Update app.py

app.py

@@ -11,7 +11,7 @@ CITATION_BUTTON_TEXT = r"""@misc{aienergyscore-leaderboard,
     howpublished = "\url{https://huggingface.co/spaces/AIEnergyScore/Leaderboard}",
 }"""
 
-# List of CSV filenames (one per task)
+# List of tasks (CSV filenames)
 tasks = [
     'asr.csv',
     'object_detection.csv',
@@ -30,64 +30,49 @@ def format_stars(score):
         score_int = int(score)
     except Exception:
         score_int = 0
-    return f'<span style="color: #3fa45bff; font-size:2em;">{"★" * score_int}</span>'
+    return f'<span style="color: #3fa45bff !important; font-size:2em !important;">{"★" * score_int}</span>'
 
 def make_link(mname):
     parts = str(mname).split('/')
     display_name = parts[1] if len(parts) > 1 else mname
     return f'[{display_name}](https://huggingface.co/{mname})'
 
-def read_csv_file(task):
-    """
-    Reads a CSV from the data/energy folder using the first column as the index
-    and strips any extraneous whitespace from the column names.
-    """
-    df = pd.read_csv('data/energy/' + task, index_col=0)
-    df.columns = df.columns.str.strip()  # remove any extra whitespace
-    return df
-
 def get_plots(task):
-    df = read_csv_file(task)
-    # Convert the numeric columns
-    df['total_gpu_energy'] = pd.to_numeric(df['total_gpu_energy'], errors='coerce')
-    df['energy_score'] = pd.to_numeric(df['energy_score'], errors='coerce').astype(int)
-    # Create a short model name for display on the y-axis.
+    df = pd.read_csv('data/energy/' + task)
+    if df.columns[0].startswith("Unnamed:"):
+        df = df.iloc[:, 1:]
+    df['energy_score'] = df['energy_score'].astype(int)
     df['Display Model'] = df['model'].apply(lambda m: m.split('/')[-1])
     
-    # Define a discrete color mapping for energy scores.
     color_map = {1: "red", 2: "orange", 3: "yellow", 4: "lightgreen", 5: "green"}
     
-    # Build the scatter plot.
     fig = px.scatter(
         df,
-        x="total_gpu_energy",
-        y="Display Model",
-        color="energy_score",
-        custom_data=['model', 'total_gpu_energy', 'energy_score'],
+        x="total_gpu_energy",  # Ensure correct column for x-axis
+        y="Display Model",  # Keep model name for y-axis
+        color="energy_score",  # Ensure correct column for point color
+        custom_data=['energy_score'],
         height=500,
         width=800,
-        color_discrete_map=color_map,
+        color_discrete_map=color_map
     )
     fig.update_traces(
-        hovertemplate=(
-            "Model: %{customdata[0]}<br>" +
-            "Total GPU Energy (Wh): %{customdata[1]:.4f}<br>" +
-            "Energy Score: %{customdata[2]}"
-        )
-    )
-    fig.update_layout(
-        xaxis_title="Total GPU Energy (Wh)",
-        yaxis_title="Model",
-        margin=dict(l=40, r=40, t=40, b=40)
+        hovertemplate="<br>".join([
+            "Model: %{y}",
+            "GPU Energy (Wh): %{x}",
+            "Energy Score: %{customdata[0]}"
+        ])
     )
+    fig.update_layout(xaxis_title="GPU Energy (Wh)", yaxis_title="Model")
     return fig
 
 def get_all_plots():
     all_df = pd.DataFrame()
     for task in tasks:
-        df = read_csv_file(task)
-        df['total_gpu_energy'] = pd.to_numeric(df['total_gpu_energy'], errors='coerce')
-        df['energy_score'] = pd.to_numeric(df['energy_score'], errors='coerce').astype(int)
+        df = pd.read_csv('data/energy/' + task)
+        if df.columns[0].startswith("Unnamed:"):
+            df = df.iloc[:, 1:]
+        df['energy_score'] = df['energy_score'].astype(int)
         df['Display Model'] = df['model'].apply(lambda m: m.split('/')[-1])
         all_df = pd.concat([all_df, df], ignore_index=True)
     all_df = all_df.drop_duplicates(subset=['model'])
@@ -96,39 +81,38 @@ def get_all_plots():
     
