H1.3 top features

Top feature 0 in H1.3: (feature 14816)

TOP ACTIVATIONS
MAX = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000

Top DFA by src position
MAX = 0.248

atre
Tokenatre
Feature activation-0.027
Top resid features:
,
Token,
Feature activation-0.062
Top resid features:
a
Token a
Feature activation+0.079
Top resid features:
former
Token former
Feature activation+0.019
Top resid features:
boss
Token boss
Feature activation+0.059
Top resid features:
of
Token of
Feature activation+0.229
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
&
Token &
Feature activation+0.000
Top resid features:
Development
Token Development
Feature activation+0.000
Top resid features:
A
Token A
Feature activation-0.018
Top resid features:
atre
Tokenatre
Feature activation-0.029
Top resid features:
,
Token,
Feature activation-0.055
Top resid features:
a
Token a
Feature activation+0.093
Top resid features:
former
Token former
Feature activation+0.052
Top resid features:
boss
Token boss
Feature activation+0.143
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
&
Token &
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation-0.005
Top resid features:
.
Token.
Feature activation+0.007
Top resid features:
A
Token A
Feature activation+0.073
Top resid features:
atre
Tokenatre
Feature activation-0.021
Top resid features:
,
Token,
Feature activation-0.014
Top resid features:
a
Token a
Feature activation+0.248
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.001
Top resid features:
.
Token.
Feature activation+0.005
Top resid features:
A
Token A
Feature activation-0.010
Top resid features:
atre
Tokenatre
Feature activation-0.031
Top resid features:
,
Token,
Feature activation-0.054
Top resid features:
a
Token a
Feature activation+0.159
Top resid features:
former
Token former
Feature activation+0.137
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.014
Top resid features:
rik
Tokenrik
Feature activation-0.078
Top resid features:
ar
Tokenar
Feature activation-0.060
Top resid features:
said
Token said
Feature activation+0.106
Top resid features:
Saturday
Token Saturday
Feature activation-0.098
Top resid features:
.
Token.
Feature activation+0.037
Top resid features:
Ċ
TokenĊ
Feature activation+0.021
Top resid features:
Ċ
TokenĊ
Feature activation+0.000
Top resid features:
The
TokenThe
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation-0.043
Top resid features:
rik
Tokenrik
Feature activation-0.084
Top resid features:
ar
Tokenar
Feature activation-0.053
Top resid features:
said
Token said
Feature activation+0.135
Top resid features:
Saturday
Token Saturday
Feature activation-0.083
Top resid features:
.
Token.
Feature activation+0.089
Top resid features:
Ċ
TokenĊ
Feature activation+0.000
Top resid features:
Ċ
TokenĊ
Feature activation+0.000
Top resid features:
The
TokenThe
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.009
Top resid features:
rik
Tokenrik
Feature activation-0.066
Top resid features:
ar
Tokenar
Feature activation-0.052
Top resid features:
said
Token said
Feature activation+0.086
Top resid features:
Saturday
Token Saturday
Feature activation-0.090
Top resid features:
.
Token.
Feature activation+0.026
Top resid features:
Ċ
TokenĊ
Feature activation+0.018
Top resid features:
Ċ
TokenĊ
Feature activation+0.031
Top resid features:
The
TokenThe
Feature activation+0.000
Top resid features:
said
Token said
Feature activation+0.067
Top resid features:
Saturday
Token Saturday
Feature activation-0.082
Top resid features:
.
Token.
Feature activation+0.043
Top resid features:
Ċ
TokenĊ
Feature activation-0.003
Top resid features:
Ċ
TokenĊ
Feature activation+0.001
Top resid features:
The
TokenThe
Feature activation+0.163
Top resid features:
defence
Token defence
Feature activation+0.000
Top resid features:
ministry
Token ministry
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation+0.000
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation-0.007
Top resid features:
headed
Token headed
Feature activation-0.030
Top resid features:
by
Token by
Feature activation+0.012
Top resid features:
V
Token V
Feature activation-0.065
Top resid features:
.
Token.
Feature activation+0.063
Top resid features:
K
TokenK
Feature activation+0.168
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.065
Top resid features:
rik
Tokenrik
Feature activation-0.008
Top resid features:
ar
Tokenar
Feature activation-0.014
Top resid features:
said
Token said
Feature activation+0.034
Top resid features:
Saturday
Token Saturday
Feature activation-0.051
Top resid features:
.
Token.
Feature activation+0.045
Top resid features:
The
TokenThe
Feature activation+0.030
Top resid features:
defence
Token defence
Feature activation+0.005
Top resid features:
ministry
Token ministry
Feature activation+0.020
Top resid features:
committee
Token committee
Feature activation-0.004
Top resid features:
headed
Token headed
Feature activation+0.136
Top resid features:
by
Token by
Feature activation+0.143
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
defence
Token defence
Feature activation-0.008
Top resid features:
ministry
Token ministry
Feature activation-0.003
Top resid features:
committee
Token committee
Feature activation-0.010
Top resid features:
headed
Token headed
Feature activation-0.021
Top resid features:
by
Token by
Feature activation+0.047
Top resid features:
V
Token V
Feature activation+0.178
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.016
Top resid features:
V
Token V
Feature activation-0.012
Top resid features:
.
Token.
Feature activation+0.041
Top resid features:
K
TokenK
Feature activation-0.070
Top resid features:
.
Token.
Feature activation+0.045
Top resid features:
A
Token A
Feature activation+0.227
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.071
Top resid features:
rik
Tokenrik
Feature activation-0.007
Top resid features:
ar
Tokenar
Feature activation-0.014
Top resid features:
said
Token said
Feature activation+0.028
Top resid features:
Saturday
Token Saturday
Feature activation-0.051
Top resid features:
.
Token.
Feature activation+0.044
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.102
Top resid features:
rik
Tokenrik
Feature activation-0.016
Top resid features:
ar
Tokenar
Feature activation-0.029
Top resid features:
said
Token said
Feature activation+0.025
Top resid features:
Saturday
Token Saturday
Feature activation-0.047
Top resid features:
.
Token.
Feature activation+0.019
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.108
Top resid features:
rik
Tokenrik
Feature activation-0.006
Top resid features:
ar
Tokenar
Feature activation-0.007
Top resid features:
said
Token said
Feature activation+0.021
Top resid features:
Saturday
Token Saturday
Feature activation-0.051
Top resid features:
.
Token.
Feature activation+0.027
Top resid features:
Saturday
Token Saturday
Feature activation-0.059
Top resid features:
.
Token.
Feature activation+0.026
Top resid features:
Ċ
TokenĊ
Feature activation-0.009
Top resid features:
Ċ
TokenĊ
Feature activation-0.006
Top resid features:
The
TokenThe
Feature activation+0.050
Top resid features:
defence
Token defence
Feature activation+0.053
Top resid features:
ministry
Token ministry
Feature activation+0.020
Top resid features:
committee
Token committee
Feature activation-0.014
Top resid features:
headed
Token headed
Feature activation-0.044
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:
Saturday
Token Saturday
Feature activation-0.059
Top resid features:
.
Token.
Feature activation+0.017
Top resid features:
Ċ
TokenĊ
Feature activation-0.008
Top resid features:
Ċ
TokenĊ
Feature activation-0.004
Top resid features:
The
TokenThe
Feature activation+0.055
Top resid features:
defence
Token defence
Feature activation+0.083
Top resid features:
ministry
Token ministry
Feature activation-0.028
Top resid features:
committee
Token committee
Feature activation-0.092
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:
said
Token said
Feature activation+0.065
Top resid features:
Saturday
Token Saturday
Feature activation-0.067
Top resid features:
.
Token.
Feature activation+0.020
Top resid features:
Ċ
TokenĊ
Feature activation-0.006
Top resid features:
Ċ
TokenĊ
Feature activation-0.001
Top resid features:
The
TokenThe
Feature activation+0.100
Top resid features:
defence
Token defence
Feature activation+0.032
Top resid features:
ministry
Token ministry
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation+0.000
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
Saturday
Token Saturday
Feature activation-0.066
Top resid features:
.
Token.
Feature activation+0.036
Top resid features:
Ċ
TokenĊ
Feature activation-0.005
Top resid features:
Ċ
TokenĊ
Feature activation-0.001
Top resid features:
The
TokenThe
Feature activation+0.078
Top resid features:
defence
Token defence
Feature activation+0.228
Top resid features:
ministry
Token ministry
Feature activation-0.071
Top resid features:
committee
Token committee
Feature activation+0.000
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:

Decoder Weights Distribution

Head 0: 0.09

Head 1: 0.08

Head 2: 0.07

Head 3: 0.11

Head 4: 0.08

Head 5: 0.08

Head 6: 0.08

Head 7: 0.07

Head 8: 0.08

Head 9: 0.08

Head 10: 0.07

Head 11: 0.10

Positive logits

Skies1.59

Lans1.44

Jess1.38

pause1.37

escalation1.29

rison1.28

Byrne1.27

miscarriage1.26

veter1.25

PTSD1.24

renheit1.22

Migration1.21

flows1.20

traumatic1.20

Surviv1.20

senal1.19

nette1.19

Lent1.18

Loot1.18

rina1.17

Negative logits

oped-1.69

)?-1.44

.).-1.44

.)-1.39

â-1.39

op-1.36

adem-1.33

ó-1.32

opal-1.32

abb-1.31

.)-1.31

prolet-1.30

.}-1.29

cyclop-1.27

).-1.25

.),-1.24

pi-1.21

Kl-1.20

?).-1.20

umb-1.20

INTERVAL 0.000 - 0.000
CONTAINS 100.000%

advantages
Token advantages
Feature activation+0.000
of
Token of
Feature activation+0.000
EP
Token EP
Feature activation+0.000
YC
TokenYC
Feature activation+0.000
/
Token/
Feature activation+0.000
Na
TokenNa
Feature activation+0.000
ples
Tokenples
Feature activation+0.000
,
Token,
Feature activation+0.000
on
Token on
Feature activation+0.000
paper
Token paper
Feature activation+0.000
,
Token,
Feature activation+0.000
I
Token I
Feature activation+0.000
wasn
Token wasn
Feature activation+0.000
âĢ
TokenâĢ
Feature activation+0.000
Ļ
TokenĻ
Feature activation+0.000
t
Tokent
Feature activation+0.000
in
Token in
Feature activation+0.000
the
Token the
Feature activation+0.000
super
Token super
Feature activation+0.000
-
Token-
Feature activation+0.000
cool
Tokencool
Feature activation+0.000
crowd
Token crowd
Feature activation+0.000
your
Token your
Feature activation+0.000
target
Token target
Feature activation+0.000
audience
Token audience
Feature activation+0.000
.
Token.
Feature activation+0.000
According
Token According
Feature activation+0.000
to
Token to
Feature activation+0.000
Ed
Token Ed
Feature activation+0.000
an
Tokenan
Feature activation+0.000
G
Token G
Feature activation+0.000
elt
Tokenelt
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
press
Token press
Feature activation+0.000
release
Token release
Feature activation+0.000
at
Token at
Feature activation+0.000
the
Token the
Feature activation+0.000
time
Token time
Feature activation+0.000
:
Token:
Feature activation+0.000
The
Token The
Feature activation+0.000
development
Token development
Feature activation+0.000
of
Token of
Feature activation+0.000
our
Token our
Feature activation+0.000
who
Token who
Feature activation+0.000
conducted
Token conducted
Feature activation+0.000
this
Token this
Feature activation+0.000
survey
Token survey
Feature activation+0.000
with
Token with
Feature activation+0.000
Republican
Token Republican
Feature activation+0.000
poll
Token poll
Feature activation+0.000
ster
Tokenster
Feature activation+0.000
Bill
Token Bill
Feature activation+0.000
McInt
Token McInt
Feature activation+0.000
ur
Tokenur
Feature activation+0.000

BOTTOM ACTIVATIONS
MIN = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000

Top feature 1 in H1.3: (feature 18095)

TOP ACTIVATIONS
MAX = 2.017

<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Long
TokenLong
Feature activation+0.717
distance
Token distance
Feature activation+2.017
running
Token running
Feature activation+1.586
requires
Token requires
Feature activation+1.069
rest
Token rest
Feature activation+0.854
and
Token and
Feature activation+0.790
hyd
Token hyd
Feature activation+0.492
them
Token them
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.118
Ċ
TokenĊ
Feature activation+0.476
Whatever
TokenWhatever
Feature activation+0.804
happens
Token happens
Feature activation+1.908
,
Token,
Feature activation+1.093
the
Token the
Feature activation+0.739
rugby
Token rugby
Feature activation+0.489
world
Token world
Feature activation+0.634
are
Token are
Feature activation+0.381
Ċ
TokenĊ
Feature activation+0.000
Advertisement
TokenAdvertisement
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.840
Ċ
TokenĊ
Feature activation+1.152
Who
TokenWho
Feature activation+1.099
wants
Token wants
Feature activation+1.864
to
Token to
Feature activation+1.662
share
Token share
Feature activation+1.287
a
Token a
Feature activation+1.234
story
Token story
Feature activation+0.910
that
Token that
Feature activation+0.580
person
Token person
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.373
Cr
TokenCr
Feature activation+0.735
ut
Tokenut
Feature activation+1.833
cher
Tokencher
Feature activation+1.183
,
Token,
Feature activation+0.992
a
Token a
Feature activation+0.893
pastor
Token pastor
Feature activation+0.547
and
Token and
Feature activation+0.473
Ċ
TokenĊ
Feature activation+0.000
Miami
TokenMiami
Feature activation+0.159
Ċ
TokenĊ
Feature activation+0.821
Ċ
TokenĊ
Feature activation+0.999
Mont
TokenMont
Feature activation+1.069
real
Tokenreal
Feature activation+1.833
Ċ
TokenĊ
Feature activation+1.340
Ċ
TokenĊ
Feature activation+1.194
Sac
TokenSac
Feature activation+1.089
rament
Tokenrament
Feature activation+1.668
o
Tokeno
Feature activation+1.206
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
For
TokenFor
Feature activation+0.429
example
Token example
Feature activation+1.800
,
Token,
Feature activation+1.257
code
Token code
Feature activation+1.204
like
Token like
Feature activation+1.040
this
Token this
Feature activation+0.996
:
Token:
Feature activation+0.805
arguments
Token arguments
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.553
One
TokenOne
Feature activation+0.953
of
Token of
Feature activation+1.775
the
Token the
Feature activation+1.509
most
Token most
Feature activation+1.414
troubling
Token troubling
Feature activation+1.247
arguments
Token arguments
Feature activation+0.973
for
Token for
Feature activation+0.583
theory
Token theory
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.520
Which
TokenWhich
Feature activation+0.920
led
Token led
Feature activation+1.772
to
Token to
Feature activation+1.247
rumours
Token rumours
Feature activation+1.107
about
Token about
Feature activation+1.146
his
Token his
Feature activation+0.641
sexuality
Token sexuality
Feature activation+0.432
ii
Tokenii
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.287
There
TokenThere
Feature activation+0.793
's
Token's
Feature activation+1.756
the
Token the
Feature activation+1.279
clue
Token clue
Feature activation+1.312
,
Token,
Feature activation+0.823
in
Token in
Feature activation+0.598
the
Token the
Feature activation+0.515
work
Token work
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.439
Par
TokenPar
Feature activation+0.941
o
Tokeno
Feature activation+1.738
,
Token,
Feature activation+1.258
a
Token a
Feature activation+1.036
robotic
Token robotic
Feature activation+0.815
seal
Token seal
Feature activation+0.732
,
Token,
Feature activation+0.373
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
CT
TokenCT
Feature activation+0.662
A
TokenA
Feature activation+1.735
âĢ
TokenâĢ
Feature activation+1.296
Ļ
TokenĻ
Feature activation+1.382
s
Tokens
Feature activation+1.105
product
Token product
Feature activation+0.993
lineup
Token lineup
Feature activation+0.809
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
At
TokenAt
Feature activation+0.527
the
Token the
Feature activation+1.657
time
Token time
Feature activation+1.721
of
Token of
Feature activation+1.139
his
Token his
Feature activation+1.248
death
Token death
Feature activation+1.096
,
Token,
Feature activation+0.666
Sullivan
Token Sullivan
Feature activation+0.302
fastest
Token fastest
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.025
Ċ
TokenĊ
Feature activation+0.403
Tem
TokenTem
Feature activation+0.830
per
Tokenper
Feature activation+1.720
atures
Tokenatures
Feature activation+1.047
there
Token there
Feature activation+0.976
are
Token are
Feature activation+0.933
projected
Token projected
Feature activation+0.352
to
Token to
Feature activation+0.264
recycled
Token recycled
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.512
If
TokenIf
Feature activation+0.676
you
Token you
Feature activation+1.719
have
Token have
Feature activation+1.532
any
Token any
Feature activation+1.397
further
Token further
Feature activation+1.107
questions
Token questions
Feature activation+1.067
please
Token please
Feature activation+0.725
system
Token system
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.417
For
TokenFor
Feature activation+0.656
years
Token years
Feature activation+1.718
,
Token,
Feature activation+1.258
the
Token the
Feature activation+1.091
1
Token 1
Feature activation+0.975
.
Token.
Feature activation+0.791
52
Token52
Feature activation+0.498
âĢ
TokenâĢ
Feature activation+0.000
Ŀ
TokenĿ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.447
S
TokenS
Feature activation+0.617
ain
Tokenain
Feature activation+1.715
z
Tokenz
Feature activation+1.229
âĢ
TokenâĢ
Feature activation+0.766
Ļ
TokenĻ
Feature activation+0.806
s
Tokens
Feature activation+0.565
Toro
Token Toro
Feature activation+0.353
ling
Tokenling
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.464
This
TokenThis
Feature activation+0.922
campaign
Token campaign
Feature activation+1.710
is
Token is
Feature activation+1.459
one
Token one
Feature activation+1.131
of
Token of
Feature activation+0.918
the
Token the
Feature activation+0.861
cle
Token cle
Feature activation+0.597
altogether
Token altogether
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.185
Ċ
TokenĊ
Feature activation+0.627
Ag
TokenAg
Feature activation+1.087
ro
Tokenro
Feature activation+1.709
chem
Tokenchem
Feature activation+1.398
icals
Tokenicals
Feature activation+1.018
maker
Token maker
Feature activation+0.879
Sy
Token Sy
Feature activation+0.717
ng
Tokenng
Feature activation+0.540
happening
Token happening
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.030
Ċ
TokenĊ
Feature activation+0.410
Self
TokenSelf
Feature activation+0.686
-
Token-
Feature activation+1.708
Dri
TokenDri
Feature activation+0.768
ving
Tokenving
Feature activation+0.677
cars
Token cars
Feature activation+0.667
became
Token became
Feature activation+0.590
a
Token a
Feature activation+0.496
accountable
Token accountable
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.064
Ċ
TokenĊ
Feature activation+0.576
Perhaps
TokenPerhaps
Feature activation+0.816
even
Token even
Feature activation+1.698
more
Token more
Feature activation+1.320
troubling
Token troubling
Feature activation+1.285
is
Token is
Feature activation+0.876
their
Token their
Feature activation+0.530
proposal
Token proposal
Feature activation+0.401

