Always set 0 as the port for distributed training (?)

[frostedoyster]

"In Linux, if you specify 0 as the port number when creating a socket, the system will automatically assign an available port from the dynamic or ephemeral port range."

Source: ChatGPT

That would allow us to remove one parameter for models that can do distributed training (e.g.

metatrain/src/metatrain/experimental/soap_bpnn/default-hypers.yaml

Line 28 in 68bd9f5

distributed_port: 39591

)
@Luthaf

[Luthaf]

That sounds like a good idea, ideally printing the port we picked in the output somewhere?

We could also leave the option to manually specify a port in the hypers, but default it to 0