NVIDIANemotron 3.5
Nemotron 3.5 Embed 8B
The 8.38B-parameter model embeds text queries and multimodal documents for retrieval. Documents can contain text, images or both.
What the release provides
Its vectors have 4,096 dimensions. The card gives a 32,768-token context. A frozen Nemotron 3 Embed 8B language component preserves the previous text embedding space while trained visual components add image retrieval.Nemotron 3.5 Embed 8B official model card
How to use it well
Practical guidance. For a scanned document collection, compare image retrieval with OCR text retrieval on the actual question set. Preserve page identity in each result. A useful vector match points to evidence; it does not establish that a generated answer is supported.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | nvidia/Nemotron-3.5-Embed-8B-BF16 |
| Release license | OpenMDW 1.1Nemotron 3.5 Embed 8B official model card |