Top DFA by src position
MAX = 1.301

<|endoftext|>
Token<|endoftext|>
Feature activation+0.990
Top resid features:
.
Token.
Feature activation+0.323
Top resid features:
Ċ
TokenĊ
Feature activation+1.210
Top resid features:
Ċ
TokenĊ
Feature activation+1.243
Top resid features:
Long
TokenLong
Feature activation+1.021
Top resid features:
distance
Token distance
Feature activation+0.421
Top resid features:
running
Token running
Feature activation+0.000
Top resid features:
requires
Token requires
Feature activation+0.000
Top resid features:
rest
Token rest
Feature activation+0.000
Top resid features:
before
Token before
Feature activation+0.267
Top resid features:
them
Token them
Feature activation+0.218
Top resid features:
.
Token.
Feature activation+0.466
Top resid features:
Ċ
TokenĊ
Feature activation+0.869
Top resid features:
Ċ
TokenĊ
Feature activation+0.871
Top resid features:
Whatever
TokenWhatever
Feature activation+1.117
Top resid features:
happens
Token happens
Feature activation+0.416
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
rugby
Token rugby
Feature activation+0.000
Top resid features:
world
Token world
Feature activation+0.000
Top resid features:
Ċ
TokenĊ
Feature activation+0.585
Top resid features:
Ċ
TokenĊ
Feature activation+0.642
Top resid features:
Advertisement
TokenAdvertisement
Feature activation+0.183
Top resid features:
Ċ
TokenĊ
Feature activation+0.747
Top resid features:
Ċ
TokenĊ
Feature activation+0.749
Top resid features:
Who
TokenWho
Feature activation+0.922
Top resid features:
wants
Token wants
Feature activation+0.438
Top resid features:
to
Token to
Feature activation+0.000
Top resid features:
share
Token share
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
story
Token story
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.873
Top resid features:
person
Token person
Feature activation+0.282
Top resid features:
.
Token.
Feature activation+0.403
Top resid features:
Ċ
TokenĊ
Feature activation+1.112
Top resid features:
Ċ
TokenĊ
Feature activation+1.129
Top resid features:
Cr
TokenCr
Feature activation+0.871
Top resid features:
ut
Tokenut
Feature activation+0.354
Top resid features:
cher
Tokencher
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.722
Top resid features:
Ċ
TokenĊ
Feature activation+0.648
Top resid features:
Ċ
TokenĊ
Feature activation+0.698
Top resid features:
Miami
TokenMiami
Feature activation+0.420
Top resid features:
Ċ
TokenĊ
Feature activation+0.770
Top resid features:
Ċ
TokenĊ
Feature activation+0.731
Top resid features:
Mont
TokenMont
Feature activation+0.642
Top resid features:
real
Tokenreal
Feature activation+0.393
Top resid features:
Ċ
TokenĊ
Feature activation+0.000
Top resid features:
Ċ
TokenĊ
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.909
Top resid features:
.
Token.
Feature activation+0.348
Top resid features:
Ċ
TokenĊ
Feature activation+1.256
Top resid features:
Ċ
TokenĊ
Feature activation+1.301
Top resid features:
For
TokenFor
Feature activation+0.847
Top resid features:
example
Token example
Feature activation+0.330
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
code
Token code
Feature activation+0.000
Top resid features:
like
Token like
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.823
Top resid features:
arguments
Token arguments
Feature activation+0.303
Top resid features:
.
Token.
Feature activation+0.389
Top resid features:
Ċ
TokenĊ
Feature activation+1.069
Top resid features:
Ċ
TokenĊ
Feature activation+1.039
Top resid features:
One
TokenOne
Feature activation+0.930
Top resid features:
of
Token of
Feature activation+0.414
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
most
Token most
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.860
Top resid features:
theory
Token theory
Feature activation+0.256
Top resid features:
.
Token.
Feature activation+0.444
Top resid features:
Ċ
TokenĊ
Feature activation+1.077
Top resid features:
Ċ
TokenĊ
Feature activation+1.054
Top resid features:
Which
TokenWhich
Feature activation+0.897
Top resid features:
led
Token led
Feature activation+0.376
Top resid features:
to
Token to
Feature activation+0.000
Top resid features:
rumours
Token rumours
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.903
Top resid features:
ii
Tokenii
Feature activation+0.222
Top resid features:
.
Token.
Feature activation+0.343
Top resid features:
Ċ
TokenĊ
Feature activation+1.020
Top resid features:
Ċ
TokenĊ
Feature activation+1.000
Top resid features:
There
TokenThere
Feature activation+0.836
Top resid features:
's
Token's
Feature activation+0.622
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
clue
Token clue
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.770
Top resid features:
ive
Tokenive
Feature activation+0.217
Top resid features:
work
Token work
Feature activation+0.220
Top resid features:
.
Token.
Feature activation+0.407
Top resid features:
Ċ
TokenĊ
Feature activation+0.979
Top resid features:
Ċ
TokenĊ
Feature activation+0.978
Top resid features:
Par
TokenPar
Feature activation+0.971
Top resid features:
o
Tokeno
Feature activation+0.388
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+1.000
Top resid features:
.
Token.
Feature activation+0.287
Top resid features:
Ċ
TokenĊ
Feature activation+1.156
Top resid features:
Ċ
TokenĊ
Feature activation+1.205
Top resid features:
CT
TokenCT
Feature activation+0.844
Top resid features:
A
TokenA
Feature activation+0.435
Top resid features:
âĢ
TokenâĢ
Feature activation+0.000
Top resid features:
Ļ
TokenĻ
Feature activation+0.000
Top resid features:
s
Tokens
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.832
Top resid features:
.
Token.
Feature activation+0.314
Top resid features:
Ċ
TokenĊ
Feature activation+0.949
Top resid features:
Ċ
TokenĊ
Feature activation+0.987
Top resid features:
At
TokenAt
Feature activation+0.976
Top resid features:
the
Token the
Feature activation+0.543
Top resid features:
time
Token time
Feature activation+0.311
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
his
Token his
Feature activation+0.000
Top resid features:
would
Token would
Feature activation+0.205
Top resid features:
warm
Token warm
Feature activation+0.138
Top resid features:
fastest
Token fastest
Feature activation+0.187
Top resid features:
.
Token.
Feature activation+0.407
Top resid features:
Ċ
TokenĊ
Feature activation+1.021
Top resid features:
Ċ
TokenĊ
Feature activation+1.034
Top resid features:
Tem
TokenTem
Feature activation+0.911
Top resid features:
per
Tokenper
Feature activation+0.264
Top resid features:
atures
Tokenatures
Feature activation+0.000
Top resid features:
there
Token there
Feature activation+0.000
Top resid features:
are
Token are
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.818
Top resid features:
recycled
Token recycled
Feature activation+0.275
Top resid features:
.
Token.
Feature activation+0.443
Top resid features:
Ċ
TokenĊ
Feature activation+1.092
Top resid features:
Ċ
TokenĊ
Feature activation+1.041
Top resid features:
If
TokenIf
Feature activation+0.824
Top resid features:
you
Token you
Feature activation+0.419
Top resid features:
have
Token have
Feature activation+0.000
Top resid features:
any
Token any
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.837
Top resid features:
system
Token system
Feature activation+0.215
Top resid features:
.
Token.
Feature activation+0.444
Top resid features:
Ċ
TokenĊ
Feature activation+1.082
Top resid features:
Ċ
TokenĊ
Feature activation+1.047
Top resid features:
For
TokenFor
Feature activation+0.944
Top resid features:
years
Token years
Feature activation+0.340
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.793
Top resid features:
.
Token.
Feature activation+0.329
Top resid features:
âĢ
TokenâĢ
Feature activation+0.201
Top resid features:
Ŀ
TokenĿ
Feature activation+0.256
Top resid features:
Ċ
TokenĊ
Feature activation+1.088
Top resid features:
Ċ
TokenĊ
Feature activation+1.089
Top resid features:
S
TokenS
Feature activation+0.736
Top resid features:
ain
Tokenain
Feature activation+0.414
Top resid features:
z
Tokenz
Feature activation+0.000
Top resid features:
âĢ
TokenâĢ
Feature activation+0.000
Top resid features:
Ļ
TokenĻ
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.814
Top resid features:
ling
Tokenling
Feature activation+0.319
Top resid features:
.
Token.
Feature activation+0.385
Top resid features:
Ċ
TokenĊ
Feature activation+1.082
Top resid features:
Ċ
TokenĊ
Feature activation+1.020
Top resid features:
This
TokenThis
Feature activation+0.852
Top resid features:
campaign
Token campaign
Feature activation+0.429
Top resid features:
is
Token is
Feature activation+0.000
Top resid features:
one
Token one
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.288
Top resid features:
market
Token market
Feature activation+0.233
Top resid features:
altogether
Token altogether
Feature activation+0.216
Top resid features:
.
Token.
Feature activation+0.407
Top resid features:
Ċ
TokenĊ
Feature activation+0.997
Top resid features:
Ċ
TokenĊ
Feature activation+1.051
Top resid features:
Ag
TokenAg
Feature activation+0.765
Top resid features:
ro
Tokenro
Feature activation+0.250
Top resid features:
chem
Tokenchem
Feature activation+0.000
Top resid features:
icals
Tokenicals
Feature activation+0.000
Top resid features:
maker
Token maker
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.731
Top resid features:
special
Token special
Feature activation+0.176
Top resid features:
was
Token was
Feature activation+0.219
Top resid features:
happening
Token happening
Feature activation+0.203
Top resid features:
.
Token.
Feature activation+0.470
Top resid features:
Ċ
TokenĊ
Feature activation+0.966
Top resid features:
Ċ
TokenĊ
Feature activation+0.955
Top resid features:
Self
TokenSelf
Feature activation+0.848
Top resid features:
-
Token-
Feature activation+0.331
Top resid features:
Dri
TokenDri
Feature activation+0.000
Top resid features:
ving
Tokenving
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.753
Top resid features:
journalists
Token journalists
Feature activation+0.292
Top resid features:
accountable
Token accountable
Feature activation+0.257
Top resid features:
.
Token.
Feature activation+0.472
Top resid features:
Ċ
TokenĊ
Feature activation+0.979
Top resid features:
Ċ
TokenĊ
Feature activation+0.956
Top resid features:
Perhaps
TokenPerhaps
Feature activation+0.854
Top resid features:
even
Token even
Feature activation+0.325
Top resid features:
more
Token more
Feature activation+0.000
Top resid features:
troubling
Token troubling
Feature activation+0.000
Top resid features:

Decoder Weights Distribution

Head 0: 0.04

Head 1: 0.11

Head 2: 0.16

Head 3: 0.11

Head 4: 0.12

Head 5: 0.04

Head 6: 0.03

Head 7: 0.07

Head 8: 0.07

Head 9: 0.07

Head 10: 0.11

Head 11: 0.06

Positive logits

yss0.67

iasm0.63

Profile0.63

idem0.61

Originally0.60

beware0.58

ensibly0.58

ebus0.56

DragonMagazine0.56

GOODMAN0.56

spoilers0.55

atis0.55

ennett0.54

atto0.54

asionally0.54

sbm0.54

STA0.53

sie0.53

eva0.53

isites0.52

Negative logits

venge-0.69

challeng-0.62

\<-0.61

..."-0.58

Sahara-0.58

unemploy-0.56

fulfillment-0.55

raping-0.55

blocking-0.52

guarding-0.52

].-0.52

vert-0.52

rouse-0.51

ruining-0.51

respectively-0.51

SPONSORED-0.51

$$$$-0.51

obstruct-0.51

roof-0.50

murderers-0.50

INTERVAL 1.815 - 2.017
CONTAINS 0.001%

them
Token them
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.118
Ċ
TokenĊ
Feature activation+0.476
Whatever
TokenWhatever
Feature activation+0.804
happens
Token happens
Feature activation+1.908
,
Token,
Feature activation+1.093
the
Token the
Feature activation+0.739
rugby
Token rugby
Feature activation+0.489
world
Token world
Feature activation+0.634
are
Token are
Feature activation+0.381
Ċ
TokenĊ
Feature activation+0.000
Miami
TokenMiami
Feature activation+0.159
Ċ
TokenĊ
Feature activation+0.821
Ċ
TokenĊ
Feature activation+0.999
Mont
TokenMont
Feature activation+1.069
real
Tokenreal
Feature activation+1.833
Ċ
TokenĊ
Feature activation+1.340
Ċ
TokenĊ
Feature activation+1.194
Sac
TokenSac
Feature activation+1.089
rament
Tokenrament
Feature activation+1.668
o
Tokeno
Feature activation+1.206
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Long
TokenLong
Feature activation+0.717
distance
Token distance
Feature activation+2.017
running
Token running
Feature activation+1.586
requires
Token requires
Feature activation+1.069
rest
Token rest
Feature activation+0.854
and
Token and
Feature activation+0.790
hyd
Token hyd
Feature activation+0.492
person
Token person
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.373
Cr
TokenCr
Feature activation+0.735
ut
Tokenut
Feature activation+1.833
cher
Tokencher
Feature activation+1.183
,
Token,
Feature activation+0.992
a
Token a
Feature activation+0.893
pastor
Token pastor
Feature activation+0.547
and
Token and
Feature activation+0.473
Ċ
TokenĊ
Feature activation+0.000
Advertisement
TokenAdvertisement
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.840
Ċ
TokenĊ
Feature activation+1.152
Who
TokenWho
Feature activation+1.099
wants
Token wants
Feature activation+1.864
to
Token to
Feature activation+1.662
share
Token share
Feature activation+1.287
a
Token a
Feature activation+1.234
story
Token story
Feature activation+0.910
that
Token that
Feature activation+0.580

INTERVAL 1.614 - 1.815
CONTAINS 0.005%

Yeah
TokenYeah
Feature activation+0.280
.
Token.
Feature activation+0.745
Ċ
TokenĊ
Feature activation+0.900
Ċ
TokenĊ
Feature activation+1.066
What
TokenWhat
Feature activation+0.959
would
Token would
Feature activation+1.644
you
Token you
Feature activation+1.493
do
Token do
Feature activation+1.221
about
Token about
Feature activation+1.036
that
Token that
Feature activation+0.504
?
Token?
Feature activation+0.465
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
In
TokenIn
Feature activation+0.551
spring
Token spring
Feature activation+1.626
and
Token and
Feature activation+1.316
summer
Token summer
Feature activation+1.193
,
Token,
Feature activation+0.933
you
Token you
Feature activation+0.804
can
Token can
Feature activation+0.725
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
For
TokenFor
Feature activation+0.429
example
Token example
Feature activation+1.800
,
Token,
Feature activation+1.257
code
Token code
Feature activation+1.204
like
Token like
Feature activation+1.040
this
Token this
Feature activation+0.996
:
Token:
Feature activation+0.805
ers
Tokeners
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.315
At
TokenAt
Feature activation+0.629
the
Token the
Feature activation+1.635
time
Token time
Feature activation+1.653
one
Token one
Feature activation+1.172
-
Token-
Feature activation+1.114
click
Tokenclick
Feature activation+1.039
download
Token download
Feature activation+0.957
Mont
TokenMont
Feature activation+1.069
real
Tokenreal
Feature activation+1.833
Ċ
TokenĊ
Feature activation+1.340
Ċ
TokenĊ
Feature activation+1.194
Sac
TokenSac
Feature activation+1.089
rament
Tokenrament
Feature activation+1.668
o
Tokeno
Feature activation+1.206
Ċ
TokenĊ
Feature activation+0.772
Ċ
TokenĊ
Feature activation+0.547
San
TokenSan
Feature activation+0.167
Francisco
Token Francisco
Feature activation+0.698

INTERVAL 1.412 - 1.614
CONTAINS 0.016%

so
Token so
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.335
South
TokenSouth
Feature activation+0.924
Korean
Token Korean
Feature activation+1.581
and
Token and
Feature activation+0.891
American
Token American
Feature activation+1.164
military
Token military
Feature activation+0.564
officers
Token officers
Feature activation+0.270
and
Token and
Feature activation+0.379
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Table
TokenTable
Feature activation+0.575
1
Token 1
Feature activation+1.549
:
Token:
Feature activation+1.489
Monthly
Token Monthly
Feature activation+1.047
Premium
Token Premium
Feature activation+0.927
for
Token for
Feature activation+1.051
a
Token a
Feature activation+0.693
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
An
TokenAn
Feature activation+0.819
In
Token In
Feature activation+1.453
Link
TokenLink
Feature activation+1.298
z
Tokenz
Feature activation+1.407
Link
Token Link
Feature activation+1.154
-
Token-
Feature activation+1.136
up
Tokenup
Feature activation+0.952
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
However
TokenHowever
Feature activation+0.646
,
Token,
Feature activation+1.443
this
Token this
Feature activation+1.145
was
Token was
Feature activation+1.207
just
Token just
Feature activation+0.999
one
Token one
Feature activation+0.820
of
Token of
Feature activation+0.621
cause
Token cause
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.102
Ċ
TokenĊ
Feature activation+0.501
Stew
TokenStew
Feature activation+0.519
art
Tokenart
Feature activation+1.590
partnered
Token partnered
Feature activation+1.048
with
Token with
Feature activation+1.060
charity
Token charity
Feature activation+0.600
fundraising
Token fundraising
Feature activation+0.556
platform
Token platform
Feature activation+0.302

