Reproducing GLEE Lite results on TAO

[LNdr28]

Hi, thanks for the work! I'm trying to reproduce your results on TAO with the GLEE Lite model. You report a TETA of 40.1, but I only get 36.1.
I use the config projects/GLEE/configs/videos/Lite/TAO_Lite.yaml and the TAO val split. I tried various versions of the TETA implementation, both from the old and new repo.
The numbers of the other video datasets (LVVIS and BURST) match up.
Could you share the exact configuration and versions you use to achieve the results from the paper?
Thanks!