suyog-ghimire/UncertaintyQA

UncertaintyQA是一个精心策划的数据集，旨在教导语言模型何时应自信回答以及何时应承认不确定性。它遵循Stanford Alpaca的指令格式，包含可回答（确定）和不可回答、不明确、不可能或未知的问题，要求模型回答“我不知道”。该数据集用于微调LLMs以校准不确定性，减少幻觉，并在模糊情况下鼓励诚实输出。数据集包含约350个可回答样本和1600个不可回答样本，总计约1800个样本。每个样本遵循Alpaca模式，包含指令、输入和输出字段。

UncertaintyQA is a curated dataset designed to teach language models when to answer confidently and when to admit uncertainty. It follows the Stanford Alpaca instruction-format and contains a mix of answerable (certain) questions and unanswerable, ill-posed, impossible, or unknown questions requiring the model to respond with "I don’t know." This dataset is intended for fine-tuning LLMs to calibrate uncertainty, reduce hallucinations, and encourage honest outputs in ambiguous situations. The dataset consists of ~350 certain/answerable samples and ~1600 uncertain/unanswerable samples, totaling ~1800 samples. Each entry follows the Alpaca schema with instruction, input, and output fields.