# M3 Attention Maps

We show two attention maps of a head using two fixed utterances. The left column shows the maps from the shorter utterance; the right column shows the longer. We also show the properties of a head: **id**, **category**, **globalness (G)**, **verticality (V)**, **Diagonality (D)**. Each metric value of a head is followed by the metric rank among all heads. For example, `G: 5.120 (8)` means the head has the globalness of **5.120** and is the **8th** global head among all heads.

## Layer 1


## Layer 2


## Layer 3


## Layer 4


## Layer 5


## Layer 6


## Layer 7


## Layer 8


## Layer 9


## Layer 10


## Layer 11


## Layer 12