Nemotron-3.5-Embed-1B-BF16
by nvidia
NVIDIA's 1.55B multimodal embedder: search PDF pages as images, text or both from a text query, in the same text space as Nemotron-3-Embed-1B
nvidia/Nemotron-3.5-Embed-1B-BF16Deploy Nemotron-3.5-Embed-1B-BF16
Single-tenantMixpeek has no managed extractor for this model. On a single-tenant deployment you upload the weights and a custom plugin serves them next to the rest of your pipeline.
Overview
Nemotron-3.5-Embed-1B-BF16 is an embedding model from NVIDIA, published on Hugging Face in October 2026 under the OpenMDW-1.1 license, which allows commercial use. It embeds a document page as an image, as text, or as both together, and embeds queries as text, so a question can find the page that answers it even when the answer sits in a table, a chart or an infographic that OCR would flatten. Each input becomes one 2,048-dimensional vector.
It adds image retrieval to Nemotron-3-Embed-1B without moving the text space. The language model was initialized from Nemotron-3-Embed-1B and kept frozen while the vision side was distilled from the 8B sibling, so the card says text embeddings land in the same space as Nemotron-3-Embed-1B. A corpus already indexed with the earlier model can add page images without re-embedding its text.
NVIDIA evaluated text retrieval in 34 languages and image retrieval in 18, on ViDoRe V3 and MIRACL-VISION with NDCG@10. The card lists those benchmarks without publishing scores, and we have not measured it.
Architecture
A Pixtral vision encoder initialized from Mistral's Ministral-3-3B-Instruct-2512 and a language model initialized from nvidia/Nemotron-3-Embed-1B-BF16, 1.55B parameters in total, BF16 weights. Token-level hidden states are mean-pooled and L2-normalized into one 2,048-dimensional vector per input; the vector can be truncated to 1,024 or 512 dimensions and re-normalized. Context is 32,768 tokens, including image tokens; page images default to a 1,120-pixel longest edge and can go up to 1,540. The model was distilled from Nemotron-3.5-Embed-8B-BF16 with cosine-distance and mean-squared-error losses on multilingual text and image retrieval data, with the language model frozen and the vision encoder and projection fine-tuned. Inputs take a query: or passage: prefix.
Mixpeek SDK Integration
# Embed each PDF page image (with its OCR text when you have it) with
# Nemotron-3.5-Embed-1B-BF16, then upsert the vectors into a Mixpeek Vector Store
# namespace you query directly. Needs a CUDA GPU (A100, H100 or GB200 per the card).
import torch
from PIL import Image
from sentence_transformers import SentenceTransformer
from mixpeek import Mixpeek
model = SentenceTransformer(
"nvidia/Nemotron-3.5-Embed-1B-BF16",
device="cuda:0",
model_kwargs={"attn_implementation": "flash_attention_2", "dtype": torch.bfloat16},
)
client = Mixpeek(api_key="YOUR_KEY")
pages = [{"id": "q3-report-p14", "image": "q3-report-p14.png", "text": "Revenue by region..."}]
vectors = model.encode_document(
[{"image": Image.open(p["image"]), "text": p["text"]} for p in pages]
)
client.namespaces.upsert(
namespace="reports",
documents=[{"id": p["id"], "dense_embedding": v.tolist(), "text": p["text"]}
for p, v in zip(pages, vectors)],
)Capabilities
- One 2,048-dimensional vector per page image, text chunk, or image plus text
- Text queries against pages that hold tables, charts and infographics
- Truncation to 1,024 or 512 dimensions for cheaper storage
- Same text embedding space as Nemotron-3-Embed-1B-BF16
- Sentence Transformers, Transformers and vLLM; OpenMDW-1.1 license
Use Cases on Mixpeek
Performance
The card lists A100 80GB, H100 80GB and GB200 as tested GPUs and recommends Flash Attention 2. An NVFP4 build of the 1B and an 8B BF16 sibling are published alongside it. No scores are published on the card and we have not measured it.
Common Pipeline Companions
Frequently Asked Questions
What is Nemotron-3.5-Embed-1B-BF16?
An NVIDIA embedding model with 1.55B parameters that turns a document page (as an image, text, or both) or a text query into one 2,048-dimensional vector, for retrieval over PDFs, slides and scanned documents. It is released under the OpenMDW-1.1 license, which allows commercial use.
Do I need to re-embed my Nemotron-3-Embed-1B text index?
According to the card, no. The language model was initialized from Nemotron-3-Embed-1B-BF16 and frozen during training, so text embeddings share the earlier model's space. You can add page-image vectors from the new model next to existing text vectors. Spot-check a few queries on your own data before relying on it.
Can queries be images?
No. The card says queries are text only. Documents can be an image, text, or an image and text together.
Can I store smaller vectors?
Yes. Truncate the 2,048-dimensional vector to its first 1,024 or 512 values and L2-normalize it again; the card says truncated vectors remain effective for retrieval.
How do I use Nemotron-3.5-Embed-1B with Mixpeek?
Embed your page images and queries with it on your own GPU and upsert the vectors into a Mixpeek Vector Store namespace, as in the example on this page. MVS can combine those dense vectors with BM25 over the page text in one hybrid query. On Enterprise, the weights can also be uploaded and run by a custom plugin.
Specification
Build a pipeline with Nemotron-3.5-Embed-1B-BF16
Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.
Run it on your own data, free