🧪 LLM Sampling Mechanics • Softmax Logits Simulator • OpenAI & Anthropic SDKs
LLM Temperature & Top-P Visualizer
Stop guessing sampling parameters. Visualize in real time how Temperature and Top-P flatten or sharpen token probability distributions before calling expensive APIs.
Interactive Hyperparameters
Temperature: 0.0 (Deterministic Argmax)
0.0 (Greedy / Pure Logic) 0.7 (Standard) 1.5 (High Entropy / Creative)
Top-P (Nucleus Sampling Cutoff): 1.0 (All Valid Tokens)
0.1 (Strict Top 10% Mass) 0.5 (Balanced Cutoff) 1.0 (No Cutoff)
Frequency Penalty: 0.0
Next-Token Softmax Probability Distribution
Context: "The capital of France is..."
Production Python API Payload