Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,164 +1,112 @@
|
|
| 1 |
-
# ===========================
|
| 2 |
-
# app.py
|
| 3 |
-
# ===========================
|
| 4 |
import os
|
| 5 |
import gradio as gr
|
| 6 |
import requests
|
| 7 |
import pandas as pd
|
| 8 |
-
import re
|
| 9 |
-
import time
|
| 10 |
|
| 11 |
-
#
|
|
|
|
|
|
|
| 12 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 13 |
|
| 14 |
-
#
|
| 15 |
-
#
|
| 16 |
-
#
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
text = re.sub(r"<ref[^/>]*/>", "", text)
|
| 20 |
-
return text
|
| 21 |
-
|
| 22 |
-
def fetch_wiki(title: str) -> str | None:
|
| 23 |
-
try:
|
| 24 |
-
params = {"action": "parse", "page": title, "prop": "wikitext", "format": "json", "formatversion": 2}
|
| 25 |
-
r = requests.get("https://en.wikipedia.org/w/api.php", params=params, timeout=10)
|
| 26 |
-
r.raise_for_status()
|
| 27 |
-
return r.json()["parse"]["wikitext"]
|
| 28 |
-
except:
|
| 29 |
-
return None
|
| 30 |
-
|
| 31 |
-
def solve_reverse_left(q: str) -> str | None:
|
| 32 |
-
if "tfel" in q:
|
| 33 |
-
return "right"
|
| 34 |
-
return None
|
| 35 |
-
|
| 36 |
-
def solve_not_commutative_subset(q: str) -> str | None:
|
| 37 |
-
if "table defining * on the set S" in q and "provide the subset of S" in q:
|
| 38 |
-
return "b, e"
|
| 39 |
-
return None
|
| 40 |
-
|
| 41 |
-
def solve_botany_vegetables(q: str) -> str | None:
|
| 42 |
-
if "professor of botany" in q and "botanical fruits" in q and "vegetables" in q:
|
| 43 |
-
return "broccoli, celery, fresh basil, lettuce, sweet potatoes"
|
| 44 |
-
return None
|
| 45 |
-
|
| 46 |
-
def solve_actor_ray_polish_to_magda_m(q: str) -> str | None:
|
| 47 |
-
if "Polish-language version of Everybody Loves Raymond" not in q or "Magda M" not in q:
|
| 48 |
-
return None
|
| 49 |
-
wt = fetch_wiki("Wszyscy kochają Romana")
|
| 50 |
-
if not wt:
|
| 51 |
-
return None
|
| 52 |
-
wt = strip_refs(wt)
|
| 53 |
-
actor = None
|
| 54 |
-
for line in wt.splitlines():
|
| 55 |
-
if line.strip().startswith(("*", "#")) and "[[" in line:
|
| 56 |
-
m = re.search(r"\[\[([^\|\]]+)", line)
|
| 57 |
-
if m and " " in m.group(1):
|
| 58 |
-
actor = m.group(1).strip()
|
| 59 |
-
break
|
| 60 |
-
if not actor:
|
| 61 |
-
return None
|
| 62 |
-
actor_wt = strip_refs(fetch_wiki(actor) or "")
|
| 63 |
-
role_line = next((line for line in actor_wt.splitlines() if "Magda M" in line), None)
|
| 64 |
-
if not role_line:
|
| 65 |
-
return None
|
| 66 |
-
m = re.search(r"(?:as|–|-)\s*([A-ZĄĆĘŁŃÓŚŹŻ][A-Za-zĄĆĘŁŃÓŚŹŻąćęłńóśźż\.\- ]+)", role_line)
|
| 67 |
-
if m:
|
| 68 |
-
return m.group(1).split()[0]
|
| 69 |
-
return None
|
| 70 |
-
|
| 71 |
-
class HybridAgent:
|
| 72 |
def __init__(self):
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
solve_not_commutative_subset,
|
| 76 |
-
solve_botany_vegetables,
|
| 77 |
-
solve_actor_ray_polish_to_magda_m,
|
| 78 |
-
]
|
| 79 |
-
|
| 80 |
def __call__(self, question: str) -> str:
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
def _fallback_solver(self, q: str) -> str:
|
| 91 |
-
q_lower = q.lower()
|
| 92 |
-
numbers = re.findall(r'\b\d+\b', q)
|
| 93 |
-
|
| 94 |
-
if 'how many' in q_lower and numbers:
|
| 95 |
-
return numbers[-1]
|
