README.md

---
library_name: sentence-transformers
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- visual-document-retrieval
- cross-modal-distillation
- knowledge-distillation
- document-retrieval
- multilingual
- nanovdr
base_model: distilbert/distilbert-base-uncased
language:
- en
- de
- fr
- es
- it
- pt
license: apache-2.0
datasets:
- openbmb/VisRAG-Ret-Train-Synthetic-data
- openbmb/VisRAG-Ret-Train-In-domain-data
- vidore/colpali_train_set
- llamaindex/vdr-multilingual-train
model-index:
- name: NanoVDR-S-Multi
  results:
  - task:
      type: retrieval
    dataset:
      name: ViDoRe v1
      type: vidore/vidore-benchmark-667173f98e70a1c0fa4d
    metrics:
    - name: NDCG@5
      type: ndcg_at_5
      value: 82.2
  - task:
      type: retrieval
    dataset:
      name: ViDoRe v2
      type: vidore/vidore-benchmark-v2
    metrics:
    - name: NDCG@5
      type: ndcg_at_5
      value: 61.9
---

<p align="center">
  <img width="560" src="banner.png" alt="NanoVDR"/>
</p>

<p align="center">
  <a href="https://github.com/nanovdr/nanovdr">GitHub</a> |
  <a href="https://huggingface.co/collections/nanovdr/nanovdr">All Models</a>
</p>

> **Paper**: Our arxiv preprint is currently on hold. Details on training methodology, ablations, and full results will be available once the paper is published.

# NanoVDR-S-Multi

**The recommended NanoVDR model for production use.**

NanoVDR-S-Multi is a **69M-parameter multilingual text-only** query encoder for visual document retrieval. It encodes text queries into the same embedding space as a frozen 2B VLM teacher ([Qwen3-VL-Embedding-2B](https://huggingface.co/Qwen/Qwen3-VL-Embedding-2B)), so you can retrieve document page images using **only a DistilBERT forward pass** — no vision model at query time.

### Highlights

- **95.1% teacher retention** — a 69M text-only model recovers 95% of a 2B VLM teacher across 22 ViDoRe datasets
- **Outperforms DSE-Qwen2 (2B)** on multilingual v2 (+6.2) and v3 (+4.1) with **32x fewer parameters**
- **Outperforms ColPali (~3B)** on multilingual v2 (+7.2) and v3 (+4.5) with **single-vector cosine** retrieval (no MaxSim)
- **Single-vector retrieval** — queries and documents share the same 2048-dim embedding space as [Qwen3-VL-Embedding-2B](https://huggingface.co/Qwen/Qwen3-VL-Embedding-2B); retrieval is a plain dot product, FAISS-compatible, **4 KB per page** (float16)
- **Lightweight on storage** — 282 MB model file; doc index costs 64× less than ColPali's multi-vector patches
- **51 ms CPU query latency** — 50x faster than DSE-Qwen2, 143x faster than ColPali
- **6 languages**: English, German, French, Spanish, Italian, Portuguese — all >92% teacher retention

---

## Results

| Model | Type | Params | ViDoRe v1 | ViDoRe v2 | ViDoRe v3 | Avg Retention |
|-------|------|--------|-----------|-----------|-----------|---------------|
| Tomoro-8B | VLM | 8.0B | 90.6 | 65.0 | 59.0 | — |
| Qwen3-VL-Emb (Teacher) | VLM | 2.0B | 84.3 | 65.3 | 50.0 | — |
| DSE-Qwen2 | VLM | 2.2B | 85.1 | 55.7 | 42.4 | — |
| ColPali | VLM | ~3B | 84.2 | 54.7 | 42.0 | — |
| **NanoVDR-S-Multi** | **Text-only** | **69M** | **82.2** | **61.9** | **46.5** | **95.1%** |

<sub>NDCG@5 (×100). v1 = 10 English datasets, v2 = 4 multilingual datasets, v3 = 8 multilingual datasets.</sub>

