GEITJE

GEITJE was a Dutch large language model created by Bram Rijgersberg through full-parameter continued pretraining of Mistral 7B on 10 billion Dutch tokens from the Gigacorpus and MADLAD-400, with chat variants fine-tuned on Dutch conversational data. It was among the first serious Dutch LLM adaptations and spawned the GEITje 7B Ultra preference-aligned continuation.

Its history is as notable as its weights: Dutch anti-piracy group BREIN pressured the developer over copyrighted books in the training data, and the model weights were voluntarily removed from Hugging Face in early 2025 rather than face litigation. Included here as a historical entry with that lifecycle caveat; do not expect downloadable weights.