update codes

app.py

@@ -5,7 +5,8 @@ import inspect
 import pandas as pd
 import time
 
-from smolagents import ApiModel
+# smolagents의 복잡한 추상 에이전트 객체를 쓰지 않고, 가장 로우레벨에서 안전하게 모델을 호출하는 클라이언트만 사용
+from huggingface_hub import InferenceClient
 
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
@@ -14,27 +15,36 @@ DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 # ----- THIS IS WHERE YOU CAN BUILD WHAT YOU WANT ------
 class BasicAgent:
     def __init__(self):
-        # Hugging Face 환경에서 로그인 토큰을 명시적으로 바인딩하여 인증 멈춤 현상 해결
-        self.model = ApiModel(
-            model_id="Qwen/Qwen2.5-Coder-32B-Instruct",
-            api_key=os.getenv("HF_TOKEN")
+        # Hugging Face Spaces 환경에 기본 제공되는 HF_TOKEN 환경 변수를 사용하여 클라이언트 인증
+        self.client = InferenceClient(
+            model="Qwen/Qwen2.5-Coder-32B-Instruct",
+            token=os.getenv("HF_TOKEN")
         )
-        print("BasicAgent initialized.")
+        print("BasicAgent initialized with native InferenceClient.")
 
     def __call__(self, question: str) -> str:
         print(f"Agent received question (first 50 chars): {question[:50]}...")
         
         try:
-            # 원본 코드처럼 즉시 추론하여 결과 텍스트를 그대로 반환
-            messages = [{"role": "user", "content": question + "\n\nProvide the final answer as briefly as possible at the very end."}]
-            response = self.model(messages)
+            refined_prompt = (
+                f"{question}\n\n"
+                f"Provide the final answer as briefly as possible at the very end. "
+                f"No explanations in the last line."
+            )
             
-            final_answer = response.content.strip()
+            # 에이전트 루프 없이 단발성 텍스트 생성을 수행하므로 구조적 런타임 에러가 발생하지 않음
+            response = self.client.text_generation(
+                refined_prompt,
+                max_new_tokens=512,
+                clean_up_tokenization_spaces=True
+            )
+            
+            final_answer = response.strip()
             print(f"Agent returning answer: {final_answer}")
             return final_answer
             
         except Exception as e:
-            print(f"Error running model: {e}")
+            print(f"Error running model via InferenceClient: {e}")
             return "unknown"
 
 def run_and_submit_all( profile: gr.OAuthProfile | None):