Token Sampler Lab

Drag the temperature dial, watch tokens get sampled live, and overflow a context window. An interactive playground for how LLMs actually pick words.

Prompt

Raw Sampling Simulated tool-call lookup

Temperature 1.00

Next-Token Probability Distribution

Output Stream — sampled tokens appear here

Context Buffer — 1,000,000 token budget

0 / 1,000,000 tokens (0.0%)

CONTEXT WINDOW EXCEEDED — NOT GRACEFULLY

Variance Readout — 10 trials