Update app.py

app.py

@@ -45,10 +45,7 @@ def install_packages():
     # Upgrade pip
     run_command("pip install --upgrade pip")
 
-     # Set environment variables for CUDA support
-    os.environ["CUDA_HOME"] = "/usr/local/cuda"  # Adjust this path if needed
-    os.environ["CMAKE_ARGS"] = "-DGGML_CUDA=on -DCUDAToolkit_ROOT=/usr/local/cuda"
-    os.environ["FORCE_CMAKE"] = "1"
+
 
     # Install all optional dependencies with CUDA support
     run_command("pip install -e .[all] --verbose")
@@ -56,7 +53,7 @@ def install_packages():
     run_command("make clean")
     
     # Reinstall the package with CUDA support
-    run_command('pip install -e . --force-reinstall --no-deps --no-cache-dir --verbose')
+    run_command('CMAKE_ARGS="-DLLAMA_CUBLAS=on -DCUDA_PATH=/usr/local/cuda-12.2 -DCUDAToolkit_ROOT=/usr/local/cuda-12.2 -DCUDAToolkit_INCLUDE_DIR=/usr/local/cuda-12/include -DCUDAToolkit_LIBRARY_DIR=/usr/local/cuda-12.2/lib64" FORCE_CMAKE=1 pip install -e .')
 
     # Install llama-cpp-agent
     run_command("pip install llama-cpp-agent")