Update README.md

README.md

@@ -1,7 +1,7 @@
 ---
 license: mit
 ---
-This dataset is a deduplicated subset of the validation split of [WinoGrande](https://huggingface.co/datasets/allenai/winogrande), as used in the paper [How Much Can We Forget about Data Contamination?](https://arxiv.org/abs/2410.03249). The deduplication was performed using [this script](https://github.com/tml-tuebingen/forgetting-contamination/blob/main/evaluation/find_duplicate_benchmark_questions.py).
+This dataset is a deduplicated subset of the XL train split of [WinoGrande](https://huggingface.co/datasets/allenai/winogrande), as used in the paper [How Much Can We Forget about Data Contamination?](https://arxiv.org/abs/2410.03249). The deduplication was performed using [this script](https://github.com/tml-tuebingen/forgetting-contamination/blob/main/evaluation/find_duplicate_benchmark_questions.py).
 
 The data fields are the same as in https://huggingface.co/datasets/allenai/winogrande, with the additional "split-id" column that can be used to partition the benchmark questions into different subsets.
 The dataset can be used as a plug-in replacement if you want to work with the deduplicated benchmark questions.