TearedModels/

conlangcrafter-cpt-bd412d52-hard-typo2

Continued-pretraining (CPT) corpus in a constructed language generated by the ConlangCrafter pipeline (language id bd412d52).

This dataset exists to give language-model fine-tuning runs a demonstrably out-of-distribution target. Because the language was synthesized by an LLM pipeline after public model pretraining cutoffs and never published in any form before, no large pretrained model has seen it. CPT against this… See the full description on the dataset page: https://huggingface.co/datasets/TearedModels/conlangcrafter-cpt-bd412d52-hard-typo2.