load_model.py

# from ultralytics import YOLOv10
import torch
from config.set_config import set_configuration

set_config_project = set_configuration()
layout_model_weights_path = set_config_project.layout_model_weights_path
no_of_threads = set_config_project.no_of_threads

# def load_model_for_process(detection_model_path=layout_model_weights_path):
#     """
#     Load model in each subprocess to avoid CUDA initialization issues
    
#     Returns:
#         Model loaded in appropriate device
#     """
#     # Your model loading logic
#     device = "cuda" if torch.cuda.is_available() else "cpu"
#     # print(f"Using device: {device}")

#     model = YOLOv10(detection_model_path).to(device)
#     class_names = model.names
#     class_names["11"] = "Table-header"   
#     class_names["12"] = "Portfolio-Company-Table"  

#     return model, class_names

import torch

from ultralytics import YOLO

# def load_model_for_process(detection_model_path=layout_model_weights_path):
#     """
#     Load model in each subprocess to avoid CUDA initialization issues
    
#     Returns:
#         Model loaded in appropriate device
#     """
#     # Your model loading logic
#     device = "cuda" if torch.cuda.is_available() else "cpu"
#     # print(f"Using device: {device}")

#     model = YOLO(detection_model_path).to(device)
#     class_names = model.names
#     class_names["11"] = "Table-header"   
#     class_names["12"] = "Portfolio-Company-Table"  
#     print("YOLOV12"*10)

#     return model, class_names


'''Below code for docling heron model'''

from transformers import RTDetrV2ForObjectDetection, RTDetrImageProcessor
# MODEL_NAME_DOCLING = "ds4sd/docling-layout-heron"
MODEL_NAME_DOCLING = layout_model_weights_path

def load_model_for_process(model_name=MODEL_NAME_DOCLING):
    """
    Load the Docling Heron model and image processor in each subprocess to avoid CUDA initialization issues.
    
    Returns:
        Tuple of (model, image_processor, class_names)
    """
    device = "cuda" if torch.cuda.is_available() else "cpu"
    print(f"Using device: {device}")

    # Load the image processor and model
    image_processor = RTDetrImageProcessor.from_pretrained(model_name)
    model = RTDetrV2ForObjectDetection.from_pretrained(model_name).to(device)

    # Define class names mapping
    class_names = {
        0: "Caption",
        1: "Footnote",
        2: "Formula",
        3: "List-item",
        4: "Page-footer",
        5: "Page-header",
        6: "Picture",
        7: "Section-header",
        8: "Table",
        9: "Text",
        10: "Title",
        11: "Document Index",
        12: "Code",
        13: "Checkbox-Selected",
        14: "Checkbox-Unselected",
        15: "Form",
        16: "Key-Value Region",
        # Additional classes for compatibility with existing pipeline
        17 : "Table-header",
        18 : "Portfolio-Company-Table"
    }

    return model, image_processor, class_names