evaluate / app.py
priyanka0394's picture
Claude Opus 4.6 (1M context)
fix: Add !important to hero capsule styles for HuggingFace compatibility
d6e477e
Raw History Blame Contribute Delete
17 kB
import base64
import gradio as gr
import requests
# Base64 encode logo for inline embedding (Gradio /file= route is unreliable)
# Read SVG and replace black fill with white for dark banner background
with open("logo-full.svg", "r") as _f:
_svg = _f.read().replace('fill="black"', 'fill="white"')
LOGO_B64 = base64.b64encode(_svg.encode()).decode()
API_URL = "https://api.trustmodel.ai/api/leaderboards/latest/"
LEADERBOARD_URL = "https://trustmodel.ai/leaderboard"
BADGE_BASE_URL = "https://trustmodel.ai/badge/model"
DIMENSIONS = [
("bias_safety", "Bias & Safety"),
("security_vulnerability", "Security & Vulnerability"),
("hallucination", "Hallucination"),
("context_awareness", "Context Awareness"),
("instruction_adherence", "Instruction Adherence"),
("reasoning", "Reasoning"),
("accuracy", "Accuracy"),
("enterprise_alignment", "Enterprise Alignment"),
("specialization", "Specialization"),
("cost_latency", "Cost & Latency"),
]
# Generate CSS classes for every possible bar width (0-100) and each color
BAR_COLORS = {
"green-hi": "#059669",
"green": "#10b981",
"yellow": "#f59e0b",
"red": "#ef4444",
}
bar_width_css = ""
for w in range(101):
bar_width_css += f".bar-w-{w} {{ width: {w}% !important; }}\n"
bar_color_css = ""
for name, hex_val in BAR_COLORS.items():
bar_color_css += f".bar-{name} {{ background-color: {hex_val} !important; }}\n"
CUSTOM_CSS = f"""
{bar_width_css}
{bar_color_css}
.header-banner {{
background: linear-gradient(180deg, hsl(210,55%,12%) 0%, hsl(210,50%,18%) 50%, hsl(210,55%,14%) 100%);
border-radius: 16px;
padding: 40px 32px;
margin-bottom: 8px;
text-align: center;
position: relative;
overflow: hidden;
}}
.header-banner::before {{
content: '';
position: absolute;
inset: 0;
background: radial-gradient(ellipse at center, hsla(190,100%,50%,0.08) 0%, transparent 70%);
pointer-events: none;
}}
.header-banner h1 {{
color: #fff !important;
font-size: 1.8rem !important;
margin: 0 0 8px 0 !important;
font-weight: 800 !important;
position: relative;
}}
.header-banner p {{
color: rgba(255,255,255,0.6) !important;
font-size: 0.95rem !important;
margin: 0 !important;
position: relative;
}}
.header-banner a {{
color: hsl(190,100%,60%) !important;
text-decoration: none !important;
font-weight: 600 !important;
}}
.header-banner a:hover {{
text-decoration: underline !important;
}}
.stat-row {{
display: flex;
gap: 10px;
justify-content: center;
margin-top: 20px;
flex-wrap: wrap;
position: relative;
}}
.stat-pill {{
background: rgba(255,255,255,0.12) !important;
border: 1px solid rgba(255,255,255,0.25) !important;
border-radius: 20px;
padding: 5px 14px;
color: rgba(255,255,255,0.9) !important;
font-size: 0.75rem;
font-weight: 500;
}}
.score-hero {{
text-align: center;
padding: 24px;
border-radius: 12px;
background: linear-gradient(135deg, #f0fdf4 0%, #ecfeff 100%);
border: 1px solid #d1fae5;
margin: 12px 0;
}}
.score-value {{
font-size: 3rem !important;
font-weight: 800 !important;
color: #059669 !important;
line-height: 1 !important;
}}
.score-label {{
font-size: 1rem !important;
color: #6b7280 !important;
margin-top: 4px !important;
}}
.trust-level {{
display: inline-block;
margin-top: 8px;
padding: 4px 14px;
border-radius: 20px;
font-size: 0.85rem;
font-weight: 600;
}}
.level-safe {{ background: #d1fae5 !important; color: #065f46 !important; }}
.level-trusted {{ background: #a7f3d0 !important; color: #064e3b !important; }}
.level-caution {{ background: #fef3c7 !important; color: #92400e !important; }}
.level-risk {{ background: #fee2e2 !important; color: #991b1b !important; }}
.dimension-bar {{
display: flex;
align-items: center;
gap: 12px;
padding: 6px 0;
}}
.dim-label {{
width: 180px;
font-size: 0.85rem !important;
color: #374151 !important;
font-weight: 500 !important;
text-align: right;
flex-shrink: 0;
}}
.dim-track {{
flex: 1;
height: 10px;
background: #e5e7eb !important;
border-radius: 5px;
overflow: hidden;
}}
.dim-fill {{
height: 100%;
border-radius: 5px;
}}
.dim-score {{
width: 45px;
font-size: 0.8rem !important;
color: #6b7280 !important;
font-weight: 600 !important;
}}
.badge-card {{
