NVIDIA Releases Nemotron 3.5 Retriever Models

NVIDIA released a collection of Nemotron 3.5 retrieval models, detailed on its Hugging Face collection page. The collection includes Nemotron-3.5-Embed-1B, Nemotron-3.5-Embed-8B, Nemotron-3.5-Rerank-8B and Nemotron-3.5-Parse.

Nemotron-3.5-Embed-8B accepts text, document images or combined image and text inputs. It supports pages containing text, tables, charts and infographics, with text retrieval evaluated across 34 languages and image retrieval across 18 languages.

The 8.38 billion parameter model produces 4,096 dimensional embedding vectors and supports a context length of 32,768 tokens. NVIDIA provides examples for Transformers, Sentence Transformers and vLLM, while use of the model is governed by the OpenMDW 1.1 license. The 8B embedding model is built on Mistral AI's Ministral 3 8B Instruct base model.