Update app.py

app.py

@@ -1,17 +1,14 @@
 import os
-import io
 import gradio as gr
 import requests
 import pandas as pd
-from datasets import load_dataset
-from huggingface_hub import snapshot_download
 from agent import GaiaAgent
 
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
 def run_and_submit_all(profile: gr.OAuthProfile | None):
     """
-    Load GAIA dataset locally, run agent on questions, submit answers
+    Fetch questions from GAIA API, run agent, submit answers
     """
     
     # Check login
@@ -22,67 +19,18 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     space_id = os.getenv("SPACE_ID")
     
     print(f"\n[run_and_submit_all] starting for user: {username}")
-    print(f"[run_and_submit_all] space_id: {space_id}")
     
-    # Download GAIA dataset
-    print("[run_and_submit_all] downloading GAIA dataset...")
+    # Fetch questions from API
+    print("[run_and_submit_all] fetching questions from API...")
     try:
-        data_dir = snapshot_download(
-            repo_id="gaia-benchmark/GAIA",
-            repo_type="dataset"
-        )
-        print(f"[run_and_submit_all] dataset downloaded to: {data_dir}")
+        questions_resp = requests.get(f"{DEFAULT_API_URL}/questions", timeout=30)
+        questions_resp.raise_for_status()
+        questions = questions_resp.json()
+        print(f"[run_and_submit_all] ✓ fetched {len(questions)} questions")
     except Exception as e:
-        print(f"[run_and_submit_all] error downloading dataset: {e}")
-        return f"Error downloading dataset: {e}", None
-    
-    # Load dataset
-    print("[run_and_submit_all] loading dataset...")
-    try:
-        dataset = load_dataset(data_dir, "2023_level1", split="validation", cache_dir=data_dir)
-        print(f"[run_and_submit_all] loaded {len(dataset)} questions")
-    except Exception as e:
-        print(f"[run_and_submit_all] error loading dataset: {e}")
-        return f"Error loading dataset: {e}", None
-    
-    # Map task IDs to file paths
-    id_to_path = {}
-    for ex in dataset:
-        if ex.get("file_path") and ex.get("file_name"):
-            full_path = os.path.join(data_dir, ex["file_path"])
-            if os.path.exists(full_path):
-                id_to_path[ex["task_id"]] = full_path
-    
-    print(f"[run_and_submit_all] mapped {len(id_to_path)} task IDs to files")
-    
-    # Target task IDs (the 20 questions)
-    target_task_ids = [
-        "8e867cd7-cff9-4e6c-867a-ff5ddc2550be",
-        "a1e91b78-d3d8-4675-bb8d-62741b4b68a6",
-        "2d83110e-a098-4ebb-9987-066c06fa42d0",
-        "cca530fc-4052-43b2-b130-b30968d8aa44",
-        "4fc2f1ae-8625-45b5-ab34-ad4433bc21f8",
-        "6f37996b-2ac7-44b0-8e68-6d28256631b4",
-        "9d191bce-651d-4746-be2d-7ef8ecadb9c2",
-        "cabe07ed-9eca-40ea-8ead-410ef5e83f91",
-        "3cef3a44-215e-4aed-8e3b-b1e3f08063b7",
-        "99c9cc74-fdc8-46c6-8f8d-3ce2d3bfeea3",
-        "305ac316-eef6-4446-960a-92d80d542f82",
-        "f918266a-b3e0-4914-865d-4faa564f1aef",
-        "3f57289b-8c60-48be-bd80-01f8099ca449",
-        "1f975693-876d-457b-a649-393859e79bf3",
-        "840bfca7-4f7b-481a-8794-c560c340185d",
-        "bda648d7-d618-4883-88f4-3466eabd860e",
-        "cf106601-ab4f-4af9-b045-5295fe67b37d",
-        "a0c07678-e491-4bbc-8f0b-07405144218f",
-        "7bd855d8-463d-4ed5-93ca-5fe35145f733",
-        "5a0c1adf-205e-4841-a666-7c3ef95def9d"
-    ]
-    
-    # Filter to target questions
-    print(f"[run_and_submit_all] filtering to {len(target_task_ids)} target questions...")
-    subset = dataset.filter(lambda ex: ex["task_id"] in target_task_ids)
-    subset = subset.to_list()
+        error_msg = f"Error fetching questions: {str(e)[:200]}"
+        print(f"[run_and_submit_all] ✗ {error_msg}")
+        return error_msg, None
     
