Update app.py

app.py

@@ -27,7 +27,7 @@ def load_model(model_id: str = MODEL_ID):
     mdl = AutoModelForCausalLM.from_pretrained(
         model_id,
         trust_remote_code=True,
-        torch_dtype="auto",
+        torch_dtype=torch.bfloat16,
         device_map="auto",
     )
 
@@ -138,7 +138,7 @@ def generate_reply(
     max_new_tokens: int = 128,
     temperature: float = 1.0,
     top_p: float = 0.8,
-    max_retries: int = 5,
+    max_retries: int = 10,
 ) -> str:
     """Implements the 4 guardrails from Appendix C.1."""
     messages = build_hf_messages(system_prompt, history_pairs)