加载中…
加载中…
PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model
🔥 Official Website: Baidu AI Studio | 📝 arXiv: Technical Report
PaddleOCR-VL is a SOTA and resource-efficient model tailored for document parsing. Its core component is PaddleOCR-VL-0.9B, a compact yet powerful vision-language model (VLM) that integrates a NaViT-style dynamic resolution visual encoder with the ERNIE-4.5-0.3B language model to enable accurate element recognition. This innovative model efficiently supports 109 languages and excels in recognizing complex elements (e.g., text, tables, formulas, and charts), while maintaining minimal resource consumption. Through comprehensive evaluations on widely used public benchmarks and in-house benchmarks, PaddleOCR-VL achieves SOTA performance in both page-level document parsing and element-level recognition. It significantly outperforms existing solutions, exhibits strong competitiveness against top-tier VLMs, and delivers fast inference speeds. These strengths make it highly suitable for practical deployment in real-world scenarios.
Compact yet Powerful VLM Architecture: We present a novel vision-language model that is specifically designed for resource-efficient inference, achieving outstanding performance in element recognition. By integrating a NaViT-style dynamic high-resolution visual encoder with the lightweight ERNIE-4.5-0.3B language model, we significantly enhance the model’s recognition capabilities and decoding efficiency. This integration maintains high accuracy while reducing computational demands, making it well-suited for efficient and practical document processing applications.
SOTA Performance on Document Parsing: PaddleOCR-VL achieves state-of-the-art performance in both page-level document parsing and element-level recognition. It significantly outperforms existing pipeline-based solutions and exhibiting strong competitiveness against leading vision-language models (VLMs) in document parsing. Moreover, it excels in recognizing complex document elements, such as text, tables, formulas, and charts, making it suitable for a wide range of challenging content types, including handwritten text and historical documents. This makes it highly versatile and suitable for a wide range of document types and scenarios.
Multilingual Support: PaddleOCR-VL Supports 109 languages, covering major global languages, including but not limited to Chinese, English, Japanese, Latin, and Korean, as well as languages with different scripts and structures, such as Russian (Cyrillic script), Arabic, Hindi (Devanagari script), and Thai. This broad language coverage substantially enhances the applicability of our system to multilingual and globalized document processing scenarios.
2025.11.07 🚀 Enabled flash-attn in the transformers library to achieve faster inference with PaddleOCR-VL-0.9B.2025.11.04 🌟 PaddleOCR-VL-0.9B is now officially supported on vLLM .2025.10.29 🤗 Supports calling the core module PaddleOCR-VL-0.9B of PaddleOCR-VL via the transformers library.2025.10.16 🚀 We release PaddleOCR-VL, — a multilingual documents parsing via a 0.9B Ultra-Compact Vision-Language Model with SOTA performance.Install PaddlePaddle and PaddleOCR:
# The following command installs the PaddlePaddle version for CUDA 12.6. For other CUDA versions and the CPU version, please refer to https://www.paddlepaddle.org.cn/en/install/quick?docurl=/documentation/docs/en/develop/install/pip/linux-pip_en.html
python -m pip install paddlepaddle-gpu==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cu126/
python -m pip install -U "paddleocr[doc-parser]>=3.4.0"
CLI usage:
paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --pipeline_version v1
Python API usage:
from paddleocr import PaddleOCRVL
pipeline = PaddleOCRVL(pipeline_version="v1")
output = pipeline.predict("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png")
for res in output:
res.print()
res.save_to_json(save_path="output")
res.save_to_markdown(save_path="output")
Start the VLM inference server:
You can start the vLLM inference service using one of two methods:
Method 1: PaddleOCR method
docker run \
--rm \
--gpus all \
--network host \
ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu \
paddleocr genai_server --model_name PaddleOCR-VL-0.9B --host 0.0.0.0 --port 8080 --backend vllm
Method 2: vLLM method
Call the PaddleOCR CLI or Python API:
paddleocr doc_parser \
-i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png \
--pipeline_version v1 \
--vl_rec_backend vllm-server \
--vl_rec_server_url http://127.0.0.1:8080/v1
from paddleocr import PaddleOCRVL
pipeline = PaddleOCRVL(pipeline_version="v1", vl_rec_backend="vllm-server", vl_rec_server_url="http://127.0.0.1:8080/v1")
output = pipeline.predict("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png")
for res in output:
res.print()
res.save_to_json(save_path="output")
res.save_to_markdown(save_path="output")
For more usage details and parameter explanations, see the documentation.
Currently, we support inference using the PaddleOCR-VL-0.9B model with the transformers library, which can recognize texts, formulas, tables, and chart elements. In the future, we plan to support full document parsing inference with transformers. Below is a simple script we provide to support inference using the PaddleOCR-VL-0.9B model with transformers.
