NotImplementedError when using PaddleOCR 3.7.0 with PaddlePaddle 3.3.1

Search before asking

• I have searched the PaddleOCR Docs and found no similar bug report.
• I have searched the PaddleOCR Issues and found no similar bug report.
• I have searched the PaddleOCR Discussions and found no similar bug report.

Bug (问题描述)

Bug Report: NotImplementedError when using PaddleOCR 3.7.0

Environment

• Operating System: Linux (Ubuntu/Debian)
• Python Version: 3.12
• PaddleOCR Version: 3.7.0
• PaddlePaddle Version: 3.3.1
• Device: CPU (GPU unavailable)

Problem Description

When running the command paddleocr ocr -i <image_path>, a NotImplementedError is thrown during the text detection phase.

Error Log

$ paddleocr ocr -i test.jpg --lang ch
Creating model: ('PP-LCNet_x1_0_doc_ori', None, None)
Model files already exist. Using cached files.
...
Resized image size (4032x3024) exceeds max_side_limit of 4000. Resizing to fit within limit.
Traceback (most recent call last):
  File "/home/cai/qwen25/bin/paddleocr", line 6, in <module>
    sys.exit(console_entry())
  File "/home/cai/qwen25/lib/python3.12/site-packages/paddleocr/__main__.py", line 26, in console_entry
    main()
  File "/home/cai/qwen25/lib/python3.12/site-packages/paddleocr/_cli.py", line 309, in main
    _execute(args)
  File "/home/cai/qwen25/lib/python3.12/site-packages/paddleocr/_cli.py", line 298, in _execute
    args.executor(args)
  File "/home/cai/qwen25/lib/python3.12/site-packages/paddleocr/_pipelines/ocr.py", line 604, in execute_with_args
    perform_simple_inference(PaddleOCR, params)
  File "/home/cai/qwen25/lib/python3.12/site-packages/paddleocr/_utils/cli.py", line 68, in perform_simple_inference
    for i, res in enumerate(result):
  File "/home/cai/qwen25/lib/python3.12/site-packages/paddlex/inference/pipelines/_parallel.py", line 161, in predict
    yield from self._pipeline.predict(
  File "/home/cai/qwen25/lib/python3.12/site-packages/paddlex/inference/pipelines/ocr/pipeline.py", line 366, in predict
    det_results = list(
  File "/home/cai/qwen25/lib/python3.12/site-packages/paddlex/inference/models/predictors/base_predictor.py", line 117, in __call__
    yield from self.apply(input, **kwargs)
  File "/home/cai/qwen25/lib/python3.12/site-packages/paddlex/inference/models/predictors/base_predictor.py", line 140, in apply
    pred = self.process(batch_data, **kwargs)
  File "/home/cai/qwen25/lib/python3.12/site-packages/paddlex/inference/models/text_detection/predictor.py", line 145, in process
    batch_preds = self.runner(x=x)
  File "/home/cai/qwen25/lib/python3.12/site-packages/paddlex/inference/models/runners/paddle_static/runner.py", line 297, in __call__
    pred = self.infer(x)
  File "/home/cai/qwen25/lib/python3.12/site-packages/paddlex/inference/models/runners/paddle_static/runner.py", line 265, in __call__
    self.predictor.run()
NotImplementedError: (Unimplemented) ConvertPirAttribute2RuntimeAttribute not support [pir::ArrayAttribute<pir::DoubleAttribute>]  (at /paddle/paddle/fluid/framework/new_executor/instruction/onednn/onednn_instruction.cc:116)

Steps to Reproduce

1. Install dependencies:

   pip install paddleocr paddlepaddle -i https://pypi.tuna.tsinghua.edu.cn/simple
   

2. Execute the command:

   paddleocr ocr -i test.jpg --lang ch
   

3. The above error occurs.

Expected Behavior

• Text in the image is recognized successfully
• Recognition results are returned in JSON format

Additional Information

• Model files have been downloaded to ~/.paddlex/official_models/ completely
• The same error occurs with different images
• The --use_angle_cls parameter is prompted as deprecated
• The identical error also appears when calling via Python API

‍♂️ Environment (运行环境)

Project Environment Configuration

Basic Environment

Item	Configuration
Operating System	Linux (Ubuntu/Debian)
Compiler	GCC 12+
Qt Version	Qt 5.15+
CMake Version	3.16+

