Update app.py

app.py

@@ -8,15 +8,16 @@ from ddgs import DDGS
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
 # -------------------------
-# GROQ API CALL
+# LLM CALL
 # -------------------------
 _last_call = 0
 
-def call_llm(api_key, prompt, system="", max_tokens=128):
+def call_llm(api_key, prompt, max_tokens=128):
+
     global _last_call
 
-    if time.time() - _last_call < 2.5:
-        time.sleep(2.5)
+    if time.time() - _last_call < 2:
+        time.sleep(2)
 
     _last_call = time.time()
 
@@ -27,23 +28,21 @@ def call_llm(api_key, prompt, system="", max_tokens=128):
         "Content-Type": "application/json"
     }
 
-    data = {
+    body = {
         "model": "llama-3.3-70b-versatile",
-        "messages": [
-            {"role": "system", "content": system},
-            {"role": "user", "content": prompt}
-        ],
+        "messages": [{"role": "user", "content": prompt}],
         "temperature": 0,
         "max_tokens": max_tokens
     }
 
-    r = requests.post(url, headers=headers, json=data, timeout=60)
+    r = requests.post(url, headers=headers, json=body, timeout=60)
 
     if r.status_code != 200:
         raise Exception(r.text)
 
     return r.json()["choices"][0]["message"]["content"].strip()
 
+
 # -------------------------
 # CLEAN ANSWER
 # -------------------------
@@ -66,33 +65,27 @@ def clean_answer(text):
 
     return text.strip('"').strip("'").strip("*")
 
+
 # -------------------------
 # WEB SEARCH
 # -------------------------
-def web_search(query):
+def search_web(query):
 
     results = []
 
-    with DDGS() as ddgs:
-        for r in ddgs.text(query, max_results=6):
-            results.append(
-                f"{r['title']} — {r['body']}"
-            )
+    try:
+        with DDGS() as ddgs:
+            for r in ddgs.text(query, max_results=5):
+                results.append(r["body"])
+    except:
+        pass
 
     return "\n".join(results)
 
+
 # -------------------------
 # AGENT
 # -------------------------
-SYSTEM = """
-You are solving GAIA benchmark questions.
-
-Rules:
-Return ONLY the final answer.
-No explanation.
-Exact match grading.
-"""
-
 class BasicAgent:
 
     def __init__(self):
@@ -100,47 +93,44 @@ class BasicAgent:
         self.key = os.getenv("GROQ_API_KEY")
 
         if not self.key:
-            raise RuntimeError("GROQ_API_KEY missing")
+            raise RuntimeError("GROQ_API_KEY not set")
 
         print("Agent ready")
 
-    # automatic retry
     def solve(self, prompt):
 
-        for attempt in range(3):
+        for _ in range(3):
 
             try:
 
-                answer = call_llm(
-                    self.key,
-                    prompt,
-                    SYSTEM,
-                    max_tokens=128
-                )
+                answer = call_llm(self.key, prompt)
 
                 answer = clean_answer(answer)
 
-                if len(answer) > 0:
+                if answer:
                     return answer
 
             except Exception as e:
+
                 print("Retry:", e)
 
             time.sleep(2)
 
         return ""
 
-    def __call__(self, question, task_id=""):
+    def __call__(self, question: str, task_id: str = "") -> str:
 
         print("Question:", question)
 
-        search = web_search(question)
+        search = search_web(question)
 
         prompt = f"""
+Answer the question exactly.
+
 Question:
 {question}
 
-Web information:
+Context:
 {search}
 
 Return ONLY the final answer.
@@ -152,13 +142,14 @@ Return ONLY the final answer.
 
         return answer
 
+
 # -------------------------
-# EVALUATION
+# RUN EVALUATION
 # -------------------------
 def run_and_submit_all(profile):
 
     if not profile:
-        return "Please login", None
+        return "Please login first", None
 
     username = profile.username
 
@@ -205,6 +196,7 @@ Correct: {result['correct_count']}
 
     return msg, pd.DataFrame(logs)
 
+
 # -------------------------
 # UI
 # -------------------------
@@ -216,9 +208,9 @@ with gr.Blocks() as demo:
 
     run_btn = gr.Button("Run Evaluation")
 
-    status = gr.Textbox()
+    status = gr.Textbox(label="Result")
 
-    table = gr.DataFrame()
+    table = gr.DataFrame(label="Answers")
 
     run_btn.click(
         run_and_submit_all,