### Per-Language Retention (v2 + v3, 19,537 queries)

| Language | #Queries | Teacher | NanoVDR-S-Multi | Retention |
|----------|----------|---------|-----------------|-----------|
| English | 6,237 | 64.0 | 60.3 | 94.3% |
| French | 2,694 | 51.0 | 47.8 | 93.6% |
| Portuguese | 2,419 | 48.7 | 46.1 | 94.6% |
| Spanish | 2,694 | 51.4 | 47.8 | 93.1% |
| Italian | 2,419 | 49.0 | 45.7 | 93.3% |
| German | 2,694 | 49.3 | 45.4 | 92.0% |

All 6 languages achieve **>92%** of the 2B teacher's performance.

---

## Quick Start

```python
from sentence_transformers import SentenceTransformer

model = SentenceTransformer("nanovdr/NanoVDR-S-Multi")

queries = [
    "What was the revenue growth in Q3 2024?",           # English
    "Quel est le chiffre d'affaires du trimestre?",       # French
    "Wie hoch war das Umsatzwachstum im dritten Quartal?", # German
    "¿Cuál fue el crecimiento de ingresos en el Q3?",     # Spanish
    "Qual foi o crescimento da receita no terceiro trimestre?", # Portuguese
    "Qual è stata la crescita dei ricavi nel terzo trimestre?", # Italian
]
query_embeddings = model.encode(queries)
print(query_embeddings.shape)  # (6, 2048)

# Cosine similarity against pre-indexed document embeddings
# scores = query_embeddings @ doc_embeddings.T
```

### Prerequisites: Document Indexing with Teacher Model

NanoVDR is a **query encoder only**. Documents must be indexed offline using the teacher VLM ([Qwen3-VL-Embedding-2B](https://huggingface.co/Qwen/Qwen3-VL-Embedding-2B)), which encodes page images into 2048-d embeddings. This is a one-time cost.

```python
# pip install transformers qwen-vl-utils torch
from scripts.qwen3_vl_embedding import Qwen3VLEmbedder

teacher = Qwen3VLEmbedder(model_name_or_path="Qwen/Qwen3-VL-Embedding-2B")

# Encode document page images
documents = [
    {"image": "page_001.png"},
    {"image": "page_002.png"},
    # ... all document pages in your corpus
]
doc_embeddings = teacher.process(documents)  # (N, 2048), L2-normalized
```

> **Note:** The `Qwen3VLEmbedder` class and full usage guide (including vLLM/SGLang acceleration) are available at the [Qwen3-VL-Embedding-2B model page](https://huggingface.co/Qwen/Qwen3-VL-Embedding-2B). Document indexing requires a GPU; once indexed, retrieval uses only CPU.

### Full Retrieval Pipeline

```python
import numpy as np
from sentence_transformers import SentenceTransformer

# doc_embeddings: (N, 2048) numpy array from teacher indexing above

# Step 1: Encode text queries with NanoVDR (CPU, ~51ms per query)
student = SentenceTransformer("nanovdr/NanoVDR-S-Multi")
query_emb = student.encode("Quel est le chiffre d'affaires?")  # shape: (2048,)

# Step 2: Retrieve via cosine similarity
scores = query_emb @ doc_embeddings.T
top_k_indices = np.argsort(scores)[-5:][::-1]
```

---

## How It Works

NanoVDR uses **asymmetric cross-modal distillation** to decouple query and document encoding:

| | Document Encoding (offline) | Query Encoding (online) |
|-|----------------------------|------------------------|
| **Model** | Qwen3-VL-Embedding-2B (frozen) | NanoVDR-S-Multi (69M) |
| **Input** | Page images | Text queries (6 languages) |
| **Output** | 2048-d embedding | 2048-d embedding |
| **Hardware** | GPU (one-time indexing) | CPU (real-time serving) |

The student is trained to **align query embeddings** with the teacher's query embeddings via pointwise cosine loss — no document embeddings or hard negatives are needed during training. At inference, student query embeddings are directly compatible with teacher document embeddings.

