exllama + triton inference server?

[arbi-dev]

would it be possible to deploy exllama with Nvidia's open-source Triton Inference Server? Triton inference server has several useful features for deploying llms in production.
https://developer.nvidia.com/triton-inference-server

As I understand it allows custom back-ends, so was wondering if this could potentially work or if anyone has tried?
https://github.com/triton-inference-server/backend