NVIDIA AI Releases Nemotron 3 Embed: An Open Embedding Collection Whose 8B Checkpoint Ranks #1 on RTEB
NVIDIA has unveiled Nemotron 3 Embed, a groundbreaking open embedding collection that has taken the AI world by storm. The 8B checkpoint of Nemotron 3 Embed has achieved an unprecedented ranking of #1 on RTEB (Retrieval Embedding Benchmark), boasting an impressive average NDCG@10 score of 78.46.
What is Nemotron 3 Embed?
Nemotron 3 Embed is a collection of three open checkpoints designed to work on the embedding layer of AI models. It targets production-scale RAG (Retrieval-Augmented Generation), agentic retrieval, code retrieval, and agent memory applications. The models are transformer encoders trained with bidirectional attention masking, and the final embedding is generated through average pooling over token-level representations.
Key Features and Performance
- The 8B-BF16 model ranks #1 on RTEB with an average NDCG@10 score of 78.46.
- Three open checkpoints are available: Nemotron-3-Embed-8B-BF16, Nemotron-3-Embed-1B-BF16, and Nemotron-3-Embed-1B-NVFP4.
- The 1B models were built using a compression pipeline, not a smaller training run.
- NVFP4 serving offers up to 2x higher throughput than BF16 and retains 99%+ of BF16 retrieval accuracy.
Use Cases and Applications
- Multilingual enterprise search
- Code retrieval
- Agent memory
- Cost-tiered RAG
Conclusion
NVIDIA’s Nemotron 3 Embed is a significant advancement in the field of AI, offering state-of-the-art performance and versatility. With its open-source nature and comprehensive documentation, it is poised to revolutionize various AI applications and empower developers to build more sophisticated models.
“The Nemotron 3 Embed collection represents a major breakthrough in embedding technology, providing developers with a powerful tool to enhance their AI applications.” – NVIDIA Research Team
| Model | Params | Emb dim | RTEB | ViDoRe-V3 text | MMTEB (Retrieval) |
|---|---|---|---|---|---|
| Nemotron-3-Embed-8B-BF16 | ~8B | 4096 | 78.46 | 60.60 | 75.45 |
| Nemotron-3-Embed-1B-BF16 | 1.14B | 2048 | 72.38 | 57.74 | 71.04 |
| Nemotron-3-Embed-1B-NVFP4 | 1.14B | 2048 | 72.00 | — | — |
Leave a Reply