A reasoning dataset made by DeepSeek R1. The reasoning data is made from punctuation-restoration tasks from Wikipedia. We have stored the reasoning in cases where the output is 100% true.
A total of 22.000 tasks where generated.
Of these a total of 7794 tasks had the correct answer and where in Norwegian. This were trimmed to 6745 to be of the same size as the English reasoning dataset.
This was split into test=250, validation=250 and train=6245
