NVIDIA NeMo Data Designer
An open-source Python framework for synthetic data generation with vLLM or cloud APIs, correlated fields, and automated validation. Apache 2.0 licensed.
NVIDIA NeMo Data Designer is a Python framework for developers and data teams creating synthetic datasets with controlled variation and relationships between fields. It can start with an empty dataset or build on your own seed data. The framework is open source under Apache 2.0.
The Python code runs locally and connects to model endpoints, including vLLM, NVIDIA Build, OpenAI, and OpenRouter. Model execution happens at the endpoint you choose. Cloud providers receive the inputs sent for generation and require API credentials; using the library locally doesn't make those requests private to your machine.
Data Designer combines statistical sampling with model-generated content, so datasets can follow chosen distributions while keeping related fields consistent. It produces text, structured records, and images, and accepts image, audio, and video context in multimodal workflows. Python and SQL checks, custom validators, and LLM judges can validate or score the results.
You can preview samples before a larger run, then resume and monitor generation. The framework handles batches and parallel requests. Local or remote MCP servers connect generation to external tools, with interaction traces available for capture. Plugins support custom columns, seed readers, and processors; a coding-agent skill works with Claude Code and Codex for dataset design and generation.
Optional telemetry sends model names and input and output token counts to NVIDIA. You can disable it independently of the data practices of your chosen model provider.
