Update app.py

app.py

@@ -3,6 +3,9 @@ import gradio as gr
 import requests
 import inspect
 import pandas as pd
+import re
+from huggingface_hub import InferenceClient
+
 
 # (Keep Constants as is)
 # --- Constants ---
@@ -10,14 +13,84 @@ DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
 # --- Basic Agent Definition ---
 # ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------
+# --- Basic Agent Definition ---
 class BasicAgent:
+    """
+    Minimal LLM-based agent for GAIA level-1 style questions.
+    Goal: >=30% (at least 6/20 exact match).
+    """
     def __init__(self):
-        print("BasicAgent initialized.")
+        print("BasicAgent initialized (LLM mode).")
+
+        # 必須先在 Space 設定 Secret：HF_TOKEN（你的 Hugging Face access token）
+        self.hf_token = os.getenv("HF_TOKEN") or os.getenv("HUGGINGFACEHUB_API_TOKEN")
+        if not self.hf_token:
+            raise RuntimeError("Missing HF_TOKEN. Please set it in Space Settings → Secrets.")
+
+        # 先用 7B 最穩最容易跑完；不夠分再升 14B/32B
+        self.model_id = os.getenv("MODEL_ID", "Qwen/Qwen2.5-7B-Instruct")
+
+        # 重要：用 router，不要用 api-inference（你之前 410 就是這個）
+        self.client = InferenceClient(
+            model=self.model_id,
+            token=self.hf_token,
+            base_url="https://router.huggingface.co",
+            timeout=120,
+        )
+
+    def _sanitize(self, text: str) -> str:
+        if not text:
+            return ""
+        t = text.strip()
+
+        # 移除 FINAL ANSWER 這種字眼（課程有說不要加）
+        t = re.sub(r"(?i)\bFINAL ANSWER\b\s*[:\-]*\s*", "", t).strip()
+
+        # 如果模型分行，取最後一行（通常答案會在最後）
+        lines = [ln.strip() for ln in t.splitlines() if ln.strip()]
+        if lines:
+            t = lines[-1]
+
+        # 去掉引號
+        t = t.strip().strip('"').strip("'").strip()
+        return t
+
     def __call__(self, question: str) -> str:
         print(f"Agent received question (first 50 chars): {question[:50]}...")
-        fixed_answer = "This is a default answer."
-        print(f"Agent returning fixed answer: {fixed_answer}")
-        return fixed_answer
+
+        system = (
+            "You are a precise question-answering assistant.\n"
+            "Return ONLY the final answer, nothing else.\n"
+            "No explanations. No extra words. No punctuation unless required.\n"
+            "If the answer is a number/date/name, output it exactly.\n"
+        )
+
+        prompt = f"{system}\nQuestion: {question}\nAnswer:"
+
+        # 用 chat completion 風格（InferenceClient 會依模型支援）
+        try:
+            out = self.client.text_generation(
+                prompt,
+                max_new_tokens=128,
+                temperature=0.0,
+                do_sample=False,
+                return_full_text=False,
+            )
+        except Exception as e:
+            # 如果 text_generation 因模型接口差異出錯，退回 chat_completion
+            print("text_generation failed, fallback to chat_completion:", e)
+            out = self.client.chat_completion(
+                messages=[
+                    {"role": "system", "content": system},
+                    {"role": "user", "content": question},
+                ],
+                max_tokens=128,
+                temperature=0.0,
+            ).choices[0].message.content
+
+        ans = self._sanitize(str(out))
+        print(f"Agent answer: {ans}")
+        return ans
 
 def run_and_submit_all( profile: gr.OAuthProfile | None):
     """