Update app.py

app.py

@@ -23,6 +23,7 @@ os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
 
 def initialize_zero_gpu():
     """Initialize Hugging Face ZeroGPU"""
+    logger.info("Initializing Hugging Face ZeroGPU...")
     try:
         # Check if CUDA is available
         if torch.cuda.is_available():
@@ -33,6 +34,13 @@ def initialize_zero_gpu():
             print("CUDA is not available. Using CPU instead.")
     except Exception as e:
         print(f"Error initializing Hugging Face ZeroGPU: {str(e)}")
+        logger.error(f"Error initializing Hugging Face ZeroGPU: {str(e)}")
+        
+    # Check for the missing autotune file
+    autotune_file = '/home/user/.triton/autotune'
+    if not os.path.exists(autotune_file):
+        logger.warning(f"Missing autotune file: {autotune_file}. This may affect performance.")
+        print(f"Warning: Missing autotune file: {autotune_file}. This may affect performance.")
 
 def initialize_environment():
     """Initialize application environment and configurations"""
@@ -377,15 +385,7 @@ def signal_handler(signum, frame):
     sys.exit(0)
 
 if __name__ == "__main__":
-   def initialize_zero_gpu():
-    """Initialize Hugging Face ZeroGPU (if available)"""
-    if torch.cuda.is_available():
-        # Install CUDA drivers and adjust library versions if necessary
-        # ... (refer to solutions above)
-        deepspeed.init_distributed()
-        print("ZeroGPU initialized successfully.")
-    else:
-        print("CUDA is not available. Using CPU instead.")
+    initialize_zero_gpu()
 
     # Register cleanup handlers
     atexit.register(cleanup)
@@ -419,4 +419,4 @@ if __name__ == "__main__":
         logger.error(f"Error launching application: {str(e)}")
         raise
     finally:
-        logger.info("Application shutdown")
+        logger.info("Application shutdown")