[!NOTE] Note: We currently recommend using the official method for inference, as it is faster and supports page-level document parsing. The example code below only supports element-level recognition.
from PIL import Image
import torch
from transformers import AutoModelForCausalLM, AutoProcessor
# ---- Settings ----
model_path = "PaddlePaddle/PaddleOCR-VL"
image_path = "test.png"
task = "ocr" # Options: 'ocr' | 'table' | 'chart' | 'formula'
# ------------------
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
PROMPTS = {
"ocr": "OCR:",
"table": "Table Recognition:",
"formula": "Formula Recognition:",
"chart": "Chart Recognition:",
}
image = Image.open(image_path).convert("RGB")
model = AutoModelForCausalLM.from_pretrained(
model_path, trust_remote_code=True, torch_dtype=torch.bfloat16
).to(DEVICE).eval()
processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)
messages = [
{"role": "user",
"content": [
{"type": "image", "image": image},
{"type": "text", "text": PROMPTS[task]},
]
}
]
inputs = processor.apply_chat_template(
messages,
tokenize=True,
add_generation_prompt=True,
return_dict=True,
return_tensors="pt"
).to(DEVICE)
outputs = model.generate(**inputs, max_new_tokens=1024)
outputs = processor.batch_decode(outputs, skip_special_tokens=True)[0]
print(outputs)
# ensure the flash-attn2 is installed
pip install flash-attn --no-build-isolation
import torch
from transformers import AutoModelForCausalLM, AutoProcessor
from PIL import Image
# ---- Settings ----
model_path = "PaddlePaddle/PaddleOCR-VL"
image_path = "test.png"
task = "ocr" # ← change to "table" | "chart" | "formula"
# ------------------
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
model = AutoModelForCausalLM.from_pretrained(
model_path,
trust_remote_code=True,
torch_dtype=torch.bfloat16,
attn_implementation="flash_attention_2",
).to(dtype=torch.bfloat16, device=DEVICE).eval()
processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)
PROMPTS = {
"ocr": "OCR:",
"table": "Table Recognition:",
"chart": "Chart Recognition:",
"formula": "Formula Recognition:",
}
messages = [
{
"role": "user",
"content": [
{"type": "image", "image": Image.open(image_path).convert("RGB")},
{"type": "text", "text": PROMPTS[task]}
]
}
]
inputs = processor.apply_chat_template(
messages,
tokenize=True,
add_generation_prompt=True,
return_dict=True,
return_tensors="pt"
).to(DEVICE)
with torch.inference_mode():
out = model.generate(
**inputs,
max_new_tokens=1024,
do_sample=False,
use_cache=True
)
outputs = processor.batch_decode(out, skip_special_tokens=True)[0]
print(outputs)
Notes:
- The metrics are from MinerU, OmniDocBench, and our own internal evaluations.
Comparison of OmniDocBench-OCR-block Performance
PaddleOCR-VL’s robust and versatile capability in handling diverse document types, establishing it as the leading method in the OmniDocBench-OCR-block performance evaluation.
Comparison of In-house-OCR Performance
In-house-OCR provides a evaluation of performance across multiple languages and text types. Our model demonstrates outstanding accuracy with the lowest edit distances in all evaluated scripts.
Comparison of In-house-Table Performance
Our self-built evaluation set contains diverse types of table images, such as Chinese, English, mixed Chinese-English, and tables with various characteristics like full, partial, or no borders, book/manual formats, lists, academic papers, merged cells, as well as low-quality, watermarked, etc. PaddleOCR-VL achieves remarkable performance across all categories.
Comparison of In-house-Formula Performance
In-house-Formula evaluation set contains simple prints, complex prints, camera scans, and handwritten formulas. PaddleOCR-VL demonstrates the best performance in every category.
Comparison of In-house-Chart Performance
The evaluation set is broadly categorized into 11 chart categories, including bar-line hybrid, pie, 100% stacked bar, area, bar, bubble, histogram, line, scatterplot, stacked area, and stacked bar. PaddleOCR-VL not only outperforms expert OCR VLMs but also surpasses some 72B-level multimodal language models.
We would like to thank ERNIE, Keye, MinerU, OmniDocBench for providing valuable code, model weights and benchmarks. We also appreciate everyone's contribution to this open-source project!
If you find PaddleOCR-VL helpful, feel free to give us a star and citation.
@misc{cui2025paddleocrvlboostingmultilingualdocument,
title={PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model},
author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Handong Zheng and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
year={2025},
eprint={2510.14528},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2510.14528},
}