about DNN using cvxpylayer to handle the output

[babuqi]

I'm using a DNN to solve a problem, and i set the batchsize 10, every sample get a 4-dimension vector, so the output of the DNN is a 10*4 tensor.
Now i want to use cvxpylayer to project the output into my feasible region, and use the adjusted output to construct a loss function to do back propragation, by the way, one constraint of my porblem is Ax>b, A is a matrix. How should set the DCO to solve the proj. task in one times, ranther than one sample by one sample, or, how to make the grad between the output of dnn and DCO connected.