45 datasets found

A self-recognition test for language models using model-generated security questions.

The Quora Question Pairs (QQP) dataset consists of 50,000 question pairs labeled with paraphrase or non-paraphrase.

Ensuring alignment with human preferences is a crucial characteristic of large language models (LLMs). Presently, the primary alignment methods, RLHF and DPO, require extensive...

The paper proposes a novel approach to aligning language models with human preferences, focusing on the use of preference optimization in reward-free RLHF.

The dataset used for training large language models, with a focus on balancing the text distribution and mitigating overfitting.

A general language assistant for aligning language models with human users

Multilingual Blending: LLM Safety Alignment Evaluation with Language Mixture

The authors evaluate the performance of their proposed conformal prediction methods for multistep feedback covariate shift (MFCS) on synthetic black-box optimization and active...

Semantic parsing maps a user-issued natural language (NL) utterance to a machine-executable meaning representation (MR), such as λ−calculus (Zettlemoyer and Collins, 2005), SQL...

The Universal Conceptual Cognitive Annotation (UCCA) dataset is a graph-based semantic annotation scheme based on typological linguistic principles.

DuReader dataset is a Chinese machine reading comprehension dataset, focusing on real-world web data

TruthX: Alleviating Hallucinations by Editing Large Language Models

A language model that demonstrates capabilities in processing and generating human-like text.

The dataset used in the paper is a multiple-choice reading comprehension dataset, which includes a passage, question, and answer. The passage is a script, and the question is a...

The Natural Questions dataset consists of questions extracted from web queries, with each question accompanied by a corresponding Wikipedia article containing the answer.

The TriviaQA dataset is a collection of questions sourced from Quiz League websites, with sentence-level supporting facts annotation.

The dataset used for the experiments across ten models– ranging from bag-of-words models to pre-trained transformers– and find that a model having higher AUC does not necessarily...

The dataset is used for training and evaluating the MS MARCO model, a question answering model.

The dataset used in the paper is the Stanford Natural Language Inference (SNLI) dataset, which consists of 549,367 premise-hypothesis pairs for train/dev/test sets and target...

The LaMini approach involves generating a large-scale instruction dataset by leveraging the outputs of a large language model, gpt-3.5-turbo.