Update app.py

app.py

@@ -33,9 +33,10 @@ if __name__ == "__main__":
     os.system("ulimit -l unlimited")
 
     hf_hub_download("TheBloke/deepseek-coder-1.3b-base-GGUF", "deepseek-coder-1.3b-base.Q5_K_M.gguf", local_dir="./")
-
-    global model_l = Llama(model_path="./deepseek-coder-1.3b-base.Q5_K_M.gguf", n_ctx=16192, n_gpu_layers=0, n_threads=2, use_mlock=True)
-    global total = 0
+    global model_l
+    global total
+    model_l = Llama(model_path="./deepseek-coder-1.3b-base.Q5_K_M.gguf", n_ctx=16192, n_gpu_layers=0, n_threads=2, use_mlock=True)
+    total = 0
     
     import uvicorn
     uvicorn.run(app, host="0.0.0.0", port=7860)