← All Modules / Module 1 / Session 1.4

Perplexity Calculator

Calculate and compare perplexity scores across different language models

Test Text Input

0
Total Tokens
0.00
Avg Log Prob
0.00
Cross-Entropy

Key Insights

  • Enter text to see perplexity calculations
  • Lower perplexity indicates better language modeling
  • Compare different models on the same test text

Model Comparison

Understanding Perplexity

  • Definition: 2^(cross-entropy) - measures model uncertainty
  • Interpretation: Average number of equally likely next words
  • Lower is better: Less confused by the text
  • Typical ranges: N-grams: 50-200, Neural: 20-50, Modern LLMs: <10
  • Comparison tool: Use same test set for fair evaluation

Word-by-Word Analysis

How Perplexity is Calculated

Step 1: For each word, the model assigns a probability P(word | context)

Step 2: Calculate log probability: log₂(P(word | context))

Step 3: Average all log probabilities across the text

Step 4: Perplexity = 2^(-average_log_probability)

Calculate perplexity to see word-by-word breakdown