update 14B

app.py

@@ -17,6 +17,7 @@ load_dotenv()
 webhook_url = os.environ.get("WEBHOOK_URL")
 
 file_name_list = [
+    '14b',
     '9b',
     '7b',
     '3b',
@@ -36,6 +37,7 @@ metric_list = [
 ]
 
 model_size_list = [
+    '~14B',
     '~9B',
     '~7B',
     '~3B',
@@ -49,25 +51,13 @@ metric_to_sheet = {
 }
 
 model_size_to_file_name = {
+    '~14B': '14b',
     '~9B': '9b',
     '~7B': '7b',
     '~3B': '3b',
     '~1.5B': '1b5',
 }
 
-css = """
-.gr-dataframe table {
-    table-layout: fixed;
-    width: 100%; /* Ensures the table fills its container */
-}
-.gr-dataframe th, .gr-dataframe td {
-    width: 100px;  /* Set the exact width of each cell */
-    overflow: hidden;  /* Ensures the content doesn't overflow */
-    text-overflow: ellipsis;  /* Adds an ellipsis (...) if the text overflows */
-    white-space: nowrap;  /* Keeps the content on a single line */
-}
-"""
-
 about_md = """
 # Uncheatable Eval
 
@@ -167,61 +157,49 @@ def update_table(period: str,
         if 'Average (The lower the better)' in combined_data.columns:
             relevant_columns = [col for col in visible_columns if
                                 col not in ['Name', 'Parameters Count (B)', 'Average (The lower the better)']]
-            combined_data['Average (The lower the better)'] = round(combined_data[relevant_columns].mean(axis=1), 3)
-
-        sorted_data = combined_data.sort_values(by=sort_by, ascending=ascending)
-        sorted_data = sorted_data.rename(columns={'Average (The lower the better)': 'Average (lower=better)'})
-        visible_columns = ['Name', 'Parameters Count (B)', 'Average (lower=better)'] + visible_columns
-        filtered_data = sorted_data[visible_columns]
-
-        filtered_data.columns = [col.replace('_', ' ') for col in filtered_data.columns]
-
-        formatter = {col: "{:.3f}" for col in filtered_data.columns if
-                     filtered_data[col].dtype in ['float64', 'float32']}
-
-        def color_column(s):
-            return ['background-color: #fffdd0' if pd.notna(x) else 'default' for x in s]
-
-        # color gradient
-        colors = ["#63be7b", "#ffffff", "#f8696b"]
-        cmap = LinearSegmentedColormap.from_list("custom_cmap", colors)
-        target_color_columns = []
-        if 'Average' in color_columns:
-            target_color_columns.append('Average (lower=better)')
-        if 'Individual Tests' in color_columns:
-            target_color_columns.extend([col for col in filtered_data.columns if
-                                         col not in ['Name', 'Parameters Count (B)', 'Average (lower=better)']])
-
-        # styler = filtered_data.style.format(formatter).background_gradient(
-        #     cmap=cmap,
-        #     subset=target_color_columns,
-        #     vmin=min_value,
-        #     vmax=max_value
-        # ).apply(color_column, subset=['Parameters Count (B)'])
-
-        # for better visualization
-        vmin = {}
-        vmax = {}
-        for column in filtered_data.columns:
-            if column in ['Name', 'Parameters Count (B)']:
-                continue
-            col_values = filtered_data[column]
-            second_largest = col_values.nlargest(2).iloc[-1]
-            vmin[column] = col_values.min()
-            vmax[column] = second_largest
-
-        target_color_columns = []
-        if 'Average' in color_columns:
-            target_color_columns.append('Average (lower=better)')
-        if 'Individual Tests' in color_columns:
-            target_color_columns.extend([col for col in filtered_data.columns if
-                                         col not in ['Name', 'Parameters Count (B)', 'Average (lower=better)']])
-
-        styler = filtered_data.style.format(formatter).map(color_cell, subset=['Parameters Count (B)'])
-        for column in target_color_columns:
-            styler = styler.background_gradient(cmap=cmap, subset=[column], vmin=vmin[column], vmax=vmax[column])
-
-        return styler
+            if len(combined_data) > 0:
+                combined_data['Average (The lower the better)'] = round(combined_data[relevant_columns].mean(axis=1), 3)
+
+        if len(combined_data) > 0:
+            sorted_data = combined_data.sort_values(by=sort_by, ascending=ascending)
+            sorted_data = sorted_data.rename(columns={'Average (The lower the better)': 'Average (lower=better)'})
+            visible_columns = ['Name', 'Parameters Count (B)', 'Average (lower=better)'] + visible_columns
+            filtered_data = sorted_data[visible_columns]
+
+            filtered_data.columns = [col.replace('_', ' ') for col in filtered_data.columns]
+
+            formatter = {col: "{:.3f}" for col in filtered_data.columns if
+                         filtered_data[col].dtype in ['float64', 'float32']}
+
+            # color gradient
+            colors = ["#63be7b", "#ffffff", "#f8696b"]
+            cmap = LinearSegmentedColormap.from_list("custom_cmap", colors)
+            vmin = {}
+            vmax = {}
+            for column in filtered_data.columns:
+                if column in ['Name', 'Parameters Count (B)']:
+                    continue
+                col_values = filtered_data[column]
+                if len(col_values) > 1:
+                    second_largest = col_values.nlargest(2).iloc[-1]
+                    vmin[column] = col_values.min()
+                    vmax[column] = second_largest
+
+            target_color_columns = []
+            if 'Average' in color_columns:
+                target_color_columns.append('Average (lower=better)')
+            if 'Individual Tests' in color_columns:
+                target_color_columns.extend([col for col in filtered_data.columns if
+                                             col not in ['Name', 'Parameters Count (B)', 'Average (lower=better)']])
+
+            styler = filtered_data.style.format(formatter)
+            for column in target_color_columns:
+                if column in vmin and column in vmax:  # Ensure that the vmin and vmax dicts contain the column
+                    styler = styler.background_gradient(cmap=cmap, subset=[column], vmin=vmin[column], vmax=vmax[column])
+
+            return styler
+        else:
+            return pd.DataFrame()
     else:
         return pd.DataFrame()
 
