Normalized and paraphrased splits of 21 standard NLP benchmarks in English, German, French, Spanish, and Italian, intended for base model pretraining.

Generation

English: paraphrased with Qwen3.5-27B-FP8 (April 2026) German: translated and refined with Qwen3.5-27B-FP8 (April 2026) French: translated and refined with Qwen3.5-27B-FP8 (May 2026) Spanish: translated and refined with Qwen3.5-27B-FP8 (May 2026) Italian: translated and refined with… See the full description on the dataset page: https://huggingface.co/datasets/AIML-TUDA/QA-base.