Spaces:
Running
Running
Update src/App.js
Browse files- src/App.js +270 -20
src/App.js
CHANGED
|
@@ -1,25 +1,275 @@
|
|
| 1 |
-
import
|
| 2 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
|
| 4 |
-
function App() {
|
| 5 |
return (
|
| 6 |
-
<div className="
|
| 7 |
-
<
|
| 8 |
-
|
| 9 |
-
<
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
className="
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
</div>
|
| 22 |
);
|
| 23 |
-
}
|
| 24 |
|
| 25 |
-
export default
|
|
|
|
| 1 |
+
import React, { useState, useEffect } from 'react';
|
| 2 |
+
|
| 3 |
+
const BenchmarkChart = () => {
|
| 4 |
+
// Real data from your CSV
|
| 5 |
+
const benchmarkData = [
|
| 6 |
+
{
|
| 7 |
+
model: "Claude 4 Sonnet",
|
| 8 |
+
direct_conversation: 26.33,
|
| 9 |
+
keyword_objective_combined: 3.13
|
| 10 |
+
},
|
| 11 |
+
{
|
| 12 |
+
model: "Claude Opus 4.1",
|
| 13 |
+
direct_conversation: 20.67,
|
| 14 |
+
keyword_objective_combined: 3.65
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
model: "Deepseek R1-0528",
|
| 18 |
+
direct_conversation: 68.67,
|
| 19 |
+
keyword_objective_combined: 48.18
|
| 20 |
+
},
|
| 21 |
+
{
|
| 22 |
+
model: "GPT 5",
|
| 23 |
+
direct_conversation: 8.33,
|
| 24 |
+
keyword_objective_combined: 3.65,
|
| 25 |
+
bio_topic_change: 23.5,
|
| 26 |
+
enhancement: 10,
|
| 27 |
+
root_problem: 4.5
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
model: "GPT 5 mini",
|
| 31 |
+
direct_conversation: 7.67,
|
| 32 |
+
keyword_objective_combined: 3.91,
|
| 33 |
+
bio_topic_change: 14.5,
|
| 34 |
+
enhancement: 5.5,
|
| 35 |
+
root_problem: 3
|
| 36 |
+
},
|
| 37 |
+
{
|
| 38 |
+
model: "GPT o3",
|
| 39 |
+
direct_conversation: 22,
|
| 40 |
+
keyword_objective_combined: 10.94
|
| 41 |
+
},
|
| 42 |
+
{
|
| 43 |
+
model: "Gemini 2.5 Pro",
|
| 44 |
+
direct_conversation: 55.67,
|
| 45 |
+
keyword_objective_combined: 41.67,
|
| 46 |
+
bio_topic_change: 53.5,
|
| 47 |
+
enhancement: 47,
|
| 48 |
+
root_problem: 26
|
| 49 |
+
},
|
| 50 |
+
{
|
| 51 |
+
model: "Grok 4",
|
| 52 |
+
direct_conversation: 68.67,
|
| 53 |
+
keyword_objective_combined: 52.6
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
model: "Llama 3.1 405B",
|
| 57 |
+
direct_conversation: 67,
|
| 58 |
+
keyword_objective_combined: 41.67
|
| 59 |
+
}
|
| 60 |
+
];
|
| 61 |
+
|
| 62 |
+
const [currentPhase, setCurrentPhase] = useState('baseline');
|
| 63 |
+
const [currentMethodIndex, setCurrentMethodIndex] = useState(0);
|
| 64 |
+
|
| 65 |
+
const synthesisMethodsOrder = ['keyword_objective_combined', 'bio_topic_change', 'enhancement', 'root_problem'];
|
| 66 |
+
|
| 67 |
+
const phases = [
|
| 68 |
+
{ key: 'baseline', label: 'Direct Conversation (Baseline)' },
|
| 69 |
+
{ key: 'additive_synthesis', label: 'Adding Synthesis Methods' }
|
| 70 |
+
];
|
| 71 |
+
|
| 72 |
+
useEffect(() => {
|
| 73 |
+
const interval = setInterval(() => {
|
| 74 |
+
setCurrentPhase(prev => prev === 'baseline' ? 'additive_synthesis' : 'baseline');
|
| 75 |
+
setCurrentMethodIndex(0); // Reset when switching phases
|
| 76 |
+
}, 12000); // Increased total cycle time to 12 seconds
|
| 77 |
+
|
| 78 |
+
return () => clearInterval(interval);
|
| 79 |
+
}, []);
|
| 80 |
+
|
| 81 |
+
useEffect(() => {
|
| 82 |
+
if (currentPhase === 'additive_synthesis') {
