Mini LLM Workbench

See how a tiny language model works, token by token.

Mini LLM Workbench — learn how tiny language models tokenize text, train in small steps, and chat with a demo bot today.

How to use Mini LLM Workbench

Type a sentence in the tokenizer playground and press Tokenize to see each word and punctuation mark become a token number.
Try a sample sentence, or type your own words, and compare known words against unknown words shown as token 0.
Open the training visualizer and step through the five stages from collecting data to trying the model out.
Move the epochs slider and watch the seeded example loss value and loss curve fall.
Ask the demo chatbot about tokens, training, loss, or models, and read its fixed-rule replies.
Download the chat transcript with the Download chat button.

Tokenizer playground with a 51-entry toy vocabulary, token IDs, and known versus unknown counts.
Five-step training-concept visualizer with plain-language explanations for each stage.
Seeded example loss curve and epochs slider, same input always gives the same curve.
Rule-based demo chatbot with canned replies about tokens, training, loss, and models.
Sample sentences, copy-tokens button, and downloadable chat transcript.
Honest on-page labels throughout: a learning tool, not a real model, no weights loaded, chatbot is rule-based.

Yes, unlimited.