first commit

Services/llamacpp/Scripts/startGeneralFast.sh

@@ -0,0 +1,20 @@
+#!/bin/bash
+
+exec /app/llama-server $GENERAL_FAST_MODEL \
+  -c $GENERAL_CONTEXT_SIZE -n 128 \
+  --temp 0.6 --top-p 0.9 --top-k 40 --repeat-penalty 1.05 \
+  --flash-attn auto --threads -1 --threads-batch -1 --threads-http -1 \
+  --jinja \
+  --timeout 600 --host 0.0.0.0 --port 8091 &
+PID=$!
+
+cleanup() {
+    echo "Stopping llama-server..."
+    kill $PID 2>/dev/null
+    wait $PID 2>/dev/null
+    exit 0
+}
+
+trap cleanup SIGTERM SIGINT
+
+wait $PID