richardyoung/ wevisdoc-4b:IQ3_M

10 1 week ago

Tencent WeVisDoc-4B: top end-to-end document parser (OmniDocBench v1.6 95.38). Page to Markdown. Images work.

vision 
ollama run richardyoung/wevisdoc-4b:IQ3_M

Details

1 week ago

31764499bc9d · 3.0GB

model
qwen3vl
·
4.41B
·
(!unknown_file_type 27!)
projector
clip
·
415M
·
F16

Readme

WeVisDoc-4B

Tencent WeVisDoc-4B: top end-to-end document parser (OmniDocBench v1.6 95.38). Page to Markdown. Images work.

🚀 Overview

GGUF build of tencent/WeVisDoc-4B, Tencent’s 4B WeVisDoc, an end-to-end document parser fine-tuned from Qwen3-VL that turns a page image into structured Markdown, with LaTeX formulas and HTML tables. It ships with its vision projector, so image input works in Ollama.

Prompts: send a page image and ask for Markdown (e.g. Parse this document page into Markdown.). Tencent’s own client and its exact prompt are in the WeVisDoc repository.

🎯 Key Features

  • OmniDocBench v1.6 Overall 95.38, first among the end-to-end parsers Tencent compared
  • Page image in, Markdown out: formulas as LaTeX, tables as HTML
  • English and Chinese
  • Apache-2.0

🏷️ Available Versions

Tag Size (+ 0.84 GB projector) BPW Notes
IQ3_M 2.1 GB 3.66 Smallest
IQ4_XS 2.5 GB 4.25 Great quality-size balance
latest / Q4_K_M 2.7 GB 4.85 Recommended
Q5_K_M 3.2 GB 5.68 Higher quality
Q6_K 3.6 GB 6.56 Very high quality
Q8_0 4.7 GB 8.5 Near-lossless

💻 Quick Start

ollama run richardyoung/wevisdoc-4b
# then drag an image file into the prompt, or send it in the API "images" field

🛠️ Use Cases

  • Converting scanned pages and PDFs (rendered to images) into Markdown
  • Formula-heavy papers (LaTeX output)
  • Tables to HTML for RAG pipelines

✅ Verified

Tested through Ollama before upload (Q4_K_M and Q8_0) on a synthetic invoice with the prompt Parse this document page into Markdown.; Q4_K_M returned: ‘## INVOICE 4471 / Date: 2026-09-28 / Widgets x3 ….. \(42.00 / TOTAL: \)86.20’.

🔧 Technical Details

  • Base Model: tencent/WeVisDoc-4B
  • Parameters: 4.4B
  • Context Length: 256K tokens
  • Quantization: GGUF via llama.cpp with an importance matrix (calibration_datav3); vision projector at F16
  • License: Apache-2.0 (inherited from the base model)

🙏 Acknowledgments

  • Base Model: Tencent
  • Quantization: llama.cpp

Built & maintained by Richard Young · DeepNeuro