INTERVAL 1.210 - 1.412
CONTAINS 0.041%

.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.738
city
Token city
Feature activation+1.456
is
Token is
Feature activation+1.267
offering
Token offering
Feature activation+1.052
up
Token up
Feature activation+1.044
to
Token to
Feature activation+0.907
$
Token $
Feature activation+0.616
1
Token1
Feature activation+0.397
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.424
Effects
TokenEffects
Feature activation+0.742
in
Token in
Feature activation+1.258
Britain
Token Britain
Feature activation+1.276
Ċ
TokenĊ
Feature activation+0.858
Ċ
TokenĊ
Feature activation+0.875
A
TokenA
Feature activation+0.858
second
Token second
Feature activation+1.555
study
Token study
Feature activation+1.290
month
Token month
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.016
Ċ
TokenĊ
Feature activation+0.424
Ms
TokenMs
Feature activation+0.625
.
Token.
Feature activation+1.389
Pi
Token Pi
Feature activation+1.097
erson
Tokenerson
Feature activation+1.162
and
Token and
Feature activation+0.837
Tea
Token Tea
Feature activation+0.689
Party
Token Party
Feature activation+0.680
Ċ
TokenĊ
Feature activation+0.840
Ċ
TokenĊ
Feature activation+1.152
Who
TokenWho
Feature activation+1.099
wants
Token wants
Feature activation+1.864
to
Token to
Feature activation+1.662
share
Token share
Feature activation+1.287
a
Token a
Feature activation+1.234
story
Token story
Feature activation+0.910
that
Token that
Feature activation+0.580
can
Token can
Feature activation+0.320
be
Token be
Feature activation+0.218
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
This
TokenThis
Feature activation+0.626
alteration
Token alteration
Feature activation+1.347
always
Token always
Feature activation+1.361
has
Token has
Feature activation+1.316
the
Token the
Feature activation+1.093
potential
Token potential
Feature activation+0.719
to
Token to
Feature activation+0.596
open
Token open
Feature activation+0.379

INTERVAL 1.008 - 1.210
CONTAINS 0.081%

make
Token make
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.067
Ċ
TokenĊ
Feature activation+0.294
You
TokenYou
Feature activation+0.441
will
Token will
Feature activation+1.075
need
Token need
Feature activation+0.730
:
Token:
Feature activation+0.570
Ċ
TokenĊ
Feature activation+0.080
Ċ
TokenĊ
Feature activation+0.055
1
Token1
Feature activation+0.000
If
TokenIf
Feature activation+0.676
you
Token you
Feature activation+1.719
have
Token have
Feature activation+1.532
any
Token any
Feature activation+1.397
further
Token further
Feature activation+1.107
questions
Token questions
Feature activation+1.067
please
Token please
Feature activation+0.725
give
Token give
Feature activation+0.506
us
Token us
Feature activation+0.203
a
Token a
Feature activation+0.017
call
Token call
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Some
TokenSome
Feature activation+0.688
of
Token of
Feature activation+1.482
these
Token these
Feature activation+1.288
stories
Token stories
Feature activation+1.376
are
Token are
Feature activation+1.135
downright
Token downright
Feature activation+0.749
shocking
Token shocking
Feature activation+0.881
,
Token,
Feature activation+0.389
some
Token some
Feature activation+0.000
are
Token are
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.095
Ċ
TokenĊ
Feature activation+0.424
C
TokenC
Feature activation+0.778
OUN
TokenOUN
Feature activation+1.373
TER
TokenTER
Feature activation+1.119
FE
TokenFE
Feature activation+0.666
IT
TokenIT
Feature activation+0.823
B
Token B
Feature activation+0.413
ILL
TokenILL
Feature activation+0.280
S
TokenS
Feature activation+0.203
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Hum
TokenHum
Feature activation+0.730
or
Tokenor
Feature activation+1.345
struck
Token struck
Feature activation+1.150
when
Token when
Feature activation+0.956
,
Token,
Feature activation+0.776
on
Token on
Feature activation+0.582
tracks
Token tracks
Feature activation+0.659
parallel
Token parallel
Feature activation+0.554

INTERVAL 0.807 - 1.008
CONTAINS 0.116%

Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.420
The
TokenThe
Feature activation+0.827
government
Token government
Feature activation+1.191
has
Token has
Feature activation+0.980
pledged
Token pledged
Feature activation+0.853
to
Token to
Feature activation+0.608
re
Token re
Feature activation+0.358
-
Token-
Feature activation+0.497
ex
Tokenex
Feature activation+0.177
amine
Tokenamine
Feature activation+0.107
Ċ
TokenĊ
Feature activation+0.000
In
TokenIn
Feature activation+0.127
1979
Token 1979
Feature activation+1.230
I
Token I
Feature activation+1.027
signed
Token signed
Feature activation+0.739
a
Token a
Feature activation+0.995
contract
Token contract
Feature activation+0.640
with
Token with
Feature activation+0.542
Harvard
Token Harvard
Feature activation+0.241
University
Token University
Feature activation+0.146
Press
Token Press
Feature activation+0.000
man
Tokenman
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.402
People
TokenPeople
Feature activation+0.565
who
Token who
Feature activation+1.267
may
Token may
Feature activation+0.953
have
Token have
Feature activation+0.841
lost
Token lost
Feature activation+0.637
their
Token their
Feature activation+0.576
jobs
Token jobs
Feature activation+0.177
are
Token are
Feature activation+0.099
Ċ
TokenĊ
Feature activation+0.000
Since
TokenSince
Feature activation+0.139
the
Token the
Feature activation+1.095
80
Token 80
Feature activation+1.357
body
Token body
Feature activation+1.045
cameras
Token cameras
Feature activation+0.983
were
Token were
Feature activation+0.874
officially
Token officially
Feature activation+0.663
brought
Token brought
Feature activation+0.386
on
Token on
Feature activation+0.372
board
Token board
Feature activation+0.100
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.301
Chinese
Token Chinese
Feature activation+0.929
Ex
Token Ex
Feature activation+1.013
clusion
Tokenclusion
Feature activation+0.846
Act
Token Act
Feature activation+0.894
which
Token which
Feature activation+0.721
later
Token later
Feature activation+0.262
morphed
Token morphed
Feature activation+0.247
into
Token into
Feature activation+0.230
a
Token a
Feature activation+0.000

INTERVAL 0.605 - 0.807
CONTAINS 0.168%

ľ
Tokenľ
Feature activation+0.639
âĢ¦
TokenâĢ¦
Feature activation+0.872
I
TokenI
Feature activation+0.578
saw
Token saw
Feature activation+1.109
this
Token this
Feature activation+0.889
toy
Token toy
Feature activation+0.721
last
Token last
Feature activation+0.561
week
Token week
Feature activation+0.345
and
Token and
Feature activation+0.094
made
Token made
Feature activation+0.000
my
Token my
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
disease
Token disease
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.378
Here
TokenHere
Feature activation+0.791
are
Token are
Feature activation+1.639
the
Token the
Feature activation+1.291
top
Token top
Feature activation+1.093
20
Token 20
Feature activation+1.088
and
Token and
Feature activation+0.738
Stock
Token Stock
Feature activation+0.859
Ċ
TokenĊ
Feature activation+0.858
Ċ
TokenĊ
Feature activation+0.992
David
TokenDavid
Feature activation+0.763
Stock
Token Stock
Feature activation+1.130
Ċ
TokenĊ
Feature activation+0.805
Ċ
TokenĊ
Feature activation+0.683
David
TokenDavid
Feature activation+0.319
Stock
Token Stock
Feature activation+0.637
Ċ
TokenĊ
Feature activation+0.249
Ċ
TokenĊ
Feature activation+0.059
news
Token news
Feature activation+0.000
continues
Token continues
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.018
Ċ
TokenĊ
Feature activation+0.396
I
TokenI
Feature activation+0.749
don
Token don
Feature activation+1.234
't
Token't
Feature activation+1.102
know
Token know
Feature activation+0.796
what
Token what
Feature activation+0.599
Bernie
Token Bernie
Feature activation+0.430
Ċ
TokenĊ
Feature activation+0.388
The
TokenThe
Feature activation+0.809
fact
Token fact
Feature activation+1.377
that
Token that
Feature activation+1.164
Harry
Token Harry
Feature activation+0.976
was
Token was
Feature activation+0.778
still
Token still
Feature activation+0.688
there
Token there
Feature activation+0.454
was
Token was
Feature activation+0.285
evidence
Token evidence
Feature activation+0.000
enough
Token enough
Feature activation+0.000

INTERVAL 0.403 - 0.605
CONTAINS 0.232%

hua
Tokenhua
Feature activation+0.000
said
Token said
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.226
H
TokenH
Feature activation+0.500
ang
Tokenang
Feature activation+0.880
zhou
Tokenzhou
Feature activation+0.636
,
Token,
Feature activation+0.483
capital
Token capital
Feature activation+0.155
of
Token of
Feature activation+0.238
El
Token El
Feature activation+0.032
isa
Tokenisa
Feature activation+0.429
Lind
Token Lind
Feature activation+0.611
str
Tokenstr
Feature activation+0.488
ö
Tokenö
Feature activation+0.651
m
Tokenm
Feature activation+0.552
Ċ
TokenĊ
Feature activation+0.227
Ċ
TokenĊ
Feature activation+0.226
5
Token5
Feature activation+0.235
.
Token.
Feature activation+0.280
Alice
Token Alice
Feature activation+0.140
relationship
Token relationship
Feature activation+0.000
with
Token with
Feature activation+0.000
Beijing
Token Beijing
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.157
Ċ
TokenĊ
Feature activation+0.487
The
TokenThe
Feature activation+0.775
new
Token new
Feature activation+1.093
generation
Token generation
Feature activation+1.330
of
Token of
Feature activation+0.814
activists
Token activists
Feature activation+0.626
Ċ
TokenĊ
Feature activation+0.372
âĢ
TokenâĢ
Feature activation+0.456
ľ
Tokenľ
Feature activation+0.491
C
TokenC
Feature activation+0.759
ann
Tokenann
Feature activation+1.087
abis
Tokenabis
Feature activation+0.596
businesses
Token businesses
Feature activation+0.483
offer
Token offer
Feature activation+0.521
meaningful
Token meaningful
Feature activation+0.075
employment
Token employment
Feature activation+0.000
at
Token at
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.147
Ph
TokenPh
Feature activation+0.645
ile
Tokenile
Feature activation+1.049
as
Tokenas
Feature activation+1.044
F
Token F
Feature activation+0.466
ogg
Tokenogg
Feature activation+0.449
is
Token is
Feature activation+0.307
a
Token a
Feature activation+0.000
rich
Token rich
Feature activation+0.000
British
Token British
Feature activation+0.000

INTERVAL 0.202 - 0.403
CONTAINS 0.288%

failure
Token failure
Feature activation+0.000
British
Token British
Feature activation+0.000
Heart
Token Heart
Feature activation+0.000
Foundation
Token Foundation
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.240
"
Token"
Feature activation+0.279
When
TokenWhen
Feature activation+0.000
you
Token you
Feature activation+0.227
have
Token have
Feature activation+0.162
insomnia
Token insomnia
Feature activation+0.000
limited
Tokenlimited
Feature activation+1.382
surveillance
Token surveillance
Feature activation+0.957
,
Token,
Feature activation+0.798
attacks
Token attacks
Feature activation+0.611
on
Token on
Feature activation+0.796
paramedics
Token paramedics
Feature activation+0.260
,
Token,
Feature activation+0.277
and
Token and
Feature activation+0.000
Arabs
Token Arabs
Feature activation+0.000
kicking
Token kicking
Feature activation+0.000
babies
Token babies
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.084
âĢ
TokenâĢ
Feature activation+0.185
ľ
Tokenľ
Feature activation+0.088
Make
TokenMake
Feature activation+0.117
this
Token this
Feature activation+0.720
right
Token right
Feature activation+0.276
.
Token.
Feature activation+0.043
We
Token We
Feature activation+0.000
are
Token are
Feature activation+0.000
better
Token better
Feature activation+0.000
than
Token than
Feature activation+0.000
Str
Token Str
Feature activation+0.994
ab
Tokenab
Feature activation+1.094
ucks
Tokenucks
Feature activation+0.788
signed
Token signed
Feature activation+0.547
letters
Token letters
Feature activation+0.653
of
Token of
Feature activation+0.278
intent
Token intent
Feature activation+0.000
âĢĶ
Token âĢĶ
Feature activation+0.000
a
Token a
Feature activation+0.000
step
Token step
Feature activation+0.000
short
Token short
Feature activation+0.000
:
Token:
Feature activation+1.010
Sl
Token Sl
Feature activation+0.634
owed
Tokenowed
Feature activation+0.550
down
Token down
Feature activation+0.573
heavy
Token heavy
Feature activation+0.464
side
Token side
Feature activation+0.236
attacks
Token attacks
Feature activation+0.022
a
Token a
Feature activation+0.000
bit
Token bit
Feature activation+0.000
for
Token for
Feature activation+0.000
judgement
Token judgement
Feature activation+0.000

INTERVAL 0.000 - 0.202
CONTAINS 99.055%

gluten
Token gluten
Feature activation+0.000
free
Token free
Feature activation+0.000
,
Token,
Feature activation+0.000
gluten
Token gluten
Feature activation+0.000
free
Token free
Feature activation+0.000
recipe
Token recipe
Feature activation+0.000
,
Token,
Feature activation+0.000
gluten
Token gluten
Feature activation+0.000
free
Token free
Feature activation+0.000
red
Token red
Feature activation+0.000
velvet
Token velvet
Feature activation+0.000
can
Token can
Feature activation+0.000
keep
Token keep
Feature activation+0.000
your
Token your
Feature activation+0.000
house
Token house
Feature activation+0.000
safe
Token safe
Feature activation+0.000
,
Token,
Feature activation+0.000
especially
Token especially
Feature activation+0.000
if
Token if
Feature activation+0.000
you
Token you
Feature activation+0.000
are
Token are
Feature activation+0.000
travelling
Token travelling
Feature activation+0.000
CIA
Token CIA
Feature activation+0.000
and
Token and
Feature activation+0.000
the
Token the
Feature activation+0.000
FBI
Token FBI
Feature activation+0.000
declined
Token declined
Feature activation+0.000
to
Token to
Feature activation+0.000
comment
Token comment
Feature activation+0.000
on
Token on
Feature activation+0.000
Brennan
Token Brennan
Feature activation+0.000
âĢ
TokenâĢ
Feature activation+0.000
Ļ
TokenĻ
Feature activation+0.000
and
Token and
Feature activation+0.000
their
Token their
Feature activation+0.000
willingness
Token willingness
Feature activation+0.000
to
Token to
Feature activation+0.000
cooperate
Token cooperate
Feature activation+0.000
,
Token,
Feature activation+0.000
among
Token among
Feature activation+0.000
several
Token several
Feature activation+0.000
other
Token other
Feature activation+0.000
factors
Token factors
Feature activation+0.000
.
Token.
Feature activation+0.000
that
Token that
Feature activation+0.000
Adrian
Token Adrian
Feature activation+0.000
za
Tokenza
Feature activation+0.000
's
Token's
Feature activation+0.000
dreadful
Token dreadful
Feature activation+0.000
hitting
Token hitting
Feature activation+0.000
hasn
Token hasn
Feature activation+0.000
't
Token't
Feature activation+0.000
messed
Token messed
Feature activation+0.000
the
Token the
Feature activation+0.000
Giants
Token Giants
Feature activation+0.000

BOTTOM ACTIVATIONS
MIN = 0.000

Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
Organisation
Token Organisation
Feature activation+0.000
,
Token,
Feature activation+0.000
which
Token which
Feature activation+0.000
is
Token is
Feature activation+0.000
to
Token to
Feature activation+0.000
recommend
Token recommend
Feature activation+0.000
guidelines
Token guidelines
Feature activation+0.000
for
Token for
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
Organisation
Token Organisation
Feature activation+0.000
,
Token,
Feature activation+0.000
which
Token which
Feature activation+0.000
is
Token is
Feature activation+0.000
to
Token to
Feature activation+0.000
recommend
Token recommend
Feature activation+0.000
guidelines
Token guidelines
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
Organisation
Token Organisation
Feature activation+0.000
,
Token,
Feature activation+0.000
which
Token which
Feature activation+0.000
is
Token is
Feature activation+0.000
to
Token to
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
Organisation
Token Organisation
Feature activation+0.000
,
Token,
Feature activation+0.000
which
Token which
Feature activation+0.000
is
Token is
Feature activation+0.000
to
Token to
Feature activation+0.000
recommend
Token recommend
Feature activation+0.000
defence
Token defence
Feature activation+0.947
ministry
Token ministry
Feature activation+0.915
committee
Token committee
Feature activation+0.659
headed
Token headed
Feature activation+0.408
by
Token by
Feature activation+0.182
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.022
Ċ
TokenĊ
Feature activation+0.352
The
TokenThe
Feature activation+0.655
defence
Token defence
Feature activation+0.947
ministry
Token ministry
Feature activation+0.915
ministry
Token ministry
Feature activation+0.915
committee
Token committee
Feature activation+0.659
headed
Token headed
Feature activation+0.408
by
Token by
Feature activation+0.182
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
committee
Token committee
Feature activation+0.659
headed
Token headed
Feature activation+0.408
by
Token by
Feature activation+0.182
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
Organisation
Token Organisation
Feature activation+0.000
,
Token,
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
Organisation
Token Organisation
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
Organisation
Token Organisation
Feature activation+0.000
,
Token,
Feature activation+0.000
which
Token which
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
Organisation
Token Organisation
Feature activation+0.000
,
Token,
Feature activation+0.000
which
Token which
Feature activation+0.000
is
Token is
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
headed
Token headed
Feature activation+0.408
by
Token by
Feature activation+0.182
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
by
Token by
Feature activation+0.182
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000

Top feature 2 in H1.3: (feature 1490)