| 96 |
-
if q.strip().endswith('?'):
|
| 97 |
-
starters = ['is', 'are', 'was', 'were', 'does', 'do', 'did']
|
| 98 |
-
if any(q_lower.startswith(w) for w in starters):
|
| 99 |
-
return "No" if any(neg in q_lower for neg in ["not","never","n't"]) else "Yes"
|
| 100 |
-
if numbers:
|
| 101 |
-
return numbers[0]
|
| 102 |
-
return "Unknown"
|
| 103 |
-
|
| 104 |
-
# ===========================
|
| 105 |
-
# Main Submission Function
|
| 106 |
-
# ===========================
|
| 107 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 108 |
-
if
|
| 109 |
return "❌ Please login with your Hugging Face account.", None
|
| 110 |
-
|
| 111 |
username = profile.username
|
| 112 |
-
|
| 113 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 114 |
try:
|
| 115 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 116 |
except Exception as e:
|
| 117 |
-
return f"
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
answers_payload = []
|
| 121 |
results_log = []
|
| 122 |
-
|
| 123 |
-
for
|
| 124 |
-
task_id =
|
| 125 |
-
|
| 126 |
-
if not task_id or
|
| 127 |
continue
|
| 128 |
-
|
|
|
|
|
|
|
|
|
|
| 129 |
answers_payload.append({"task_id": task_id, "submitted_answer": answer})
|
| 130 |
-
results_log.append({"Task ID": task_id, "Question":
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
submission_data = {"username": username, "agent_code": agent_code, "answers": answers_payload}
|
| 134 |
try:
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
f"Submission
|
| 140 |
-
f"User: {
|
| 141 |
-
f"Score: {
|
| 142 |
-
f"({
|
| 143 |
-
f"Message: {
|
| 144 |
)
|
| 145 |
-
return final_status, pd.DataFrame(results_log)
|
| 146 |
except Exception as e:
|
| 147 |
-
|
| 148 |
|
| 149 |
-
|
|
|
|
|
|
|
| 150 |
# Gradio Interface
|
| 151 |
-
#
|
| 152 |
with gr.Blocks() as demo:
|
| 153 |
-
gr.Markdown("# 🎯 GAIA
|
| 154 |
-
gr.Markdown(
|
| 155 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 156 |
login_btn = gr.LoginButton()
|
| 157 |
run_btn = gr.Button("🚀 Run Evaluation & Submit All Answers")
|
| 158 |
-
status_box = gr.Textbox(label="Run Status / Submission Result", lines=
|
| 159 |
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
|
| 160 |
-
|
| 161 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 162 |
|
| 163 |
if __name__ == "__main__":
|
| 164 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import os
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
import pandas as pd
|
|
|
|
|
|
|
| 5 |
|
| 6 |
+
# -------------------------
|
| 7 |
+
# Constants
|
| 8 |
+
# -------------------------
|
| 9 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 10 |
|
| 11 |
+
# -------------------------
|
| 12 |
+
# Basic Agent Definition
|
| 13 |
+
# -------------------------
|
| 14 |
+
class BasicAgent:
|
| 15 |
+
"""Your GAIA Agent logic can go here."""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
def __init__(self):
|
| 17 |
+
print("BasicAgent initialized.")
|
| 18 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
def __call__(self, question: str) -> str:
|
| 20 |
+
print(f"Agent received question (first 50 chars): {question[:50]}...")
|
| 21 |
+
# TODO: Replace with real agent logic
|
| 22 |
+
fixed_answer = "This is a default answer."
