model inference but self.training is save true

[morestart]

i use self.training param to judge what data return.

return x if self.training else (torch.cat(z, 1), x)

but when i load my model, i use debug mode find that the self.training is save True.

self.model = CustomModel.load_from_checkpoint(model_path)
self.model.training = False

i use above code change model.training status, but its not work

this is my inference full code:

class CustomModelInference:
    def __init__(
            self,
            model_path: str,
            conf_thres: float = 0.25,
            iou_thres: float = 0.45,
            max_det: int = 1000,
            device: str = 'cuda:0',
            need_classes: list | None = None
    ):
        self.conf_thres = conf_thres
        self.iou_thres = iou_thres
        self.max_det = max_det
        self.device = device
        self.need_classes = need_classes

        self.model = CustomModel.load_from_checkpoint(model_path)
        self.model.training = False
        self.model.to(device)
        self.stride = int(self.model.stride.max())
        self.names = self.model.names
        self.imgsz = self.model.imgsz

    @torch.no_grad()
    def infer(self, img: np.ndarray):
        imgsz = check_img_size(self.imgsz, s=self.stride)
        cudnn.benchmark = True
        img = letterbox(img, imgsz, stride=self.stride, auto=True)[0]
        # img = np.stack(img, 0)
        if len(img.shape) == 3:
            img = img[None]
        img = img[..., ::-1].transpose((0, 3, 1, 2))
        img = np.ascontiguousarray(img)
        img = torch.from_numpy(img).to(self.device)
        img = img.float()
        img = img / 255.0

        out, train_out = self.model(img)

[morestart]

fine, i know the answer, i have to set model.eval()......

[awaelchli]

training mode is the default in pytorch, so this seems to be correct.

import torch 
class MyModel(torch.nn.Module):
    pass

model = MyModel()
model.training
Out[14]: True