INDEX
Explanations
attends to image-related tokens from associated text tokens
New Auto-Interp
Head Attr Weights
0:0.08
1:0.10
2:0.10
3:0.22
4:0.14
5:0.04
6:0.13
7:0.15
Negative Logits
UnusedPrivate
-0.33
horabuena
-0.27
WriteTagHelper
-0.25
'\\;'
-0.24
idigung
-0.24
MenuGroup
-0.24
遷
-0.23
IconData
-0.22
ിയ
-0.21
бав
-0.21
POSITIVE LOGITS
OOTDTY
0.35
чему
0.34
/\.(
0.33
CWE
0.33
وتسجيلات
0.32
fieldNum
0.32
naught
0.32
quants
0.31
Wikimedijinoj
0.31
)");
0.31
Activations Density 0.089%