background: #f8fafc;
border: 2px dashed #cbd5e1;
border-radius: 12px;
padding: 20px;
text-align: center;
margin: 12px 0;
}}
.badge-card img {{
margin: 12px auto;
display: block;
}}
.badge-label {{
font-size: 0.85rem !important;
color: #64748b !important;
margin-bottom: 8px;
font-weight: 600 !important;
}}
.badge-hint {{
font-size: 0.75rem !important;
color: #94a3b8 !important;
margin-top: 8px;
}}
.footer-bar {{
text-align: center;
padding: 16px;
margin-top: 8px;
font-size: 0.8rem;
color: #9ca3af;
}}
.footer-bar a {{
color: #6b7280 !important;
text-decoration: none !important;
margin: 0 8px;
}}
.footer-bar a:hover {{
color: #0ea5e9 !important;
}}
.footer-link {{
font-size: 0.75rem !important;
color: #94a3b8 !important;
text-decoration: none !important;
font-weight: 500 !important;
transition: color 0.2s ease !important;
}}
.footer-link:hover {{
color: #0ea5e9 !important;
}}
.lookup-row {{
display: flex;
align-items: flex-end !important;
gap: 12px;
}}
.compare-card {{
text-align: center;
padding: 16px 24px;
border-radius: 12px;
background: #f8fafc;
border: 1px solid #e2e8f0;
min-width: 160px;
}}
.compare-score {{
font-size: 2rem !important;
font-weight: 800 !important;
color: #059669 !important;
}}
.compare-name {{
font-size: 0.8rem !important;
color: #64748b !important;
font-weight: 600 !important;
}}
"""
def fetch_leaderboard():
try:
response = requests.get(API_URL, timeout=10)
response.raise_for_status()
data = response.json()
return data.get("leaderboard", [])
except Exception:
return []
def get_model_names(leaderboard):
return [entry["model"] for entry in leaderboard]
def format_score(score):
if score is None:
return "N/A"
return f"{score:.1f}"
def get_trust_level(score):
if score >= 90:
return "Highly Trusted", "level-trusted"
elif score >= 75:
return "Generally Safe", "level-safe"
elif score >= 60:
return "Use with Caution", "level-caution"
else:
return "High Risk", "level-risk"
def get_bar_color_hex(score):
if score >= 90:
return "#22c55e"
elif score >= 75:
return "#3b82f6"
elif score >= 60:
return "#f97316"
else:
return "#ef4444"
def lookup_model(model_name, leaderboard):
if not model_name:
return "Please select a model.", "", ""
entry = next((e for e in leaderboard if e["model"] == model_name), None)
if not entry:
return "Model not found.", "", ""
scores = entry.get("category_scores", {})
composite = entry.get("composite_score", 0)
trust_level, level_class = get_trust_level(composite)
vendor = entry.get("vendor", "Unknown")
total_tests = entry.get("total_tests", 0)
slug = entry["model"].lower().replace(" ", "-").replace("(", "").replace(")", "")
badge_url = f"{BADGE_BASE_URL}/{slug}.svg"
badge_code = f'[![TrustScore]({badge_url})](https://trustmodel.ai/leaderboard)'
# Score hero
result = f"""<div class="score-hero">
<div class="score-value">{format_score(composite)}</div>
<div class="score-label">out of 100 &middot; {vendor} &middot; {total_tests} tests</div>
<div class="trust-level {level_class}">{trust_level}</div>
</div>\n\n"""
# Dimension bars with inline styles
result += "<h3>Dimension Scores</h3>\n"
for key, label in DIMENSIONS:
score = scores.get(key)
if score is not None:
color = get_bar_color_hex(score)
w = int(round(score))
result += f'''<div style="display:flex;align-items:center;gap:12px;padding:6px 0;">
<span style="width:180px;font-size:0.85rem;color:#374151;font-weight:500;text-align:right;flex-shrink:0;">{label}</span>
<span style="flex:1;height:10px;background:#e5e7eb;border-radius:5px;overflow:hidden;"><span style="display:block;height:100%;width:{w}%;background-color:{color};border-radius:5px;"></span></span>
<span style="width:45px;font-size:0.8rem;color:#6b7280;font-weight:600;">{score:.1f}</span>
</div>\n'''
# Badge preview
result += f"""\n<div class="badge-card">
<div class="badge-label">Badge Preview</div>
<a href="https://trustmodel.ai/leaderboard"><img src="{badge_url}" alt="TrustScore: {format_score(composite/10)}/10"></a>
<div class="badge-hint">Copy the markdown below to add this badge to your model card</div>
</div>"""
return result, badge_code, badge_url
def compare_models(model_1, model_2, model_3, leaderboard):
models = [m for m in [model_1, model_2, model_3] if m]
if len(models) < 2:
return "Please select at least 2 models to compare."
