diff --git a/systems/x86_64-linux/halo/llama-server-27B.nix b/systems/x86_64-linux/halo/llama-server-27B.nix index ab25831..d86cee9 100644 --- a/systems/x86_64-linux/halo/llama-server-27B.nix +++ b/systems/x86_64-linux/halo/llama-server-27B.nix @@ -46,6 +46,7 @@ "-c 524288" "--fit on" "--slot-save-path %C/llama-server/kv-slots-27B" + "--cache-ram 0" ]; Restart = "on-failure"; RestartSec = 10;