TOP ACTIVATIONS
MAX = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000

Top DFA by src position
MAX = 0.204

atre
Tokenatre
Feature activation-0.004
Top resid features:
,
Token,
Feature activation-0.000
Top resid features:
a
Token a
Feature activation+0.086
Top resid features:
former
Token former
Feature activation+0.048
Top resid features:
boss
Token boss
Feature activation+0.009
Top resid features:
of
Token of
Feature activation+0.162
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
&
Token &
Feature activation+0.000
Top resid features:
Development
Token Development
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.010
Top resid features:
.
Token.
Feature activation+0.022
Top resid features:
A
Token A
Feature activation+0.041
Top resid features:
atre
Tokenatre
Feature activation-0.004
Top resid features:
,
Token,
Feature activation+0.011
Top resid features:
a
Token a
Feature activation+0.090
Top resid features:
former
Token former
Feature activation+0.056
Top resid features:
boss
Token boss
Feature activation+0.062
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.003
Top resid features:
.
Token.
Feature activation+0.035
Top resid features:
A
Token A
Feature activation+0.075
Top resid features:
atre
Tokenatre
Feature activation+0.031
Top resid features:
,
Token,
Feature activation+0.055
Top resid features:
a
Token a
Feature activation+0.161
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.027
Top resid features:
A
Token A
Feature activation+0.036
Top resid features:
atre
Tokenatre
Feature activation+0.003
Top resid features:
,
Token,
Feature activation+0.014
Top resid features:
a
Token a
Feature activation+0.104
Top resid features:
former
Token former
Feature activation+0.143
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.112
Top resid features:
rik
Tokenrik
Feature activation+0.031
Top resid features:
ar
Tokenar
Feature activation+0.041
Top resid features:
said
Token said
Feature activation+0.137
Top resid features:
Saturday
Token Saturday
Feature activation-0.049
Top resid features:
.
Token.
Feature activation+0.074
Top resid features:
Ċ
TokenĊ
Feature activation+0.048
Top resid features:
Ċ
TokenĊ
Feature activation+0.000
Top resid features:
The
TokenThe
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.139
Top resid features:
rik
Tokenrik
Feature activation+0.011
Top resid features:
ar
Tokenar
Feature activation+0.045
Top resid features:
said
Token said
Feature activation+0.169
Top resid features:
Saturday
Token Saturday
Feature activation-0.036
Top resid features:
.
Token.
Feature activation+0.164
Top resid features:
Ċ
TokenĊ
Feature activation+0.000
Top resid features:
Ċ
TokenĊ
Feature activation+0.000
Top resid features:
The
TokenThe
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.095
Top resid features:
rik
Tokenrik
Feature activation+0.034
Top resid features:
ar
Tokenar
Feature activation+0.039
Top resid features:
said
Token said
Feature activation+0.113
Top resid features:
Saturday
Token Saturday
Feature activation-0.049
Top resid features:
.
Token.
Feature activation+0.059
Top resid features:
Ċ
TokenĊ
Feature activation+0.037
Top resid features:
Ċ
TokenĊ
Feature activation+0.050
Top resid features:
The
TokenThe
Feature activation+0.000
Top resid features:
said
Token said
Feature activation+0.100
Top resid features:
Saturday
Token Saturday
Feature activation-0.017
Top resid features:
.
Token.
Feature activation+0.082
Top resid features:
Ċ
TokenĊ
Feature activation+0.052
Top resid features:
Ċ
TokenĊ
Feature activation+0.040
Top resid features:
The
TokenThe
Feature activation+0.142
Top resid features:
defence
Token defence
Feature activation+0.000
Top resid features:
ministry
Token ministry
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation+0.000
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
defence
Token defence
Feature activation+0.020
Top resid features:
ministry
Token ministry
Feature activation-0.000
Top resid features:
committee
Token committee
Feature activation+0.044
Top resid features:
headed
Token headed
Feature activation+0.012
Top resid features:
by
Token by
Feature activation+0.023
Top resid features:
V
Token V
Feature activation+0.151
Top resid features:
.
Token.
Feature activation+0.062
Top resid features:
K
TokenK
Feature activation+0.061
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
defence
Token defence
Feature activation+0.013
Top resid features:
ministry
Token ministry
Feature activation-0.008
Top resid features:
committee
Token committee
Feature activation+0.027
Top resid features:
headed
Token headed
Feature activation+0.008
Top resid features:
by
Token by
Feature activation+0.021
Top resid features:
V
Token V
Feature activation+0.131
Top resid features:
.
Token.
Feature activation+0.091
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
The
TokenThe
Feature activation+0.070
Top resid features:
defence
Token defence
Feature activation+0.047
Top resid features:
ministry
Token ministry
Feature activation+0.005
Top resid features:
committee
Token committee
Feature activation+0.031
Top resid features:
headed
Token headed
Feature activation-0.064
Top resid features:
by
Token by
Feature activation+0.135
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
defence
Token defence
Feature activation+0.023
Top resid features:
ministry
Token ministry
Feature activation-0.008
Top resid features:
committee
Token committee
Feature activation+0.041
Top resid features:
headed
Token headed
Feature activation-0.000
Top resid features:
by
Token by
Feature activation+0.037
Top resid features:
V
Token V
Feature activation+0.204
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.020
Top resid features:
V
Token V
Feature activation+0.053
Top resid features:
.
Token.
Feature activation+0.044
Top resid features:
K
TokenK
Feature activation-0.042
Top resid features:
.
Token.
Feature activation+0.045
Top resid features:
A
Token A
Feature activation+0.151
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.080
Top resid features:
rik
Tokenrik
Feature activation+0.016
Top resid features:
ar
Tokenar
Feature activation+0.023
Top resid features:
said
Token said
Feature activation+0.041
Top resid features:
Saturday
Token Saturday
Feature activation-0.010
Top resid features:
.
Token.
Feature activation+0.058
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.095
Top resid features:
rik
Tokenrik
Feature activation+0.012
Top resid features:
ar
Tokenar
Feature activation+0.024
Top resid features:
said
Token said
Feature activation+0.033
Top resid features:
Saturday
Token Saturday
Feature activation-0.011
Top resid features:
.
Token.
Feature activation+0.041
Top resid features:
.
Token.
Feature activation+0.033
Top resid features:
K
TokenK
Feature activation+0.004
Top resid features:
.
Token.
Feature activation+0.035
Top resid features:
A
Token A
Feature activation+0.046
Top resid features:
atre
Tokenatre
Feature activation+0.063
Top resid features:
,
Token,
Feature activation+0.119
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
said
Token said
Feature activation+0.069
Top resid features:
Saturday
Token Saturday
Feature activation-0.005
Top resid features:
.
Token.
Feature activation+0.070
Top resid features:
Ċ
TokenĊ
Feature activation+0.029
Top resid features:
Ċ
TokenĊ
Feature activation+0.025
Top resid features:
The
TokenThe
Feature activation+0.084
Top resid features:
defence
Token defence
Feature activation+0.069
Top resid features:
ministry
Token ministry
Feature activation-0.006
Top resid features:
committee
Token committee
Feature activation+0.011
Top resid features:
headed
Token headed
Feature activation-0.029
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
said
Token said
Feature activation+0.078
Top resid features:
Saturday
Token Saturday
Feature activation-0.012
Top resid features:
.
Token.
Feature activation+0.056
Top resid features:
Ċ
TokenĊ
Feature activation+0.034
Top resid features:
Ċ
TokenĊ
Feature activation+0.029
Top resid features:
The
TokenThe
Feature activation+0.120
Top resid features:
defence
Token defence
Feature activation+0.068
Top resid features:
ministry
Token ministry
Feature activation-0.041
Top resid features:
committee
Token committee
Feature activation+0.061
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
said
Token said
Feature activation+0.091
Top resid features:
Saturday
Token Saturday
Feature activation-0.018
Top resid features:
.
Token.
Feature activation+0.077
Top resid features:
Ċ
TokenĊ
Feature activation+0.058
Top resid features:
Ċ
TokenĊ
Feature activation+0.050
Top resid features:
The
TokenThe
Feature activation+0.184
Top resid features:
defence
Token defence
Feature activation-0.043
Top resid features:
ministry
Token ministry
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation+0.000
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
Saturday
Token Saturday
Feature activation-0.018
Top resid features:
.
Token.
Feature activation+0.075
Top resid features:
Ċ
TokenĊ
Feature activation+0.037
Top resid features:
Ċ
TokenĊ
Feature activation+0.031
Top resid features:
The
TokenThe
Feature activation+0.132
Top resid features:
defence
Token defence
Feature activation+0.156
Top resid features:
ministry
Token ministry
Feature activation-0.085
Top resid features:
committee
Token committee
Feature activation+0.000
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:

Decoder Weights Distribution

Head 0: 0.08

Head 1: 0.08

Head 2: 0.08

Head 3: 0.11

Head 4: 0.08

Head 5: 0.09

Head 6: 0.10

Head 7: 0.09

Head 8: 0.08

Head 9: 0.07

Head 10: 0.07

Head 11: 0.08

Positive logits

Bloomberg1.47

billion1.47

Borders1.45

migration1.44

Brexit1.42

Dangerous1.41

franc1.37

Billion1.36

actionGroup1.35

Migration1.35

Clover1.34

manifesto1.34

forecasting1.32

Sonia1.32

ufact1.32

Laun1.32

Flavoring1.32

forecasts1.31

Brexit1.31

Bans1.30

Negative logits

Interstitial-1.76

interstitial-1.66

ather-1.56

osh-1.53

ּ-1.53

orge-1.50

aneously-1.48

olor-1.48

etry-1.47

]);-1.46

owler-1.46

)))-1.46

,'"-1.45

nosis-1.43

-1.43

-1.41

lements-1.40

rape-1.39

gew-1.38

asting-1.37

INTERVAL 0.000 - 0.000
CONTAINS 100.000%

capacity
Token capacity
Feature activation+0.000
like
Token like
Feature activation+0.000
steel
Token steel
Feature activation+0.000
and
Token and
Feature activation+0.000
cement
Token cement
Feature activation+0.000
)
Token)
Feature activation+0.000
following
Token following
Feature activation+0.000
the
Token the
Feature activation+0.000
initial
Token initial
Feature activation+0.000
surge
Token surge
Feature activation+0.000
in
Token in
Feature activation+0.000
groups
Token groups
Feature activation+0.000
.
Token.
Feature activation+0.000
Officials
Token Officials
Feature activation+0.000
provided
Token provided
Feature activation+0.000
no
Token no
Feature activation+0.000
answer
Token answer
Feature activation+0.000
when
Token when
Feature activation+0.000
asked
Token asked
Feature activation+0.000
how
Token how
Feature activation+0.000
they
Token they
Feature activation+0.000
might
Token might
Feature activation+0.000
the
Token the
Feature activation+0.000
essential
Token essential
Feature activation+0.000
thing
Token thing
Feature activation+0.000
we
Token we
Feature activation+0.000
needed
Token needed
Feature activation+0.000
in
Token in
Feature activation+0.000
life
Token life
Feature activation+0.000
apart
Token apart
Feature activation+0.000
from
Token from
Feature activation+0.000
food
Token food
Feature activation+0.000
,
Token,
Feature activation+0.000
to
Token to
Feature activation+0.000
a
Token a
Feature activation+0.000
2
Token 2
Feature activation+0.000
-
Token-
Feature activation+0.000
0
Token0
Feature activation+0.000
deficit
Token deficit
Feature activation+0.000
âĢĶ
Token âĢĶ
Feature activation+0.000
a
Token a
Feature activation+0.000
superior
Token superior
Feature activation+0.000
opponent
Token opponent
Feature activation+0.000
,
Token,
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
to
Token to
Feature activation+0.000
appear
Token appear
Feature activation+0.000
as
Token as
Feature activation+0.000
a
Token a
Feature activation+0.000
witness
Token witness
Feature activation+0.000
at
Token at
Feature activation+0.000
the
Token the
Feature activation+0.000
Sc
Token Sc
Feature activation+0.000
opes
Tokenopes
Feature activation+0.000
trial
Token trial
Feature activation+0.000

BOTTOM ACTIVATIONS
MIN = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000

Top feature 3 in H1.3: (feature 1597)

TOP ACTIVATIONS
MAX = 1.347

owner
Token owner
Feature activation+0.576
called
Token called
Feature activation+0.649
911
Token 911
Feature activation+0.597
.
Token.
Feature activation+0.790
When
Token When
Feature activation+0.912
an
Token an
Feature activation+1.347
officer
Token officer
Feature activation+1.129
arrived
Token arrived
Feature activation+1.076
,
Token,
Feature activation+0.943
Th
Token Th
Feature activation+0.818
ones
Tokenones
Feature activation+0.707
called
Token called
Feature activation+0.649
911
Token 911
Feature activation+0.597
.
Token.
Feature activation+0.790
When
Token When
Feature activation+0.912
an
Token an
Feature activation+1.347
officer
Token officer
Feature activation+1.129
arrived
Token arrived
Feature activation+1.076
,
Token,
Feature activation+0.943
Th
Token Th
Feature activation+0.818
ones
Tokenones
Feature activation+0.707
avan
Tokenavan
Feature activation+0.578
911
Token 911
Feature activation+0.597
.
Token.
Feature activation+0.790
When
Token When
Feature activation+0.912
an
Token an
Feature activation+1.347
officer
Token officer
Feature activation+1.129
arrived
Token arrived
Feature activation+1.076
,
Token,
Feature activation+0.943
Th
Token Th
Feature activation+0.818
ones
Tokenones
Feature activation+0.707
avan
Tokenavan
Feature activation+0.578
h
Tokenh
Feature activation+0.575
Ļ
TokenĻ
Feature activation+0.287
t
Tokent
Feature activation+0.434
analogous
Token analogous
Feature activation+0.583
.
Token.
Feature activation+0.549
Unlike
Token Unlike
Feature activation+0.704
the
Token the
Feature activation+0.992
Asian
Token Asian
Feature activation+0.939
countries
Token countries
Feature activation+0.779
that
Token that
Feature activation+0.700
got
Token got
Feature activation+0.542
in
Token in
Feature activation+0.351
theory
Token theory
Feature activation+0.000
.
Token.
Feature activation+0.000
If
Token If
Feature activation+0.064
at
Token at
Feature activation+0.628
any
Token any
Feature activation+0.751
point
Token point
Feature activation+0.973
a
Token a
Feature activation+0.858
chromosome
Token chromosome
Feature activation+0.609
was
Token was
Feature activation+0.473
spontaneously
Token spontaneously
Feature activation+0.354
lost
Token lost
Feature activation+0.316
.
Token.
Feature activation+0.790
When
Token When
Feature activation+0.912
an
Token an
Feature activation+1.347
officer
Token officer
Feature activation+1.129
arrived
Token arrived
Feature activation+1.076
,
Token,
Feature activation+0.943
Th
Token Th
Feature activation+0.818
ones
Tokenones
Feature activation+0.707
avan
Tokenavan
Feature activation+0.578
h
Tokenh
Feature activation+0.575
was
Token was
Feature activation+0.469
t
Tokent
Feature activation+0.434
analogous
Token analogous
Feature activation+0.583
.
Token.
Feature activation+0.549
Unlike
Token Unlike
Feature activation+0.704
the
Token the
Feature activation+0.992
Asian
Token Asian
Feature activation+0.939
countries
Token countries
Feature activation+0.779
that
Token that
Feature activation+0.700
got
Token got
Feature activation+0.542
in
Token in
Feature activation+0.351
trouble
Token trouble
Feature activation+0.397
audience
Token audience
Feature activation+0.000
.
Token.
Feature activation+0.030
According
Token According
Feature activation+0.165
to
Token to
Feature activation+0.611
Ed
Token Ed
Feature activation+0.806
an
Tokenan
Feature activation+0.926
G
Token G
Feature activation+0.883
elt
Tokenelt
Feature activation+0.552
,
Token,
Feature activation+0.303
if
Token if
Feature activation+0.253
you
Token you
Feature activation+0.213
ML
TokenML
Feature activation+0.000
export
Token export
Feature activation+0.000
.
Token.
Feature activation+0.000
As
Token As
Feature activation+0.314
someone
Token someone
Feature activation+0.866
who
Token who
Feature activation+0.922
has
Token has
Feature activation+0.775
to
Token to
Feature activation+0.611
send
Token send
Feature activation+0.740
out
Token out
Feature activation+0.731
charts
Token charts
Feature activation+0.508
When
Token When
Feature activation+0.087
Louise
Token Louise
Feature activation+0.596
arrived
Token arrived
Feature activation+0.720
at
Token at
Feature activation+0.853
Cogn
Token Cogn
Feature activation+0.784
ac
Tokenac
Feature activation+0.919
,
Token,
Feature activation+0.565
Ant
Token Ant
Feature activation+0.498
oin
Tokenoin
Feature activation+0.576
ette
Tokenette
Feature activation+0.205
was
Token was
Feature activation+0.130
!
Token!
Feature activation+0.578
You
Token You
Feature activation+0.518
are
Token are
Feature activation+0.745
really
Token really
Feature activation+0.856
damn
Token damn
Feature activation+0.879
powerful
Token powerful
Feature activation+0.919
.
Token.
Feature activation+0.737
Ċ
TokenĊ
Feature activation+0.078
Ċ
TokenĊ
Feature activation+0.000
Now
TokenNow
Feature activation+0.000
,
Token,
Feature activation+0.008
has
Token has
Feature activation+0.304
decided
Token decided
Feature activation+0.407
.
Token.
Feature activation+0.412
He
Token He
Feature activation+0.567
knows
Token knows
Feature activation+0.726
where
Token where
Feature activation+0.918
he
Token he
Feature activation+0.657
's
Token's
Feature activation+0.491
going
Token going
Feature activation+0.449
to
Token to
Feature activation+0.329
college
Token college
Feature activation+0.257
The
Token The
Feature activation+0.130
owner
Token owner
Feature activation+0.576
called
Token called
Feature activation+0.649
911
Token 911
Feature activation+0.597
.
Token.
Feature activation+0.790
When
Token When
Feature activation+0.912
an
Token an
Feature activation+1.347
officer
Token officer
Feature activation+1.129
arrived
Token arrived
Feature activation+1.076
,
Token,
Feature activation+0.943
Th
Token Th
Feature activation+0.818
not
Token not
Feature activation+0.617
protect
Token protect
Feature activation+0.773
anyone
Token anyone
Feature activation+0.670
.
Token.
Feature activation+0.577
They
Token They
Feature activation+0.569
get
Token get
Feature activation+0.899
their
Token their
Feature activation+0.796
j
Token j
Feature activation+0.771
oll
Tokenoll
Feature activation+0.769
ies
Tokenies
Feature activation+0.404
murdering
Token murdering
Feature activation+0.531
him
Token him
Feature activation+0.000
.
Token.
Feature activation+0.000
For
Token For
Feature activation+0.280
the
Token the
Feature activation+0.700
most
Token most
Feature activation+0.764
part
Token part
Feature activation+0.899
,
Token,
Feature activation+0.635
this
Token this
Feature activation+0.554
view
Token view
Feature activation+0.526
is
Token is
Feature activation+0.414
correct
Token correct
Feature activation+0.367
.
Token.
Feature activation+0.000
For
Token For
Feature activation+0.000
example
Token example
Feature activation+0.134
,
Token,
Feature activation+0.417
on
Token on
Feature activation+0.676
one
Token one
Feature activation+0.887
questionnaire
Token questionnaire
Feature activation+0.758
,
Token,
Feature activation+0.511
respondents
Token respondents
Feature activation+0.548
are
Token are
Feature activation+0.415
asked
Token asked
Feature activation+0.464
.
Token.
Feature activation+0.000
The
Token The
Feature activation+0.000
company
Token company
Feature activation+0.407
staying
Token staying
Feature activation+0.614
away
Token away
Feature activation+0.785
from
Token from
Feature activation+0.887
home
Token home
Feature activation+0.787
consoles
Token consoles
Feature activation+0.612
after
Token after
Feature activation+0.700
the
Token the
Feature activation+0.374
Saturn
Token Saturn
Feature activation+0.159
Ark
Token Ark
Feature activation+0.026
ency
Tokenency
Feature activation+0.038
.
Token.
Feature activation+0.177
It
Token It
Feature activation+0.310
quickly
Token quickly
Feature activation+0.773
propag
Token propag
Feature activation+0.886
ated
Tokenated
Feature activation+0.553
to
Token to
Feature activation+0.608
all
Token all
Feature activation+0.617
of
Token of
Feature activation+0.367
our
Token our
Feature activation+0.256
.
Token.
Feature activation+0.030
According
Token According
Feature activation+0.165
to
Token to
Feature activation+0.611
Ed
Token Ed
Feature activation+0.806
an
Tokenan
Feature activation+0.926
G
Token G
Feature activation+0.883
elt
Tokenelt
Feature activation+0.552
,
Token,
Feature activation+0.303
if
Token if
Feature activation+0.253
you
Token you
Feature activation+0.213
don
Token don
Feature activation+0.198
Cruz
Token Cruz
Feature activation+0.000
.
Token.
Feature activation+0.000
Given
Token Given
Feature activation+0.228
the
Token the
Feature activation+0.586
choice
Token choice
Feature activation+0.822
between
Token between
Feature activation+0.883
Cruz
Token Cruz
Feature activation+0.733
and
Token and
Feature activation+0.533
Clinton
Token Clinton
Feature activation+0.426
right
Token right
Feature activation+0.460
now
Token now
Feature activation+0.345

