Version 0.4

app.py

@@ -11,7 +11,7 @@ from model_page import plot_model_stats
 
 # Configure matplotlib to prevent memory warnings and set dark background
 matplotlib.rcParams['figure.facecolor'] = '#000000'
-matplotlib.rcParams['axes.facecolor'] = '#000000' 
+matplotlib.rcParams['axes.facecolor'] = '#000000'
 matplotlib.rcParams['savefig.facecolor'] = '#000000'
 plt.ioff()  # Turn off interactive mode to prevent figure accumulation
 
@@ -35,36 +35,36 @@ def load_css():
 
 # Create the Gradio interface with sidebar and dark theme
 with gr.Blocks(title="Model Test Results Dashboard", css=load_css()) as demo:
-    
+
     with gr.Row():
-        # Sidebar for model selection  
+        # Sidebar for model selection
         with gr.Column(scale=1, elem_classes=["sidebar"]):
             gr.Markdown("# 🤖 TCID", elem_classes=["sidebar-title"])
-            
+
             # Description with integrated last update time
             if Ci_results.last_update_time:
                 description_text = f"**Transformer CI Dashboard**\n\n*Result overview by model and hardware (last updated: {Ci_results.last_update_time})*\n"
             else:
                 description_text = f"**Transformer CI Dashboard**\n\n*Result overview by model and hardware (loading...)*\n"
             description_display = gr.Markdown(description_text, elem_classes=["sidebar-description"])
-            
+
             # Summary button at the top
             summary_button = gr.Button(
-                "summary\n📊", 
+                "summary\n📊",
                 variant="primary",
                 size="lg",
                 elem_classes=["summary-button"]
             )
-            
+
             # Model selection header
             gr.Markdown(f"**Select model ({len(Ci_results.available_models)}):**", elem_classes=["model-header"])
-            
+
             # Scrollable container for model buttons
             with gr.Column(scale=1, elem_classes=["model-container"]):
                 # Create individual buttons for each model
                 model_buttons = []
                 model_choices = [model.lower() for model in Ci_results.available_models] if Ci_results.available_models else ["auto", "bert", "clip", "llama"]
-                
+
                 for model_name in model_choices:
                     btn = gr.Button(
                         model_name,
@@ -73,31 +73,31 @@ with gr.Blocks(title="Model Test Results Dashboard", css=load_css()) as demo:
                         elem_classes=["model-button"]
                     )
                     model_buttons.append(btn)
-            
+
             # CI job links at bottom of sidebar
             ci_links_display = gr.Markdown("🔗 **CI Jobs:** *Loading...*", elem_classes=["sidebar-links"])
-        
+
         # Main content area
         with gr.Column(scale=4, elem_classes=["main-content"]):
             # Summary display (default view)
             summary_display = gr.Plot(
                 value=create_summary_page(Ci_results.df, Ci_results.available_models),
-                label="", 
+                label="",
                 format="png",
                 elem_classes=["plot-container"],
                 visible=True
             )
-            
+
             # Detailed view components (hidden by default)
             with gr.Column(visible=False, elem_classes=["detail-view"]) as detail_view:
-                
+
                 # Create the plot output
                 plot_output = gr.Plot(
-                    label="", 
+                    label="",
                     format="png",
                     elem_classes=["plot-container"]
                 )
-                
+
                 # Create two separate failed tests displays in a row layout
                 with gr.Row():
                     with gr.Column(scale=1):
@@ -118,7 +118,7 @@ with gr.Blocks(title="Model Test Results Dashboard", css=load_css()) as demo:
                             container=False,
                             elem_classes=["failed-tests"]
                         )
-    
+
     # Set up click handlers for model buttons
     for i, btn in enumerate(model_buttons):
         model_name = model_choices[i]
@@ -129,12 +129,12 @@ with gr.Blocks(title="Model Test Results Dashboard", css=load_css()) as demo:
             fn=lambda: [gr.update(visible=False), gr.update(visible=True)],
             outputs=[summary_display, detail_view]
         )
-    
+