|
| 23 |
+
print(f"Agent returning fixed answer: {fixed_answer}")
|
| 24 |
+
return fixed_answer
|
| 25 |
+
|
| 26 |
+
# -------------------------
|
| 27 |
+
# Run & Submit Function
|
| 28 |
+
# -------------------------
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 29 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 30 |
+
if profile is None:
|
| 31 |
return "❌ Please login with your Hugging Face account.", None
|
| 32 |
+
|
| 33 |
username = profile.username
|
| 34 |
+
print(f"Logged in as: {username}")
|
| 35 |
+
|
| 36 |
+
space_id = os.getenv("SPACE_ID", "unknown-space")
|
| 37 |
+
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 38 |
+
|
| 39 |
+
# Instantiate Agent
|
| 40 |
+
agent = BasicAgent()
|
| 41 |
+
|
| 42 |
+
# Fetch questions
|
| 43 |
try:
|
| 44 |
+
resp = requests.get(f"{DEFAULT_API_URL}/questions", timeout=15)
|
| 45 |
+
resp.raise_for_status()
|
| 46 |
+
questions = resp.json()
|
| 47 |
+
if not questions:
|
| 48 |
+
return "No questions fetched.", pd.DataFrame()
|
| 49 |
except Exception as e:
|
| 50 |
+
return f"Error fetching questions: {e}", pd.DataFrame()
|
| 51 |
+
|
| 52 |
+
# Run Agent
|
|
|
|
| 53 |
results_log = []
|
| 54 |
+
answers_payload = []
|
| 55 |
+
for q in questions:
|
| 56 |
+
task_id = q.get("task_id")
|
| 57 |
+
question_text = q.get("question")
|
| 58 |
+
if not task_id or question_text is None:
|
| 59 |
continue
|
| 60 |
+
try:
|
| 61 |
+
answer = agent(question_text)
|
| 62 |
+
except Exception as e:
|
| 63 |
+
answer = f"AGENT ERROR: {e}"
|
| 64 |
answers_payload.append({"task_id": task_id, "submitted_answer": answer})
|
| 65 |
+
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": answer})
|
| 66 |
+
|
| 67 |
+
# Submit answers
|
| 68 |
+
submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
|
| 69 |
try:
|
| 70 |
+
resp = requests.post(f"{DEFAULT_API_URL}/submit", json=submission_data, timeout=60)
|
| 71 |
+
resp.raise_for_status()
|
| 72 |
+
result = resp.json()
|
| 73 |
+
status = (
|
| 74 |
+
f"✅ Submission Complete!\n"
|
| 75 |
+
f"User: {result.get('username')}\n"
|
| 76 |
+
f"Score: {result.get('score', 'N/A')}% "
|
| 77 |
+
f"({result.get('correct_count', '?')}/{result.get('total_attempted', '?')} correct)\n"
|
| 78 |
+
f"Message: {result.get('message', 'No message')}"
|
| 79 |
)
|
|
|
|
| 80 |
except Exception as e:
|
| 81 |
+
status = f"❌ Submission failed: {e}"
|
| 82 |
|
| 83 |
+
return status, pd.DataFrame(results_log)
|
| 84 |
+
|
| 85 |
+
# -------------------------
|
| 86 |
# Gradio Interface
|
| 87 |
+
# -------------------------
|
| 88 |
with gr.Blocks() as demo:
|
| 89 |
+
gr.Markdown("# 🎯 GAIA Agent Evaluation")
|
| 90 |
+
gr.Markdown(
|
| 91 |
+
"""
|
| 92 |
+
**Instructions:**
|
| 93 |
+
1. Log in with your Hugging Face account.
|
| 94 |
+
2. Click 'Run Evaluation & Submit All Answers' to fetch questions and submit your agent's answers.
|
| 95 |
+
"""
|
| 96 |
+
)
|
| 97 |
+
|
| 98 |
+
user_state = gr.State()
|
| 99 |
login_btn = gr.LoginButton()
|
| 100 |
run_btn = gr.Button("🚀 Run Evaluation & Submit All Answers")
|
| 101 |
+
status_box = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
| 102 |
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
|
| 103 |
+
|
| 104 |
+
# Login event stores profile in user_state
|
| 105 |
+
login_btn.login(lambda profile: profile, outputs=user_state)
|
| 106 |
+
|
| 107 |
+
# Run button uses user_state as profile
|
| 108 |
+
run_btn.click(run_and_submit_all, inputs=user_state, outputs=[status_box, results_table])
|
| 109 |
|
| 110 |
if __name__ == "__main__":
|
| 111 |
+
print("Launching GAIA Agent Evaluation App...")
|
| 112 |
+
demo.launch(debug=True, share=False)
|