Suvach

Suvach是一个专为评估Hindi语言的问答（QA）模型而设计的新型基准数据集。该数据集由先进的LLMs生成，包含超过100,000个问答对，平均每个问题包含1200个tokens。数据来源于Hindi语的维基百科页面，通过预处理和分块技术生成问题和答案。Suvach旨在通过避免机器翻译的潜在缺陷，为Indic语言模型提供一个更强大的评估环境，适用于教育、医疗和治理等多个领域，解决低资源语言评估的难题。Suvach is a novel benchmark dataset specifically designed for evaluating Hindi-language question answering (QA) models. Generated by state-of-the-art LLMs, it contains over 100,000 QA pairs, with an average of 1200 tokens per question. The dataset is sourced from Hindi-language Wikipedia pages, where questions and answers are created through preprocessing and chunking techniques. Suvach aims to provide a more robust evaluation environment for Indic language models by eliminating potential flaws in machine translation, addressing the challenges of low-resource language evaluation across multiple domains including education, healthcare, and governance.

搜集汇总

背景与挑战

背景概述

Suvach是一个专为评估Hindi语言问答模型设计的新型基准数据集，由先进LLMs生成，包含超过100,000个问答对，平均每个问题1200 tokens，数据来源于Hindi维基百科。它旨在避免机器翻译的缺陷，为Indic语言模型提供更强大的评估环境，适用于教育、医疗和治理等多个低资源语言领域。