     # Summary button click handler
     def show_summary_and_update_links():
         """Show summary page and update CI links."""
         return create_summary_page(Ci_results.df, Ci_results.available_models), get_description_text(), get_ci_links()
-    
+
     summary_button.click(
         fn=show_summary_and_update_links,
         outputs=[summary_display, description_display, ci_links_display]
@@ -142,7 +142,7 @@ with gr.Blocks(title="Model Test Results Dashboard", css=load_css()) as demo:
         fn=lambda: [gr.update(visible=True), gr.update(visible=False)],
         outputs=[summary_display, detail_view]
     )
-    
+
     # Function to get current description text
     def get_description_text():
         """Get description text with integrated last update time."""
@@ -150,7 +150,7 @@ with gr.Blocks(title="Model Test Results Dashboard", css=load_css()) as demo:
             return f"**Transformer CI Dashboard**\n\n*Result overview by model and hardware (last updated: {Ci_results.last_update_time})*\n"
         else:
             return f"**Transformer CI Dashboard**\n\n*Result overview by model and hardware (loading...)*\n"
-    
+
     # Function to get CI job links
     def get_ci_links():
         """Get CI job links from the most recent data."""
@@ -158,16 +158,16 @@ with gr.Blocks(title="Model Test Results Dashboard", css=load_css()) as demo:
             # Check if df exists and is not empty
             if Ci_results.df is None or Ci_results.df.empty:
                 return "🔗 **CI Jobs:** *Loading...*"
-            
+
             # Get links from any available model (they should be the same for all models in a run)
             amd_multi_link = None
             amd_single_link = None
             nvidia_multi_link = None
             nvidia_single_link = None
-            
+
             for model_name in Ci_results.df.index:
                 row = Ci_results.df.loc[model_name]
-                
+
                 # Extract AMD links
                 if pd.notna(row.get('job_link_amd')) and (not amd_multi_link or not amd_single_link):
                     amd_link_raw = row.get('job_link_amd')
@@ -176,7 +176,7 @@ with gr.Blocks(title="Model Test Results Dashboard", css=load_css()) as demo:
                             amd_multi_link = amd_link_raw['multi']
                         if 'single' in amd_link_raw and not amd_single_link:
                             amd_single_link = amd_link_raw['single']
-                
+
                 # Extract NVIDIA links
                 if pd.notna(row.get('job_link_nvidia')) and (not nvidia_multi_link or not nvidia_single_link):
                     nvidia_link_raw = row.get('job_link_nvidia')
@@ -185,13 +185,13 @@ with gr.Blocks(title="Model Test Results Dashboard", css=load_css()) as demo:
                             nvidia_multi_link = nvidia_link_raw['multi']
                         if 'single' in nvidia_link_raw and not nvidia_single_link:
                             nvidia_single_link = nvidia_link_raw['single']
-                
+
                 # Break if we have all links
                 if amd_multi_link and amd_single_link and nvidia_multi_link and nvidia_single_link:
                     break
-            
+
             links_md = "🔗 **CI Jobs:**\n\n"
-            
+
             # AMD links
             if amd_multi_link or amd_single_link:
                 links_md += "**AMD:**\n"
@@ -200,7 +200,7 @@ with gr.Blocks(title="Model Test Results Dashboard", css=load_css()) as demo:
                 if amd_single_link:
                     links_md += f"• [Single GPU]({amd_single_link})\n"
                 links_md += "\n"
-            
+
             # NVIDIA links
             if nvidia_multi_link or nvidia_single_link:
                 links_md += "**NVIDIA:**\n"
@@ -208,15 +208,15 @@ with gr.Blocks(title="Model Test Results Dashboard", css=load_css()) as demo:
                     links_md += f"• [Multi GPU]({nvidia_multi_link})\n"
                 if nvidia_single_link:
                     links_md += f"• [Single GPU]({nvidia_single_link})\n"
-            
+
             if not (amd_multi_link or amd_single_link or nvidia_multi_link or nvidia_single_link):
                 links_md += "*No links available*"
-                
+
             return links_md
         except Exception as e:
             logger.error(f"getting CI links: {e}")
             return "🔗 **CI Jobs:** *Error loading links*"
-    
+
 
