dci-bench
收藏资源简介:
DCI-Bench是一个包含信息检索(IR)和推理密集型检索任务(BRIGHT)评估基准的数据集。该数据集分为两个子集:1) IR评估(Sample-50),包含2wikimultihopqa、bamboogle、beir_arguana、beir_scifact、hotpotqa、musique、nq和triviaqa等测试集;2) BRIGHT基准,包含bright_biology、bright_earth_science、bright_economics和bright_robotics等训练集。该数据集主要用于评估信息检索和复杂推理任务的性能。
DCI-Bench is a dataset containing evaluation benchmarks for Information Retrieval (IR) and reasoning-intensive retrieval tasks (BRIGHT). The dataset is divided into two subsets: 1) IR evaluation (Sample-50), which includes test sets such as 2wikimultihopqa, bamboogle, beir_arguana, beir_scifact, hotpotqa, musique, nq, and triviaqa; 2) BRIGHT benchmark, which includes training sets such as bright_biology, bright_earth_science, bright_economics, and bright_robotics. The dataset is primarily used to evaluate the performance of information retrieval and complex reasoning tasks.




