How does llama-server pick which MoE experts go on the GPU and which stay on the

<!-- SC_OFF --><div class="md"><p>I know llama.cpp can use -cmoe switch to load non shared experts in MoE models to RAM and run by CPU and left the other weights and KV cache on

<!-- SC_OFF --><div class="md"><p>Is there an easy way to know if a model is using CPU/RAM (and not only GPU/VRAM)?<br /> (I think standard verbose output, which got shorter, says

<!-- SC_OFF --><div class="md"><p>I have been using llama.cpp to run some models recently. For example, I've been running GLM-4.7-Flash with this command <code>.\llama-server.exe

<!-- SC_OFF --><div class="md"><p>currently in order to use MTP one needs to enable it in the starting argument of llama server.</p> <p>--spec-type draft-mtp<br />

<!-- SC_OFF --><div class="md"><p>Hi,</p> <p>I'm running a local server with three AMD MI50. Tensor parallelism is not an option since it's very slow with PCIe 3.0 and those cards