Keep the iGPU out of powersave while llama-server runs, and pair Laguna S 2.1 with Poolside's DFlash draft (f16 draft KV, n-max 4).
Keep the iGPU out of powersave while llama-server runs, and pair Laguna S 2.1 with Poolside's DFlash draft (f16 draft KV, n-max 4).