     # Initialize agent
     print("[run_and_submit_all] initializing agent...")
@@ -92,35 +40,38 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     results_log = []
     answers_payload = []
     
-    for i, item in enumerate(subset):
+    for i, item in enumerate(questions):
         task_id = item.get("task_id")
         question = item.get("question")
-        file_name = item.get("file_name")
-        file_path = id_to_path.get(task_id)
+        file_name = item.get("file_name", "")
         
-        print(f"\n[run_and_submit_all] [{i+1}/{len(subset)}] task_id={task_id}")
+        print(f"\n[run_and_submit_all] [{i+1}/{len(questions)}] task_id={task_id}")
         print(f"  question: {question[:80]}...")
-        print(f"  file: {file_name}")
+        print(f"  file: {file_name if file_name else '(none)'}")
         
         file_content = ""
         
-        # Load file if it exists
-        if file_path and os.path.exists(file_path):
+        # Try to fetch file if it exists
+        if file_name:
             try:
-                print(f"  reading file: {file_path}")
-                if file_name.endswith((".txt", ".py", ".json", ".csv", ".md")):
-                    with open(file_path, "r", encoding="utf-8") as f:
-                        file_content = f.read()
-                        print(f"  loaded {len(file_content)} chars from text file")
+                print(f"  fetching file: {file_name}...")
+                file_resp = requests.get(
+                    f"{DEFAULT_API_URL}/files/{task_id}",
+                    timeout=30
+                )
+                if file_resp.status_code == 200:
+                    # Try to decode as text first
+                    try:
+                        file_content = file_resp.text[:5000]
+                        print(f"  loaded {len(file_content)} chars from file")
+                    except:
+                        # If binary, note it
+                        file_content = f"[Binary file: {file_name}, {len(file_resp.content)} bytes]"
+                        print(f"  loaded binary file")
                 else:
-                    # Binary file (image, audio, etc.) - read as base64 for potential use
-                    with open(file_path, "rb") as f:
-                        file_bytes = f.read()
-                        # Just note that we have a binary file
-                        file_content = f"[Binary file: {file_name}, {len(file_bytes)} bytes]"
-                        print(f"  loaded binary file ({len(file_bytes)} bytes)")
+                    print(f"  file fetch returned {file_resp.status_code} (skipping)")
             except Exception as e:
-                print(f"  error reading file: {e}")
+                print(f"  error fetching file: {e}")
         
         # Run agent
         try:
@@ -145,7 +96,11 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     }
     
     try:
-        response = requests.post(f"{DEFAULT_API_URL}/submit", json=submission_data, timeout=60)
+        response = requests.post(
+            f"{DEFAULT_API_URL}/submit",
+            json=submission_data,
+            timeout=60
+        )
         response.raise_for_status()
         result = response.json()
         
@@ -160,27 +115,26 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         return status_msg, pd.DataFrame(results_log)
         
     except Exception as e:
-        error_msg = f"❌ Submission failed: {e}"
+        error_msg = f"❌ Submission failed: {str(e)[:200]}"
         print(error_msg)
         return error_msg, pd.DataFrame(results_log)
 
 
 # Gradio UI
 with gr.Blocks() as demo:
-    gr.Markdown("# GAIA Agent — Mistral + Local Dataset")
+    gr.Markdown("# GAIA Agent — Mistral")
     gr.Markdown("""
     **How it works:**
     1. Click "Login with Hugging Face"
     2. Click "Run Evaluation"
-    3. Wait for the agent to process all questions
-    4. See your score!
-    
-    **What's happening:**
-    - Downloads GAIA benchmark dataset locally
-    - Runs Mistral agent on each question
-    - Loads attached files from local filesystem
-    - Uses DuckDuckGo for web search
-    - Submits all answers for scoring
+    3. Agent processes all 20 questions
+    4. See your score instantly!
+    
+    **Features:**
+    - Uses Mistral model via Groq API
+    - Web search via DuckDuckGo (free, no keys)
+    - Fetches files from GAIA API
+    - Automatic answer submission
     """)
     
     gr.LoginButton()