Deployed 88a2a79 with MkDocs version: 1.6.1

scicode-bench · Nov 4, 2024 · b445229 · b445229
1 parent e634166
commit b445229
Showing 1 changed file with 32 additions and 32 deletions.
diff --git a/leaderboard/index.html b/leaderboard/index.html
@@ -552,88 +552,88 @@ <h1 id="scicode-leaderboard">SciCode Leaderboard</h1>
 <tr>
 <th>Models</th>
 <th>Main Problem Resolve Rate</th>
-<th><span style="background-color:lightgrey">Subproblem</span></th>
+<th><span style="color:grey">Subproblem</span></th>
 </tr>
 </thead>
 <tbody>
 <tr>
 <td>🥇 OpenAI o1-preview</td>
-<td><div align="center">7.7</div></td>
-<td><div align="center" style="background-color:lightgrey">28.5</div></td>
+<td><div align="center"><strong>7.7</strong></div></td>
+<td><div align="center" style="color:grey">28.5</div></td>
 </tr>
 <tr>
 <td>🥈 Claude3.5-Sonnet</td>
-<td><div align="center">4.6</div></td>
-<td><div align="center" style="background-color:lightgrey">26.0</div></td>
+<td><div align="center"><strong>4.6</strong></div></td>
+<td><div align="center" style="color:grey">26.0</div></td>
 </tr>
 <tr>
 <td>🥉 Claude3.5-Sonnet (new)</td>
-<td><div align="center">4.6</div></td>
-<td><div align="center" style="background-color:lightgrey">25.3</div></td>
+<td><div align="center"><strong>4.6</strong></div></td>
+<td><div align="center" style="color:grey">25.3</div></td>
 </tr>
 <tr>
 <td>Deepseek-Coder-v2</td>
-<td><div align="center">3.1</div></td>
-<td><div align="center" style="background-color:lightgrey">21.2</div></td>
+<td><div align="center"><strong>3.1</strong></div></td>
+<td><div align="center" style="color:grey">21.2</div></td>
 </tr>
 <tr>
 <td>GPT-4o</td>
-<td><div align="center">1.5</div></td>
-<td><div align="center" style="background-color:lightgrey">25.0</div></td>
+<td><div align="center"><strong>1.5</strong></div></td>
+<td><div align="center" style="color:grey">25.0</div></td>
 </tr>
 <tr>
 <td>GPT-4-Turbo</td>
-<td><div align="center">1.5</div></td>
-<td><div align="center" style="background-color:lightgrey">22.9</div></td>
+<td><div align="center"><strong>1.5</strong></div></td>
+<td><div align="center" style="color:grey">22.9</div></td>
 </tr>
 <tr>
 <td>OpenAI o1-mini</td>
-<td><div align="center">1.5</div></td>
-<td><div align="center" style="background-color:lightgrey">22.2</div></td>
+<td><div align="center"><strong>1.5</strong></div></td>
+<td><div align="center" style="color:grey">22.2</div></td>
 </tr>
 <tr>
 <td>Gemini 1.5 Pro</td>
-<td><div align="center">1.5</div></td>
-<td><div align="center" style="background-color:lightgrey">21.9</div></td>
+<td><div align="center"><strong>1.5</strong></div></td>
+<td><div align="center" style="color:grey">21.9</div></td>
 </tr>
 <tr>
 <td>Claude3-Opus</td>
-<td><div align="center">1.5</div></td>
-<td><div align="center" style="background-color:lightgrey">21.5</div></td>
+<td><div align="center"><strong>1.5</strong></div></td>
+<td><div align="center" style="color:grey">21.5</div></td>
 </tr>
 <tr>
 <td>Llama-3.1-405B-Chat</td>
-<td><div align="center">1.5</div></td>
-<td><div align="center" style="background-color:lightgrey">19.8</div></td>
+<td><div align="center"><strong>1.5</strong></div></td>
+<td><div align="center" style="color:grey">19.8</div></td>
 </tr>
 <tr>
 <td>Claude3-Sonnet</td>
-<td><div align="center">1.5</div></td>
-<td><div align="center" style="background-color:lightgrey">17.0</div></td>
+<td><div align="center"><strong>1.5</strong></div></td>
+<td><div align="center" style="color:grey">17.0</div></td>
 </tr>
 <tr>
 <td>Qwen2-72B-Instruct</td>
-<td><div align="center">1.5</div></td>
-<td><div align="center" style="background-color:lightgrey">17.0</div></td>
+<td><div align="center"><strong>1.5</strong></div></td>
+<td><div align="center" style="color:grey">17.0</div></td>
 </tr>
 <tr>
 <td>Llama-3.1-70B-Chat</td>
-<td><div align="center">0.0</div></td>
-<td><div align="center" style="background-color:lightgrey">17.0</div></td>
+<td><div align="center"><strong>0.0</strong></div></td>
+<td><div align="center" style="color:grey">17.0</div></td>
 </tr>
 <tr>
 <td>Mixtral-8x22B-Instruct</td>
-<td><div align="center">0.0</div></td>
-<td><div align="center" style="background-color:lightgrey">16.3</div></td>
+<td><div align="center"><strong>0.0</strong></div></td>
+<td><div align="center" style="color:grey">16.3</div></td>
 </tr>
 <tr>
 <td>Llama-3-70B-Chat</td>
-<td><div align="center">0.0</div></td>
-<td><div align="center" style="background-color:lightgrey">14.6</div></td>
+<td><div align="center"><strong>0.0</strong></div></td>
+<td><div align="center" style="color:grey">14.6</div></td>
 </tr>
 </tbody>
 </table>
-<p>Note: If the models tie in the Main Problem resolve rate, we will then compare the Subproblems.</p>
+<p><strong>Note: If the models tie in the Main Problem resolve rate, we will then compare the Subproblems.</strong></p>
 <!-- Once you've added the results to the submission repository,
      bring back the table here -->
 <!-- include-markdown "leaderboard_table.md" -->