Spaces:
Sleeping
Sleeping
Commit ·
ca5b23e
1
Parent(s): 9c28362
fix: Use production API, match real dimension keys, remove mock data
Browse filesSwitch to production API (api.trustmodel.ai), update dimension keys to match
real response (bias_safety, context_awareness, etc.), remove mock data fallback,
remove EU AI Act score and score range (N/A in production), hide badge section
until lookup is triggered.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
app.py
CHANGED
|
@@ -1,64 +1,32 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
import requests
|
| 3 |
|
| 4 |
-
API_URL = "https://api
|
| 5 |
LEADERBOARD_URL = "https://trustmodel.ai/leaderboard"
|
| 6 |
BADGE_BASE_URL = "https://trustmodel.ai/badge/model"
|
| 7 |
|
| 8 |
DIMENSIONS = [
|
| 9 |
-
("
|
| 10 |
-
("
|
| 11 |
-
("bias", "Bias Detection"),
|
| 12 |
("hallucination", "Hallucination"),
|
| 13 |
-
("
|
| 14 |
-
("
|
| 15 |
-
("
|
| 16 |
("accuracy", "Accuracy"),
|
| 17 |
-
("
|
| 18 |
-
("
|
|
|
|
| 19 |
]
|
| 20 |
|
| 21 |
-
# Mock data for development (0-100 scale matching real API) — remove when leaderboard API has data
|
| 22 |
-
MOCK_LEADERBOARD = [
|
| 23 |
-
{"id": "1", "model": "GPT-4o", "vendor": "OpenAI", "composite_score": 92.67,
|
| 24 |
-
"category_scores": {"safety": 88.44, "vulnerability": 96.44, "bias": 90.99, "hallucination": 99.31, "holistic_reasoning": 100.0, "holistic_programming": 87.72, "holistic_creativity": 92.18, "accuracy": 96.59, "latency": 76.26, "cost": 82.14},
|
| 25 |
-
"eu_ai_act_score": 91.6, "eu_ai_act_status": "PASS", "total_tests": 920,
|
| 26 |
-
"score_distribution": {"min": 77.32, "max": 107.62, "median": 94.99, "std_dev": 14.53}},
|
| 27 |
-
{"id": "2", "model": "Claude 3.5 Sonnet", "vendor": "Anthropic", "composite_score": 94.12,
|
| 28 |
-
"category_scores": {"safety": 95.10, "vulnerability": 93.20, "bias": 94.80, "hallucination": 96.30, "holistic_reasoning": 98.60, "holistic_programming": 91.00, "holistic_creativity": 93.40, "accuracy": 95.60, "latency": 80.50, "cost": 78.30},
|
| 29 |
-
"eu_ai_act_score": 93.7, "eu_ai_act_status": "PASS", "total_tests": 920,
|
| 30 |
-
"score_distribution": {"min": 78.30, "max": 108.50, "median": 95.10, "std_dev": 12.80}},
|
| 31 |
-
{"id": "3", "model": "Gemini 1.5 Pro", "vendor": "Google", "composite_score": 89.45,
|
| 32 |
-
"category_scores": {"safety": 86.00, "vulnerability": 91.50, "bias": 87.60, "hallucination": 92.90, "holistic_reasoning": 95.20, "holistic_programming": 85.70, "holistic_creativity": 88.80, "accuracy": 93.10, "latency": 82.30, "cost": 75.90},
|
| 33 |
-
"eu_ai_act_score": 88.2, "eu_ai_act_status": "PASS", "total_tests": 920,
|
| 34 |
-
"score_distribution": {"min": 72.50, "max": 103.20, "median": 90.10, "std_dev": 15.20}},
|
| 35 |
-
{"id": "4", "model": "Llama 3.1 70B", "vendor": "Meta", "composite_score": 82.30,
|
| 36 |
-
