Celery tasks and Torch multiprocessing

I have a script with neural networks here is example of init

import torch
from torchvision.models.detection import fasterrcnn_resnet50_fpn
from torchvision.models.detection import FasterRCNN_ResNet50_FPN_Weights

class Detector():
....
def _initialize_detector(self):
        """Initialize object detection model."""
        model = fasterrcnn_resnet50_fpn(weights=FasterRCNN_ResNet50_FPN_Weights.DEFAULT)
        model.to(device).eval()
        return model

it is working fine, but when I create celery task

@celery_app.task
def video_detect_task(
    video_file: str,
    bvh_filename: str,
):
    """Celery task for video_detect_task module.

    :param video_path: path to input video file
    :param bvh_filename: path to save bvh file
    """
    task_id = current_task.request.id
    try:
        model = Detector.load(Config())
        frames_num = get_video_duration(video_file)
        logger.info('frames: ',extra = {'number ':frames_num})
        skeleton_name, _ = os.path.splitext(os.path.basename(bvh_filename))
        skeleton_full_path = os.path.dirname(bvh_filename) + "/" + skeleton_name
        save_character_path = ".".join([skeleton_full_path, 'bvh'])

        status = model.process(
            video_file=video_file,
            bvh_file_name=save_character_path
        )

then it code not working it cant even load the model by logs I see this:

22.5%-12-22 00:10:27,496: WARNING/ForkPoolWorker-2] 
22.5%-12-22 00:10:27,496: WARNING/ForkPoolWorker-2] 
22.6%-12-22 00:10:27,497: WARNING/ForkPoolWorker-2] 
22.7%-12-22 00:10:27,497: WARNING/ForkPoolWorker-2] 
22.8%-12-22 00:10:27,497: WARNING/ForkPoolWorker-2]

Thank you!

BUT when I do

torch.set_num_threads(1)

celery task starts works also okay only long. All computetaion oc CPU.

I think it is some problem with Celery and Troech multiprocessing

I run celery with this command

command: ["celery", "-A", "video_detect_task.tasks.celery_app.celery_app", "worker", "-Q", "video_detect_task_queue", "--loglevel=info"]