diff --git a/.devops/handler.py b/.devops/handler.py
index a1fa0cbde..14249b3f7 100644
--- a/.devops/handler.py
+++ b/.devops/handler.py
@@ -1,8 +1,10 @@
+import subprocess
 import runpod
 import os
 import time
 
-sleep_time = int(os.environ.get('SLEEP_TIME', 1))
+llama_cmd = os.environ.get('LLAMA_CMD', "/server --host 0.0.0.0 --threads 8 -ngl 999 -np 8 -cb -m model.gguf -c 16384")
+subprocess.Popen(llama_cmd.split(' '))
 
 ## load your model(s) into vram here
 
diff --git a/.devops/main-cuda.Dockerfile b/.devops/main-cuda.Dockerfile
index fa0934e62..44747f000 100644
--- a/.devops/main-cuda.Dockerfile
+++ b/.devops/main-cuda.Dockerfile
@@ -27,7 +27,7 @@ ENV LLAMA_CUDA_MMV_Y=2
 ENV LLAMA_CUDA_DMMV_X=64
 ENV LLAMA_CUDA_F16=true
 
-RUN make -j
+RUN make
 
 # Accept the build argument into an environment variable
 ARG MODEL_URL