Pretrained resnet to multilabeling classification

I have a question. I operate on the dateset where every image can be classify to more than one class. The input (label) for every element looks like a vector: [0,1,0,0,1] what means that some element belongs to two different classes. I would like to use pretrained resnet for this task but I am not sure how to implement it. Which loss function is the best for that kind of problem?

Hello, crossentropyloss is a good place to start: CrossEntropyLoss — PyTorch 1.8.1 documentation

Depending on the level of noise in your dataset however you may need to utilize a more customized loss function

Yep, but what about resnet? How make it to return 1x6 output?