     # Auto-update CI links when the interface loads
     demo.load(

data.py

@@ -49,7 +49,7 @@ def get_distant_data() -> pd.DataFrame:
     # Retrieve NVIDIA dataframe
     nvidia_src = "hf://datasets/hf-internal-testing/transformers_daily_ci/**/ci_results_run_models_gpu/model_results.json"
     files_nvidia = sorted(fs.glob(nvidia_src), reverse=True)
-    # NOTE: should this be removeprefix instead of lstrip? 
+    # NOTE: should this be removeprefix instead of lstrip?
     nvidia_path = files_nvidia[0].lstrip('datasets/hf-internal-testing/transformers_daily_ci/')
     nvidia_path = "https://huggingface.co/datasets/hf-internal-testing/transformers_daily_ci/raw/main/" + nvidia_path
     df_nvidia = read_one_dataframe(nvidia_path, "nvidia")

model_page.py

@@ -41,47 +41,47 @@ MAX_FAILURE_ITEMS = 10
 def _create_pie_chart(ax: plt.Axes, device_label: str, filtered_stats: dict) -> None:
     """Create a pie chart for device statistics."""
     if not filtered_stats:
-        ax.text(0.5, 0.5, 'No test results', 
+        ax.text(0.5, 0.5, 'No test results',
                horizontalalignment='center', verticalalignment='center',
                transform=ax.transAxes, fontsize=14, color='#888888',
                fontfamily='monospace', weight='normal')
-        ax.set_title(device_label, fontsize=DEVICE_TITLE_FONT_SIZE, weight='bold', 
+        ax.set_title(device_label, fontsize=DEVICE_TITLE_FONT_SIZE, weight='bold',
                     pad=DEVICE_TITLE_PAD, color=TITLE_COLOR, fontfamily='monospace')
         ax.axis('off')
         return
-        
+
     chart_colors = [COLORS[category] for category in filtered_stats.keys()]
-    
+
     # Create minimal pie chart - full pie, no donut effect
     wedges, texts, autotexts = ax.pie(
-        filtered_stats.values(), 
+        filtered_stats.values(),
         labels=[label.lower() for label in filtered_stats.keys()],  # Lowercase for minimal look
         colors=chart_colors,
-        autopct=lambda pct: f'{int(pct * sum(filtered_stats.values()) / 100)}',
+        autopct=lambda pct: f'{round(pct * sum(filtered_stats.values()) / 100)}',
         startangle=PIE_START_ANGLE,
         explode=None,  # No separation
         shadow=False,
         wedgeprops=dict(edgecolor='#1a1a1a', linewidth=BORDER_LINE_WIDTH),  # Minimal borders
-        textprops={'fontsize': 12, 'weight': 'normal', 
+        textprops={'fontsize': 12, 'weight': 'normal',
                   'color': LABEL_COLOR, 'fontfamily': 'monospace'}
     )
-    
+
     # Enhanced percentage text styling for better readability
     for autotext in autotexts:
         autotext.set_color(BLACK)  # Black text for better contrast
         autotext.set_weight('bold')
         autotext.set_fontsize(14)
         autotext.set_fontfamily('monospace')
-    
+
     # Minimal category labels
     for text in texts:
         text.set_color(LABEL_COLOR)
         text.set_weight('normal')
         text.set_fontsize(13)
         text.set_fontfamily('monospace')
-    
+
     # Device label closer to chart and bigger
-    ax.set_title(device_label, fontsize=DEVICE_TITLE_FONT_SIZE, weight='normal', 
+    ax.set_title(device_label, fontsize=DEVICE_TITLE_FONT_SIZE, weight='normal',
                 pad=DEVICE_TITLE_PAD, color=TITLE_COLOR, fontfamily='monospace')
 
