🚀 v1.0.0 — Initial Public Release

🧠 Transformer Explainer v1.0.0

The first public release of the interactive LLM Attention Simulation.

✨ Features

• 🔤 Token Embedding Visualization — Unicode-aware tokenization with positional encoding
• 🔑 Q/K/V Projection Inspector — Examine Query, Key, Value vectors per attention head
• 🗺️ Masked Self-Attention Heatmap — Interactive attention matrix with causal masking
• 📊 Next-Token Probability Distribution — Real-time softmax output with candidate tokens
• ⚡ Autoregressive Text Generation — Step-by-step token generation with attention updates
• 🎛️ Adjustable Sampling — Temperature (0.3–1.5), Top-k (1–12), generation length controls
• 👁️ Multi-Head Attention View — Switch between heads to compare learned patterns

🏗️ Simulated Architecture

Param	Value
Model	GPT-2 Small
Layers	12
Attention Heads	4
Hidden Size	16
Causal Mask	Active

🛠️ Tech

• Pure HTML + CSS + Vanilla JavaScript
• Zero dependencies, no build step
• Deployed on Vercel

🔗 Links

• 🚀 Live Demo: simulasillm.vercel.app
• 📄 Research Paper: paper-llm-attention.vercel.app