Update app.py

app.py

@@ -7,34 +7,31 @@ import spaces
 warnings.filterwarnings("ignore")
 
 
-print("Loading Whisper (transcription)...")
-whisper_model = whisper.load_model("tiny")
-
-model_name = "Qwen/Qwen2.5-0.5B-Instruct"
-print(f"Loading {model_name} (reflection)...")
-generator = pipeline(
-    "text-generation",
-    model=model_name,
-    device_map="auto",
-    dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
-)
-
-print("Models ready.")
 
 @spaces.GPU
 def transcribe(audio_path: str) -> str:
-    """Transcribe audio file to text using Whisper."""
     if audio_path is None:
         return ""
-    result = whisper_model.transcribe(audio_path)
+    print("Loading Whisper inside GPU...")
+    model = whisper.load_model("tiny")
+    result = model.transcribe(audio_path)
+    del model
+    torch.cuda.empty_cache() if torch.cuda.is_available() else None
     return result["text"].strip()
 
+
 @spaces.GPU
 def reflect(transcript: str) -> str:
-    """Generate a journaling reflection using Qwen."""
     if not transcript:
-        return "No transcript found — try recording again."
-
+        return "No transcript..."
+
+    print("Loading Qwen inside GPU...")
+    generator = pipeline(
+        "text-generation",
+        model="Qwen/Qwen2.5-0.5B-Instruct",
+        device_map="auto",
+        torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+    )
     prompt = f"""You are a warm, thoughtful journaling companion. Your tone is human and gentle, never clinical or robotic.
 
 The user just shared this voice journal entry:
@@ -72,7 +69,8 @@ Keep the full response under 160 words. Be kind, specific, and real."""
     else:
         # Fallback: strip the prompt
         reply = generated[len(prompt):]
-
+    del generator
+    torch.cuda.empty_cache() if torch.cuda.is_available() else None    
     return reply.strip()