entries = []
for name in models:
entry = next((e for e in leaderboard if e["model"] == name), None)
if entry:
entries.append(entry)
if len(entries) < 2:
return "Could not find selected models."
# Score cards
result = '<div style="display:flex;gap:16px;justify-content:center;margin-bottom:20px;flex-wrap:wrap;">'
for e in entries:
composite = e.get("composite_score", 0)
trust_level, level_class = get_trust_level(composite)
score_color = get_bar_color_hex(composite)
result += f"""<div style="text-align:center;padding:16px 24px;border-radius:12px;background:#f8fafc;border:1px solid #e2e8f0;min-width:160px;">
<div style="font-size:0.8rem;color:#64748b;font-weight:600;">{e['model']}</div>
<div style="font-size:2rem;font-weight:800;color:{score_color};">{format_score(composite)}</div>
<span class="trust-level {level_class}">{trust_level}</span>
</div>"""
result += "</div>\n\n"
# Comparison table as HTML
result += '<h3>Dimension Comparison</h3>'
result += '<table style="width:100%;border-collapse:collapse;font-size:0.85rem;margin-top:12px;">'
result += '<thead><tr style="border-bottom:2px solid #e2e8f0;text-align:left;">'
result += '<th style="padding:8px 12px;">Dimension</th>'
for e in entries:
result += f'<th style="padding:8px 12px;text-align:center;">{e["model"]}</th>'
result += '</tr></thead><tbody>'
result += '<tr style="background:#f0fdf4;font-weight:700;border-bottom:1px solid #e2e8f0;">'
result += '<td style="padding:8px 12px;">TrustScore</td>'
for e in entries:
cs = e.get("composite_score", 0)
result += f'<td style="padding:8px 12px;text-align:center;color:{get_bar_color_hex(cs)};">{format_score(cs)}</td>'
result += '</tr>'
for key, label in DIMENSIONS:
scores_list = []
for e in entries:
score = e.get("category_scores", {}).get(key)
scores_list.append(score)
best = max((s for s in scores_list if s is not None), default=None)
result += '<tr style="border-bottom:1px solid #f1f5f9;">'
result += f'<td style="padding:8px 12px;color:#374151;">{label}</td>'
for score in scores_list:
color = get_bar_color_hex(score) if score is not None else "#6b7280"
is_best = score is not None and score == best and len(entries) > 1
weight = "700" if is_best else "600"
result += f'<td style="padding:8px 12px;text-align:center;font-weight:{weight};color:{color};">{format_score(score)}</td>'
result += '</tr>'
result += '</tbody></table>'
best_entry = max(entries, key=lambda e: e.get("composite_score", 0))
result += f'<p style="margin-top:16px;font-weight:700;font-size:0.95rem;">Winner: {best_entry["model"]} with TrustScore {format_score(best_entry.get("composite_score", 0))}/100</p>'
return result
# Fetch leaderboard data on startup
leaderboard_data = fetch_leaderboard()
model_names = get_model_names(leaderboard_data)
model_count = len(model_names)
theme = gr.themes.Soft(
primary_hue="cyan",
secondary_hue="blue",
neutral_hue="slate",
font=("Inter", "system-ui", "sans-serif"),
)
# Base64 encode favicon SVG for head injection
with open("favicon.svg", "r") as _f:
FAVICON_B64 = base64.b64encode(_f.read().encode()).decode()
FAVICON_HEAD = f'<link rel="icon" type="image/svg+xml" href="data:image/svg+xml;base64,{FAVICON_B64}">'
with gr.Blocks(title="TrustModel — AI Evaluation", css=CUSTOM_CSS, theme=theme, head=FAVICON_HEAD) as demo:
# Header
gr.HTML(f"""<div class="header-banner">
<img src="data:image/svg+xml;base64,{LOGO_B64}" alt="TrustModel" style="height:40px;margin:0 auto 14px;display:block;position:relative;">
