LingoIITGN/Triveni

该数据集是由Vaani和Flickr30k两个主要来源的数据组合而成，用于支持多语言和多模态模型的预训练。Vaani数据集包含约16,000小时的自发、图像提示言语，来自84.6K名讲者的约9.6百万句话，覆盖54种印度语言。Flickr30k数据集是一个用于句子级图像描述的标准基准，包含158,000个描述和244,000个共指链。Indic Multimodal Fine-Tuning Dataset是一个多语言、注释数据集，用于在印度语境中细化多模态模型，支持英语、印地语和Hinglish语三种语言的图像描述，包含11,406张图像和三种语言的注释。

The dataset is a combination of data from two major sources—Vaani and Flickr30k—to support multilingual and multimodal model pretraining. The Vaani dataset includes approximately 16,000 hours of spontaneous, image-prompted speech, with about 9.6 million utterances from 84.6K speakers across 54 Indian languages. The Flickr30k dataset is a standard benchmark for sentence-based image description, containing 158,000 captions and 244,000 coreference chains. The Indic Multimodal Fine-Tuning Dataset is a multilingual, annotated dataset developed for fine-tuning multimodal models in the Indian context, supporting captions in English, Hindi, and Hinglish across diverse image categories.