.className = 'changelog-arrow';
+  arrow.textContent = '▶';
+  
+  const label = document.createElement('span');
+  label.textContent = 'Changelog';
+  label.className = 'changelog-label';
+
+  const content = document.createElement('div');
+  content.className = 'changelog-content';
+  content.style.display = 'none';
+
+  changelogHeader.appendChild(arrow);
+  changelogHeader.appendChild(label);
+  changelogContainer.appendChild(changelogHeader);
+  changelogContainer.appendChild(content);
+
+  // Toggle changelog visibility
+  changelogHeader.addEventListener('click', () => {
+    const isVisible = content.style.display !== 'none';
+    content.style.display = isVisible ? 'none' : 'block';
+    arrow.textContent = isVisible ? '▶' : '▼';
+  });
+
+  return { container: changelogContainer, content };
+}
+
 export function initLeaderboardResults(containerId) {
   const container = document.getElementById(containerId);
   if (!container) return;
@@ -120,17 +195,12 @@ export function initLeaderboardResults(containerId) {
   titleElement.textContent = 'FineTasks Leaderboard';
   titleElement.className = 'leaderboard-title';
 
+  // Create changelog
+  const { container: changelogContainer, content: changelogContent } = createChangelog();
+
   const tableContainer = document.createElement('div');
   tableContainer.className = 'table-container';
   
-  const languageLabel = document.createElement('label');
-  languageLabel.textContent = 'Language: ';
-  const languageDropdown = createDropdown(Object.keys(languageMap), updateLanguageTable);
-
-  const extraColumnLabel = document.createElement('label');
-  extraColumnLabel.textContent = 'Task: ';
-  const extraColumnDropdown = createDropdown(['None'], updateTable);
-
   let leaderboardDataTable;
   let currentData = [];
 
@@ -139,12 +209,14 @@ export function initLeaderboardResults(containerId) {
   captionElement.className = 'table-caption';
   captionElement.textContent = container.dataset.caption || '';
 
+  // Define update functions first
   async function updateLanguageTable() {
+    const selectedVersion = versionDropdown.value;
     const selectedLanguage = languageDropdown.value;
     const csvFile = languageMap[selectedLanguage];
 
     try {
-      const response = await fetch(`data/os_models/${csvFile}`);
+      const response = await fetch(`data/os_models/${selectedVersion}/${csvFile}`);
       if (!response.ok) {
         throw new Error(`HTTP error! status: ${response.status}`);
       }
@@ -154,7 +226,6 @@ export function initLeaderboardResults(containerId) {
         ? results.sort((a, b) => a.avg_rank_macro - b.avg_rank_macro)
         : results.sort((a, b) => b.agg_score_macro - a.agg_score_macro);
 
-      // Update extra column dropdown options
       if (selectedLanguage !== 'All Languages') {
         const columnOptions = ['None'].concat(Object.keys(currentData[0]).filter(key => 
           !['runname', 'seed', 'steps', 'agg_score_micro', 'rank', 'avg_rank_macro', ''].includes(key)
@@ -176,6 +247,7 @@ export function initLeaderboardResults(containerId) {
       }
 
       updateTable();
+      updateChangelog();
     } catch (error) {
       console.error('Error fetching CSV:', error);
       tableContainer.innerHTML = `<p>Error loading data: ${error.message}</p>`;
@@ -184,7 +256,7 @@ export function initLeaderboardResults(containerId) {
 
   function updateTable() {
     const extraColumn = languageDropdown.value === 'All Languages' ? 'All Languages' : 
-                        (extraColumnDropdown.value === 'None' ? null : extraColumnDropdown.value);
+                       (extraColumnDropdown.value === 'None' ? null : extraColumnDropdown.value);
     
     tableContainer.innerHTML = '';
     const tableWrapper = createResultsTable(currentData, extraColumn);
@@ -201,17 +273,71 @@ export function initLeaderboardResults(containerId) {
       sortable: true,
       fixedHeight: true,
       labels: {
-        info: ''  // This removes the "Showing 1 to X of Y entries" text
+        info: ''
       }
     });
 
-    // Adjust column widths after the table is created
     setTimeout(adjustColumnWidths, 0);
   }
 
