EVIE-Preview-4.5B

Next-Generation Multilingual Visual Document Retrieval with Ultra-Compact Token Embeddings

Overview • Architecture • Benchmark Results • Quick Start • Reproducing • Training Details • Citation

---

Overview

EVIE-Preview-4.5B is a state-of-the-art multilingual Visual Document Retrieval (VDR) model built upon Qwen3.5-4B. It employs ColBERT-style late interaction with native 128-dimensional multi-vector token embeddings (4.54B parameters, BF16).

By combining native GatedDeltaNet linear-attention and full-attention hybrid modeling with a compact visual projection, EVIE achieves top-tier performance across ViDoRe V1+V2 and ViDoRe V3 while generating compact 128D multi-vector representations.

Note: This is a preview release. The next iteration of EVIE is pending release.

Key Highlights

• 🎯 Ultra-Compact 128D Token Vectors: Native 128-dimensional multi-vector representations keep each token vector narrow while preserving strong retrieval quality.
• 🏆 SOTA on ViDoRe Benchmarks: Outperforms larger 8B models on ViDoRe V3 (leading 7 of 8 public domains) and delivers top average accuracy on ViDoRe V1+V2 (85.93 nDCG@5).
• 🌐 Robust Multilingual & Multi-Format: Strong zero-shot generalization across diverse languages (EN, FR, DE, IT, ES, PT, ZH, etc.) and visual formats (charts, tables, scientific reports, financial filings).
• ⚡ Seamless ColPali Compatibility: Fully integrated with the standard colpali-engine ecosystem and late-interaction scoring pipelines.

---

Architecture

  Text Query ───────► ColQwen3_5 (BiDir Attn) ─────► Query Token Embeddings (128D)
                                                                 │
                                                       Late Interaction (MaxSim) ──► Relevance Score
                                                                 │
Document Image ─────► ColQwen3_5 (Dynamic Vision) ──► Doc Token Embeddings (128D)

1. Vision-Language Backbone: Built on Qwen3.5-4B with interleaved linear and full attention layers.
2. Compact Projection: Projects contextual token states directly into 128-dimensional representations.
3. Late-Interaction Retrieval: Calculates similarity via token-level MaxSim operator across query tokens and document visual tokens.

---

Benchmark Results

Model Footprint

Model	Native Uncompressed Dim / Vector	Relative Width / Vector
EVIE-Preview-4.5B	128	1.0×
colqwen3.5-4.5B-v3	320	2.5×
jina-embeddings-v4	128	1.0×
nemotron-colembed-vl-4b-v2	2560	20×
llama-nemotron-colembed-vl-3b-v2	3072	24×
nemotron-colembed-vl-8b-v2	4096	32×

Relative width compares the native uncompressed width of one token vector with EVIE's 128D output. It is not a total index-size comparison: total storage also depends on vectors per page, data precision, projection or compression settings, and index overhead.

---

ViDoRe V3: 8 Public Domains (nDCG@10)

Evaluated across 8 domains with queries spanning 6 languages (EN, FR, DE, IT, PT, ES):

Model	Avg	CompSci	Energy	Finance (EN)	Finance (FR)	HR	Industrial	Pharma	Physics
EVIE-Preview-4.5B	64.40	80.33	71.45	67.68	53.01	65.65	57.48	68.66	50.96
nemotron-colembed-vl-8b-v2	63.54	79.30	69.82	67.29	51.54	66.32	56.03	67.19	50.84
tomoro-colqwen3-embed-8b	61.60	75.35	68.41	65.08	49.10	63.98	54.41	66.36	50.13
nemotron-colembed-vl-4b-v2	61.42	78.56	67.48	65.02	49.01	62.39	53.91	66.10	48.86
tomoro-colqwen3-embed-4b	60.16	75.44	66.43	63.84	46.83	60.09	53.58	65.74	49.32
llama-nemotron-colembed-vl-3b-v2	59.70	77.09	64.88	64.23	44.41	62.28	51.71	66.04	46.93
colnomic-embed-multimodal-7b	57.64	76.20	63.58	56.57	45.46	58.67	50.13	62.26	48.25
jina-embeddings-v4	57.54	71.81	63.50	59.30	46.10	59.53	50.38	63.09	46.63

🏆 Result: EVIE leads in 7 out of 8 public domains on ViDoRe V3.

---

ViDoRe V1 + V2 (nDCG@5)

