specific method · filed under data curation

Knowledge distillation for synthetic data

Synthetic data is generated by distilling trajectories, solutions, or translations from strong teacher models or agent systems.

Also called knowledge distillation, distillation.

sources: 3
models: 2
lab adopt it: 1
strongest: used

How sources treat it

One count per evidence span, weakest treatment to strongest.

used 3

Documented in

NVIDIA-Nemotron-3-Ultra-550B-A55B NVIDIA-Nemotron-3.5-Lightning-30B-A3B

Evidence

3 spans quoted from the sources, strongest treatment first.

During post-training, we generate synthetic data by distilling trajectories, solutions, and translations from strong teacher models and agent systems

useddata curationin Nemotron 3.5 LightningNVIDIA

nemotron-3.5-lightning-30b-a3b Model by NVIDIA | NVIDIA NIM·NVIDIA·2026-09-02

we generate synthetic data by distilling trajectories, solutions, and translations from strong teacher models

useddata curationin Nemotron 3 UltraNVIDIA

nvidia/nemotron-3-ultra-550b-a55b·NVIDIA·2026-07-23

we generate synthetic data by distilling trajectories, solutions, and translations from strong teacher models and agent systems, often grounded in real tasks or documents and aggressively filtered for quality.

useddata curationin Nemotron 3.5 LightningNVIDIA

nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 · Hugging Face·NVIDIA·2026-08-14

Filed alongside

Other methods under data curation :: synthetic data.

Agentic data synthesis pipeline Automated data synthesis and environment-construction pipelines Automatic synthesis of task-oriented RL environments Containerized coding-environment construction with self-testing and trace removal Counterfactual data augmentation Recycling filtered documents into image-text pairs Automated batch synthesis of RL training data Coding environments from GitHub issue–PR pairs Container buildability and verifiability check Corpus rephrasing with fidelity verification Dynamic multi-agent synthetic-data generation loop End-to-end synthetic environment generation Environment, toolset, task, and solution synthesis pipeline Failure-case and negative-feedback-driven environment generation Heterogeneous answer-generation agents Iterative task-difficulty escalation Knowledge-graph-guided task synthesis Large-scale environment synthesis and curation Long-context data synthesis by permutation and concatenation Matching synthesis-pipeline complexity to teacher capability Metadata-conditioned synthetic generation Mocked tools for agent environments Modular synthetic-data pipeline composition Multi-agent collaborative environment construction