Update start.py

start.py

@@ -1,3 +1,15 @@
 import subprocess
+import os
 
-subprocess.run("uvicorn app:app --host 0.0.0.0 --port 7860", shell=True)
+# Uninstall the existing llama-cpp-python package
+subprocess.run("pip uninstall -y llama-cpp-python", shell=True)
+
+# Set the required environment variables for installing with CUDA support
+os.environ["CMAKE_ARGS"] = "-DLLAMA_CUBLAS=on"
+os.environ["FORCE_CMAKE"] = "1"
+
+# Install llama-cpp-python with CUDA support
+subprocess.run("pip install llama-cpp-python", shell=True)
+
+# Start the Hugging Face Space
+subprocess.run("uvicorn app:app --host 0.0.0.0 --port 7860", shell=True)