Reproducing Multigrain subset results

Hi, I'm trying to reproduce the baseline following the step provided in the below link.
( https://github.com/facebookresearch/isc2021/blob/main/docs/4_multigrain_subset.md )
I strictly followed the steps without any modification of the given code.
However, for my case, the result seems far different from the doc.

It supposed to be:
Average Precision: 0.56314
Recall at P90 : 0.41374
Threshold at P90 : -1.51692
Recall at rank 1: 0.60970
Recall at rank 10: 0.70827

But I got:
Track 2 running matching of 4991 queries in 4991 database (1500D descriptors), max_results=500000.
Evaluating 499999 predictions (4991 GT matches)
Average Precision: 0.09594
Recall at P90 : 0.07113
Threshold at P90 : -1.52351
Recall at rank 1: 0.10258
Recall at rank 10: 0.12022

I really have no idea on the incorrect result.