Update app.py

app.py

@@ -13,6 +13,7 @@ import spaces
 
 # Model configuration
 MODEL_NAME = "yasserrmd/SinaReason-Magistral-2509"
+MODEL_NAME = "yasserrmd/SinaReason-Magistral-2509-bnb-4bit"
 DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 
 # Medical system prompt as recommended by the model card
@@ -142,6 +143,7 @@ class SinaReasonMedicalChat:
         for new_token in streamer:
             partial_response += new_token
 
+            prinr(partial_response)
             # Extract thinking and response
             thinking, response = self.extract_thinking_and_response(partial_response)
 
@@ -364,5 +366,6 @@ with gr.Blocks(css=css, title="SinaReason Medical Reasoning", theme=gr.themes.So
 # Launch configuration for HF Spaces
 if __name__ == "__main__":
     demo.launch(
+         debug=True,
         show_error=True
     )