416 datasets found

AdvBench is a dataset for evaluating the safety of large language models.

Social Chemistry 101 is a dataset that encompasses diverse social norms.

ETHICS is a benchmark for evaluating a language model's knowledge of fundamental ethical concepts.

CrowS-Pairs is a challenge dataset for measuring social biases in masked language models.

ALI-Agent is an evaluation framework that leverages the autonomous abilities of LLM-powered agents to probe adaptive and long-tail risks in target LLMs.

Opportunity activity recognition dataset contains questions answerable using Wikidata as the knowledge graph, focusing on questions with a single entity and relation.

Disc-medllm: Bridging general large language models and real-world medical consultation.

A large-scale Chinese benchmark for evaluating medical large language models. The dataset consists of 280,839 samples, with 74 tasks, and covers 24 departments and 150 diseases.

TruthX: Alleviating Hallucinations by Editing Large Language Models

The dataset used for training and evaluation of the proposed RRHF paradigm.

The dataset used in this paper is a set of repairs for a knowledge base, with each repair being a maximal R-consistent subset of facts.

Math23k is the most commonly used Chinese dataset in MWP solving. It contains 23,162 problems with 21,162 training problems, 1,000 validation problems and 1,000 testing problems.

DocRED is a large-scale human-annotated dataset for document-level RE, which is constructed from Wikipedia and Wikidata.

DBpedia is a public knowledge graph which is derived from structured information in Wikipedia, mainly infoboxes.

RadQA: A question answering dataset to improve comprehension of radiology reports

Large-scale passage retrieval aims to fetch relevant passages from a million- or billion-scale collection for a given query to meet users’ information needs, serving as an...

Large-scale passage retrieval aims to fetch relevant passages from a million- or billion-scale collection for a given query to meet users’ information needs, serving as an...

This paper introduces the TemporalAlignmentQA (TAQA) dataset, which contains 20,148 time-sensitive questions and their answers for each year from 2000 to 2023.

The declaration dataset is generated from the annotations from GQA dataset, containing questions and their corresponding declarative sentences.

We use the VQA v2.0 dataset for the evaluation of our proposed joint model, where the answers are balanced in order to minimize the effectiveness of learning dataset priors.