|
| 83 |
+
const methodInterval = setInterval(() => {
|
| 84 |
+
setCurrentMethodIndex(prev => {
|
| 85 |
+
const nextIndex = prev + 1;
|
| 86 |
+
// Stay at final state (all methods added) for longer
|
| 87 |
+
if (nextIndex > synthesisMethodsOrder.length) {
|
| 88 |
+
return synthesisMethodsOrder.length; // Stay at max for longer
|
| 89 |
+
}
|
| 90 |
+
return nextIndex;
|
| 91 |
+
});
|
| 92 |
+
}, 2000); // Slower progression - 2 seconds per method
|
| 93 |
+
return () => clearInterval(methodInterval);
|
| 94 |
+
}
|
| 95 |
+
}, [currentPhase]);
|
| 96 |
+
|
| 97 |
+
const getCurrentValue = (modelData, phase) => {
|
| 98 |
+
if (phase === 'baseline') {
|
| 99 |
+
return modelData.direct_conversation || 0;
|
| 100 |
+
} else if (phase === 'additive_synthesis') {
|
| 101 |
+
let cumulativeValue = modelData.direct_conversation || 0;
|
| 102 |
+
|
| 103 |
+
// Add each synthesis method's contribution up to currentMethodIndex
|
| 104 |
+
for (let i = 0; i < currentMethodIndex; i++) {
|
| 105 |
+
const method = synthesisMethodsOrder[i];
|
| 106 |
+
if (modelData[method] !== undefined) {
|
| 107 |
+
cumulativeValue += modelData[method];
|
| 108 |
+
}
|
| 109 |
+
}
|
| 110 |
+
|
| 111 |
+
return cumulativeValue;
|
| 112 |
+
}
|
| 113 |
+
return 0;
|
| 114 |
+
};
|
| 115 |
+
|
| 116 |
+
const getCurrentMethodsAdded = (modelData, phase) => {
|
| 117 |
+
if (phase === 'baseline') return ['Direct Conversation'];
|
| 118 |
+
|
| 119 |
+
const methods = ['Direct Conversation'];
|
| 120 |
+
for (let i = 0; i < currentMethodIndex; i++) {
|
| 121 |
+
const method = synthesisMethodsOrder[i];
|
| 122 |
+
if (modelData[method] !== undefined) {
|
| 123 |
+
methods.push(method.replace(/_/g, ' ').replace(/\b\w/g, l => l.toUpperCase()));
|
| 124 |
+
}
|
| 125 |
+
}
|
| 126 |
+
return methods;
|
| 127 |
+
};
|
| 128 |
+
|
| 129 |
+
const getBarColor = (modelData, phase) => {
|
| 130 |
+
if (phase === 'baseline') {
|
| 131 |
+
return 'from-blue-500 to-blue-600';
|
| 132 |
+
} else {
|
| 133 |
+
// Green gradient for additive synthesis
|
| 134 |
+
return 'from-green-500 to-green-600';
|
| 135 |
+
}
|
| 136 |
+
};
|
| 137 |
|
|
|
|
| 138 |
return (
|
| 139 |
+
<div className="min-h-screen bg-gradient-to-br from-slate-900 to-slate-800 p-8">
|
| 140 |
+
<div className="max-w-6xl mx-auto">
|
| 141 |
+
{/* Header */}
|
| 142 |
+
<div className="text-center mb-12">
|
| 143 |
+
<h1 className="text-4xl font-bold text-white mb-4">
|
| 144 |
+
LLM Safety Benchmark Results
|
| 145 |
+
</h1>
|
| 146 |
+
<p className="text-slate-300 text-lg">
|
| 147 |
+
SafetyBench Aug 2025 - Success Rate Comparison
|
| 148 |
+
</p>
|
| 149 |
+
|
| 150 |
+
{/* Methodology Disclaimer */}
|
| 151 |
+
<div className="mt-6 p-4 bg-yellow-900/30 border border-yellow-500/30 rounded-lg max-w-4xl mx-auto">
|
| 152 |
+
<div className="flex items-start space-x-3">
|
| 153 |
+
<div className="text-yellow-400 mt-1">⚠️</div>
|
| 154 |
+
<div className="text-left">
|
| 155 |
+
<p className="text-yellow-200 font-semibold mb-2">Methodology Note</p>
|
| 156 |
+
<p className="text-yellow-100 text-sm leading-relaxed">
|
| 157 |
+
<strong>Additive Visualization:</strong> This chart shows cumulative impact by progressively adding each synthesis method's individual success rate.
