NeMo Data Designer tech report: 25 trillion tokens generated via open-source synthetic data framework
code_star · x · 2026-09-18
The team behind NeMo Data Designer (grown out of Gretel, now under NVIDIA) released a tech report (arXiv:2609.17699) for their open-source multimodal synthetic data generation framework.
Key points:
- Declarative configuration: humans or agents define each dataset column — text, code, structured outputs, images, embeddings, and statistical samplers steering diversity — extensible via a plugin system.
- Configurations are inspectable artifacts supporting sharing and reproducibility; a preview-and-revision loop lets users validate small samples before full-scale generation.
- Runtime handles dependency resolution, scheduling calls to user model endpoints, and retries.
- Over 25 trillion tokens have run through the framework to date, with case studies spanning structured, agentic, multimodal, and domain-specialized datasets, including ones used in Nemotron model development.
- The author recounts the pre-acquisition grind at Gretel and notes the design choices turned out ideal for agentic dataset building in 2026.
Related event: NVIDIA Open-Sources NeMo Data Designer, 25 Trillion Tokens Generated(3 posts)→
More from coding & agent
- Butterfly traces multi-agent failures with graphs, 62.5% accuracy vs 53.8% for log prompting — JeremyCMorgan · 2026-11-04
- Free course covers the 4 hard problems of production generative media agents — Al_Grigor · 2026-11-04
- Quota reset night: Codex back to 100% while Claude shows just 10% left — rudrank · 2026-11-04
- Self-improving software is inevitable: closing the agent-observability loop — bibryam · 2026-11-04
- Building a 7-GPU 132GB VRAM local inference rig: consolidate on EPYC or split across two systems? — infectiousstupidity · 2026-11-04
- Founder of 50-person, 12-country team says he talks to AI agents more than humans — Scobleizer · 2026-11-04