Pytorch->ONXX->TRT incompatibility: TensorRT not supporting INT64 but PyTorch indexing forces INT64

So I have noticed that TRT does not support INT64 which is a big problem when it comes to NLU models, since most of them use indexing to get embeddings, and Pytorch forces indexing of tensors to be done using Long/INT64.

Is there any workaround or fix for this? does this mean I cannot use TensorRT for NLP/NLU?