HTML Extraction

Automatic HTML code extraction from LLM responses.

Overview

When LLM responses contain HTML code blocks, PraisonAI Bench automatically extracts and saves them as .html files, organized by model.

How It Works

┌─────────────────┐
│  LLM Response   │  Contains ```html ... ```
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  Auto-Detection │  Detects HTML blocks
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  Extraction     │  Extracts clean HTML
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  Save to File   │  output/{model}/{test}.html
└─────────────────┘

Supported Formats

PraisonAI Bench auto-detects HTML in multiple formats:

Format	Example
Markdown-wrapped	html ...
Truncated blocks	Incomplete responses
Raw HTML	Direct HTML content

Output Organization

HTML files are organized by model:

output/
├── gpt-4o/
│   ├── rotating_cube_simulation.html
│   └── particle_system.html
├── xai/
│   └── grok-code-fast-1/
│       ├── terrain_simulation.html
│       └── solar_system.html
└── benchmark_results_20250829_160426.json

Extract from Existing Results

Extract HTML from previously saved benchmark results:

# Extract HTML from existing benchmark results
praisonaibench --extract output/benchmark_results_20250829_160426.json

# Works with any benchmark results JSON file
praisonaibench --extract my_results.json

Use Cases

Perfect for benchmarking: - Three.js Visualizations - 3D graphics and animations - React Components - UI component generation - Interactive Demos - Web applications - Charts & Graphs - Data visualization - Game Development - Browser games