Elixir Nx EXLA Support

[marcnnn]

Hello,

Would you be able to help me get Elixir Nx EXLA support?

In the elixir-nx/nx#1504 for Apple support, there are workarounds for limitations.
Could you help me list the limitation that need to be addressed for intel support?
And hint me to the right part of documentation, how to get started?
In a research cluster, I could try it on hardware.
But I am basically just a user of ML libraries but would love to try the big VRAM that the Intel GPUs have.
Since soon Nx supports sharding and distribution out of the box, it would be amazing to try big models distributed on up to 32 Intel GPUs with 125 GB VRAM and Infiniband connection.

Thanks in Advance