 
@@ -95,47 +95,47 @@ def plot_model_stats(df: pd.DataFrame, model_name: str) -> tuple[plt.Figure, str
         failures_amd = failures_nvidia = {}
     else:
         row = df.loc[model_name]
-        
+
         # Extract and process model data
         amd_stats, nvidia_stats = extract_model_data(row)[:2]
-        
+
         # Filter out categories with 0 values for cleaner visualization
         amd_filtered = {k: v for k, v in amd_stats.items() if v > 0}
         nvidia_filtered = {k: v for k, v in nvidia_stats.items() if v > 0}
-        
+
         # Generate failure info directly from dataframe
         failures_amd = dict(row.get('failures_amd', {}))
         failures_nvidia = dict(row.get('failures_nvidia', {}))
 
     # failure_xxx = {"single": [test, ...], "multi": [...]}
     # test = {"line": test_name. "trace": error_msg}
-    
+
     # Always create figure with two subplots side by side with padding
     fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(FIGURE_WIDTH_DUAL, FIGURE_HEIGHT_DUAL), facecolor=BLACK)
     ax1.set_facecolor(BLACK)
     ax2.set_facecolor(BLACK)
-    
+
     # Create both pie charts with device labels
     _create_pie_chart(ax1, "amd", amd_filtered)
     _create_pie_chart(ax2, "nvidia", nvidia_filtered)
-    
+
     # Add subtle separation line between charts - stops at device labels level
     line_x = 0.5
-    fig.add_artist(plt.Line2D([line_x, line_x], [0.0, SEPARATOR_LINE_Y_END], 
-                              color='#333333', linewidth=SEPARATOR_LINE_WIDTH, 
+    fig.add_artist(plt.Line2D([line_x, line_x], [0.0, SEPARATOR_LINE_Y_END],
+                              color='#333333', linewidth=SEPARATOR_LINE_WIDTH,
                               alpha=SEPARATOR_ALPHA, transform=fig.transFigure))
-    
+
     # Add central shared title for model name
-    fig.suptitle(f'{model_name.lower()}', fontsize=32, weight='bold', 
+    fig.suptitle(f'{model_name.lower()}', fontsize=32, weight='bold',
                 color='#CCCCCC', fontfamily='monospace', y=MODEL_TITLE_Y)
-    
+
     # Clean layout with padding and space for central title
     plt.tight_layout()
     plt.subplots_adjust(top=SUBPLOT_TOP, wspace=SUBPLOT_WSPACE)
-    
+
     amd_failed_info = prepare_textbox_content(failures_amd, 'AMD', bool(amd_filtered))
     nvidia_failed_info = prepare_textbox_content(failures_nvidia, 'NVIDIA', bool(nvidia_filtered))
-    
+
     return fig, amd_failed_info, nvidia_failed_info
 
 
@@ -147,7 +147,7 @@ def prepare_textbox_content(failures: dict[str, list], device: str, data_availab
     # Catch the case where there are no failures
     if not failures:
         return generate_underlined_line(f"No failures for {device}")
-    
+
     # Summary of failures
     single_failures = failures.get("single", [])
     multi_failures = failures.get("multi", [])
@@ -166,7 +166,7 @@ def prepare_textbox_content(failures: dict[str, list], device: str, data_availab
             name = name.split("::")[-1]
             info_lines.append(name)
         info_lines.append("\n")
-    
+
     # Add multi-gpu failures
     if multi_failures:
         info_lines.append(generate_underlined_line("Multi GPU failures:"))

styles.css

@@ -56,7 +56,7 @@ div[data-testid="column"]:has(.sidebar) {
     transition: all 0.4s cubic-bezier(0.4, 0, 0.2, 1) !important;
     position: relative !important;
     overflow: hidden !important;
-    box-shadow: 
+    box-shadow:
         0 4px 15px rgba(0, 0, 0, 0.3),
         inset 0 1px 0 rgba(255, 255, 255, 0.2) !important;
     font-weight: 600 !important;
@@ -110,8 +110,8 @@ div[data-testid="column"]:has(.sidebar) {
 
