get_score.py 727 B

12345678910111213141516171819202122232425262728
  1. import argparse
  2. import json
  3. def main():
  4. parser = argparse.ArgumentParser(description="Get agent's gaia score")
  5. parser.add_argument('--file', type=str, help="Path to the agent's output.jsonl")
  6. args = parser.parse_args()
  7. this_log = args.file
  8. outs = []
  9. with open(this_log, 'r') as f:
  10. lines = f.readlines()
  11. for line in lines:
  12. outs.append(json.loads(line))
  13. print(f'Reading {this_log}')
  14. print(f'Metadata:\n {outs[0]["metadata"]}')
  15. total = 0
  16. success = 0
  17. for out in outs:
  18. total += 1
  19. if out['test_result']['score']:
  20. success += 1
  21. print(f'Success rate: {success}/{total} = {success/total}')
  22. if __name__ == '__main__':
  23. main()