Prediction value of ResNet 3D Pytorch Multilabel classification is too small

Hi,
i trained ResNet3D using dataset multilabel and im using BCELoss with sigmoid layer i have 31% topKcategorical accuracy (i have 23 labels), i trying to predict test-set but model given all labels a proba too small(0.0…),
example:
[0.061, 0.032, 0.036, 0.009, 0.024, 0.009, 0.025, 0.017, 0.021,
0.014, 0.013, 0.017, 0.078, 0.01 , 0.068, 0.014, 0.068, 0.035,
0.04 , 0.044, 0.03 , 0.04 , 0.015]
any help ?