final answer extraction from the solution for aime24 dataset

[maziye-azad]

Hi, thank you for your great work.
how the final answer in extracted from thinking and the solution for comparison with the ground truth in datasets like aime24?
thank you

[Muennighoff]

see https://github.com/simplescaling/s1/blob/main/eval/lm-evaluation-harness/lm_eval/tasks/aime/utils.py