---

## Training

| | Value |
|--|-------|
| Base model | `distilbert/distilbert-base-uncased` (66M) |
| Projector | 2-layer MLP: 768 → 768 → 2048 (2.4M params) |
| Total params | 69M |
| Objective | Pointwise cosine alignment with teacher query embeddings |
| Training data | 1.49M pairs — 711K original + 778K translated queries |
| Languages | EN (original) + DE, FR, ES, IT, PT (translated via [Helsinki-NLP Opus-MT](https://huggingface.co/Helsinki-NLP)) |
| Epochs | 10 |
| Batch size | 1,024 (effective) |
| Learning rate | 3e-4 (OneCycleLR, 3% warmup) |
| Hardware | 1× H200 GPU |
| Training time | ~10 GPU-hours |
| Embedding caching | ~1 GPU-hour (teacher encodes all queries in text mode) |

### Multilingual Augmentation Pipeline

1. Extract 489K English queries from the 711K training set
2. Translate each to 5 languages using Helsinki-NLP Opus-MT → 778K translated queries
3. Re-encode translated queries with the frozen teacher in text mode (15 min on H200)
4. Combine: 711K original + 778K translated = **1.49M training pairs**
5. Train with halved epochs (10 vs 20) and slightly higher lr (3e-4 vs 2e-4) to match total steps

---

## Efficiency

| | NanoVDR-S-Multi | DSE-Qwen2 | ColPali | Tomoro-8B |
|--|-----------------|-----------|---------|-----------|
| Parameters | **69M** | 2,209M | ~3B | 8,000M |
| Query latency (CPU, B=1) | **51 ms** | 2,539 ms | 7,300 ms | GPU only |
| Checkpoint size | **274 MB** | 8.8 GB | 11.9 GB | 35.1 GB |
| Index type | Single-vector | Single-vector | Multi-vector | Multi-vector |
| Scoring | Cosine | Cosine | MaxSim | MaxSim |
| Index storage (500K pages) | **4.1 GB** | 3.1 GB | 128 GB | 128 GB |

---

## Model Variants

NanoVDR-S-Multi is the **recommended model**. The other variants are provided for research and ablation purposes.

| Model | Backbone | Params | v1 | v2 | v3 | Retention | Latency | Recommended |
|-------|----------|--------|----|----|----|-----------|---------| ------------|
| **[NanoVDR-S-Multi](https://huggingface.co/nanovdr/NanoVDR-S-Multi)** | **DistilBERT** | **69M** | **82.2** | **61.9** | **46.5** | **95.1%** | **51 ms** | **Yes** |
| [NanoVDR-S](https://huggingface.co/nanovdr/NanoVDR-S) | DistilBERT | 69M | 82.2 | 60.5 | 43.5 | 92.4% | 51 ms | EN-only |
| [NanoVDR-M](https://huggingface.co/nanovdr/NanoVDR-M) | BERT-base | 112M | 82.1 | 62.2 | 44.7 | 94.0% | 101 ms | Ablation |
| [NanoVDR-L](https://huggingface.co/nanovdr/NanoVDR-L) | ModernBERT | 151M | 82.4 | 61.5 | 44.2 | 93.4% | 109 ms | Ablation |

## Key Properties

| Property | Value |
|----------|-------|
| Output dimension | 2048 (aligned with Qwen3-VL-Embedding-2B) |
| Max sequence length | 512 tokens |
| Supported languages | EN, DE, FR, ES, IT, PT |
| Similarity function | Cosine similarity |
| Pooling | Mean pooling |
| Normalization | L2-normalized |

## Citation

```bibtex
@article{nanovdr2026,
  title={NanoVDR: Distilling a 2B Vision-Language Retriever into a 70M Text-Only Encoder for Visual Document Retrieval},
  author={Liu, Zhuchenyang and Zhang, Yao and Xiao, Yu},
  journal={arXiv preprint arXiv:2502.XXXXX},
  year={2026}
}
```

## License

Apache 2.0