fix: correct dtype argument in model loading and enhance demo launch options

app.py

@@ -46,7 +46,7 @@ chat_tokenizer = AutoTokenizer.from_pretrained(CHAT_MODEL_NAME, padding_side="le
 if chat_tokenizer.pad_token is None:
     chat_tokenizer.pad_token = chat_tokenizer.eos_token
 
-chat_model = AutoModelForCausalLM.from_pretrained(CHAT_MODEL_NAME, torch_dtype=torch.bfloat16)
+chat_model = AutoModelForCausalLM.from_pretrained(CHAT_MODEL_NAME, dtype=torch.bfloat16)
 
 chat_model.to(device)  # type: ignore
 chat_model.eval()
@@ -57,7 +57,7 @@ print("✓ Chat model loaded")
 print(f"Loading classifier: {CLASSIFIER_MODEL_NAME}")
 
 cls_tokenizer = AutoTokenizer.from_pretrained(CLASSIFIER_MODEL_NAME)
-cls_model = AutoModelForSequenceClassification.from_pretrained(CLASSIFIER_MODEL_NAME, torch_dtype=torch.bfloat16)
+cls_model = AutoModelForSequenceClassification.from_pretrained(CLASSIFIER_MODEL_NAME, dtype=torch.bfloat16)
 
 cls_model.to(device)
 cls_model.eval()
@@ -117,6 +117,7 @@ def generate(submission: list[dict[str, str]], team_id: str) -> list[dict[str, s
             **inputs,
             max_new_tokens=MAX_NEW_TOKENS,
             do_sample=False,
+            temperature=None,
             repetition_penalty=REPETITION_PENALTY,
             pad_token_id=chat_tokenizer.pad_token_id,
             eos_token_id=chat_tokenizer.eos_token_id,
@@ -193,5 +194,5 @@ with gr.Blocks() as demo:
 if __name__ == "__main__":
     print("\n=== Launching Application ===")
     demo.queue(default_concurrency_limit=None, api_open=True)
-    demo.launch()
+    demo.launch(show_error=True)
     print("✓ Application running")