Datasets:
Fix templating bug in English type: 0 needles
#3
by jonsoft - opened
The English type:0 needles have a mistake in the answer format of the retrieval_question. This confuses LLM scorers, and they sometimes award the model 7/10 instead of 10/10 because they did not answer using the (incorrect) question format.
This PR fixes the mistake, improving evaluations based on this dataset.