Model	Avg	ArxivQA	DocVQA	InfoVQA	ShiftProj	SynAI	SynEnergy	SynGov	SynHealth	Tabfquad	Tatdqa	BioMed	ESGHL	ESG	Econ
EVIE-Preview-4.5B	85.9	91.5	62.9	93.0	94.0	100.0	99.0	98.9	98.9	97.5	81.6	71.0	80.1	66.3	68.3
Ops-Colqwen3-4B	84.9	91.8	66.5	94.0	90.8	99.6	97.3	98.0	99.6	93.6	82.4	65.5	78.6	66.0	64.5
nemotron-colembed-vl-8b-v2	84.8	93.1	68.1	94.6	93.3	100.0	97.9	98.9	99.6	97.7	83.4	66.2	73.2	60.6	60.8
nemotron-colembed-vl-4b-v2	83.9	92.0	67.4	93.3	92.3	99.3	96.2	98.0	98.5	98.1	81.2	64.3	71.4	61.5	60.8
colqwen3.5-4.5B-v3	83.7	91.9	66.6	93.6	90.2	100.0	97.1	97.3	98.9	95.9	84.0	65.3	73.8	58.0	59.9
llama-nemotron-colembed-vl-3b-v2	83.6	90.4	67.2	94.7	92.0	100.0	98.0	98.0	98.9	97.3	81.0	63.2	73.1	58.6	58.6
tomoro-colqwen3-embed-8b	83.5	91.2	66.4	94.5	87.9	99.3	96.7	97.6	99.1	94.2	80.9	65.5	76.0	60.7	59.5
EvoQwen2.5-VL-Retriever-7B-v1	83.4	91.5	65.1	94.1	88.8	99.6	96.6	96.3	98.9	93.6	82.3	65.2	77.0	59.7	59.1
tomoro-colqwen3-embed-4b	83.2	90.6	66.3	94.3	87.4	99.3	96.9	97.2	99.6	94.3	79.9	65.4	74.6	62.4	56.3
llama-nemoretriever-colembed-3b-v1	83.1	88.4	66.2	94.9	90.7	99.6	96.6	97.8	99.3	95.9	80.6	62.7	75.4	57.4	57.8
SauerkrautLM-ColQwen3-8b-v0.1	82.9	93.8	64.7	94.5	90.4	98.6	96.5	96.8	99.3	92.2	84.0	63.3	70.8	57.9	58.0

Avg: Unweighted mean across all 14 tasks. Tasks 1–10: ViDoRe V1. Tasks 11–14: ViDoRe V2.

---

Quick Start

Installation

pip install -r requirements.txt

Python Inference

import torch
from PIL import Image
from colpali_engine.models import ColQwen3_5, ColQwen3_5Processor

# Use the Hugging Face model repository or local directory
model_id = "tencent/EVIE-Preview-4.5B"

# 1. Load model and enable bidirectional attention
model = ColQwen3_5.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="cuda",
    attn_implementation="flash_attention_2",
).eval()
model.enable_bidirectional_attention()

# 2. Load processor
processor = ColQwen3_5Processor.from_pretrained(model_id)

# 3. Prepare inputs
images = [Image.open("document_page.png")]
queries = ["What key insights are presented on this page?"]

image_batch = processor.process_images(images).to(model.device)
query_batch = processor.process_queries(queries).to(model.device)

# 4. Generate multi-vector embeddings and score
with torch.inference_mode():
    image_embeddings = model(**image_batch)
    model.rope_deltas = None  # required before query forward
    query_embeddings = model(**query_batch)

scores = processor.score(query_embeddings, image_embeddings)
print("Late-interaction retrieval scores:", scores)

⚠️ Important: Both model.enable_bidirectional_attention() and resetting model.rope_deltas = None prior to query forward passes are required to replicate the checkpoint's full retrieval performance.

CLI Scoring Tool

You can also use the included infer.py script directly:

python infer.py --query "Quarterly revenue report" --image document_page_1.png --image document_page_2.png

---

Reproducing

Run the end-to-end evaluation benchmark across all visible GPUs:

bash reproduce.sh

Notes

• Automatic Dataset Download: On the first run, reproduce.sh automatically invokes download_data.py to fetch the 22 public ViDoRe datasets (~55 GB) from Hugging Face.
• Custom Dataset Path: To reuse an existing dataset directory, pass it directly:

  bash reproduce.sh /path/to/vidore
  

• Target Aggregates:

ViDoRe V1        nDCG@5    91.72  (10 tasks)
ViDoRe V2        nDCG@5    71.44  (4 tasks)
ViDoRe V1+V2     nDCG@5    85.93  (14 tasks)
ViDoRe V3 public nDCG@10   64.40  (8 domains x 6 languages)

---

Training Details

EVIE was trained on approximately 0.8 million high-quality image-query pairs covering multilingual documents, technical reports, complex financial tables, infographics, and document visual QA.

Hard Negative Mining & Data Filtering

• Dynamic Mining & Verification: Hard negatives are actively mined using intermediate retrievers and re-verified:
  • Candidates that accurately answer the query are promoted to positives.
  • Partially relevant or ambiguous candidates are masked out of the loss.
  • Only strictly irrelevant pages are retained as true hard negatives.
• Quality Filtering: Rows containing empty queries, corrupted images, or degraded text are systematically discarded.

---

Acknowledgements

• Built upon the ColPali Engine developed by Illuin Technology.
• Powered by the Qwen3.5-4B vision-language backbone.
• Evaluated on the ViDoRe Benchmark family.

---

Citation

@misc{tencent2026evie,
  title        = {EVIE-Preview-4.5B},
  author       = {{Tencent}},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/tencent/EVIE-Preview-4.5B}},
  note         = {Multilingual visual document retrieval with compact multi-vector embeddings}
}