CPU Inference only works on a single core

[KodeMunkie]

I run my (test) setup on 2 Mini PCs, running Ubuntu Server one an Intel N95, the other and Intel N100. Both clock at ~3.4Ghz and each has 4 cores (4 real cores, no hyperthreading on this CPU).

It seems according to HTOP that CPU inference only runs on a 1 core per node instead of all 4 cores.

At first I thought this was either a design decision to use maximum core clock speed but now I'm not certain.

Please can you advise on how I can either config and run multiple CPU instances of exo on the same PC (1 exo instance per core or just document how to use linux taskset with exo), or alternatively provide a config/environment flag that specifies the number of real cores and/or CPU affinity?

[stillbanbo]

Same issue for two arm linux devices.

Used "CLANG=1 python main.py".

[m-schenker]

I am affected by this issue as well. I installed Exo on Fedora 41 and it uses a single core only. Like this Exo is unfortunately not usable for me.

[Azmodeszer]

Any news here? Having the same problem. It's running on two 20-thread VMs and only using half a thread on each, basically. :D

Unsure if it should just use 100% CPU out-of-the-box and this is a bug or if there's an option to enable multithreading somewhere.