+  function updateChangelog() {
+    const selectedVersion = versionDropdown.value;
+    changelogContent.textContent = versionChangelog[selectedVersion];
+  }
+
+  // Add this function to get URL parameters
+  function getUrlParameter(name) {
+    const urlParams = new URLSearchParams(window.location.search);
+    return urlParams.get(name);
+  }
+
+  // Add this function to set URL parameters
+  function updateUrlParameter(key, value) {
+    const urlParams = new URLSearchParams(window.location.search);
+    if (value) {
+      urlParams.set(key, value);
+    } else {
+      urlParams.delete(key);
+    }
+    const newUrl = `${window.location.pathname}${urlParams.toString() ? '?' + urlParams.toString() : ''}`;
+    window.history.pushState({ path: newUrl }, '', newUrl);
+  }
+
+  // Get initial language from URL
+  const urlLanguage = getUrlParameter('language');
+  const initialLanguage = urlLanguage && Object.keys(languageMap).includes(urlLanguage) 
+    ? urlLanguage 
+    : 'All Languages';
+
+  // Create dropdowns with initial values
+  const languageLabel = document.createElement('label');
+  languageLabel.textContent = 'Language: ';
+  const languageDropdown = createDropdown(
+    Object.keys(languageMap), 
+    (e) => {
+      updateLanguageTable();
+      updateUrlParameter('language', e.target.value === 'All Languages' ? null : e.target.value);
+    }, 
+    initialLanguage
+  );
+
+  const extraColumnLabel = document.createElement('label');
+  extraColumnLabel.textContent = 'Task: ';
+  const extraColumnDropdown = createDropdown(['None'], updateTable);
+
+  const versionLabel = document.createElement('label');
+  versionLabel.textContent = 'Version: ';
+  const versionDropdown = createDropdown(Object.keys(versionMap), updateLanguageTable);
+
+  // Create controls
   const controls = document.createElement('div');
   controls.className = 'controls leaderboard-controls fine-tasks-controls';
 
+  const versionControlGroup = document.createElement('div');
+  versionControlGroup.className = 'control-group';
+  versionControlGroup.appendChild(versionLabel);
+  versionControlGroup.appendChild(versionDropdown);
+
   const languageControlGroup = document.createElement('div');
   languageControlGroup.className = 'control-group';
   languageControlGroup.appendChild(languageLabel);
@@ -222,16 +348,20 @@ export function initLeaderboardResults(containerId) {
   extraColumnControlGroup.appendChild(extraColumnLabel);
   extraColumnControlGroup.appendChild(extraColumnDropdown);
 
+  controls.appendChild(versionControlGroup);
   controls.appendChild(languageControlGroup);
   controls.appendChild(extraColumnControlGroup);
 
+  // Add elements to container in new order
   container.appendChild(titleElement);
   container.appendChild(tableContainer);
-  container.appendChild(captionElement); // Add caption below the table
+  container.appendChild(captionElement);
   container.appendChild(controls);
+  container.appendChild(changelogContainer);
 
-  // Initialize with All Languages data
-  languageDropdown.value = 'All Languages';
+  // Initialize with URL language if present
+  versionDropdown.value = 'v2';
+  languageDropdown.value = initialLanguage;
   updateLanguageTable();
 }
 

app/src/style.css

@@ -385,3 +385,43 @@ html, body {
 body {
     -webkit-text-size-adjust: 100%;
 }
+
+/* Changelog styles */
+
+.changelog-container {
+  margin-top: 20px;
+  font-size: 14px;
+}
+
+.changelog-header {
+  cursor: pointer;
+  display: flex;
+  align-items: center;
+  gap: 12px; /* Increased space between arrow and label */
+  padding: 8px;
+  background-color: #f5f5f5;
+  border-radius: 4px;
+}
+
+.changelog-header:hover {
+  background-color: #ebebeb;
+}
+
+.changelog-arrow {
+  font-size: 12px;
+  color: #666;
+}
+
+.changelog-label {
+  font-weight: 600;
+  color: #333;
+}
+
+.changelog-content {
+  margin-top: 8px;
+  padding: 12px;
+  background-color: #fff;
+  border: 1px solid #eee;
+  border-radius: 4px;
+  line-height: 1.4;
+}