"category_scores": {"safety": 78.00, "vulnerability": 84.80, "bias": 80.20, "hallucination": 85.30, "holistic_reasoning": 88.50, "holistic_programming": 79.00, "holistic_creativity": 77.90, "accuracy": 86.40, "latency": 90.10, "cost": 95.20},
|
| 37 |
-
"eu_ai_act_score": 80.9, "eu_ai_act_status": "PASS", "total_tests": 920,
|
| 38 |
-
"score_distribution": {"min": 65.10, "max": 98.80, "median": 83.20, "std_dev": 16.50}},
|
| 39 |
-
{"id": "5", "model": "Mistral Large", "vendor": "Mistral AI", "composite_score": 85.60,
|
| 40 |
-
"category_scores": {"safety": 82.60, "vulnerability": 87.20, "bias": 84.30, "hallucination": 88.50, "holistic_reasoning": 90.80, "holistic_programming": 83.30, "holistic_creativity": 82.20, "accuracy": 89.60, "latency": 85.40, "cost": 88.70},
|
| 41 |
-
"eu_ai_act_score": 84.5, "eu_ai_act_status": "PASS", "total_tests": 920,
|
| 42 |
-
"score_distribution": {"min": 70.50, "max": 101.00, "median": 86.40, "std_dev": 14.90}},
|
| 43 |
-
{"id": "6", "model": "Cohere Command R+", "vendor": "Cohere", "composite_score": 78.90,
|
| 44 |
-
"category_scores": {"safety": 75.00, "vulnerability": 80.50, "bias": 77.70, "hallucination": 81.90, "holistic_reasoning": 84.20, "holistic_programming": 76.60, "holistic_creativity": 74.50, "accuracy": 82.00, "latency": 88.30, "cost": 91.50},
|
| 45 |
-
"eu_ai_act_score": 77.5, "eu_ai_act_status": "NEEDS REVIEW", "total_tests": 920,
|
| 46 |
-
"score_distribution": {"min": 60.80, "max": 95.50, "median": 79.80, "std_dev": 17.10}},
|
| 47 |
-
]
|
| 48 |
-
|
| 49 |
-
|
| 50 |
def fetch_leaderboard():
|
| 51 |
-
"""Fetch the latest leaderboard data from TrustModel API.
|
| 52 |
try:
|
| 53 |
response = requests.get(API_URL, timeout=10)
|
| 54 |
response.raise_for_status()
|
| 55 |
data = response.json()
|
| 56 |
-
|
| 57 |
-
if leaderboard:
|
| 58 |
-
return leaderboard
|
| 59 |
except Exception:
|
| 60 |
-
|
| 61 |
-
return MOCK_LEADERBOARD
|
| 62 |
|
| 63 |
|
| 64 |
def get_model_names(leaderboard):
|
|
@@ -101,7 +69,6 @@ def lookup_model(model_name, leaderboard):
|
|
| 101 |
result = f"## {entry['model']}\n"
|
| 102 |
result += f"**Vendor:** {entry.get('vendor', 'Unknown')}\n\n"
|
| 103 |
result += f"### TrustScore: {format_score(composite)}/100 ({trust_level})\n\n"
|
| 104 |
-
result += f"**EU AI Act Score:** {format_score(entry.get('eu_ai_act_score', 0))} — {entry.get('eu_ai_act_status', 'N/A')}\n\n"
|
| 105 |
result += f"**Total Tests:** {entry.get('total_tests', 0)}\n\n"
|
| 106 |
result += "### Dimension Scores\n\n"
|
| 107 |
result += "| Dimension | Score |\n|-----------|-------|\n"
|
|
@@ -110,10 +77,6 @@ def lookup_model(model_name, leaderboard):
|
|
| 110 |
score = scores.get(key)
|
| 111 |
result += f"| {label} | {format_score(score)} |\n"
|
| 112 |
|
| 113 |
-
dist = entry.get("score_distribution", {})
|
| 114 |
-
if dist:
|
| 115 |
-
result += f"\n**Score Range:** {format_score(dist.get('min'))} — {format_score(dist.get('max'))} (median: {format_score(dist.get('median'))})\n"
