|
|
import matplotlib.pyplot as plt |
|
|
import numpy as np |
|
|
|
|
|
import gradio as gr |
|
|
|
|
|
|
|
|
|
|
|
MODELS = { |
|
|
"llama" : {"passed": 14, "failed": 1, "skipped": 6, "error": 0}, |
|
|
"gemma3" : {"passed": 42, "failed": 6, "skipped": 12, "error": 0}, |
|
|
"csm" : {"passed": 0, "failed": 0, "skipped": 0, "error": 1}, |
|
|
} |
|
|
|
|
|
def plot_model_stats(model_name: str) -> plt.Figure: |
|
|
"""Draws a pie chart of model's passed, failed, skipped, and error stats.""" |
|
|
model_stats = MODELS[model_name] |
|
|
|
|
|
|
|
|
colors = { |
|
|
'passed': '#4CAF50', |
|
|
'failed': '#E53E3E', |
|
|
'skipped': '#FFD54F', |
|
|
'error': '#8B0000' |
|
|
} |
|
|
|
|
|
|
|
|
filtered_stats = {k: v for k, v in model_stats.items() if v > 0} |
|
|
|
|
|
if not filtered_stats: |
|
|
|
|
|
fig, ax = plt.subplots(figsize=(10, 8), facecolor='#000000') |
|
|
ax.set_facecolor('#000000') |
|
|
ax.text(0.5, 0.5, 'No test results available', |
|
|
horizontalalignment='center', verticalalignment='center', |
|
|
transform=ax.transAxes, fontsize=16, color='#888888', |
|
|
fontfamily='monospace', weight='normal') |
|
|
ax.set_xlim(0, 1) |
|
|
ax.set_ylim(0, 1) |
|
|
ax.axis('off') |
|
|
return fig |
|
|
|
|
|
|
|
|
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(18, 9), facecolor='#000000') |
|
|
ax1.set_facecolor('#000000') |
|
|
ax2.set_facecolor('#000000') |
|
|
|
|
|
|
|
|
chart_colors = [colors[category] for category in filtered_stats.keys()] |
|
|
|
|
|
def create_pie_chart(ax, device_label): |
|
|
|
|
|
wedges, texts, autotexts = ax.pie( |
|
|
filtered_stats.values(), |
|
|
labels=[label.lower() for label in filtered_stats.keys()], |
|
|
colors=chart_colors, |
|
|
autopct='%1.1f%%', |
|
|
startangle=90, |
|
|
explode=None, |
|
|
shadow=False, |
|
|
wedgeprops=dict(edgecolor='#1a1a1a', linewidth=0.5), |
|
|
textprops={'fontsize': 12, 'weight': 'normal', 'color': '#CCCCCC', 'fontfamily': 'monospace'} |
|
|
) |
|
|
|
|
|
|
|
|
for autotext in autotexts: |
|
|
autotext.set_color('#000000') |
|
|
autotext.set_weight('bold') |
|
|
autotext.set_fontsize(14) |
|
|
autotext.set_fontfamily('monospace') |
|
|
|
|
|
|
|
|
for text in texts: |
|
|
text.set_color('#AAAAAA') |
|
|
text.set_weight('normal') |
|
|
text.set_fontsize(13) |
|
|
text.set_fontfamily('monospace') |
|
|
|
|
|
|
|
|
ax.set_title(device_label, |
|
|
fontsize=28, weight='bold', pad=2, color='#FFFFFF', |
|
|
fontfamily='monospace') |
|
|
|
|
|
|
|
|
create_pie_chart(ax1, "AMD") |
|
|
create_pie_chart(ax2, "Nvidia") |
|
|
|
|
|
|
|
|
line_x = 0.5 |
|
|
fig.add_artist(plt.Line2D([line_x, line_x], [0.15, 0.8], |
|
|
color='#333333', linewidth=1, alpha=0.5, |
|
|
transform=fig.transFigure)) |
|
|
|
|
|
|
|
|
total_tests = sum(model_stats.values()) |
|
|
fig.suptitle(f'{model_name.lower()} β’ {total_tests} tests', |
|
|
fontsize=18, weight='normal', color='#CCCCCC', |
|
|
fontfamily='monospace', y=0.95) |
|
|
|
|
|
|
|
|
plt.tight_layout() |
|
|
plt.subplots_adjust(top=0.85, wspace=0.4) |
|
|
|
|
|
return fig |
|
|
|
|
|
def get_model_stats_summary(model_name: str) -> tuple: |
|
|
"""Get summary stats for a model (total tests, success rate, status indicator).""" |
|
|
stats = MODELS[model_name] |
|
|
total = sum(stats.values()) |
