22 lines
No EOL
491 B
Python
22 lines
No EOL
491 B
Python
from loaders import load_mind2web_tasks
|
||
from runner import run_many_tasks
|
||
|
||
# твой агент (как у тебя уже есть)
|
||
from agent import agent
|
||
|
||
|
||
def main():
|
||
tasks = load_mind2web_tasks(
|
||
"../Mind2Web/test_task/test_task_2.json",
|
||
limit=2
|
||
)
|
||
|
||
results = run_many_tasks(tasks, agent)
|
||
|
||
print("\n=== FINAL RESULTS ===")
|
||
print(f"Total: {len(results)}")
|
||
print(f"Success: {sum(r.success for r in results)}")
|
||
|
||
|
||
if __name__ == "__main__":
|
||
main() |