home_3b_v3 model very slow on Intel N100 homelab

[Bittabola]

Hi all,

I’ve just installed home-llm via HACS, was able to create a Voice Assistant with it. However, it’s very very slow to respond.

My hardware:

• HAOS installed in Proxmox
• Intel N100 CPU, all 4 cores exposed to VM
• 12GB RAM exposed to VM
• 49 entities exposed in HA

My simple questions like What’s the status of blinds is taking 220 seconds to get a response. When I run qwen2.5-coder:1.5b via Ollama, it’s much much faster.

Is there anything wrong with my setup?

[acon96]

There have been various reports of the N100 CPU having weird issues with llama.cpp (the backend). Can you enable debug logs and provide those? It should print out what optimizations are enabled on that CPU, which I assume doesn't include AVX or other optimized SIMD instructions.