Language Model Simulator
See the idea behind predicting the next word. This is a tiny teaching model, not a large language model: it counts which token follows which in a short text, then writes by sampling from those odds. It runs in your browser and makes no network calls.
What you can do
- Words or characters as tokens, each with its ID
- The odds of every next token, before and after temperature
- Sampling at any temperature, step by step
- Built-in training texts, or paste your own
- Shows when the model has to fall back to a shorter context