Top DFA by src position
MAX = 0.710

The
Token The
Feature activation+0.451
Top resid features:
owner
Token owner
Feature activation+0.168
Top resid features:
called
Token called
Feature activation+0.162
Top resid features:
911
Token 911
Feature activation+0.054
Top resid features:
.
Token.
Feature activation+0.182
Top resid features:
When
Token When
Feature activation+0.623
Top resid features:
an
Token an
Feature activation+0.267
Top resid features:
officer
Token officer
Feature activation+0.000
Top resid features:
arrived
Token arrived
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
Th
Token Th
Feature activation+0.000
Top resid features:
The
Token The
Feature activation+0.384
Top resid features:
owner
Token owner
Feature activation+0.129
Top resid features:
called
Token called
Feature activation+0.156
Top resid features:
911
Token 911
Feature activation+0.062
Top resid features:
.
Token.
Feature activation+0.169
Top resid features:
When
Token When
Feature activation+0.513
Top resid features:
an
Token an
Feature activation+0.281
Top resid features:
officer
Token officer
Feature activation+0.003
Top resid features:
arrived
Token arrived
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
Th
Token Th
Feature activation+0.000
Top resid features:
The
Token The
Feature activation+0.331
Top resid features:
owner
Token owner
Feature activation+0.123
Top resid features:
called
Token called
Feature activation+0.121
Top resid features:
911
Token 911
Feature activation+0.049
Top resid features:
.
Token.
Feature activation+0.184
Top resid features:
When
Token When
Feature activation+0.453
Top resid features:
an
Token an
Feature activation+0.247
Top resid features:
officer
Token officer
Feature activation+0.108
Top resid features:
arrived
Token arrived
Feature activation+0.055
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
Th
Token Th
Feature activation+0.000
Top resid features:
âĢ
TokenâĢ
Feature activation+0.013
Top resid features:
Ļ
TokenĻ
Feature activation+0.098
Top resid features:
t
Tokent
Feature activation+0.106
Top resid features:
analogous
Token analogous
Feature activation+0.115
Top resid features:
.
Token.
Feature activation+0.109
Top resid features:
Unlike
Token Unlike
Feature activation+0.606
Top resid features:
the
Token the
Feature activation+0.226
Top resid features:
Asian
Token Asian
Feature activation+0.000
Top resid features:
countries
Token countries
Feature activation+0.000
Top resid features:
that
Token that
Feature activation+0.000
Top resid features:
got
Token got
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.385
Top resid features:
viable
Token viable
Feature activation+0.063
Top resid features:
theory
Token theory
Feature activation+0.133
Top resid features:
.
Token.
Feature activation+0.381
Top resid features:
If
Token If
Feature activation+0.615
Top resid features:
at
Token at
Feature activation+0.268
Top resid features:
any
Token any
Feature activation+0.350
Top resid features:
point
Token point
Feature activation+0.202
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
chromosome
Token chromosome
Feature activation+0.000
Top resid features:
The
Token The
Feature activation+0.343
Top resid features:
owner
Token owner
Feature activation+0.124
Top resid features:
called
Token called
Feature activation+0.123
Top resid features:
911
Token 911
Feature activation+0.044
Top resid features:
.
Token.
Feature activation+0.182
Top resid features:
When
Token When
Feature activation+0.358
Top resid features:
an
Token an
Feature activation+0.219
Top resid features:
officer
Token officer
Feature activation+0.091
Top resid features:
arrived
Token arrived
Feature activation+0.097
Top resid features:
,
Token,
Feature activation+0.079
Top resid features:
Th
Token Th
Feature activation+0.000
Top resid features:
âĢ
TokenâĢ
Feature activation+0.031
Top resid features:
Ļ
TokenĻ
Feature activation+0.115
Top resid features:
t
Tokent
Feature activation+0.094
Top resid features:
analogous
Token analogous
Feature activation+0.097
Top resid features:
.
Token.
Feature activation+0.150
Top resid features:
Unlike
Token Unlike
Feature activation+0.524
Top resid features:
the
Token the
Feature activation+0.228
Top resid features:
Asian
Token Asian
Feature activation-0.018
Top resid features:
countries
Token countries
Feature activation+0.000
Top resid features:
that
Token that
Feature activation+0.000
Top resid features:
got
Token got
Feature activation+0.000
Top resid features:
know
Token know
Feature activation+0.109
Top resid features:
your
Token your
Feature activation+0.146
Top resid features:
target
Token target
Feature activation+0.090
Top resid features:
audience
Token audience
Feature activation+0.116
Top resid features:
.
Token.
Feature activation+0.293
Top resid features:
According
Token According
Feature activation+0.426
Top resid features:
to
Token to
Feature activation+0.276
Top resid features:
Ed
Token Ed
Feature activation+0.292
Top resid features:
an
Tokenan
Feature activation+0.171
Top resid features:
G
Token G
Feature activation+0.000
Top resid features:
elt
Tokenelt
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.406
Top resid features:
ML
TokenML
Feature activation+0.105
Top resid features:
export
Token export
Feature activation+0.123
Top resid features:
.
Token.
Feature activation+0.346
Top resid features:
As
Token As
Feature activation+0.710
Top resid features:
someone
Token someone
Feature activation+0.455
Top resid features:
who
Token who
Feature activation+0.199
Top resid features:
has
Token has
Feature activation+0.000
Top resid features:
to
Token to
Feature activation+0.000
Top resid features:
send
Token send
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.427
Top resid features:
Con
Token Con
Feature activation+0.163
Top resid features:
te
Tokente
Feature activation+0.090
Top resid features:
.
Token.
Feature activation+0.287
Top resid features:
When
Token When
Feature activation+0.483
Top resid features:
Louise
Token Louise
Feature activation+0.121
Top resid features:
arrived
Token arrived
Feature activation+0.186
Top resid features:
at
Token at
Feature activation+0.297
Top resid features:
Cogn
Token Cogn
Feature activation+0.148
Top resid features:
ac
Tokenac
Feature activation+0.142
Top resid features:
.
Token.
Feature activation+0.212
Top resid features:
And
Token And
Feature activation+0.257
Top resid features:
with
Token with
Feature activation+0.193
Top resid features:
reason
Token reason
Feature activation+0.224
Top resid features:
!
Token!
Feature activation+0.162
Top resid features:
You
Token You
Feature activation+0.367
Top resid features:
are
Token are
Feature activation+0.211
Top resid features:
really
Token really
Feature activation+0.130
Top resid features:
damn
Token damn
Feature activation+0.162
Top resid features:
powerful
Token powerful
Feature activation+0.096
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.348
Top resid features:
decide
Token decide
Feature activation+0.062
Top resid features:
.
Token.
Feature activation+0.223
Top resid features:
Check
Token Check
Feature activation+0.156
Top resid features:
that
Token that
Feature activation+0.213
Top resid features:
:
Token:
Feature activation+0.132
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.361
Top resid features:
winter
Token winter
Feature activation+0.207
Top resid features:
morning
Token morning
Feature activation+0.140
Top resid features:
.
Token.
Feature activation+0.314
Top resid features:
The
Token The
Feature activation+0.490
Top resid features:
owner
Token owner
Feature activation+0.159
Top resid features:
called
Token called
Feature activation+0.196
Top resid features:
911
Token 911
Feature activation+0.060
Top resid features:
.
Token.
Feature activation+0.085
Top resid features:
When
Token When
Feature activation+0.324
Top resid features:
do
Token do
Feature activation+0.144
Top resid features:
not
Token not
Feature activation+0.106
Top resid features:
protect
Token protect
Feature activation+0.089
Top resid features:
anyone
Token anyone
Feature activation+0.084
Top resid features:
.
Token.
Feature activation+0.065
Top resid features:
They
Token They
Feature activation+0.439
Top resid features:
get
Token get
Feature activation+0.150
Top resid features:
their
Token their
Feature activation+0.000
Top resid features:
j
Token j
Feature activation+0.000
Top resid features:
oll
Tokenoll
Feature activation+0.000
Top resid features:
ies
Tokenies
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.333
Top resid features:
alternative
Token alternative
Feature activation+0.115
Top resid features:
to
Token to
Feature activation+0.151
Top resid features:
him
Token him
Feature activation+0.171
Top resid features:
.
Token.
Feature activation+0.236
Top resid features:
For
Token For
Feature activation+0.651
Top resid features:
the
Token the
Feature activation+0.356
Top resid features:
most
Token most
Feature activation+0.293
Top resid features:
part
Token part
Feature activation+0.017
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
this
Token this
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.382
Top resid features:
.
Token.
Feature activation+0.166
Top resid features:
For
Token For
Feature activation+0.529
Top resid features:
example
Token example
Feature activation+0.287
Top resid features:
,
Token,
Feature activation+0.284
Top resid features:
on
Token on
Feature activation+0.346
Top resid features:
one
Token one
Feature activation+0.316
Top resid features:
questionnaire
Token questionnaire
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.366
Top resid features:
business
Token business
Feature activation+0.053
Top resid features:
.
Token.
Feature activation+0.315
Top resid features:
The
Token The
Feature activation+0.621
Top resid features:
company
Token company
Feature activation+0.173
Top resid features:
staying
Token staying
Feature activation+0.320
Top resid features:
away
Token away
Feature activation+0.257
Top resid features:
from
Token from
Feature activation+0.205
Top resid features:
home
Token home
Feature activation+0.000
Top resid features:
js
Tokenjs
Feature activation+0.138
Top resid features:
at
Token at
Feature activation+0.171
Top resid features:
Ark
Token Ark
Feature activation+0.079
Top resid features:
ency
Tokenency
Feature activation+0.109
Top resid features:
.
Token.
Feature activation+0.206
Top resid features:
It
Token It
Feature activation+0.451
Top resid features:
quickly
Token quickly
Feature activation+0.246
Top resid features:
propag
Token propag
Feature activation+0.133
Top resid features:
ated
Tokenated
Feature activation+0.000
Top resid features:
to
Token to
Feature activation+0.000
Top resid features:
all
Token all
Feature activation+0.000
Top resid features:
know
Token know
Feature activation+0.108
Top resid features:
your
Token your
Feature activation+0.141
Top resid features:
target
Token target
Feature activation+0.090
Top resid features:
audience
Token audience
Feature activation+0.123
Top resid features:
.
Token.
Feature activation+0.249
Top resid features:
According
Token According
Feature activation+0.499
Top resid features:
to
Token to
Feature activation+0.257
Top resid features:
Ed
Token Ed
Feature activation+0.224
Top resid features:
an
Tokenan
Feature activation+0.171
Top resid features:
G
Token G
Feature activation+0.015
Top resid features:
elt
Tokenelt
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.381
Top resid features:
or
Token or
Feature activation+0.099
Top resid features:
Ted
Token Ted
Feature activation+0.123
Top resid features:
Cruz
Token Cruz
Feature activation+0.104
Top resid features:
.
Token.
Feature activation+0.299
Top resid features:
Given
Token Given
Feature activation+0.484
Top resid features:
the
Token the
Feature activation+0.365
Top resid features:
choice
Token choice
Feature activation+0.352
Top resid features:
between
Token between
Feature activation+0.099
Top resid features:
Cruz
Token Cruz
Feature activation+0.000
Top resid features:
and
Token and
Feature activation+0.000
Top resid features:

Decoder Weights Distribution

Head 0: 0.05

Head 1: 0.11

Head 2: 0.16

Head 3: 0.10

Head 4: 0.12

Head 5: 0.05

Head 6: 0.03

Head 7: 0.07

Head 8: 0.07

Head 9: 0.07

Head 10: 0.11

Head 11: 0.06

Positive logits

DragonMagazine0.77

eland0.69

!--0.63

hindsight0.61

borne0.61

however0.61

utenberg0.60

mist0.58

considerations0.55

instances0.54

experience0.54

analog0.54

yk0.54

FML0.53

iasm0.53

ivals0.52

zero0.52

Plan0.52

lington0.51

overt0.51

Negative logits

challeng-0.80

testament-0.78

embr-0.69

Ukrain-0.67

SPONSORED-0.66

$.-0.66

Tuls-0.65

sacrific-0.65

srfAttach-0.63

''.-0.63

'.-0.62

advoc-0.62

-0.62

indo-0.62

antioxid-0.62

\.-0.61

territ-0.61

".-0.60

disadvant-0.60

Palestin-0.59

INTERVAL 1.212 - 1.347
CONTAINS 0.000%

owner
Token owner
Feature activation+0.576
called
Token called
Feature activation+0.649
911
Token 911
Feature activation+0.597
.
Token.
Feature activation+0.790
When
Token When
Feature activation+0.912
an
Token an
Feature activation+1.347
officer
Token officer
Feature activation+1.129
arrived
Token arrived
Feature activation+1.076
,
Token,
Feature activation+0.943
Th
Token Th
Feature activation+0.818
ones
Tokenones
Feature activation+0.707

INTERVAL 1.077 - 1.212
CONTAINS 0.000%

called
Token called
Feature activation+0.649
911
Token 911
Feature activation+0.597
.
Token.
Feature activation+0.790
When
Token When
Feature activation+0.912
an
Token an
Feature activation+1.347
officer
Token officer
Feature activation+1.129
arrived
Token arrived
Feature activation+1.076
,
Token,
Feature activation+0.943
Th
Token Th
Feature activation+0.818
ones
Tokenones
Feature activation+0.707
avan
Tokenavan
Feature activation+0.578

INTERVAL 0.943 - 1.077
CONTAINS 0.000%

theory
Token theory
Feature activation+0.000
.
Token.
Feature activation+0.000
If
Token If
Feature activation+0.064
at
Token at
Feature activation+0.628
any
Token any
Feature activation+0.751
point
Token point
Feature activation+0.973
a
Token a
Feature activation+0.858
chromosome
Token chromosome
Feature activation+0.609
was
Token was
Feature activation+0.473
spontaneously
Token spontaneously
Feature activation+0.354
lost
Token lost
Feature activation+0.316
Ļ
TokenĻ
Feature activation+0.287
t
Tokent
Feature activation+0.434
analogous
Token analogous
Feature activation+0.583
.
Token.
Feature activation+0.549
Unlike
Token Unlike
Feature activation+0.704
the
Token the
Feature activation+0.992
Asian
Token Asian
Feature activation+0.939
countries
Token countries
Feature activation+0.779
that
Token that
Feature activation+0.700
got
Token got
Feature activation+0.542
in
Token in
Feature activation+0.351
911
Token 911
Feature activation+0.597
.
Token.
Feature activation+0.790
When
Token When
Feature activation+0.912
an
Token an
Feature activation+1.347
officer
Token officer
Feature activation+1.129
arrived
Token arrived
Feature activation+1.076
,
Token,
Feature activation+0.943
Th
Token Th
Feature activation+0.818
ones
Tokenones
Feature activation+0.707
avan
Tokenavan
Feature activation+0.578
h
Tokenh
Feature activation+0.575
.
Token.
Feature activation+0.790
When
Token When
Feature activation+0.912
an
Token an
Feature activation+1.347
officer
Token officer
Feature activation+1.129
arrived
Token arrived
Feature activation+1.076
,
Token,
Feature activation+0.943
Th
Token Th
Feature activation+0.818
ones
Tokenones
Feature activation+0.707
avan
Tokenavan
Feature activation+0.578
h
Tokenh
Feature activation+0.575
was
Token was
Feature activation+0.469

