Language models · a library of possibilities

RuseN-Gram

Let a book teach a language model. Switch from a detective story to a sonnet, then follow the counts behind every next word.

Change book · Alice’s Adventures in Wonderland

Choose a voice

12 complete source editions · English

Novels & stories

Poetry collections

Alice’s Adventures in Wonderland

Lewis Carroll · 170 KB source edition

Sampling path

Position 0 / 0

Sample a token to begin. Click any token later to inspect the context after it.

Each step updates the context and next-token distribution. You can also choose a token directly from the table.

Loading / computing locally…

Source & edition · Download original text
Public domain in the USA — Project Gutenberg edition. Original license retained in the download.

How much context?

Larger n retains longer phrases and can reproduce passages. At n = 1 the model only knows word frequencies. Smoothing gives every known token a chance, including unlikely continuations.

Inspect / edit the training text

Gutenberg boilerplate, front matter and identified back matter are excluded from training. Literary boundaries are recorded for each edition. Blank lines delimit paragraphs; transitions never cross them. Words are lowercased, punctuation and line breaks are separate tokens.

Count-based generation runs in a local worker. No text is sent to a model service. A fixed corpus, order, smoothing and seed reproduce the same continuation in single steps or batches. Changing the starting context, seed, order or smoothing starts a new path; changing batch size does not. This model learns local phrasing, not meaning.