Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,4 +1,5 @@
|
|
| 1 |
import os
|
|
|
|
| 2 |
import requests
|
| 3 |
import pandas as pd
|
| 4 |
import gradio as gr
|
|
@@ -36,24 +37,26 @@ def solve_question(agent, question: str) -> str:
|
|
| 36 |
raw_ans = agent.run(question)
|
| 37 |
ans = str(raw_ans).strip()
|
| 38 |
|
| 39 |
-
#
|
| 40 |
-
if
|
| 41 |
return "0"
|
| 42 |
-
if
|
|
|
|
|
|
|
| 43 |
return "0"
|
| 44 |
|
| 45 |
-
#
|
| 46 |
if "\n" in ans:
|
| 47 |
lines = [line.strip() for line in ans.split("\n") if line.strip()]
|
| 48 |
if lines:
|
| 49 |
-
ans = lines[-1]
|
| 50 |
|
| 51 |
-
#
|
| 52 |
ans = ans.strip(" .\"'`:;")
|
| 53 |
|
| 54 |
-
#
|
| 55 |
-
if len(ans) >
|
| 56 |
-
ans = ans[:
|
| 57 |
|
| 58 |
return ans if ans else "0"
|
| 59 |
except Exception as e:
|
|
@@ -92,7 +95,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 92 |
results_log = []
|
| 93 |
answers_payload = []
|
| 94 |
|
| 95 |
-
yield "Запуск прогона задач...
|
| 96 |
|
| 97 |
for item in questions_data:
|
| 98 |
task_id = item.get("task_id")
|
|
@@ -104,29 +107,46 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 104 |
answers_payload.append({"task_id": task_id, "submitted_answer": ans})
|
| 105 |
results_log.append({"Task ID": task_id, "Question": q, "Submitted Answer": ans})
|
| 106 |
|
| 107 |
-
# Обновляем табли
|
| 108 |
yield f"Выполнено задач: {len(results_log)} из {len(questions_data)}...", pd.DataFrame(results_log)
|
| 109 |
|
| 110 |
-
yield "Все задачи
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
|
| 115 |
-
|
| 116 |
-
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 130 |
|
| 131 |
with gr.Blocks() as demo:
|
| 132 |
gr.Markdown("# GAIA Smart Agent Runner")
|
|
|
|
| 1 |
import os
|
| 2 |
+
import time
|
| 3 |
import requests
|
| 4 |
import pandas as pd
|
| 5 |
import gradio as gr
|
|
|
|
| 37 |
raw_ans = agent.run(question)
|
| 38 |
ans = str(raw_ans).strip()
|
| 39 |
|
| 40 |
+
# Жесткая защита от утечек трейсов и громоздких объектов
|
| 41 |
+
if not ans or ans.lower() == "none":
|
| 42 |
return "0"
|
| 43 |
+
if "call_" in ans or "python_interpreter" in ans or "Error" in ans:
|
| 44 |
+
return "0"
|
| 45 |
+
if ans.startswith("[{") or ans.startswith("{'"):
|
| 46 |
return "0"
|
| 47 |
|
| 48 |
+
# Берём последнюю непустую строчку, если ответ многострочный
|
| 49 |
if "\n" in ans:
|
| 50 |
lines = [line.strip() for line in ans.split("\n") if line.strip()]
|
| 51 |
if lines:
|
| 52 |
+
ans = lines[-1]
|
| 53 |
|
| 54 |
+
# Чистим кавычки и мусор по краям
|
| 55 |
ans = ans.strip(" .\"'`:;")
|
| 56 |
|
| 57 |
+
# Строгое ограничение длины ответа для сервера оценки (не более 60 символов)
|
| 58 |
+
if len(ans) > 60:
|
| 59 |
+
ans = ans[:60].strip()
|
| 60 |
|
| 61 |
return ans if ans else "0"
|
| 62 |
except Exception as e:
|
|
|
|
| 95 |
results_log = []
|
| 96 |
answers_payload = []
|
| 97 |
|
| 98 |
+
yield "Запуск прогона задач... Агент начал выполнение.", pd.DataFrame(results_log)
|
| 99 |
|
| 100 |
for item in questions_data:
|
| 101 |
task_id = item.get("task_id")
|
|
|
|
| 107 |
answers_payload.append({"task_id": task_id, "submitted_answer": ans})
|
| 108 |
results_log.append({"Task ID": task_id, "Question": q, "Submitted Answer": ans})
|
| 109 |
|
| 110 |
+
# Обновляем таблицу в реальном времени через yield
|
| 111 |
yield f"Выполнено задач: {len(results_log)} из {len(questions_data)}...", pd.DataFrame(results_log)
|
| 112 |
|
| 113 |
+
yield "Все задачи решены. Отправка пакета на сервер оценки (с защитой от 500 ошибки)...", pd.DataFrame(results_log)
|
| 114 |
+
|
| 115 |
+
# Попытка отправки с ретраями на случай падения бэкенда курса
|
| 116 |
+
sub_res = None
|
| 117 |
+
success = False
|
| 118 |
+
for attempt in range(3):
|
| 119 |
+
try:
|
| 120 |
+
sub_res = requests.post(
|
| 121 |
+
f"{DEFAULT_API_URL}/submit",
|
| 122 |
+
json={"username": username, "agent_code": agent_code, "answers": answers_payload},
|
| 123 |
+
timeout=180
|
| 124 |
+
)
|
| 125 |
+
if sub_res.status_code == 200:
|
| 126 |
+
success = True
|
| 127 |
+
break
|
| 128 |
+
else:
|
| 129 |
+
time.sleep(3)
|
| 130 |
+
except Exception:
|
| 131 |
+
time.sleep(3)
|
| 132 |
+
|
| 133 |
+
if success and sub_res:
|
| 134 |
+
try:
|
| 135 |
+
data = sub_res.json()
|
| 136 |
+
status = (
|
| 137 |
+
f"Submission Successful!\n"
|
| 138 |
+
f"User: {data.get('username')}\n"
|
| 139 |
+
f"Overall Score: {data.get('score', 'N/A')}%\n"
|
| 140 |
+
f"Message: {data.get('message', '')}"
|
| 141 |
+
)
|
| 142 |
+
yield status, pd.DataFrame(results_log)
|
| 143 |
+
return
|
| 144 |
+
except Exception:
|
| 145 |
+
pass
|
| 146 |
+
|
| 147 |
+
# Если даже после ретраев сервер вернул ошибку
|
| 148 |
+
err_text = sub_res.text if sub_res else "Connection timeout/error"
|
| 149 |
+
yield f"Submission status: 500 Server Error (Сервер курса перегружен). Попробуй нажать кнопку повторно через пару минут.\nDetails: {err_text}", pd.DataFrame(results_log)
|
| 150 |
|
| 151 |
with gr.Blocks() as demo:
|
| 152 |
gr.Markdown("# GAIA Smart Agent Runner")
|