Tiny_Schiller: A German Literary Corpus for Small Language Models

Tiny_schiller is a single-file corpus of 2.07 megabytes, containing eleven public-domain dramas by Schiller. It serves as a direct alternative to Karpathy's tiny_shakespeare, aimed at facilitating small language model prototyping, fine-tuning, education, and research within German literature. The corpus is derived from DraCor's GerDraCor export (CC0) and has been processed using deterministic parser engineering, negating the need for pre-training parser adjustments. It features character-level, GPT-2 byte-pair encoding, cl100k_base tokenization splits, an instruction-formatted dialogue-completion split, and 89 persona splits at the character level, all accessible via a single HuggingFace command. This provides a streamlined way to utilize German literary texts, addressing a gap in existing, more extensive German literary corpora that require extensive preprocessing.