|
|
passed = stats['passed'] |
|
|
success_rate = (passed / total * 100) if total > 0 else 0 |
|
|
|
|
|
|
|
|
if success_rate >= 80: |
|
|
status_class = "success-high" |
|
|
elif success_rate >= 50: |
|
|
status_class = "success-medium" |
|
|
else: |
|
|
status_class = "success-low" |
|
|
|
|
|
return total, success_rate, status_class |
|
|
|
|
|
|
|
|
dark_theme_css = """ |
|
|
/* Global dark theme */ |
|
|
.gradio-container { |
|
|
background-color: #000000 !important; |
|
|
color: white !important; |
|
|
} |
|
|
|
|
|
/* Remove borders from all components */ |
|
|
.gr-box, .gr-form, .gr-panel { |
|
|
border: none !important; |
|
|
background-color: #000000 !important; |
|
|
} |
|
|
|
|
|
/* Sidebar styling */ |
|
|
.sidebar { |
|
|
background: linear-gradient(145deg, #111111, #1a1a1a) !important; |
|
|
border: none !important; |
|
|
padding: 25px !important; |
|
|
box-shadow: inset 2px 2px 5px rgba(0, 0, 0, 0.3) !important; |
|
|
margin: 0 !important; |
|
|
height: 100vh !important; |
|
|
position: fixed !important; |
|
|
left: 0 !important; |
|
|
top: 0 !important; |
|
|
width: 300px !important; |
|
|
} |
|
|
|
|
|
/* Enhanced model button styling */ |
|
|
.model-button { |
|
|
background: linear-gradient(135deg, #2a2a2a, #1e1e1e) !important; |
|
|
color: white !important; |
|
|
border: 2px solid transparent !important; |
|
|
margin: 2px 0 !important; |
|
|
border-radius: 5px !important; |
|
|
padding: 8px 12px !important; |
|
|
transition: all 0.4s cubic-bezier(0.4, 0, 0.2, 1) !important; |
|
|
position: relative !important; |
|
|
overflow: hidden !important; |
|
|
box-shadow: |
|
|
0 4px 15px rgba(0, 0, 0, 0.2), |
|
|
inset 0 1px 0 rgba(255, 255, 255, 0.1) !important; |
|
|
font-weight: 600 !important; |
|
|
font-size: 16px !important; |
|
|
text-transform: uppercase !important; |
|
|
letter-spacing: 0.5px !important; |
|
|
font-family: monospace !important; |
|
|
} |
|
|
|
|
|
.model-button:hover { |
|
|
background: linear-gradient(135deg, #3a3a3a, #2e2e2e) !important; |
|
|
color: #74b9ff !important; |
|
|
} |
|
|
|
|
|
.model-button:active { |
|
|
background: linear-gradient(135deg, #2a2a2a, #1e1e1e) !important; |
|
|
color: #5a9bd4 !important; |
|
|
} |
|
|
|
|
|
/* Model stats badge */ |
|
|
.model-stats { |
|
|
display: flex !important; |
|
|
justify-content: space-between !important; |
|
|
align-items: center !important; |
|
|
margin-top: 8px !important; |
|
|
font-size: 12px !important; |
|
|
opacity: 0.8 !important; |
|
|
} |
|
|
|
|
|
.stats-badge { |
|
|
background: rgba(116, 185, 255, 0.2) !important; |
|
|
padding: 4px 8px !important; |
|
|
border-radius: 10px !important; |
|
|
font-weight: 500 !important; |
|
|
font-size: 11px !important; |
|
|
color: #74b9ff !important; |
|
|
} |
|
|
|
|
|
.success-indicator { |
|
|
width: 8px !important; |
|
|
height: 8px !important; |
|
|
border-radius: 50% !important; |
|
|
display: inline-block !important; |
|
|
margin-right: 6px !important; |
|
|
} |
|
|
|
|
|
.success-high { background-color: #4CAF50 !important; } |
|
|
.success-medium { background-color: #FF9800 !important; } |
|
|
.success-low { background-color: #F44336 !important; } |
|
|
|
|
|
/* Regular button styling for non-model buttons */ |