INTERVAL 0.808 - 0.943
CONTAINS 0.005%

CIA
Token CIA
Feature activation+0.038
.
Token.
Feature activation+0.334
gov
Tokengov
Feature activation+0.203
The
Token The
Feature activation+0.353
A
Token A
Feature activation+0.629
-
Token-
Feature activation+0.829
12
Token12
Feature activation+0.578
reconnaissance
Token reconnaissance
Feature activation+0.490
aircraft
Token aircraft
Feature activation+0.348
was
Token was
Feature activation+0.279
built
Token built
Feature activation+0.218
criticism
Token criticism
Feature activation+0.000
.
Token.
Feature activation+0.000
His
Token His
Feature activation+0.304
attitude
Token attitude
Feature activation+0.678
under
Token under
Feature activation+0.679
criticism
Token criticism
Feature activation+0.814
,
Token,
Feature activation+0.424
as
Token as
Feature activation+0.377
I
Token I
Feature activation+0.244
found
Token found
Feature activation+0.067
,
Token,
Feature activation+0.008
Instead
Token Instead
Feature activation+0.000
there
Token there
Feature activation+0.402
occurred
Token occurred
Feature activation+0.390
the
Token the
Feature activation+0.577
Â
Token Â
Feature activation+0.587
ĵ
Tokenĵ
Feature activation+0.865
post
Tokenpost
Feature activation+0.699
-
Token-
Feature activation+0.645
war
Tokenwar
Feature activation+0.478
boom
Token boom
Feature activation+0.559
,
Token,
Feature activation+0.323
ML
TokenML
Feature activation+0.000
export
Token export
Feature activation+0.000
.
Token.
Feature activation+0.000
As
Token As
Feature activation+0.314
someone
Token someone
Feature activation+0.866
who
Token who
Feature activation+0.922
has
Token has
Feature activation+0.775
to
Token to
Feature activation+0.611
send
Token send
Feature activation+0.740
out
Token out
Feature activation+0.731
charts
Token charts
Feature activation+0.508
Fog
Token Fog
Feature activation+0.000
el
Tokenel
Feature activation+0.000
.
Token.
Feature activation+0.000
Its
Token Its
Feature activation+0.191
actual
Token actual
Feature activation+0.606
meaning
Token meaning
Feature activation+0.867
is
Token is
Feature activation+0.659
supposed
Token supposed
Feature activation+0.677
to
Token to
Feature activation+0.553
invoke
Token invoke
Feature activation+0.568
the
Token the
Feature activation+0.325

INTERVAL 0.673 - 0.808
CONTAINS 0.034%

him
Token him
Feature activation+0.000
relentlessly
Token relentlessly
Feature activation+0.000
.
Token.
Feature activation+0.000
Whatever
Token Whatever
Feature activation+0.002
he
Token he
Feature activation+0.532
does
Token does
Feature activation+0.711
or
Token or
Feature activation+0.741
says
Token says
Feature activation+0.774
or
Token or
Feature activation+0.655
doesn
Token doesn
Feature activation+0.709
Â
TokenÂ
Feature activation+0.559
like
Token like
Feature activation+0.000
criticism
Token criticism
Feature activation+0.000
.
Token.
Feature activation+0.000
His
Token His
Feature activation+0.304
attitude
Token attitude
Feature activation+0.678
under
Token under
Feature activation+0.679
criticism
Token criticism
Feature activation+0.814
,
Token,
Feature activation+0.424
as
Token as
Feature activation+0.377
I
Token I
Feature activation+0.244
found
Token found
Feature activation+0.067
t
Tokent
Feature activation+0.000
mess
Token mess
Feature activation+0.000
around
Token around
Feature activation+0.000
.
Token.
Feature activation+0.000
Then
Token Then
Feature activation+0.088
go
Token go
Feature activation+0.673
out
Token out
Feature activation+0.706
by
Token by
Feature activation+0.673
some
Token some
Feature activation+0.800
nice
Token nice
Feature activation+0.682
new
Token new
Feature activation+0.541
.
Token.
Feature activation+0.000
They
Token They
Feature activation+0.225
ambush
Token ambush
Feature activation+0.649
him
Token him
Feature activation+0.617
and
Token and
Feature activation+0.632
make
Token make
Feature activation+0.682
his
Token his
Feature activation+0.589
strip
Token strip
Feature activation+0.524
.
Token.
Feature activation+0.474
As
Token As
Feature activation+0.525
he
Token he
Feature activation+0.779
reports
Token reports
Feature activation+0.000
.
Token.
Feature activation+0.000
In
Token In
Feature activation+0.081
her
Token her
Feature activation+0.602
work
Token work
Feature activation+0.821
as
Token as
Feature activation+0.749
a
Token a
Feature activation+0.681
refugee
Token refugee
Feature activation+0.505
activist
Token activist
Feature activation+0.416
,
Token,
Feature activation+0.296
Gö
Token Gö
Feature activation+0.212

INTERVAL 0.539 - 0.673
CONTAINS 0.103%

.
Token.
Feature activation+0.000
Since
Token Since
Feature activation+0.329
2014
Token 2014
Feature activation+0.564
,
Token,
Feature activation+0.529
at
Token at
Feature activation+0.615
least
Token least
Feature activation+0.623
five
Token five
Feature activation+0.661
chains
Token chains
Feature activation+0.473
have
Token have
Feature activation+0.418
opened
Token opened
Feature activation+0.386
locations
Token locations
Feature activation+0.220
arrived
Token arrived
Feature activation+1.076
,
Token,
Feature activation+0.943
Th
Token Th
Feature activation+0.818
ones
Tokenones
Feature activation+0.707
avan
Tokenavan
Feature activation+0.578
h
Tokenh
Feature activation+0.575
was
Token was
Feature activation+0.469
sitting
Token sitting
Feature activation+0.508
in
Token in
Feature activation+0.246
the
Token the
Feature activation+0.091
car
Token car
Feature activation+0.000
now
Token now
Feature activation+0.172
,
Token,
Feature activation+0.180
we
Token we
Feature activation+0.445
need
Token need
Feature activation+0.542
to
Token to
Feature activation+0.530
get
Token get
Feature activation+0.623
you
Token you
Feature activation+0.542
some
Token some
Feature activation+0.550
rest
Token rest
Feature activation+0.440
."
Token."
Feature activation+0.324
Neptune
Token Neptune
Feature activation+0.138
expected
Token expected
Feature activation+0.000
.
Token.
Feature activation+0.000
Until
Token Until
Feature activation+0.000
you
Token you
Feature activation+0.397
buy
Token buy
Feature activation+0.551
your
Token your
Feature activation+0.599
next
Token next
Feature activation+0.568
computer
Token computer
Feature activation+0.552
in
Token in
Feature activation+0.491
three
Token three
Feature activation+0.420
years
Token years
Feature activation+0.493
But
Token But
Feature activation+0.307
it
Token it
Feature activation+0.667
is
Token is
Feature activation+0.673
a
Token a
Feature activation+0.713
recurring
Token recurring
Feature activation+0.805
role
Token role
Feature activation+0.539
,
Token,
Feature activation+0.436
and
Token and
Feature activation+0.373
since
Token since
Feature activation+0.350
the
Token the
Feature activation+0.208
introduction
Token introduction
Feature activation+0.120

INTERVAL 0.404 - 0.539
CONTAINS 0.209%

time
Token time
Feature activation+0.445
.
Token.
Feature activation+0.454
You
Token You
Feature activation+0.240
agree
Token agree
Feature activation+0.471
to
Token to
Feature activation+0.415
receive
Token receive
Feature activation+0.410
occasional
Token occasional
Feature activation+0.375
updates
Token updates
Feature activation+0.244
and
Token and
Feature activation+0.032
special
Token special
Feature activation+0.000
offers
Token offers
Feature activation+0.000
Further
Token Further
Feature activation+0.108
proof
Token proof
Feature activation+0.573
that
Token that
Feature activation+0.610
all
Token all
Feature activation+0.641
styles
Token styles
Feature activation+0.572
of
Token of
Feature activation+0.440
music
Token music
Feature activation+0.408
are
Token are
Feature activation+0.338
universal
Token universal
Feature activation+0.173
.
Token.
Feature activation+0.169
It
Token It
Feature activation+0.064
suggest
Token suggest
Feature activation+0.505
this
Token this
Feature activation+0.423
is
Token is
Feature activation+0.435
by
Token by
Feature activation+0.352
design
Token design
Feature activation+0.489
.
Token.
Feature activation+0.420
Asked
Token Asked
Feature activation+0.251
whether
Token whether
Feature activation+0.680
government
Token government
Feature activation+0.707
agencies
Token agencies
Feature activation+0.541
in
Token in
Feature activation+0.298
the
Token the
Feature activation+0.519
"
Token "
Feature activation+0.499
Report
TokenReport
Feature activation+0.173
Abuse
Token Abuse
Feature activation+0.179
"
Token"
Feature activation+0.281
button
Token button
Feature activation+0.431
to
Token to
Feature activation+0.314
make
Token make
Feature activation+0.288
a
Token a
Feature activation+0.156
difference
Token difference
Feature activation+0.026
.
Token.
Feature activation+0.000
repeat
Token repeat
Feature activation+0.000
.
Token.
Feature activation+0.000
This
Token This
Feature activation+0.000
is
Token is
Feature activation+0.210
why
Token why
Feature activation+0.541
I
Token I
Feature activation+0.450
âĢ
TokenâĢ
Feature activation+0.500
Ļ
TokenĻ
Feature activation+0.354
m
Tokenm
Feature activation+0.419
saying
Token saying
Feature activation+0.466
no
Token no
Feature activation+0.420

INTERVAL 0.269 - 0.404
CONTAINS 0.284%

:
Token:
Feature activation+0.000
Win
Token Win
Feature activation+0.000
everything
Token everything
Feature activation+0.000
.
Token.
Feature activation+0.000
The
Token The
Feature activation+0.206
Yankees
Token Yankees
Feature activation+0.279
made
Token made
Feature activation+0.366
a
Token a
Feature activation+0.444
good
Token good
Feature activation+0.384
-
Token-
Feature activation+0.409
faith
Tokenfaith
Feature activation+0.368
her
Token her
Feature activation+0.000
ass
Token ass
Feature activation+0.000
.
Token.
Feature activation+0.000
And
Token And
Feature activation+0.099
the
Token the
Feature activation+0.346
next
Token next
Feature activation+0.348
break
Token break
Feature activation+0.473
,
Token,
Feature activation+0.208
G
Token G
Feature activation+0.091
wen
Tokenwen
Feature activation+0.201
would
Token would
Feature activation+0.045
.
Token.
Feature activation+0.000
I
Token I
Feature activation+0.000
falsely
Token falsely
Feature activation+0.114
believed
Token believed
Feature activation+0.228
I
Token I
Feature activation+0.204
had
Token had
Feature activation+0.296
won
Token won
Feature activation+0.367
the
Token the
Feature activation+0.241
battle
Token battle
Feature activation+0.206
,
Token,
Feature activation+0.014
when
Token when
Feature activation+0.000
requested
Token requested
Feature activation+0.087
but
Token but
Feature activation+0.278
did
Token did
Feature activation+0.268
not
Token not
Feature activation+0.321
receive
Token receive
Feature activation+0.431
more
Token more
Feature activation+0.389
specific
Token specific
Feature activation+0.416
or
Token or
Feature activation+0.382
additional
Token additional
Feature activation+0.415
information
Token information
Feature activation+0.333
from
Token from
Feature activation+0.194
man
Token man
Feature activation+0.000
command
Token command
Feature activation+0.047
.
Token.
Feature activation+0.101
Second
Token Second
Feature activation+0.074
,
Token,
Feature activation+0.310
if
Token if
Feature activation+0.303
there
Token there
Feature activation+0.374
is
Token is
Feature activation+0.242
no
Token no
Feature activation+0.264
man
Token man
Feature activation+0.210
page
Token page
Feature activation+0.000

INTERVAL 0.135 - 0.269
CONTAINS 0.418%

and
Token and
Feature activation+0.000
industrial
Token industrial
Feature activation+0.000
.
Token.
Feature activation+0.000
Things
Token Things
Feature activation+0.000
are
Token are
Feature activation+0.235
weird
Token weird
Feature activation+0.243
but
Token but
Feature activation+0.252
they
Token they
Feature activation+0.116
could
Token could
Feature activation+0.052
always
Token always
Feature activation+0.090
be
Token be
Feature activation+0.000
lithium
Token lithium
Feature activation+0.447
,
Token,
Feature activation+0.350
reserves
Token reserves
Feature activation+0.244
could
Token could
Feature activation+0.328
last
Token last
Feature activation+0.337
for
Token for
Feature activation+0.263
an
Token an
Feature activation+0.229
estimated
Token estimated
Feature activation+0.106
185
Token 185
Feature activation+0.122
years
Token years
Feature activation+0.116
.
Token.
Feature activation+0.052
limits
Token limits
Feature activation+0.000
.
Token.
Feature activation+0.000
E
Token E
Feature activation+0.000
.
Token.
Feature activation+0.201
g
Tokeng
Feature activation+0.119
.
Token.
Feature activation+0.266
traditional
Token traditional
Feature activation+0.382
401
Token 401
Feature activation+0.326
(
Token(
Feature activation+0.148
k
Tokenk
Feature activation+0.184
)
Token)
Feature activation+0.088
Think
Token Think
Feature activation+0.104
of
Token of
Feature activation+0.493
the
Token the
Feature activation+0.530
Hoover
Token Hoover
Feature activation+0.474
Dam
Token Dam
Feature activation+0.387
,
Token,
Feature activation+0.253
Mount
Token Mount
Feature activation+0.104
Rush
Token Rush
Feature activation+0.030
more
Tokenmore
Feature activation+0.000
,
Token,
Feature activation+0.000
the
Token the
Feature activation+0.000
mainstream
Token mainstream
Feature activation+0.000
.
Token.
Feature activation+0.000
Today
Token Today
Feature activation+0.033
,
Token,
Feature activation+0.227
these
Token these
Feature activation+0.227
symb
Token symb
Feature activation+0.203
iotic
Tokeniotic
Feature activation+0.147
technologies
Token technologies
Feature activation+0.250
have
Token have
Feature activation+0.127
earned
Token earned
Feature activation+0.078
them
Token them
Feature activation+0.018

INTERVAL 0.000 - 0.135
CONTAINS 98.947%

be
Token be
Feature activation+0.000
spread
Token spread
Feature activation+0.000
more
Token more
Feature activation+0.000
fairly
Token fairly
Feature activation+0.000
and
Token and
Feature activation+0.000
shared
Token shared
Feature activation+0.000
across
Token across
Feature activation+0.000
the
Token the
Feature activation+0.000
communities
Token communities
Feature activation+0.000
around
Token around
Feature activation+0.000
Western
Token Western
Feature activation+0.000
both
Token both
Feature activation+0.000
airlines
Token airlines
Feature activation+0.000
later
Token later
Feature activation+0.073
confirmed
Token confirmed
Feature activation+0.030
they
Token they
Feature activation+0.009
did
Token did
Feature activation+0.000
not
Token not
Feature activation+0.000
operate
Token operate
Feature activation+0.000
in
Token in
Feature activation+0.000
Ukrainian
Token Ukrainian
Feature activation+0.000
airspace
Token airspace
Feature activation+0.000
its
Token its
Feature activation+0.000
administrative
Token administrative
Feature activation+0.000
functions
Token functions
Feature activation+0.000
.
Token.
Feature activation+0.000
Because
Token Because
Feature activation+0.000
of
Token of
Feature activation+0.000
a
Token a
Feature activation+0.000
shortage
Token shortage
Feature activation+0.000
of
Token of
Feature activation+0.000
building
Token building
Feature activation+0.000
material
Token material
Feature activation+0.000
Illuminati
Token Illuminati
Feature activation+0.000
high
Token high
Feature activation+0.000
priest
Token priest
Feature activation+0.000
ess
Tokeness
Feature activation+0.000
.
Token.
Feature activation+0.000
In
Token In
Feature activation+0.134
fact
Token fact
Feature activation+0.708
,
Token,
Feature activation+0.519
I
Token I
Feature activation+0.434
don
Token don
Feature activation+0.501
âĢ
TokenâĢ
Feature activation+0.482
our
Token our
Feature activation+0.000
testing
Token testing
Feature activation+0.000
process
Token process
Feature activation+0.000
now
Token now
Feature activation+0.000
involves
Token involves
Feature activation+0.000
R
Token R
Feature activation+0.000
SC
TokenSC
Feature activation+0.000
unit
Token unit
Feature activation+0.000
/
Token/
Feature activation+0.000
integ
Tokeninteg
Feature activation+0.000
ration
Tokenration
Feature activation+0.000

BOTTOM ACTIVATIONS
MIN = 0.000

,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
Organisation
Token Organisation
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.057
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.057
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.057
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.057
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.057
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
ministry
Token ministry
Feature activation+0.057
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.057
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.057
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.057
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.057
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000

Top feature 4 in H1.3: (feature 10623)

