bay-calibration-llm-evaluators/

llmeval2-annotated-latest

	LLMEval²-Select Dataset




	
	Introduction

The LLMEval²-Select dataset is a curated subset of the original LLMEval² dataset introduced by Zhang et al. (2023). The original LLMEval² dataset comprises 2,553 question-answering instances, each annotated with human preferences. Each instance consists of a question paired with two answers. To construct LLMEval²-Select, Zeng et al. (2024) followed these steps:

Labelled each instance with the human-preferred answer. Removed all… See the full description on the dataset page: https://huggingface.co/datasets/bay-calibration-llm-evaluators/llmeval2-annotated-latest.