add llama-3.1 + visual improvements

app.py

@@ -50,7 +50,8 @@ def load_model(folders, tab_name, msg_lengths):
 
     for i, folder in enumerate(folders):
         model_name = folder.split('/')[-1]
-        results['Rank'].append(i)
+        if 'fine-tune' in model_name:
+            model_name += ' 🛠️'
         results['Model'].append(model_name)
         for task in msg_lengths:
             if not os.path.isfile(f'{folder}/{tab_name}/{task}.csv'):
@@ -60,13 +61,19 @@ def load_model(folders, tab_name, msg_lengths):
                 results[msg_lengths[task]].append(int(df['result'].sum() / len(df) * 100))
 
         mean_score.append(-np.mean([float(results[msg_lengths[task]][i]) for task in list(msg_lengths.keys())[:10]]))
-    for rank, i in enumerate(np.argsort(mean_score)):
-        results['Rank'][i] = rank + 1
 
-    res_df = pd.DataFrame(results).sort_values(['Rank'])
-    # print(res_df.head())
-    res_df['Avg ≤32k'] = res_df[res_df.columns[2:7]].astype(float).fillna(0).mean(axis=1).astype(int)
-    res_df['Avg ≤128k'] = res_df[res_df.columns[2:9]].astype(float).fillna(0).mean(axis=1).astype(int)
+    res_df = pd.DataFrame(results)
+    lengths = list(msg_lengths.values())
+    res_df['mean_score'] = mean_score
+    res_df['num_lengths'] = -(res_df[lengths].astype(float) > 0).sum(axis=1)
+    res_df = res_df[res_df.num_lengths != 0]
+    res_df.sort_values(['num_lengths', 'mean_score'], inplace=True)
+    res_df['Rank'] = range(1, res_df.shape[0] + 1)
+
+    res_df['Avg ≤32k'] = res_df[lengths[:5]].astype(float).fillna(0).mean(axis=1).astype(int)
+    res_df['Avg ≤128k'] = res_df[lengths[:7]].astype(float).fillna(0).mean(axis=1).astype(int)
+    ordered_columns = ['Rank', 'Model', 'Avg ≤32k', 'Avg ≤128k'] + lengths
+    res_df = res_df[ordered_columns]
     return res_df
     
 def build_leaderboard_tab(folders):