Experts per token option missing at model load screen for GGUF (reg and _hff loader)

[David-AU-github]

Describe the bug

No option to select "number" of experts to use for MOE models, in GGUF format.

Is there an existing issue for this?

• I have searched the existing issues

Reproduction

Attempt to load a MOE GGUF model - either loader -> llamacpp OR llamacpp_HF

Screenshot

No response

Logs

n/a

System Info

Windows 11, Nvidia 4060Ti Vram.
Also reported by other users, attempt to load any of my MOE models (10), from my repo.
(DavidAU at hug..face).

NOTE: Moes are built using NEWEST llamacpp versions, not older llamacpp/may be "broken" moes.

For llama-server.exe I use this hack:

./llama-server -m i:/llm/David_AU/testfiles/Grand-Horror-MOE-4X8-series1-Q4_K_S.gguf -c 4096 -ngl 99 --override-kv llama.expert_used_count=int:4  

Example using PowerShell.