|
|
.gr-button:not(.model-button) { |
|
|
background-color: #222222 !important; |
|
|
color: white !important; |
|
|
border: 1px solid #444444 !important; |
|
|
margin: 5px 0 !important; |
|
|
border-radius: 8px !important; |
|
|
transition: all 0.3s ease !important; |
|
|
} |
|
|
|
|
|
.gr-button:not(.model-button):hover { |
|
|
background-color: #333333 !important; |
|
|
border-color: #666666 !important; |
|
|
} |
|
|
|
|
|
/* Plot container */ |
|
|
.plot-container { |
|
|
background-color: #000000 !important; |
|
|
border: none !important; |
|
|
} |
|
|
|
|
|
/* Text elements */ |
|
|
h1, h2, h3, p, .markdown { |
|
|
color: white !important; |
|
|
} |
|
|
|
|
|
/* Sidebar header enhancement */ |
|
|
.sidebar h1 { |
|
|
background: linear-gradient(45deg, #74b9ff, #a29bfe) !important; |
|
|
-webkit-background-clip: text !important; |
|
|
-webkit-text-fill-color: transparent !important; |
|
|
background-clip: text !important; |
|
|
text-align: center !important; |
|
|
margin-bottom: 15px !important; |
|
|
font-size: 28px !important; |
|
|
font-weight: 700 !important; |
|
|
font-family: monospace !important; |
|
|
} |
|
|
|
|
|
/* Sidebar description text */ |
|
|
.sidebar p { |
|
|
text-align: center !important; |
|
|
margin-bottom: 20px !important; |
|
|
line-height: 1.5 !important; |
|
|
font-size: 14px !important; |
|
|
font-family: monospace !important; |
|
|
} |
|
|
|
|
|
.sidebar strong { |
|
|
color: #74b9ff !important; |
|
|
font-weight: 600 !important; |
|
|
font-family: monospace !important; |
|
|
} |
|
|
|
|
|
.sidebar em { |
|
|
color: #a29bfe !important; |
|
|
font-style: normal !important; |
|
|
opacity: 0.9 !important; |
|
|
font-family: monospace !important; |
|
|
} |
|
|
|
|
|
/* Remove all borders globally */ |
|
|
* { |
|
|
border-color: transparent !important; |
|
|
} |
|
|
|
|
|
/* Main content area */ |
|
|
.main-content { |
|
|
background-color: #000000 !important; |
|
|
padding: 20px !important; |
|
|
margin-left: 300px !important; |
|
|
} |
|
|
""" |
|
|
|
|
|
|
|
|
with gr.Blocks(title="Model Test Results Dashboard", css=dark_theme_css) as demo: |
|
|
|
|
|
with gr.Row(): |
|
|
|
|
|
with gr.Column(scale=1, elem_classes=["sidebar"]): |
|
|
gr.Markdown("# π€ AI Models") |
|
|
gr.Markdown("**Select a model to analyze test results**\n\n*Interactive dashboard with detailed metrics*") |
|
|
|
|
|
|
|
|
model_buttons = [] |
|
|
for model_name in MODELS.keys(): |
|
|
btn = gr.Button( |
|
|
f"{model_name.lower()}", |
|
|
variant="secondary", |
|
|
size="lg", |
|
|
elem_classes=["model-button"] |
|
|
) |
|
|
model_buttons.append(btn) |
|
|
|
|
|
|
|
|
with gr.Column(scale=4, elem_classes=["main-content"]): |
|
|
gr.Markdown("# π Test Results Dashboard") |
|
|
|
|
|
|
|
|
plot_output = gr.Plot( |
|
|
label="", |
|
|
format="png", |
|
|
elem_classes=["plot-container"] |
|
|
) |
|
|
|
|
|
|
|
|
for i, (model_name, button) in enumerate(zip(MODELS.keys(), model_buttons)): |
|
|
button.click( |
|
|
fn=lambda name=model_name: plot_model_stats(name), |
|
|
outputs=plot_output |
|
|
) |
|
|
|
|
|
|
|
|
demo.load( |
|
|
fn=lambda: plot_model_stats(list(MODELS.keys())[0]), |
|
|
outputs=plot_output |
|
|
) |
|
|
|
|
|
if __name__ == "__main__": |
|
|
demo.launch() |
|
|
|