[PyCuda-Torch] Can i inference concurrency input tensor on One GPU?

It means the all tensors is not one batch.
for example, pytorch server had been received two batches from each other different application at same time.

Can i concurrently process on model of pytorch-cuda ?

Thanks!