tiny_schiller

A drop-in German drama corpus you plug straight into a small language model, right here in your browser or in your own project, in just a few lines of code.

»Der Mensch spielt nur, wo er in voller Bedeutung des Worts Mensch ist, und er ist nur da ganz Mensch, wo er spielt.« Friedrich Schiller

Step 1 · Load the corpus

One line reaches the data. No parser, no tokenizer to train, no container stack.

Step 1 · Load the corpus

What you just loaded

0characters

0dramas

0personas

2.07 MB · eleven Schiller dramas · 89 named personas — all in a single file.

Step 1 · Load the corpus

Tokenized three ways

characterGPT-2 BPEcl100k_base click to compare

Each token stream ships precomputed, so a small language model reads the corpus directly.
No tokenizer to build, no cleaning to redo; the file is ready to train on.

Step 1 · Load the corpus

What that one line replaced

Everything you would normally do before a single token reaches a model:

○Pick an edition (DTA? DraCor? Gutenberg-DE?)
○Write a TEI/XML parser and handle the namespaces
○Reconcile three speaker-tag styles
○Fix legacy umlaut and quote-glyph encodings
○Strip narrow no-break spaces (U+202F)
○Normalize whitespace and rebuild the corpus
○Compare tokenizers and precompute the splits

tiny_schiller ships every step above already done. One line, and all the data you need is digested and ready to be used on your machine.

Step 2 · Run the model

The model comes online right here in your browser — a char-level net trained on that very file. No server, no API key.

tiny_schiller: Der Mensch spielt nur, wo er in voller Bedeutung des Worts Mensch ist,

A character-level model trained only on Schiller's dramas, running on your CPU alone — it invents, it does not quote.