Question, llama cpp x4 and x16 usage

[FerLuisxd]

I could not find in the documentation how to set the x4 or x16 sampling for the vision model.

[tc-mb]

Hi, the 4x/16x mixed visual token compression mode is not yet exposed in the current llama.cpp build — only the default compression rate is wired up. We'll submit a small follow-up PR to llama.cpp soon to add a switch for this. Will update here once it's merged. Thanks for the heads-up!