Virtual Environment

# Virtual environment path
/home/cai/qwen25

# Activation command
source /home/cai/qwen25/bin/activate

Python Dependencies

Package Name	Version	Purpose
paddleocr	3.7.0	OCR recognition (compatibility issue pending fix)
paddlepaddle	3.3.1	PaddlePaddle deep learning framework
numpy	2.3.5	Numerical computation
opencv-contrib-python	4.10.0.84	Image processing

AI Service Configuration

LLM Service: SiliconFlow

• Model: Qwen/Qwen2.5-7B-Instruct
• API Key: Configured via environment variable LLM_API_KEY

OCR Service: Tesseract (Default)

• Language Packs: chi_sim+eng (Chinese & English mixed recognition)
• Installation Command: sudo apt install tesseract-ocr tesseract-ocr-chi-sim

Project Structure

/home/cai/Desktop/build/
├── ai-error-book/           # Core logic module
│   └── src/
│       ├── ai_api.h         # AI API implementation
│       ├── config.h         # Configuration file
│       ├── logger.h         # Logging system
│       └── paddle_ocr.py    # PaddleOCR script
└── ai-error-book-qt/        # Qt graphical interface
    ├── build/               # Compilation output directory
    ├── main.cpp             # Entry file
    ├── mainwindow.cpp/.h    # Main window
    └── ocrwidget.cpp/.h     # OCR recognition component

Startup Instructions

# Navigate to compilation directory
cd /home/cai/Desktop/build/ai-error-book-qt/build

# Set API key and launch the program
export LLM_API_KEY="your_api_key"
./ai-error-book-qt

Environment Variable Description

Variable Name	Purpose	Default Value
LLM_API_KEY	SiliconFlow API key	None (Required)
OCR_PROVIDER	OCR service provider	tesseract
LOG_LEVEL	Logging level	INFO

Minimal Reproducible Example (最小可复现问题的Demo)

Minimal Reproducible Example

Below is a minimal reproducible script to demonstrate the PaddleOCR bug:

#!/usr/bin/env python3
"""
Minimal Reproducible Example for PaddleOCR bug
GitHub Issue: NotImplementedError when using PaddleOCR 3.7.0
"""

import sys
from paddleocr import PaddleOCR

def main():
    if len(sys.argv) < 2:
        print("Usage: python reproduce_bug.py <image_path>")
        sys.exit(1)
    
    image_path = sys.argv[1]
    
    try:
        # Initialize OCR with default settings
        ocr = PaddleOCR(lang='ch')
        
        # Try to recognize text
        print(f"[INFO] Trying to recognize image: {image_path}")
        result = ocr.predict(image_path)
        
        print("[INFO] Recognition successful!")
        print(f"[INFO] Number of text boxes: {len(result)}")
        
    except Exception as e:
        print(f"[ERROR] {type(e).__name__}: {str(e)}")
        import traceback
        traceback.print_exc()
        sys.exit(1)

if __name__ == "__main__":
    main()

Usage

# 1. Activate the virtual environment
source /home/cai/qwen25/bin/activate

# 2. Create the test script
cat > reproduce_bug.py << 'EOF'
#!/usr/bin/env python3
from paddleocr import PaddleOCR
import sys

ocr = PaddleOCR(lang='ch')
result = ocr.predict(sys.argv[1])
print(f"Success: {len(result)} text boxes found")
EOF

# 3. Run the test (replace with your actual image path)
python reproduce_bug.py /home/cai/Desktop/微信图片_20260613212028_1_10.jpg

Expected Output (Normal Execution)

[INFO] Trying to recognize image: test.jpg
[INFO] Recognition successful!
[INFO] Number of text boxes: 42

Actual Output (Error Occurred)

[INFO] Trying to recognize image: test.jpg
[ERROR] NotImplementedError: (Unimplemented) ConvertPirAttribute2RuntimeAttribute not support [pir::ArrayAttribute<pir::DoubleAttribute>]
Traceback (most recent call last):
  File "reproduce_bug.py", line 16, in main
    result = ocr.predict(image_path)
  ...

---

Test File

You can generate a simple test image with the script below if needed:

# create_test_image.py
from PIL import Image, ImageDraw, ImageFont

img = Image.new('RGB', (400, 200), color='white')
draw = ImageDraw.Draw(img)
draw.text((20, 50), "Hello World!", fill='black')
draw.text((20, 100), "你好世界！", fill='black')
img.save('test.jpg')

This minimal example helps developers quickly troubleshoot the problem.