A Config File Turns Synthetic Data Into a Reproducible Pipeline
Synthetic data work often breaks down into opaque scripts and one-off prompts. NeMo Data Designer replaces that sprawl with an inspectable declarative spec for text, code, structured outputs, images, embeddings, and statistical samplers, then adds preview-and-revise loops, dependency scheduling, retries, and plugins. The payoff is a workflow humans and agents can share, audit, and rerun across multimodal datasets.