NeMo Data Designer：用于多模态合成数据生成的可扩展框架
英文标题：NeMo Data Designer: An Extensible Framework for Multimodal Synthetic Data Generation
研究团队推出了开源通用多模态合成数据生成（SDG）框架 NeMo Data Designer（NDD）。NDD 采用直观的声明式配置格式，支持人类与 AI Agent 灵活定义涵盖文本、代码、结构化输出、图像、嵌入向量及统计采样器的数据集列，并拥有灵活的插件系统。该框架核心内置预览与修改迭代闭环，具备自动依赖解析、端点调用调度与失败重试等运行时能力。NDD 已被成功应用于 Nemotron 系列模型开发与企业级生产落地，显著加速了结构化、Agent 与多模态等复杂场景的数据流水线构建。
2.1 KB
640 B
612 B
380 B
9.1 KB
32.7 KB
22.8 KB
24.2 KB
34.7 KB
中文
[提交于 2026 年 9 月 15 日]
标题：NeMo Data Designer: An Extensible Framework for Multimodal Synthetic Data Generation
作者：Johnny Greco , Nabin Mulepati , Andre Manoel , Eric Tramel , Kirit Thadaka , Mike Knepper , Dhruv Nathawani , Dane Corneil , Yev Meyer , Alex Watson , Maarten Van Segbroeck
查看由 Johnny Greco 及其他 10 位作者撰写的题为《NeMo Data Designer: An Extensible Framework for Multimodal Synthetic Data Generation》的论文 PDF
摘要：我们提出了 NeMo Data Designer（NDD），这是一个用于多模态合成数据生成（SDG）的开源通用框架。NDD 旨在提供直观的使用体验，它采用声明式配置格式，使人类和/或 Agent 用户能够定义每个数据集列；列类型涵盖文本、代码、结构化输出、图像、嵌入向量以及用于显式调控数据集多样性的统计采样器。用户还可以借助该框架灵活的插件系统引入额外的列类型和功能。NDD 的配置本身即为可检查的工件，支持工作流共享与可复现性。鉴于 SDG 本质上是一个迭代过程，NDD 在其核心工作流中构建了“预览与修改”循环，允许用户先生成并检查少量记录，优化规范，然后再进行全量生成。在运行时，NDD 负责解析依赖项、调度对用户提供的模型端点的调用，并在请求失败时自动重试。我们详细介绍了 NDD 的架构与编程模型，并展示了涵盖结构化任务、Agent 任务、多模态任务和领域专有任务的案例研究，其中包括在 Nemotron 模型开发以及企业生产部署中所使用的数据集。
提交历史
来自：Maarten Van Segbroeck [ 查看邮箱 ]
[v1] 2026 年 9 月 15 日 星期二 18:11:27 UTC (2,206 KB)
English / 原文
[Submitted on 15 Sep 2026]
Title:NeMo Data Designer: An Extensible Framework for Multimodal Synthetic Data Generation
Authors:Johnny Greco
,
,
,
,
,
,
,
,
,
,
View a PDF of the paper titled NeMo Data Designer: An Extensible Framework for Multimodal Synthetic Data Generation, by Johnny Greco and 10 other authors
Abstract:We present NeMo Data Designer (NDD), an open-source, general-purpose framework for multi-modal synthetic data generation (SDG). Designed to be intuitive to use, NDD provides a declarative configuration format in which human and/or agent users define each dataset column, with column types spanning text, code, structured outputs, images, embeddings, and statistical samplers that are explicitly configured to steer dataset diversity. Additional column types and functionality can be introduced using the framework's flexible plugin system. NDD's configuration is an inspectable artifact, supporting workflow sharing and reproducibility. SDG is an inherently iterative process. NDD therefore builds a preview-and-revision loop into its core workflow, allowing users to generate and inspect a small number of records, refine the specification, and rerun generation at full scale. At runtime, NDD resolves dependencies, schedules calls to user-provided model endpoints, and retries failed requests. We describe NDD's architecture and programming model and present case studies spanning structured, agentic, multimodal, and domain-specialized tasks, including datasets used in Nemotron model development and in production enterprise deployments.
Submission history
From: Maarten Van Segbroeck [
]
[v1]
Tue, 15 Sep 2026 18:11:27 UTC (2,206 KB)