Update README.md

README.md

@@ -28,3 +28,5 @@ The data has not been shuffled (you can either use `dataset.shuffle(...)`,
 or download the shuffled version [here](https://huggingface.co/datasets/sradc/chunked-shuffled-wikipedia20220301en-bookcorpusopen),
 which will be faster to iterate over).
 
+This dataset is processed for convenience, at the expense of losing some percentage of the tokens due to truncation, 
+(assuming the training minibatches are truncated to 128 tokens).