<h1>AI Evaluation Space</h1>
<p>Look up TrustScores, compare AI models, and get badge embed code for your model card.</p>
<div class="stat-row">
<span class="stat-pill">Free — No Account Needed</span>
<span class="stat-pill">{model_count} Foundation Models</span>
<span class="stat-pill">OpenAI · Google · Anthropic · Meta · MistralAI</span>
</div>
<p style="margin-top:12px !important;font-size:0.85rem !important;">
<a href="https://trustmodel.ai">TrustModel.ai</a>
<a href="https://trustmodel.ai/developers">API &amp; SDKs</a> &nbsp;&middot;&nbsp;
<a href="https://trustmodel.ai/leaderboard">Full Leaderboard</a>
</p>
</div>""")
with gr.Tab("Model Lookup"):
gr.Markdown("Select a model to view its TrustScore across all 10 trust dimensions and get a badge for your model card.")
if not model_names:
gr.HTML('<p style="color:#ef4444;font-weight:600;text-align:center;">Unable to load models. Please refresh the page.</p>')
model_dropdown = gr.Dropdown(
choices=model_names,
label="Select a Model",
interactive=True,
)
result_md = gr.HTML()
badge_section = gr.Column(visible=False)
with badge_section:
gr.Markdown("### Add to Your Model Card")
badge_code = gr.Code(
label="Copy this markdown and paste into your HuggingFace model card",
language=None,
interactive=False,
)
gr.HTML('<div style="text-align:center;margin-top:12px;"><a href="https://trustmodel.ai/leaderboard" target="_blank" style="display:inline-block;padding:8px 24px;background:#0ea5e9;color:#fff;border-radius:8px;font-size:0.9rem;font-weight:600;text-decoration:none;">View Full Rankings on TrustModel.ai &rarr;</a></div>')
def on_lookup(model_name):
result, badge, badge_url = lookup_model(model_name, leaderboard_data)
show = bool(badge)
return result, badge, gr.update(visible=show)
model_dropdown.change(
fn=on_lookup,
inputs=[model_dropdown],
outputs=[result_md, badge_code, badge_section],
)
with gr.Tab("Compare Models"):
gr.Markdown("Select 2 or 3 models to compare their TrustScores side by side. Best score per dimension is highlighted.")
with gr.Row():
compare_1 = gr.Dropdown(choices=model_names, label="Model 1", interactive=True)
compare_2 = gr.Dropdown(choices=model_names, label="Model 2", interactive=True)
compare_3 = gr.Dropdown(choices=model_names, label="Model 3", interactive=True)
compare_btn = gr.Button("Compare Models", variant="primary")
compare_result = gr.HTML()
compare_btn.click(
fn=lambda m1, m2, m3: compare_models(m1, m2, m3, leaderboard_data),
inputs=[compare_1, compare_2, compare_3],
outputs=[compare_result],
)
# Footer
gr.HTML("""<div style="position:fixed;bottom:0;left:0;right:0;background:#fff;border-top:1px solid #e2e8f0;padding:12px 0;text-align:center;z-index:100;">
<div style="display:flex;align-items:center;justify-content:center;gap:8px;flex-wrap:wrap;">
<strong style="font-size:0.8rem;color:#334155;">TrustModel Inc.</strong>
<a href="https://trustmodel.ai" target="_blank" class="footer-link">Website</a>
<span style="font-size:0.75rem;color:#d1d5db;">&middot;</span>
<a href="https://trustmodel.ai/leaderboard" target="_blank" class="footer-link">Leaderboard</a>
<span style="font-size:0.75rem;color:#d1d5db;">&middot;</span>
<a href="https://trustmodel.ai/developers" target="_blank" class="footer-link">Developers</a>
<span style="font-size:0.75rem;color:#d1d5db;">&middot;</span>
<a href="https://trustmodel.ai/contact" target="_blank" class="footer-link">Contact</a>
</div>
</div>""")
demo.launch(theme=theme)