|
| 116 |
-
|
| 117 |
slug = entry["model"].lower().replace(" ", "-")
|
| 118 |
badge_code = f'[](https://trustmodel.ai/leaderboard)'
|
| 119 |
|
|
@@ -148,11 +111,6 @@ def compare_models(model_1, model_2, model_3, leaderboard):
|
|
| 148 |
result += f" **{format_score(e.get('composite_score', 0))}** |"
|
| 149 |
result += "\n"
|
| 150 |
|
| 151 |
-
result += "| **EU AI Act** |"
|
| 152 |
-
for e in entries:
|
| 153 |
-
result += f" {format_score(e.get('eu_ai_act_score', 0))} |"
|
| 154 |
-
result += "\n"
|
| 155 |
-
|
| 156 |
for key, label in DIMENSIONS:
|
| 157 |
result += f"| {label} |"
|
| 158 |
for e in entries:
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import requests
|
| 3 |
|
| 4 |
+
API_URL = "https://api.trustmodel.ai/api/leaderboards/latest/"
|
| 5 |
LEADERBOARD_URL = "https://trustmodel.ai/leaderboard"
|
| 6 |
BADGE_BASE_URL = "https://trustmodel.ai/badge/model"
|
| 7 |
|
| 8 |
DIMENSIONS = [
|
| 9 |
+
("bias_safety", "Bias & Safety"),
|
| 10 |
+
("security_vulnerability", "Security & Vulnerability"),
|
|
|
|
| 11 |
("hallucination", "Hallucination"),
|
| 12 |
+
("context_awareness", "Context Awareness"),
|
| 13 |
+
("instruction_adherence", "Instruction Adherence"),
|
| 14 |
+
("reasoning", "Reasoning"),
|
| 15 |
("accuracy", "Accuracy"),
|
| 16 |
+
("enterprise_alignment", "Enterprise Alignment"),
|
| 17 |
+
("specialization", "Specialization"),
|
| 18 |
+
("cost_latency", "Cost & Latency"),
|
| 19 |
]
|
| 20 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
def fetch_leaderboard():
|
| 22 |
+
"""Fetch the latest leaderboard data from TrustModel API."""
|
| 23 |
try:
|
| 24 |
response = requests.get(API_URL, timeout=10)
|
| 25 |
response.raise_for_status()
|
| 26 |
data = response.json()
|
| 27 |
+
return data.get("leaderboard", [])
|
|
|
|
|
|
|
| 28 |
except Exception:
|
| 29 |
+
return []
|
|
|
|
| 30 |
|
| 31 |
|
| 32 |
def get_model_names(leaderboard):
|
|
|
|
| 69 |
result = f"## {entry['model']}\n"
|
| 70 |
result += f"**Vendor:** {entry.get('vendor', 'Unknown')}\n\n"
|
| 71 |
result += f"### TrustScore: {format_score(composite)}/100 ({trust_level})\n\n"
|
|
|
|
| 72 |
result += f"**Total Tests:** {entry.get('total_tests', 0)}\n\n"
|
| 73 |
result += "### Dimension Scores\n\n"
|
| 74 |
result += "| Dimension | Score |\n|-----------|-------|\n"
|
|
|
|
| 77 |
score = scores.get(key)
|
| 78 |
result += f"| {label} | {format_score(score)} |\n"
|
| 79 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
slug = entry["model"].lower().replace(" ", "-")
|
| 81 |
badge_code = f'[](https://trustmodel.ai/leaderboard)'
|
| 82 |
|
|
|
|
| 111 |
result += f" **{format_score(e.get('composite_score', 0))}** |"
|
| 112 |
result += "\n"
|
| 113 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 114 |
for key, label in DIMENSIONS:
|
| 115 |
result += f"| {label} |"
|
| 116 |
for e in entries:
|