Modified cache to hopefully reduce poor hit quality

qwen3.5-397B-A17B-IQ4_KSS.sh

@@ -1,7 +1,7 @@
 #!/bin/bash
 #grab container image name
 MODEL_NAME="Qwen3.5-397B-A17B"
-CONT_NAME="ik-$MODEL_NAME"
+CONT_NAME="LLM-$MODEL_NAME"
 
 MODEL="huggingface/hub/models--ubergarm--Qwen3.5-397B-A17B-GGUF/snapshots/bb9c5e48cdcad8716ef0dfff1f48f43b34b89cb2/IQ4_KSS/Qwen3.5-397B-A17B-IQ4_KSS-00001-of-00006.gguf"
 MODEL_FOLDER="/home/jared/.cache"
@@ -20,7 +20,7 @@ docker container create --name $CONT_NAME --network llms -p $PORT:$PORT -v $MODE
 	--health-interval 5s \
 	--health-timeout 5s \
 	--health-start-period 20s \
-	$IK_IMAGE \
+	$IMAGE \
 	/ik_llama.cpp/build/bin/llama-server \
 	--model /model/$MODEL \
 	--alias $MODEL_NAME \
@@ -28,8 +28,11 @@ docker container create --name $CONT_NAME --network llms -p $PORT:$PORT -v $MODE
 	-ngl 999 \
 	--n-cpu-moe 38 \
 	--parallel 2 --threads 32 \
+	-cram 32768 -cram-n-min 128 -crs .9 --ctx-checkpoints 512 \
 	--host $CONT_NAME \
 	--port $PORT --jinja --no-mmap
 
 echo "Done!"
 echo $CONT_NAME
+
+docker container start -a $CONT_NAME