TOP ACTIVATIONS
MAX = 1.266

mole
Token mole
Feature activation+0.000
rate
Token rate
Feature activation+0.000
help
Token help
Feature activation+0.450
humans
Token humans
Feature activation+0.873
live
Token live
Feature activation+0.977
longer
Token longer
Feature activation+1.266
?
Token?
Feature activation+0.874
<|endoftext|>
Token<|endoftext|>
Feature activation+0.430
Its
TokenIts
Feature activation+0.557
atmosphere
Token atmosphere
Feature activation+0.478
is
Token is
Feature activation+0.321
without
Token without
Feature activation+0.739
giving
Token giving
Feature activation+0.680
them
Token them
Feature activation+0.704
sob
Token sob
Feature activation+1.036
ri
Tokenri
Feature activation+1.115
ety
Tokenety
Feature activation+1.264
tests
Token tests
Feature activation+1.110
.
Token.
Feature activation+0.623
Ċ
TokenĊ
Feature activation+0.343
Ċ
TokenĊ
Feature activation+0.130
Through
TokenThrough
Feature activation+0.326
centuries
Token centuries
Feature activation+0.007
proved
Token proved
Feature activation+0.337
press
Token press
Feature activation+0.830
urized
Tokenurized
Feature activation+0.948
cab
Token cab
Feature activation+1.203
ins
Tokenins
Feature activation+1.235
for
Token for
Feature activation+0.796
passengers
Token passengers
Feature activation+0.933
as
Token as
Feature activation+0.678
an
Token an
Feature activation+0.374
airplane
Token airplane
Feature activation+0.501
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
stock
Token stock
Feature activation+0.000
trading
Token trading
Feature activation+0.053
game
Token game
Feature activation+0.520
platform
Token platform
Feature activation+0.741
Stock
Token Stock
Feature activation+1.217
f
Tokenf
Feature activation+1.200
use
Tokenuse
Feature activation+1.067
is
Token is
Feature activation+0.801
in
Token in
Feature activation+0.336
advanced
Token advanced
Feature activation+0.452
over
Token over
Feature activation+0.000
centuries
Token centuries
Feature activation+0.007
proved
Token proved
Feature activation+0.337
press
Token press
Feature activation+0.830
urized
Tokenurized
Feature activation+0.948
cab
Token cab
Feature activation+1.203
ins
Tokenins
Feature activation+1.235
for
Token for
Feature activation+0.796
passengers
Token passengers
Feature activation+0.933
as
Token as
Feature activation+0.678
an
Token an
Feature activation+0.374
stock
Token stock
Feature activation+0.000
trading
Token trading
Feature activation+0.053
game
Token game
Feature activation+0.520
platform
Token platform
Feature activation+0.741
Stock
Token Stock
Feature activation+1.217
f
Tokenf
Feature activation+1.200
use
Tokenuse
Feature activation+1.067
is
Token is
Feature activation+0.801
in
Token in
Feature activation+0.336
advanced
Token advanced
Feature activation+0.452
talks
Token talks
Feature activation+0.579
race
Tokenrace
Feature activation+0.458
back
Token back
Feature activation+0.605
into
Token into
Feature activation+0.584
bear
Token bear
Feature activation+0.942
market
Token market
Feature activation+0.996
territory
Token territory
Feature activation+1.182
...
Token...
Feature activation+0.990
Ċ
TokenĊ
Feature activation+0.631
Ċ
TokenĊ
Feature activation+0.411
And
TokenAnd
Feature activation+0.471
it
Token it
Feature activation+0.320
up
Token up
Feature activation+0.588
his
Token his
Feature activation+0.675
long
Token long
Feature activation+0.733
form
Tokenform
Feature activation+1.125
birth
Token birth
Feature activation+0.969
certificate
Token certificate
Feature activation+1.182
.
Token.
Feature activation+0.606
But
Token But
Feature activation+0.530
since
Token since
Feature activation+0.402
conspiracy
Token conspiracy
Feature activation+0.396
theorists
Token theorists
Feature activation+0.387
it
Tokenit
Feature activation+0.493
b
Token b
Feature activation+0.678
ien
Tokenien
Feature activation+0.893
se
Token se
Feature activation+0.817
trou
Token trou
Feature activation+1.113
ver
Tokenver
Feature activation+1.181
au
Token au
Feature activation+1.175
mo
Token mo
Feature activation+1.114
ins
Tokenins
Feature activation+1.080
une
Token une
Feature activation+1.037
person
Token person
Feature activation+0.928
b
Token b
Feature activation+0.678
ien
Tokenien
Feature activation+0.893
se
Token se
Feature activation+0.817
trou
Token trou
Feature activation+1.113
ver
Tokenver
Feature activation+1.181
au
Token au
Feature activation+1.175
mo
Token mo
Feature activation+1.114
ins
Tokenins
Feature activation+1.080
une
Token une
Feature activation+1.037
person
Token person
Feature activation+0.928
ne
Tokenne
Feature activation+0.746
Options
TokenOptions
Feature activation+0.550
r
Token r
Feature activation+0.686
Trans
TokenTrans
Feature activation+0.915
former
Tokenformer
Feature activation+1.016
r
Token r
Feature activation+1.012
Trans
TokenTrans
Feature activation+1.166
former
Tokenformer
Feature activation+1.150
::
Token ::
Feature activation+0.886
Pand
Token Pand
Feature activation+0.694
oc
Tokenoc
Feature activation+0.654
->
Token ->
Feature activation+0.456
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
wine
Token wine
Feature activation+0.000
made
Token made
Feature activation+0.000
them
Token them
Feature activation+0.221
laugh
Token laugh
Feature activation+0.740
hyster
Token hyster
Feature activation+1.164
ically
Tokenically
Feature activation+0.958
and
Token and
Feature activation+0.624
Yang
Token Yang
Feature activation+0.702
gigg
Token gigg
Feature activation+0.831
led
Tokenled
Feature activation+0.641
Char
Token Char
Feature activation+0.512
lam
Tokenlam
Feature activation+0.675
agne
Tokenagne
Feature activation+0.838
Th
Token Th
Feature activation+0.979
a
Tokena
Feature activation+1.006
God
Token God
Feature activation+1.162
and
Token and
Feature activation+0.507
Angela
Token Angela
Feature activation+0.636
Y
Token Y
Feature activation+0.506
ee
Tokenee
Feature activation+0.536
,
Token,
Feature activation+0.080
shift
Token shift
Feature activation+0.000
meant
Token meant
Feature activation+0.298
far
Token far
Feature activation+0.654
more
Token more
Feature activation+0.763
settlements
Token settlements
Feature activation+0.881
ended
Token ended
Feature activation+1.160
up
Token up
Feature activation+1.105
in
Token in
Feature activation+0.659
court
Token court
Feature activation+0.727
,
Token,
Feature activation+0.419
where
Token where
Feature activation+0.277
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
spider
Token spider
Feature activation+0.000
wrapping
Token wrapping
Feature activation+0.000
silk
Token silk
Feature activation+0.467
protein
Token protein
Feature activation+0.820
domains
Token domains
Feature activation+1.158
in
Token in
Feature activation+0.653
fibre
Token fibre
Feature activation+0.836
properties
Token properties
Feature activation+0.802
"
Token"
Feature activation+0.685
by
Token by
Feature activation+0.422
r
Token r
Feature activation+0.686
Trans
TokenTrans
Feature activation+0.915
former
Tokenformer
Feature activation+1.016
r
Token r
Feature activation+1.012
Trans
TokenTrans
Feature activation+1.166
former
Tokenformer
Feature activation+1.150
::
Token ::
Feature activation+0.886
Pand
Token Pand
Feature activation+0.694
oc
Tokenoc
Feature activation+0.654
->
Token ->
Feature activation+0.456
Comp
Token Comp
Feature activation+0.275
Network
Token Network
Feature activation+0.069
Start
Token Start
Feature activation+0.398
End
Token End
Feature activation+0.686
Total
Token Total
Feature activation+0.698
View
Token View
Feature activation+0.900
ers
Tokeners
Feature activation+1.133
(
Token (
Feature activation+0.377
000
Token000
Feature activation+0.405
)
Token)
Feature activation+0.296
View
Token View
Feature activation+0.339
ers
Tokeners
Feature activation+0.699
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
speakers
Token speakers
Feature activation+0.000
talk
Token talk
Feature activation+0.160
politics
Token politics
Feature activation+0.523
so
Token so
Feature activation+0.856
far
Token far
Feature activation+1.127
,
Token,
Feature activation+0.508
with
Token with
Feature activation+0.297
economist
Token economist
Feature activation+0.421
Glenn
Token Glenn
Feature activation+0.553
Hubbard
Token Hubbard
Feature activation+0.593
dred
Token dred
Feature activation+0.203
ged
Tokenged
Feature activation+0.307
up
Token up
Feature activation+0.588
his
Token his
Feature activation+0.675
long
Token long
Feature activation+0.733
form
Tokenform
Feature activation+1.125
birth
Token birth
Feature activation+0.969
certificate
Token certificate
Feature activation+1.182
.
Token.
Feature activation+0.606
But
Token But
Feature activation+0.530
since
Token since
Feature activation+0.402
its
Token its
Feature activation+0.072
P
Token P
Feature activation+0.344
aley
Tokenaley
Feature activation+0.601
Fest
TokenFest
Feature activation+0.756
debut
Token debut
Feature activation+0.862
Saturday
Token Saturday
Feature activation+1.120
with
Token with
Feature activation+0.576
the
Token the
Feature activation+0.358
cast
Token cast
Feature activation+0.434
and
Token and
Feature activation+0.098
creators
Token creators
Feature activation+0.110

Top DFA by src position
MAX = 1.092

<|endoftext|>
Token<|endoftext|>
Feature activation+1.032
Top resid features:
mole
Token mole
Feature activation+0.382
Top resid features:
rate
Token rate
Feature activation+0.383
Top resid features:
help
Token help
Feature activation+0.447
Top resid features:
humans
Token humans
Feature activation+0.536
Top resid features:
live
Token live
Feature activation+0.524
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.788
Top resid features:
agents
Token agents
Feature activation+0.245
Top resid features:
go
Token go
Feature activation+0.318
Top resid features:
home
Token home
Feature activation+0.316
Top resid features:
without
Token without
Feature activation+0.267
Top resid features:
giving
Token giving
Feature activation+0.391
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.985
Top resid features:
over
Token over
Feature activation+0.203
Top resid features:
centuries
Token centuries
Feature activation+0.356
Top resid features:
proved
Token proved
Feature activation+0.502
Top resid features:
press
Token press
Feature activation+0.370
Top resid features:
urized
Tokenurized
Feature activation+0.408
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+1.027
Top resid features:
stock
Token stock
Feature activation+0.352
Top resid features:
trading
Token trading
Feature activation+0.460
Top resid features:
game
Token game
Feature activation+0.624
Top resid features:
platform
Token platform
Feature activation+0.699
Top resid features:
Stock
Token Stock
Feature activation+0.612
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+1.073
Top resid features:
over
Token over
Feature activation+0.231
Top resid features:
centuries
Token centuries
Feature activation+0.490
Top resid features:
proved
Token proved
Feature activation+0.543
Top resid features:
press
Token press
Feature activation+0.440
Top resid features:
urized
Tokenurized
Feature activation+0.444
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.937
Top resid features:
stock
Token stock
Feature activation+0.320
Top resid features:
trading
Token trading
Feature activation+0.406
Top resid features:
game
Token game
Feature activation+0.528
Top resid features:
platform
Token platform
Feature activation+0.571
Top resid features:
Stock
Token Stock
Feature activation+0.620
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.842
Top resid features:
ever
Token ever
Feature activation+0.240
Top resid features:
ret
Token ret
Feature activation+0.236
Top resid features:
race
Tokenrace
Feature activation+0.244
Top resid features:
back
Token back
Feature activation+0.367
Top resid features:
into
Token into
Feature activation+0.302
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.786
Top resid features:
Obama
Token Obama
Feature activation+0.227
Top resid features:
finally
Token finally
Feature activation+0.234
Top resid features:
dred
Token dred
Feature activation+0.256
Top resid features:
ged
Tokenged
Feature activation+0.234
Top resid features:
up
Token up
Feature activation+0.286
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.863
Top resid features:
ite
Tokenite
Feature activation+0.182
Top resid features:
il
Token il
Feature activation+0.312
Top resid features:
do
Token do
Feature activation+0.271
Top resid features:
it
Tokenit
Feature activation+0.202
Top resid features:
b
Token b
Feature activation+0.303
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.842
Top resid features:
ite
Tokenite
Feature activation+0.174
Top resid features:
il
Token il
Feature activation+0.257
Top resid features:
do
Token do
Feature activation+0.298
Top resid features:
it
Tokenit
Feature activation+0.177
Top resid features:
b
Token b
Feature activation+0.269
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.760
Top resid features:
H
TokenH
Feature activation+0.132
Top resid features:
aky
Tokenaky
Feature activation+0.230
Top resid features:
ll
Tokenll
Feature activation+0.143
Top resid features:
Writer
TokenWriter
Feature activation+0.293
Top resid features:
Options
TokenOptions
Feature activation+0.279
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+1.092
Top resid features:
wine
Token wine
Feature activation+0.378
Top resid features:
made
Token made
Feature activation+0.473
Top resid features:
them
Token them
Feature activation+0.461
Top resid features:
laugh
Token laugh
Feature activation+0.704
Top resid features:
hyster
Token hyster
Feature activation+0.614
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.843
Top resid features:
interview
Token interview
Feature activation+0.320
Top resid features:
featuring
Token featuring
Feature activation+0.390
Top resid features:
Char
Token Char
Feature activation+0.298
Top resid features:
lam
Tokenlam
Feature activation+0.312
Top resid features:
agne
Tokenagne
Feature activation+0.340
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.953
Top resid features:
That
TokenThat
Feature activation+0.225
Top resid features:
shift
Token shift
Feature activation+0.347
Top resid features:
meant
Token meant
Feature activation+0.368
Top resid features:
far
Token far
Feature activation+0.364
Top resid features:
more
Token more
Feature activation+0.331
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.966
Top resid features:
spider
Token spider
Feature activation+0.357
Top resid features:
wrapping
Token wrapping
Feature activation+0.453
Top resid features:
silk
Token silk
Feature activation+0.591
Top resid features:
protein
Token protein
Feature activation+0.641
Top resid features:
domains
Token domains
Feature activation+0.707
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.664
Top resid features:
H
TokenH
Feature activation+0.115
Top resid features:
aky
Tokenaky
Feature activation+0.226
Top resid features:
ll
Tokenll
Feature activation+0.117
Top resid features:
Writer
TokenWriter
Feature activation+0.313
Top resid features:
Options
TokenOptions
Feature activation+0.309
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.934
Top resid features:
Episode
Token Episode
Feature activation+0.331
Top resid features:
Network
Token Network
Feature activation+0.332
Top resid features:
Start
Token Start
Feature activation+0.383
Top resid features:
End
Token End
Feature activation+0.365
Top resid features:
Total
Token Total
Feature activation+0.432
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+1.083
Top resid features:
speakers
Token speakers
Feature activation+0.384
Top resid features:
talk
Token talk
Feature activation+0.558
Top resid features:
politics
Token politics
Feature activation+0.520
Top resid features:
so
Token so
Feature activation+0.634
Top resid features:
far
Token far
Feature activation+0.506
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.822
Top resid features:
Obama
Token Obama
Feature activation+0.244
Top resid features:
finally
Token finally
Feature activation+0.261
Top resid features:
dred
Token dred
Feature activation+0.323
Top resid features:
ged
Tokenged
Feature activation+0.339
Top resid features:
up
Token up
Feature activation+0.362
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.888
Top resid features:
world
Tokenworld
Feature activation+0.210
Top resid features:
made
Token made
Feature activation+0.335
Top resid features:
its
Token its
Feature activation+0.163
Top resid features:
P
Token P
Feature activation+0.237
Top resid features:
aley
Tokenaley
Feature activation+0.320
Top resid features:

Decoder Weights Distribution

Head 0: 0.04

Head 1: 0.08

Head 2: 0.16

Head 3: 0.10

Head 4: 0.12

Head 5: 0.05

Head 6: 0.04

Head 7: 0.08

Head 8: 0.07

Head 9: 0.08

Head 10: 0.11

Head 11: 0.07

Positive logits

latable0.62

icago0.62

today0.58

0.58

0.57

estyle0.57

duino0.57

0.57

[…]0.57

Saturday0.57

Friday0.57

geries0.56

amid0.56

Ever0.56

onto0.55

California0.55

chery0.55

https0.55

estyles0.54

packs0.53

Negative logits

srfAttach-0.68

Winged-0.67

aforementioned-0.66

IRD-0.60

iqueness-0.60

quickShipAvailable-0.60

subscript-0.59

zsche-0.57

Appendix-0.56

ETHOD-0.55

ANCE-0.55

lieutenant-0.55

implication-0.55

question-0.55

trustee-0.55

endum-0.54

acknowled-0.53

incorpor-0.53

Lovely-0.52

Primordial-0.52

INTERVAL 1.139 - 1.266
CONTAINS 0.002%

it
Tokenit
Feature activation+0.493
b
Token b
Feature activation+0.678
ien
Tokenien
Feature activation+0.893
se
Token se
Feature activation+0.817
trou
Token trou
Feature activation+1.113
ver
Tokenver
Feature activation+1.181
au
Token au
Feature activation+1.175
mo
Token mo
Feature activation+1.114
ins
Tokenins
Feature activation+1.080
une
Token une
Feature activation+1.037
person
Token person
Feature activation+0.928
without
Token without
Feature activation+0.739
giving
Token giving
Feature activation+0.680
them
Token them
Feature activation+0.704
sob
Token sob
Feature activation+1.036
ri
Tokenri
Feature activation+1.115
ety
Tokenety
Feature activation+1.264
tests
Token tests
Feature activation+1.110
.
Token.
Feature activation+0.623
Ċ
TokenĊ
Feature activation+0.343
Ċ
TokenĊ
Feature activation+0.130
Through
TokenThrough
Feature activation+0.326
shift
Token shift
Feature activation+0.000
meant
Token meant
Feature activation+0.298
far
Token far
Feature activation+0.654
more
Token more
Feature activation+0.763
settlements
Token settlements
Feature activation+0.881
ended
Token ended
Feature activation+1.160
up
Token up
Feature activation+1.105
in
Token in
Feature activation+0.659
court
Token court
Feature activation+0.727
,
Token,
Feature activation+0.419
where
Token where
Feature activation+0.277
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
spider
Token spider
Feature activation+0.000
wrapping
Token wrapping
Feature activation+0.000
silk
Token silk
Feature activation+0.467
protein
Token protein
Feature activation+0.820
domains
Token domains
Feature activation+1.158
in
Token in
Feature activation+0.653
fibre
Token fibre
Feature activation+0.836
properties
Token properties
Feature activation+0.802
"
Token"
Feature activation+0.685
by
Token by
Feature activation+0.422
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
wine
Token wine
Feature activation+0.000
made
Token made
Feature activation+0.000
them
Token them
Feature activation+0.221
laugh
Token laugh
Feature activation+0.740
hyster
Token hyster
Feature activation+1.164
ically
Tokenically
Feature activation+0.958
and
Token and
Feature activation+0.624
Yang
Token Yang
Feature activation+0.702
gigg
Token gigg
Feature activation+0.831
led
Tokenled
Feature activation+0.641