|
| 158 |
+
Values >100% represent theoretical maximum vulnerability discovery when combining multiple attack vectors.
|
| 159 |
+
Results are based on SafetyBench Aug 2025 testing methodology and should be interpreted as relative performance indicators.
|
| 160 |
+
</p>
|
| 161 |
+
</div>
|
| 162 |
+
</div>
|
| 163 |
+
</div>
|
| 164 |
+
|
| 165 |
+
<div className="mt-4 p-4 bg-slate-800 rounded-lg inline-block">
|
| 166 |
+
<p className="text-white font-semibold">
|
| 167 |
+
Current View: {phases.find(p => p.key === currentPhase)?.label}
|
| 168 |
+
</p>
|
| 169 |
+
{currentPhase === 'additive_synthesis' && currentMethodIndex > 0 && (
|
| 170 |
+
<p className="text-slate-300 text-sm mt-1">
|
| 171 |
+
Adding Method {currentMethodIndex}: {synthesisMethodsOrder[currentMethodIndex - 1]?.replace(/_/g, ' ').replace(/\b\w/g, l => l.toUpperCase())}
|
| 172 |
+
</p>
|
| 173 |
+
)}
|
| 174 |
+
</div>
|
| 175 |
+
</div>
|
| 176 |
+
|
| 177 |
+
{/* Chart Container */}
|
| 178 |
+
<div className="bg-white rounded-2xl shadow-2xl p-8">
|
| 179 |
+
<div className="space-y-6">
|
| 180 |
+
{benchmarkData.map((modelData, index) => {
|
| 181 |
+
const currentValue = getCurrentValue(modelData, currentPhase);
|
| 182 |
+
const baselineValue = modelData.direct_conversation;
|
| 183 |
+
const maxValue = 100; // Increased max scale since we're adding values
|
| 184 |
+
const barWidth = (currentValue / maxValue) * 100;
|
| 185 |
+
const methodsAdded = getCurrentMethodsAdded(modelData, currentPhase);
|
| 186 |
+
const totalGain = currentValue - baselineValue;
|
| 187 |
+
|
| 188 |
+
return (
|
| 189 |
+
<div key={modelData.model} className="relative">
|
| 190 |
+
{/* Model Name and Methods */}
|
| 191 |
+
<div className="flex items-center justify-between mb-2">
|
| 192 |
+
<div>
|
| 193 |
+
<h3 className="font-semibold text-gray-800 text-lg">
|
| 194 |
+
{modelData.model}
|
| 195 |
+
</h3>
|
| 196 |
+
<p className="text-sm text-gray-600">
|
| 197 |
+
{methodsAdded.join(' + ')}
|
| 198 |
+
</p>
|
| 199 |
+
</div>
|
| 200 |
+
<div className="text-right">
|
| 201 |
+
<span className="text-2xl font-bold text-gray-700">
|
| 202 |
+
{currentValue.toFixed(1)}%
|
| 203 |
+
</span>
|
| 204 |
+
{currentPhase === 'additive_synthesis' && totalGain > 0 && (
|
| 205 |
+
<div className="text-sm font-semibold text-green-600">
|
| 206 |
+
+{totalGain.toFixed(1)}% total gain
|
| 207 |
+
</div>
|
| 208 |
+
)}
|
| 209 |
+
</div>
|
| 210 |
+
</div>
|
| 211 |
+
|
| 212 |
+
{/* Progress Bar */}
|
| 213 |
+
<div className="relative h-12 bg-gray-200 rounded-full overflow-hidden">
|
| 214 |
+
<div
|
| 215 |
+
className={`h-full bg-gradient-to-r ${getBarColor(modelData, currentPhase)} rounded-full transition-all duration-[1800ms] ease-in-out flex items-center justify-end pr-4`}
