Update app.py

app.py

@@ -1,64 +1,49 @@
-import requests
-from datasets import load_dataset
-from transformers import pipeline
-
-# ---------------------------
-# CONFIG
-# ---------------------------
-SCORING_API = "https://agents-course-unit4-scoring.hf.space"
-MODEL_NAME = "google/flan-t5-base"
-
-# ---------------------------
-# Load model
-# ---------------------------
-print("Loading model...")
-qa = pipeline("text2text-generation", model=MODEL_NAME, max_new_tokens=64)
-
-# ---------------------------
-# Fetch the 20 questions
-# ---------------------------
-print("Fetching GAIA questions...")
-questions = requests.get(f"{SCORING_API}/questions").json()
-
-task_ids = [q["task_id"] for q in questions]
-
-# ---------------------------
-# Load GAIA validation dataset
-# ---------------------------
-print("Loading GAIA validation set...")
-dataset = load_dataset(
-    "gaia-benchmark/GAIA",
-    "2023_level1",
-    split="validation"
-)
-
-# Map task_id → correct answer
-ground_truth = {
-    item["task_id"]: item["Final answer"]
-    for item in dataset
-    if item["task_id"] in task_ids
-}
-
-# ---------------------------
-# Evaluate
-# ---------------------------
-correct = 0
-
-for q in questions:
-    task_id = q["task_id"]
-    question = q["question"]
-    true_answer = ground_truth.get(task_id, "").strip().lower()
-
-    model_output = qa(question)[0]["generated_text"].strip().lower()
-
-    match = model_output == true_answer
-    correct += int(match)
-
-    print("\n" + "="*80)
-    print(f"QUESTION:\n{question}")
-    print(f"\nEXPECTED:\n{true_answer}")
-    print(f"\nMODEL:\n{model_output}")
-    print(f"\nMATCH: {'✅' if match else '❌'}")
-
-print("\n" + "="*80)
-print(f"FINAL SCORE: {correct}/20")
+class BasicAgent:
+
+    def __init__(self):
+
+        print("Initializing GAIA Agent")
+
+        from smolagents import CodeAgent
+        from smolagents import DuckDuckGoSearchTool
+        from smolagents import PythonInterpreterTool
+        from smolagents import InferenceClientModel
+
+        model = InferenceClientModel(
+            model_id="meta-llama/Meta-Llama-3-8B-Instruct"
+        )
+
+        self.agent = CodeAgent(
+            tools=[
+                DuckDuckGoSearchTool(),
+                PythonInterpreterTool()
+            ],
+            model=model,
+            max_steps=6
+        )
+
+    def __call__(self, question: str) -> str:
+
+        try:
+
+            print("Question:", question)
+
+            result = self.agent.run(question)
+
+            if result is None:
+                return ""
+
+            answer = str(result).strip()
+
+            # Clean answer for exact-match grading
+            answer = answer.replace("FINAL ANSWER:", "")
+            answer = answer.replace("Answer:", "")
+            answer = answer.replace("The answer is", "")
+
+            return answer.split("\n")[0].strip()
+
+        except Exception as e:
+
+            print("Agent error:", e)
+
+            return ""