Update app.py

app.py

@@ -22,6 +22,7 @@ logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 
 # Lifespan context manager for startup/shutdown
+# Note: If lifespan fails, we'll use @app.on_event("startup") as fallback
 @asynccontextmanager
 async def lifespan(app: FastAPI):
     """
@@ -30,8 +31,13 @@ async def lifespan(app: FastAPI):
     """
     # Startup: Load model
     logger.info("🚀 Starting up Wav2Vec2 Emotion Detection API...")
-    load_model()
-    logger.info("✅ Startup complete - Model loaded!")
+    try:
+        load_model()
+        logger.info("✅ Startup complete - Model loaded!")
+    except Exception as e:
+        logger.error(f"❌ Model loading failed during startup: {e}")
+        logger.warning("⚠️ App will continue, but emotion detection may not work")
+        logger.warning("⚠️ Model will be loaded lazily on first request")
     yield
     # Shutdown: Cleanup (if needed)
     logger.info("🛑 Shutting down...")
@@ -44,6 +50,23 @@ app = FastAPI(
     lifespan=lifespan
 )
 
+# Fallback startup event (in case lifespan doesn't work)
+@app.on_event("startup")
+async def startup_event():
+    """
+    Fallback startup event if lifespan doesn't work.
+    This ensures model loading happens even if lifespan fails.
+    """
+    global model, processor, feature_extractor
+    if model is None:
+        logger.info("🔄 Lifespan didn't load model, trying fallback startup...")
+        try:
+            load_model()
+            logger.info("✅ Model loaded via fallback startup!")
+        except Exception as e:
+            logger.error(f"❌ Fallback model loading failed: {e}")
+            logger.warning("⚠️ Model will be loaded lazily on first request")
+
 # Configure CORS - Allow requests from React frontend
 # For public API, allow all origins (common for ML APIs)
 # Using allow_origins=["*"] for maximum compatibility
@@ -79,9 +102,15 @@ def load_model():
     """
     Load the Wav2Vec2-Emotion model and processor from Hugging Face.
     This function is called once at startup to initialize the model.
+    If called again (lazy loading), it will skip if already loaded.
     """
     global model, processor, feature_extractor
     
+    # Skip if already loaded
+    if model is not None:
+        logger.info("✅ Model already loaded, skipping...")
+        return
+    
     try:
         logger.info("🔄 Loading Wav2Vec2-Emotion model from Hugging Face...")
         logger.info("Model: superb/wav2vec2-base-superb-er")
@@ -117,7 +146,7 @@ def load_model():
                     raise
         
         # Load the model
-        logger.info("📦 Loading model...")
+        logger.info("📦 Loading model (this may take a minute)...")
         model = Wav2Vec2ForSequenceClassification.from_pretrained(model_name)
         
         # Set model to evaluation mode (not training)
@@ -129,6 +158,8 @@ def load_model():
     except Exception as e:
         logger.error(f"❌ Error loading model: {str(e)}")
         logger.error(f"📋 Full error: {repr(e)}")
+        import traceback
+        logger.error(f"📋 Traceback:\n{traceback.format_exc()}")
         raise
 
 
@@ -362,6 +393,20 @@ async def predict_emotion_endpoint(
     Returns:
         JSON response with emotion, confidence, and probability distribution
     """
+    # Lazy loading: If model wasn't loaded at startup, load it now
+    global model, processor
+    if model is None or processor is None:
+        logger.warning("⚠️ Model not loaded, attempting lazy loading...")
+        try:
+            load_model()
+            logger.info("✅ Model loaded successfully on first request!")
+        except Exception as e:
+            logger.error(f"❌ Failed to load model: {e}")
+            raise HTTPException(
+                status_code=503,
+                detail=f"Model not available. Please wait a moment and try again. Error: {str(e)}"
+            )
+    
     try:
         # Read uploaded audio file
         audio_bytes = await audio.read()