|
| 216 |
+
style={{ width: `${Math.max(barWidth, 5)}%` }}
|
| 217 |
+
>
|
| 218 |
+
<div className="text-white font-semibold text-sm">
|
| 219 |
+
{currentValue > 8 ? `${currentValue.toFixed(1)}%` : ''}
|
| 220 |
+
</div>
|
| 221 |
+
</div>
|
| 222 |
+
</div>
|
| 223 |
+
|
| 224 |
+
{/* Method Breakdown */}
|
| 225 |
+
{currentPhase === 'additive_synthesis' && currentMethodIndex > 0 && (
|
| 226 |
+
<div className="mt-2 text-xs text-gray-500 space-y-1">
|
| 227 |
+
<div>Baseline: {baselineValue.toFixed(1)}%</div>
|
| 228 |
+
{synthesisMethodsOrder.slice(0, currentMethodIndex).map(method => {
|
| 229 |
+
if (modelData[method] !== undefined) {
|
| 230 |
+
return (
|
| 231 |
+
<div key={method}>
|
| 232 |
+
+ {method.replace(/_/g, ' ')}: {modelData[method].toFixed(1)}%
|
| 233 |
+
</div>
|
| 234 |
+
);
|
| 235 |
+
}
|
| 236 |
+
return null;
|
| 237 |
+
})}
|
| 238 |
+
</div>
|
| 239 |
+
)}
|
| 240 |
+
</div>
|
| 241 |
+
);
|
| 242 |
+
})}
|
| 243 |
+
</div>
|
| 244 |
+
|
| 245 |
+
{/* Legend */}
|
| 246 |
+
<div className="mt-8 flex justify-center space-x-6 flex-wrap">
|
| 247 |
+
<div className="flex items-center space-x-2">
|
| 248 |
+
<div className="w-4 h-4 bg-gradient-to-r from-blue-500 to-blue-600 rounded"></div>
|
| 249 |
+
<span className="text-gray-700">Baseline (Direct Conversation)</span>
|
| 250 |
+
</div>
|
| 251 |
+
<div className="flex items-center space-x-2">
|
| 252 |
+
<div className="w-4 h-4 bg-gradient-to-r from-green-500 to-green-600 rounded"></div>
|
| 253 |
+
<span className="text-gray-700">Additive Synthesis Methods</span>
|
| 254 |
+
</div>
|
| 255 |
+
</div>
|
| 256 |
+
</div>
|
| 257 |
+
|
| 258 |
+
{/* Footer Info */}
|
| 259 |
+
<div className="mt-8 text-center text-slate-400 space-y-2">
|
| 260 |
+
<p>Animation cycles every 12 seconds: baseline (6s) → progressive method addition (2s each) → final results (4s)</p>
|
| 261 |
+
<p className="text-sm">
|
| 262 |
+
Data from SafetyBench Aug 2025 • Synthesis methods test different attack vectors
|
| 263 |
+
</p>
|
| 264 |
+
<div className="text-xs mt-4 max-w-4xl mx-auto space-y-1">
|
| 265 |
+
<p><strong>Interpretation:</strong> Each synthesis method represents a different approach to testing model vulnerabilities</p>
|
| 266 |
+
<p><strong>Additive Display:</strong> Shows theoretical maximum impact when combining all available synthesis techniques</p>
|
| 267 |
+
<p><strong>Baseline:</strong> Direct conversation represents standard prompting without augmentation techniques</p>
|
| 268 |
+
</div>
|
| 269 |
+
</div>
|
| 270 |
+
</div>
|
| 271 |
</div>
|
| 272 |
);
|
| 273 |
+
};
|
| 274 |
|
| 275 |
+
export default BenchmarkChart;
|