Top-P Magic

Vocabulary selection control

Explore how top-p affects AI vocabulary selection and word diversity.
See how nucleus sampling shapes response quality and variety in real-time.

← Back to Home
🎛️ Top-P Parameter Demo
📝 Prompt:
🔧 Settings:
  • Fixed: temperature = 0.7
  • Variable: top_p (0.1 → 1.0)
🎯 What you'll see:

How top-p affects vocabulary restriction vs exploration

Try different prompts to see how top-p affects vocabulary selection!
🎯 Understanding Top-P (Nucleus Sampling)
🔒 Low Top-P (0.1-0.3)

Only considers the most probable words (top 10-30%). Conservative vocabulary choices.

⚖️ Medium Top-P (0.5-0.9)

Balanced approach considering 50-90% of likely words. Good for most applications.

🌐 High Top-P (0.9-1.0)

Considers nearly all vocabulary. Maximum word diversity but maintains relevance.

💡 Use Cases

Low: Formal writing, technical docs | High: Creative writing, poetry, diverse descriptions