tostideluxekaas/toxic-dpo-v0.2-dutch

Toxic DPO v0.2 - Dutch Translation

Dataset Description

This is a direct machine-translated Dutch version of the original dataset unalignment/toxic-dpo-v0.2.

Translation method: English → Dutch using the Helsinki-NLP/opus-mt-en-nl model from the MarianMTModel translations. No manual edits, additions or filtering were applied besides the automated translation. Data set is checked on NULL values and duplicates. All fields (prompt, chosen, rejected) were translated automatically from english to dutch. This translation can have grammar mistakes so if used professionally, inference using a grammar model/software is adviced.

Full credit for data creation and/or collection goes to the original creators at unalignment. I added nothing to the data as I only translated the set using ML-models.

Intended Use & License

Free to use under the same license as the original dataset. A reference in your model-card is always welcome,though not required. Intended for research into uncensored preference optimization, DPO/ORPO training, or Dutch-language NLP.

Important Disclaimer & Responsible Use

WARNING This dataset contains toxic, offensive, harmful, explicit, biased or potentially illegal content. It is not suitable for production use, minors, or general-purpose applications. Use at your own risk. The author is not responsible for any misuse or generated outputs.

Responsible use is required. Apply strong filtering and safety measures when training models.

Loading the dataset

from datasets import load_dataset

ds = load_dataset("parquet", 
                  data_files="unalignmenttoxic-dpo-v0.2-dutch.parquet", 
                  split="train")