Flow - 4x faster than typing
Mini toolsNemotron 3 Embed 1B BF16
By NVIDIA
Model family: Nemotron
Nemotron-3-Embed-1B-BF16 is a 1.14B parameter multilingual text embedding model from NVIDIA, produced by pruning a Ministral-3-3B-Instruct based encoder and distilling from the Nemotron-3-Embed-8B-BF16 teacher model using combined cosine and mean squared error distillation losses. It has a hidden size of 2048 and supports input sequences up to 32768 tokens, outputting dense embeddings for retrieval, semantic similarity, and RAG workflows. Evaluated across 34 languages, it targets multilingual and cross-lingual retrieval such as question and answer applications over large text corpora. It is positioned as the high efficiency standard in the Nemotron 3 Embed collection, designed for cost and latency sensitive production retrieval serving while retaining strong accuracy relative to the flagship model.
New
Text
Overview
Nemotron-3-Embed-1B-BF16 is NVIDIA's high efficiency multilingual text embedding model, pruned and distilled from the 8B flagship. It produces dense vector embeddings for retrieval, semantic search, and RAG across 34 languages, built for low latency, cost sensitive production serving.
About NVIDIA
Industry:
Computer Hardware Manufacturing
Company Size:
42000
Location:
Santa Clara, California, US
Website:
nvidia.com
Last updated: July 17, 2026
