diff --git a/hw/straylight/default.nix b/hw/straylight/default.nix
index 39e9ae5..4e9273e 100644
--- a/hw/straylight/default.nix
+++ b/hw/straylight/default.nix
@@ -19,7 +19,8 @@ let
     cache-type-k = q8_0
     cache-type-v = q8_0
     n-gpu-layers = 999
-    no-mmap = true
+    ; llama.cpp 0.4.1 folded --no-mmap into --load-mode (none = read into memory, no mmap)
+    load-mode = none
     ; llama.cpp splits ctx-size across parallel slots: 262144 / 2 = 131072 per slot.
     ctx-size = 262144
     parallel = 2