@@ -334,7 +312,8 @@ def create_scaling_plot(all_data, period):
     y_tick_text = [f"{val:.1f}" for val in y_tick_vals]
 
     fig.update_xaxes(tickvals=np.log(x_tick_vals), ticktext=x_tick_text, title='Params(B)')
-    fig.update_yaxes(tickvals=np.log(y_tick_vals), ticktext=y_tick_text, title='Compression Rate (%)', autorange='reversed')
+    fig.update_yaxes(tickvals=np.log(y_tick_vals), ticktext=y_tick_text, title='Compression Rate (%)',
+                     autorange='reversed')
 
     fig.update_layout(
         xaxis=dict(showgrid=True, zeroline=False),
@@ -346,20 +325,26 @@ def create_scaling_plot(all_data, period):
     return fig
 
 
-all_data = {}
-time_list = []
-for folder in get_folders_matching_format('data'):
-    folder_name = os.path.basename(folder)
-    time_list.append(folder_name)
-    if all_data.get(folder) is None:
-        all_data[folder_name] = {}
-    for file_name in file_name_list:
-        if all_data.get(file_name) is None:
-            all_data[folder_name][file_name] = {}
-        for sheet_name in sheet_name_list:
-            final_file_name = os.path.join(folder, file_name)
-            all_data[folder_name][file_name][sheet_name] = rename_columns(
-                pd.read_excel(final_file_name + '.xlsx', sheet_name=sheet_name))
+def read_all_data(folder_name):
+    all_data = {}
+    time_list = []
+    for folder in get_folders_matching_format(folder_name):
+        folder_name = os.path.basename(folder)
+        time_list.append(folder_name)
+        if all_data.get(folder) is None:
+            all_data[folder_name] = {}
+        for file_name in file_name_list:
+            if all_data.get(file_name) is None:
+                all_data[folder_name][file_name] = {}
+            for sheet_name in sheet_name_list:
+                final_file_name = os.path.join(folder, file_name)
+                all_data[folder_name][file_name][sheet_name] = rename_columns(
+                    pd.read_excel(final_file_name + '.xlsx', sheet_name=sheet_name))
+
+    return all_data, time_list
+
+
+all_data, time_list = read_all_data('data')
 
 initial_fig = create_scaling_plot(all_data, time_list[-1])
 
@@ -385,10 +370,12 @@ css = '''
 
 '''
 
+TITLE_HTML = '<h1 style="text-align:center"><span style="font-size:1.3em">🏆 LLM Compression Leaderboard</span></h1>'
+SUBTITLE_HTML = "<h1 style='text-align:center'><span style='font-size:0.8em'>Welcome to Uncheatable Eval LLM Compression Leaderboard, where fancy fine-tuning and cheating won’t work 🚫; only compute 💻, data 📊, and real innovation 🔥 can prevail!</span></h1>"
+
 with gr.Blocks(css=css) as demo:
-    gr.HTML('<h1 style="text-align:center"><span style="font-size:1.3em">🏆 LLM Compression Leaderboard</span></h1>')
-    gr.HTML(
-        "<h1 style='text-align:center'><span style='font-size:0.8em'>Welcome to Uncheatable Eval LLM Compression Leaderboard, where fancy fine-tuning and cheating won’t work 🚫; only compute 💻, data 📊, and real innovation 🔥 can prevail!</span></h1>")
+    gr.HTML(TITLE_HTML)
+    gr.HTML(SUBTITLE_HTML)
     with gr.Tabs() as tabs:
         with gr.Tab("🏆 Leaderboard"):
             with gr.Row():
@@ -430,6 +417,7 @@ with gr.Blocks(css=css) as demo:
         with gr.Tab("📈 Scaling Law"):
             period_selector_2 = gr.Dropdown(label="Period", choices=time_list, value=time_list[0])
 
+
             def update_plot(period):
                 new_fig = create_scaling_plot(all_data, period)
                 return new_fig