From ff29e724426ec1d376d721883e93296d02d3aed7 Mon Sep 17 00:00:00 2001
From: wmantly <wmantly@users.noreply.git.theta42.com>
Date: Fri, 18 Sep 2026 02:07:18 +0000
Subject: [PATCH] fix(memory): tune cache-ram to 12288MB to prevent host RAM
 OOM-kills on 48GB servers

---
 scripts/start-server.sh | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/scripts/start-server.sh b/scripts/start-server.sh
index 90f4571..d221bd3 100755
--- a/scripts/start-server.sh
+++ b/scripts/start-server.sh
@@ -42,10 +42,9 @@ exec /opt/llama.cpp-xrip/build-nccl/bin/llama-server \
   -c 131072 \
   --parallel 1 \
   --cache-prompt \
-  --cache-reuse 64 \
   --cache-idle-slots \
   --slot-prompt-similarity 0.10 \
-  --cache-ram 32768 \
+  --cache-ram 12288 \
   --slot-save-path /var/cache/llama-slots \
   --split-mode tensor \
   --flash-attn on \