Clarification on MAP and Checkpoint Selection

[xuullin]

in particular, I noticed that MAP values are reported with error margins (e.g., “MAP 67.4 ± 0.3, 66.3 ± 0.6”). I would like to ask for clarification on the following:

Is the MAP value computed from a single trained model, where different checkpoints were selected to calculate the results?