NVIDIANemotron 3.5
Nemotron 3.5 Embed 1B
The 1.55B-parameter embedding model maps text queries and text or image documents into vectors for retrieval.
What the release provides
It produces 2,048-dimensional normalized vectors and supports shorter vectors with renormalization. Its language component stays frozen from Nemotron 3 Embed 1B, preserving that text embedding space while adding image retrieval.Nemotron 3.5 Embed 1B official model card
How to use it well
Practical guidance. Keep the query and document prefixes with the index configuration. If you reduce vector size, apply the same transformation to documents and queries. Review retrieval results before connecting the index to a model that writes answers.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | nvidia/Nemotron-3.5-Embed-1B-BF16 |
| Release license | OpenMDW 1.1; card identifies Apache-licensed Ministral ancestryNemotron 3.5 Embed 1B official model card |