RTMW metrics

[pbtgit]

I am evaluating RTMW on COCKTAIL14 according to this:
https://mmpose.readthedocs.io/en/latest/model_zoo/wholebody_2d_keypoint.html#cocktail14-dataset

I run test script (in docker) according to this:
python tools/test.py ./configs/wholebody_2d_keypoint/rtmpose/cocktail14/rtmw-m_8xb1024-270e_cocktail14-256x192.py data/rtmw-dw-l-m_simcc-cocktail14_270e-256x192-20231122.pth

However the the resulting AP/AR metrics does not match the ones published:
I get
Evaluate annotation type keypoints_body
DONE (t=9.50s).
Accumulating evaluation results...
DONE (t=0.31s).
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.721

Documentation states 0.676

What can be the cause for this?