 /* Specific control for markdown content */
 .sidebar .markdown,
-.sidebar h1, 
-.sidebar h2, 
+.sidebar h1,
+.sidebar h2,
 .sidebar h3,
 .sidebar p {
     max-width: 100% !important;
@@ -261,7 +261,7 @@ div[data-testid="column"]:has(.model-container) {
     transition: all 0.4s cubic-bezier(0.4, 0, 0.2, 1) !important;
     position: relative !important;
     overflow: hidden !important;
-    box-shadow: 
+    box-shadow:
         0 4px 15px rgba(0, 0, 0, 0.3),
         inset 0 1px 0 rgba(255, 255, 255, 0.2) !important;
     font-weight: 600 !important;
@@ -634,4 +634,4 @@ h1, h2, h3, p, .markdown {
 @keyframes resetScroll {
     0% { scroll-behavior: auto; }
     100% { scroll-behavior: auto; }
-} 
+}

summary_page.py

@@ -25,7 +25,7 @@ NVIDIA_BAR_OFFSET = 0.54  # NVIDIA bar offset ratio
 # Colors
 COLORS = {
     'passed': '#4CAF50',
-    'failed': '#E53E3E', 
+    'failed': '#E53E3E',
     'skipped': '#FFD54F',
     'error': '#8B0000',
     'empty': "#5B5B5B"
@@ -47,14 +47,14 @@ def draw_text_and_bar(
 ) -> None:
     """Draw a horizontal bar chart for given stats and its label on the left."""
     # Text
-    label_x = column_left_position - LABEL_OFFSET 
+    label_x = column_left_position - LABEL_OFFSET
     ax.text(
-        label_x, y_bar, label, ha='right', va='center', color='#CCCCCC', fontsize=LABEL_FONT_SIZE, 
+        label_x, y_bar, label, ha='right', va='center', color='#CCCCCC', fontsize=LABEL_FONT_SIZE,
         fontfamily='monospace', fontweight='normal'
     )
     # Bar
     total = sum(stats.values())
-    if total > 0:            
+    if total > 0:
         left = column_left_position
         for category in ['passed', 'failed', 'skipped', 'error']:
             if stats[category] > 0:
@@ -69,44 +69,44 @@ def create_summary_page(df: pd.DataFrame, available_models: list[str]) -> plt.Fi
     if df.empty:
         fig, ax = plt.subplots(figsize=(16, 8), facecolor='#000000')
         ax.set_facecolor('#000000')
-        ax.text(0.5, 0.5, 'No data available', 
+        ax.text(0.5, 0.5, 'No data available',
                 horizontalalignment='center', verticalalignment='center',
                 transform=ax.transAxes, fontsize=20, color='#888888',
                 fontfamily='monospace', weight='normal')
         ax.axis('off')
         return fig
-    
+
     # Calculate dimensions for N-column layout
     model_count = len(available_models)
     rows = (model_count + COLUMNS - 1) // COLUMNS  # Ceiling division
-    
+
     # Figure dimensions - wider for columns, height based on rows
     height_per_row = min(MIN_HEIGHT_PER_ROW, MAX_HEIGHT / max(rows, 1))
     figure_height = min(MAX_HEIGHT, rows * height_per_row + FIGURE_PADDING)
-    
+
     fig, ax = plt.subplots(figsize=(FIGURE_WIDTH, figure_height), facecolor='#000000')
     ax.set_facecolor('#000000')
-    
+
     visible_model_count = 0
     max_y = 0
-    
+
     for i, model_name in enumerate(available_models):
         if model_name not in df.index:
             continue
-            
+
         row = df.loc[model_name]
-        
+
         # Extract and process model data
         amd_stats, nvidia_stats = extract_model_data(row)[:2]
-            
+
         # Calculate position in 4-column grid
         col = visible_model_count % COLUMNS
         row = visible_model_count // COLUMNS