     fig = px.scatter(
         all_df,
-        x="total_gpu_energy",
+        x="total_gpu_energy",  # Ensure correct column for x-axis
         y="Display Model",
-        color="energy_score",
-        custom_data=['model', 'total_gpu_energy', 'energy_score'],
+        color="energy_score",  # Ensure correct column for point color
+        custom_data=['energy_score'],
         height=500,
         width=800,
-        color_discrete_map=color_map,
+        color_discrete_map=color_map
     )
     fig.update_traces(
-        hovertemplate=(
-            "Model: %{customdata[0]}<br>" +
-            "Total GPU Energy (Wh): %{customdata[1]:.4f}<br>" +
-            "Energy Score: %{customdata[2]}"
-        )
-    )
-    fig.update_layout(
-        xaxis_title="Total GPU Energy (Wh)",
-        yaxis_title="Model",
-        margin=dict(l=40, r=40, t=40, b=40)
+        hovertemplate="<br>".join([
+            "Model: %{y}",
+            "GPU Energy (Wh): %{x}",
+            "Energy Score: %{customdata[0]}"
+        ])
     )
+    fig.update_layout(xaxis_title="GPU Energy (Wh)", yaxis_title="Model")
     return fig
 
 def get_model_names(task):
     """
-    For a given task, load the energy CSV and return a DataFrame with the following columns:
+    For a given task, load the energy CSV and return a dataframe with the following columns:
       - Model (a markdown link)
-      - GPU Energy (Wh) formatted to 4 decimal places
+      - GPU Energy (Wh) formatted as a string with 4 decimal places
       - Score (a star rating based on energy_score)
-    For text_generation.csv only, also include the "Class" column if present.
+    For text_generation.csv only, also add the "Class" column from the CSV.
+    The final column order is: Model, GPU Energy (Wh), Score, [Class].
     """
-    df = read_csv_file(task)
-    df['total_gpu_energy'] = pd.to_numeric(df['total_gpu_energy'], errors='coerce')
-    df['energy_score'] = pd.to_numeric(df['energy_score'], errors='coerce').astype(int)
+    df = pd.read_csv('data/energy/' + task)
+    if df.columns[0].startswith("Unnamed:"):
+        df = df.iloc[:, 1:]
+    df['energy_score'] = df['energy_score'].astype(int)
+    # Format the energy as a string with 4 decimals
     df['GPU Energy (Wh)'] = df['total_gpu_energy'].apply(lambda x: f"{x:.4f}")
     df['Model'] = df['model'].apply(make_link)
     df['Score'] = df['energy_score'].apply(format_stars)
@@ -139,30 +123,31 @@ def get_model_names(task):
     else:
         df = df[['Model', 'GPU Energy (Wh)', 'Score']]
         
-    # Sort by the numeric GPU energy value.
-    df = df.sort_values(by='total_gpu_energy')
+    df = df.sort_values(by='GPU Energy (Wh)')
     return df
 
 def get_all_model_names():
     """
-    Combine data from all tasks and return a leaderboard DataFrame with:
-      - Model, GPU Energy (Wh), Score.
+    Combine data from all tasks and return a leaderboard table with:
+      - Model, GPU Energy (Wh), Score
     Duplicate models are dropped.
     """
     all_df = pd.DataFrame()
     for task in tasks:
-        df = read_csv_file(task)
-        df['total_gpu_energy'] = pd.to_numeric(df['total_gpu_energy'], errors='coerce')
-        df['energy_score'] = pd.to_numeric(df['energy_score'], errors='coerce').astype(int)
+        df = pd.read_csv('data/energy/' + task)
+        if df.columns[0].startswith("Unnamed:"):
+            df = df.iloc[:, 1:]
+        df['energy_score'] = df['energy_score'].astype(int)
         df['GPU Energy (Wh)'] = df['total_gpu_energy'].apply(lambda x: f"{x:.4f}")
         df['Model'] = df['model'].apply(make_link)
         df['Score'] = df['energy_score'].apply(format_stars)
         all_df = pd.concat([all_df, df], ignore_index=True)
     all_df = all_df.drop_duplicates(subset=['model'])
-    all_df = all_df.sort_values(by='total_gpu_energy')
+    all_df = all_df.sort_values(by='GPU Energy (Wh)')
     return all_df[['Model', 'GPU Energy (Wh)', 'Score']]
 
 # Build the Gradio interface.
+# The css argument below makes all tables (e.g. leaderboard) use a fixed layout with narrower columns.
 demo = gr.Blocks(css="""
 .gr-dataframe table {
     table-layout: fixed;
@@ -189,7 +174,7 @@ Click through the tasks below to see how different models measure up in terms of
     with gr.Tabs():
         with gr.TabItem("Text Generation 💬"):
             with gr.Row():
-                with gr.Column(scale=2):
+                with gr.Column(scale=1.3):
                     plot = gr.Plot(get_plots('text_generation.csv'))
                 with gr.Column(scale=1):
                     table = gr.Dataframe(get_model_names('text_generation.csv'), datatype="markdown")
@@ -272,6 +257,8 @@ Click through the tasks below to see how different models measure up in terms of
             lines=10,
             show_copy_button=True,
         )
-    gr.Markdown("Last updated: February 2025")
+    gr.Markdown(
+        """Last updated: February 2025"""
+    )
 
-demo.launch()
+demo.launch()