-        
+
         # Calculate horizontal position for this column
         col_left = col * COLUMN_WIDTH + BAR_MARGIN
         col_center = col * COLUMN_WIDTH + COLUMN_WIDTH / 2
-        
+
         # Calculate vertical position for this row - start from top
         vertical_spacing = height_per_row
         y_base = (VERTICAL_SPACING_RATIO + row) * vertical_spacing
@@ -114,22 +114,22 @@ def create_summary_page(df: pd.DataFrame, available_models: list[str]) -> plt.Fi
         y_amd_bar = y_base + vertical_spacing * AMD_BAR_OFFSET       # AMD bar
         y_nvidia_bar = y_base + vertical_spacing * NVIDIA_BAR_OFFSET    # NVIDIA bar
         max_y = max(max_y, y_nvidia_bar + vertical_spacing * 0.3)
-        
+
         # Model name centered above the bars in this column
-        ax.text(col_center, y_model_name, model_name.lower(), 
-               ha='center', va='center', color='#FFFFFF', 
+        ax.text(col_center, y_model_name, model_name.lower(),
+               ha='center', va='center', color='#FFFFFF',
                fontsize=MODEL_NAME_FONT_SIZE, fontfamily='monospace', fontweight='bold')
-        
+
         # AMD label and bar in this column
         bar_height = min(0.4, vertical_spacing * BAR_HEIGHT_RATIO)
         # Draw AMD bar
         draw_text_and_bar("amd", amd_stats, y_amd_bar, col_left, bar_height, ax)
         # Draw NVIDIA bar
         draw_text_and_bar("nvidia", nvidia_stats, y_nvidia_bar, col_left, bar_height, ax)
-        
+
         # Increment counter for next visible model
         visible_model_count += 1
-    
+
     # Style the axes to be completely invisible and span full width
     ax.set_xlim(-5, 105)  # Slightly wider to accommodate labels
     ax.set_ylim(0, max_y)
@@ -142,7 +142,7 @@ def create_summary_page(df: pd.DataFrame, available_models: list[str]) -> plt.Fi
     ax.set_xticks([])
     ax.set_yticks([])
     ax.yaxis.set_inverted(True)
-    
+
     # Remove all margins to make figure stick to top
-    plt.tight_layout()    
+    plt.tight_layout()
     return fig

utils.py

@@ -5,11 +5,11 @@ from datetime import datetime
 
 class TimestampFormatter(logging.Formatter):
     """Custom formatter that matches the existing timestamp format used in print statements."""
-    
+
     def format(self, record):
         # Create timestamp in the same format as existing print statements
         timestamp = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
-        
+
         # Format the message with timestamp prefix
         if record.levelno == logging.WARNING:
             return f"WARNING: {record.getMessage()}"
@@ -22,23 +22,23 @@ class TimestampFormatter(logging.Formatter):
 def setup_logger(name="tcid", level=logging.INFO):
     """Set up logger with custom timestamp formatting to match existing print format."""
     logger = logging.getLogger(name)
-    
+
     # Avoid adding multiple handlers if logger already exists
     if logger.handlers:
         return logger
-    
+
     logger.setLevel(level)
-    
+
     # Create console handler
     handler = logging.StreamHandler(sys.stdout)
     handler.setLevel(level)
-    
+
     # Set custom formatter
     formatter = TimestampFormatter()
     handler.setFormatter(formatter)
-    
+
     logger.addHandler(handler)
-    
+
     return logger