INTERVAL 1.013 - 1.139
CONTAINS 0.005%

he
Tokenhe
Feature activation+0.783
ure
Tokenure
Feature activation+0.797
use
Tokenuse
Feature activation+0.730
pour
Token pour
Feature activation+0.859
av
Token av
Feature activation+0.928
oir
Tokenoir
Feature activation+1.052
dro
Token dro
Feature activation+0.876
it
Tokenit
Feature activation+0.688
a
Token a
Feature activation+0.373
vot
Token vot
Feature activation+0.260
re
Tokenre
Feature activation+0.043
Season
Token Season
Feature activation+0.000
Ticket
Token Ticket
Feature activation+0.042
Member
Token Member
Feature activation+0.300
App
Token App
Feature activation+0.557
reciation
Tokenreciation
Feature activation+0.942
Night
Token Night
Feature activation+1.069
,
Token,
Feature activation+0.471
featuring
Token featuring
Feature activation+0.487
discounted
Token discounted
Feature activation+0.574
concessions
Token concessions
Feature activation+0.630
for
Token for
Feature activation+0.219
Sh
TokenSh
Feature activation+0.210
ack
Tokenack
Feature activation+0.588
Fore
Token Fore
Feature activation+0.676
cast
Tokencast
Feature activation+0.499
store
Token store
Feature activation+0.936
clos
Token clos
Feature activation+1.064
ings
Tokenings
Feature activation+1.094
:
Token:
Feature activation+0.788
450
Token 450
Feature activation+0.699
to
Token to
Feature activation+0.311
550
Token 550
Feature activation+0.337
Network
Token Network
Feature activation+0.069
Start
Token Start
Feature activation+0.398
End
Token End
Feature activation+0.686
Total
Token Total
Feature activation+0.698
View
Token View
Feature activation+0.900
ers
Tokeners
Feature activation+1.133
(
Token (
Feature activation+0.377
000
Token000
Feature activation+0.405
)
Token)
Feature activation+0.296
View
Token View
Feature activation+0.339
ers
Tokeners
Feature activation+0.699
se
Token se
Feature activation+0.817
trou
Token trou
Feature activation+1.113
ver
Tokenver
Feature activation+1.181
au
Token au
Feature activation+1.175
mo
Token mo
Feature activation+1.114
ins
Tokenins
Feature activation+1.080
une
Token une
Feature activation+1.037
person
Token person
Feature activation+0.928
ne
Tokenne
Feature activation+0.746
capable
Token capable
Feature activation+0.691
d
Token d
Feature activation+0.615

INTERVAL 0.886 - 1.013
CONTAINS 0.017%

300
Token300
Feature activation+0.000
student
Token student
Feature activation+0.000
loan
Token loan
Feature activation+0.326
borrowers
Token borrowers
Feature activation+0.770
said
Token said
Feature activation+1.055
they
Token they
Feature activation+0.978
âĢ
TokenâĢ
Feature activation+0.846
Ļ
TokenĻ
Feature activation+0.756
d
Tokend
Feature activation+0.808
seen
Token seen
Feature activation+0.805
advertisements
Token advertisements
Feature activation+1.077
Broadcasting
Token Broadcasting
Feature activation+0.345
Corporation
Token Corporation
Feature activation+0.289
last
Token last
Feature activation+0.374
week
Token week
Feature activation+0.498
showed
Token showed
Feature activation+0.705
Australians
Token Australians
Feature activation+0.891
overwhelmingly
Token overwhelmingly
Feature activation+0.888
believe
Token believe
Feature activation+0.777
focusing
Token focusing
Feature activation+0.682
on
Token on
Feature activation+0.203
the
Token the
Feature activation+0.004
s
Tokens
Feature activation+0.006
want
Token want
Feature activation+0.199
more
Token more
Feature activation+0.373
balanced
Token balanced
Feature activation+0.635
expert
Token expert
Feature activation+0.762
groups
Token groups
Feature activation+0.907
(
Token (
Feature activation+0.412
Photo
TokenPhoto
Feature activation+0.404
:
Token:
Feature activation+0.230
h
Token h
Feature activation+0.318
.
Token.
Feature activation+0.078
on
Token on
Feature activation+0.000
why
Token why
Feature activation+0.170
work
Token work
Feature activation+0.485
ethic
Token ethic
Feature activation+0.578
beats
Token beats
Feature activation+0.825
talent
Token talent
Feature activation+0.901
.
Token.
Feature activation+0.419
Just
Token Just
Feature activation+0.411
38
Token 38
Feature activation+0.372
minutes
Token minutes
Feature activation+0.311
,
Token,
Feature activation+0.026
Writer
TokenWriter
Feature activation+0.398
Options
TokenOptions
Feature activation+0.550
r
Token r
Feature activation+0.686
Trans
TokenTrans
Feature activation+0.915
former
Tokenformer
Feature activation+1.016
r
Token r
Feature activation+1.012
Trans
TokenTrans
Feature activation+1.166
former
Tokenformer
Feature activation+1.150
::
Token ::
Feature activation+0.886
Pand
Token Pand
Feature activation+0.694
oc
Tokenoc
Feature activation+0.654

INTERVAL 0.760 - 0.886
CONTAINS 0.044%

(
Token(
Feature activation+0.211
36
Token36
Feature activation+0.309
)
Token)
Feature activation+0.188
alias
Token alias
Feature activation+0.503
Go
Token Go
Feature activation+0.828
chu
Tokenchu
Feature activation+0.805
Pe
Token Pe
Feature activation+0.840
hel
Tokenhel
Feature activation+0.696
wan
Tokenwan
Feature activation+0.848
not
Token not
Feature activation+0.679
to
Token to
Feature activation+0.157
ais
Tokenais
Feature activation+0.148
ric
Token ric
Feature activation+0.233
as
Tokenas
Feature activation+0.390
do
Token do
Feature activation+0.617
mund
Token mund
Feature activation+0.829
o
Tokeno
Feature activation+0.784
e
Token e
Feature activation+0.707
se
Token se
Feature activation+0.733
us
Tokenus
Feature activation+0.589
invest
Token invest
Feature activation+0.836
iment
Tokeniment
Feature activation+0.884
knows
Token knows
Feature activation+0.078
something
Token something
Feature activation+0.332
about
Token about
Feature activation+0.381
fifty
Token fifty
Feature activation+0.725
technological
Token technological
Feature activation+0.746
trends
Token trends
Feature activation+0.780
,
Token,
Feature activation+0.403
with
Token with
Feature activation+0.167
thirty
Token thirty
Feature activation+0.412
of
Token of
Feature activation+0.000
those
Token those
Feature activation+0.083
-
Token-
Feature activation+0.172
leader
Tokenleader
Feature activation+0.338
Chris
Token Chris
Feature activation+0.620
Stein
Token Stein
Feature activation+0.874
followed
Token followed
Feature activation+0.932
their
Token their
Feature activation+0.810
muse
Token muse
Feature activation+0.920
to
Token to
Feature activation+0.469
four
Token four
Feature activation+0.531
stellar
Token stellar
Feature activation+0.548
No
Token No
Feature activation+0.662
:
Token:
Feature activation+0.000
Marathon
Token Marathon
Feature activation+0.252
won
Token won
Feature activation+0.518
't
Token't
Feature activation+0.507
hurt
Token hurt
Feature activation+0.599
recovery
Token recovery
Feature activation+0.811
Replay
Token Replay
Feature activation+0.953
More
Token More
Feature activation+1.048
Videos
Token Videos
Feature activation+0.909
...
Token ...
Feature activation+0.736
MUST
Token MUST
Feature activation+0.588

INTERVAL 0.633 - 0.760
CONTAINS 0.101%

tered
Tokentered
Feature activation+0.000
Catholic
Token Catholic
Feature activation+0.488
church
Token church
Feature activation+0.786
a
Token a
Feature activation+0.531
few
Token few
Feature activation+0.556
minutes
Token minutes
Feature activation+0.706
from
Token from
Feature activation+0.404
downtown
Token downtown
Feature activation+0.526
.
Token.
Feature activation+0.158
There
Token There
Feature activation+0.030
,
Token,
Feature activation+0.000
work
Token work
Feature activation+0.000
can
Token can
Feature activation+0.000
add
Token add
Feature activation+0.087
additional
Token additional
Feature activation+0.110
profile
Token profile
Feature activation+0.317
cards
Token cards
Feature activation+0.647
.
Token.
Feature activation+0.125
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Image
TokenImage
Feature activation+0.062
:
Token:
Feature activation+0.000
why
Token why
Feature activation+0.222
journey
Token journey
Feature activation+0.589
man
Tokenman
Feature activation+0.478
gol
Token gol
Feature activation+0.569
fer
Tokenfer
Feature activation+0.534
Dan
Token Dan
Feature activation+0.728
Olsen
Token Olsen
Feature activation+0.594
was
Token was
Feature activation+0.266
able
Token able
Feature activation+0.150
to
Token to
Feature activation+0.000
start
Token start
Feature activation+0.000
mand
Token mand
Feature activation+0.649
ats
Tokenats
Feature activation+0.715
glob
Token glob
Feature activation+0.963
aux
Tokenaux
Feature activation+0.957
qui
Token qui
Feature activation+0.994
pe
Token pe
Feature activation+0.724
u
Tokenu
Feature activation+0.653
vent
Tokenvent
Feature activation+0.748
Ã
Token Ã
Feature activation+0.388
ª
Tokenª
Feature activation+0.082
tre
Tokentre
Feature activation+0.320
District
Token District
Feature activation+0.256
decided
Token decided
Feature activation+0.653
to
Token to
Feature activation+0.353
phase
Token phase
Feature activation+0.575
out
Token out
Feature activation+0.610
all
Token all
Feature activation+0.675
swings
Token swings
Feature activation+0.760
because
Token because
Feature activation+0.563
what
Token what
Feature activation+0.392
if
Token if
Feature activation+0.082
a
Token a
Feature activation+0.000

INTERVAL 0.506 - 0.633
CONTAINS 0.227%

has
Token has
Feature activation+0.000
revealed
Token revealed
Feature activation+0.090
a
Token a
Feature activation+0.171
world
Token world
Feature activation+0.394
more
Token more
Feature activation+0.426
intertwined
Token intertwined
Feature activation+0.544
than
Token than
Feature activation+0.498
at
Token at
Feature activation+0.115
any
Token any
Feature activation+0.171
time
Token time
Feature activation+0.141
in
Token in
Feature activation+0.000
student
Token student
Feature activation+0.000
âĢ
TokenâĢ
Feature activation+0.000
Ļ
TokenĻ
Feature activation+0.000
s
Tokens
Feature activation+0.000
work
Token work
Feature activation+0.266
grossly
Token grossly
Feature activation+0.574
inferior
Token inferior
Feature activation+0.632
,
Token,
Feature activation+0.096
and
Token and
Feature activation+0.000
tell
Token tell
Feature activation+0.060
him
Token him
Feature activation+0.154
drones
Token drones
Feature activation+0.227
to
Token to
Feature activation+0.055
spot
Token spot
Feature activation+0.083
migrant
Token migrant
Feature activation+0.367
boats
Token boats
Feature activation+0.495
trying
Token trying
Feature activation+0.565
to
Token to
Feature activation+0.155
cross
Token cross
Feature activation+0.245
the
Token the
Feature activation+0.108
Mediterranean
Token Mediterranean
Feature activation+0.000
.
Token.
Feature activation+0.000
Ŀ
TokenĿ
Feature activation+0.128
Pl
Token Pl
Feature activation+0.460
ast
Tokenast
Feature activation+0.528
ino
Tokenino
Feature activation+0.590
heard
Token heard
Feature activation+0.796
about
Token about
Feature activation+0.599
other
Token other
Feature activation+0.605
publishers
Token publishers
Feature activation+0.731
needing
Token needing
Feature activation+0.605
help
Token help
Feature activation+0.443
.
Token.
Feature activation+0.000
Canada
Token Canada
Feature activation+0.000
rules
Token rules
Feature activation+0.000
!
Token!
Feature activation+0.000
Private
Token Private
Feature activation+0.218
sector
Token sector
Feature activation+0.398
debt
Token debt
Feature activation+0.632
in
Token in
Feature activation+0.268
Canada
Token Canada
Feature activation+0.464
has
Token has
Feature activation+0.351
more
Token more
Feature activation+0.286
than
Token than
Feature activation+0.283

INTERVAL 0.380 - 0.506
CONTAINS 0.453%

she
Token she
Feature activation+0.000
will
Token will
Feature activation+0.000
join
Token join
Feature activation+0.213
a
Token a
Feature activation+0.109
live
Token live
Feature activation+0.334
debate
Token debate
Feature activation+0.486
response
Token response
Feature activation+0.673
via
Token via
Feature activation+0.653
her
Token her
Feature activation+0.543
social
Token social
Feature activation+0.694
media
Token media
Feature activation+0.642
a
Token a
Feature activation+0.186
broken
Token broken
Feature activation+0.306
turn
Token turn
Feature activation+0.643
based
Token based
Feature activation+0.643
version
Token version
Feature activation+0.772
of
Token of
Feature activation+0.475
Space
Token Space
Feature activation+0.514
Invaders
Token Invaders
Feature activation+0.544
on
Token on
Feature activation+0.326
a
Token a
Feature activation+0.118
check
Token check
Feature activation+0.089
Obama
TokenObama
Feature activation+0.256
Care
TokenCare
Feature activation+0.353
Lite
Token Lite
Feature activation+0.492
',
Token',
Feature activation+0.498
he
Token he
Feature activation+0.521
added
Token added
Feature activation+0.494
:
Token:
Feature activation+0.277
'
Token '
Feature activation+0.267
It
TokenIt
Feature activation+0.136
is
Token is
Feature activation+0.000
ObamaCare
Token ObamaCare
Feature activation+0.002
plan
Token plan
Feature activation+0.253
that
Token that
Feature activation+0.183
will
Token will
Feature activation+0.086
actually
Token actually
Feature activation+0.154
create
Token create
Feature activation+0.211
jobs
Token jobs
Feature activation+0.399
?'
Token?'
Feature activation+0.367
(
Token (
Feature activation+0.000
Fred
TokenFred
Feature activation+0.000
Chart
Token Chart
Feature activation+0.000
rand
Tokenrand
Feature activation+0.112
astical
Tokenastical
Feature activation+0.000
images
Token images
Feature activation+0.000
about
Token about
Feature activation+0.189
the
Token the
Feature activation+0.058
Ob
Token Ob
Feature activation+0.233
amas
Tokenamas
Feature activation+0.411
and
Token and
Feature activation+0.075
shows
Token shows
Feature activation+0.264
them
Token them
Feature activation+0.416
for
Token for
Feature activation+0.157
the
Token the
Feature activation+0.015

INTERVAL 0.253 - 0.380
CONTAINS 0.811%

appy
Tokenappy
Feature activation+0.268
and
Token and
Feature activation+0.084
battles
Token battles
Feature activation+0.293
are
Token are
Feature activation+0.304
fast
Token fast
Feature activation+0.462
-
Token-
Feature activation+0.325
paced
Tokenpaced
Feature activation+0.342
.
Token.
Feature activation+0.042
You
Token You
Feature activation+0.000
can
Token can
Feature activation+0.000
buy
Token buy
Feature activation+0.000
always
Token always
Feature activation+0.000
finding
Token finding
Feature activation+0.235
the
Token the
Feature activation+0.140
positives
Token positives
Feature activation+0.268
in
Token in
Feature activation+0.091
life
Token life
Feature activation+0.265
.
Token.
Feature activation+0.112
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
âĢ
TokenâĢ
Feature activation+0.000
ľ
Tokenľ
Feature activation+0.000
victory
Token victory
Feature activation+0.482
?
Token?
Feature activation+0.328
Do
Token Do
Feature activation+0.330
they
Token they
Feature activation+0.255
choose
Token choose
Feature activation+0.280
new
Token new
Feature activation+0.358
leadership
Token leadership
Feature activation+0.280
in
Token in
Feature activation+0.000
the
Token the
Feature activation+0.000
hopes
Token hopes
Feature activation+0.000
of
Token of
Feature activation+0.000
region
Token region
Feature activation+0.000
."
Token."
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Sp
TokenSp
Feature activation+0.008
yer
Tokenyer
Feature activation+0.265
concluded
Token concluded
Feature activation+0.112
that
Token that
Feature activation+0.030
one
Token one
Feature activation+0.220
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
to
Token to
Feature activation+0.000
an
Token an
Feature activation+0.000
object
Token object
Feature activation+0.027
of
Token of
Feature activation+0.000
Wall
Token Wall
Feature activation+0.196
Street
Token Street
Feature activation+0.363
's
Token's
Feature activation+0.113
attention
Token attention
Feature activation+0.000
.
Token.
Feature activation+0.000
C
Token C
Feature activation+0.000
ME
TokenME
Feature activation+0.000

INTERVAL 0.127 - 0.253
CONTAINS 1.338%

Te
Token Te
Feature activation+0.329
I
Token I
Feature activation+0.342
ka
Tokenka
Feature activation+0.328
(
Token (
Feature activation+0.000
lower
Tokenlower
Feature activation+0.000
North
Token North
Feature activation+0.187
Island
Token Island
Feature activation+0.264
).
Token).
Feature activation+0.000
It
Token It
Feature activation+0.004
was
Token was
Feature activation+0.000
held
Token held
Feature activation+0.000
regulation
Token regulation
Feature activation+0.000
from
Token from
Feature activation+0.000
which
Token which
Feature activation+0.000
the
Token the
Feature activation+0.000
early
Token early
Feature activation+0.000
Internet
Token Internet
Feature activation+0.177
benefited
Token benefited
Feature activation+0.257
just
Token just
Feature activation+0.365
a
Token a
Feature activation+0.160
few
Token few
Feature activation+0.095
decades
Token decades
Feature activation+0.209
alo
Tokenalo
Feature activation+0.082
ed
Tokened
Feature activation+0.134
ranks
Token ranks
Feature activation+0.198
of
Token of
Feature activation+0.105
the
Token the
Feature activation+0.083
scientific
Token scientific
Feature activation+0.135
peer
Token peer
Feature activation+0.292
-
Token-
Feature activation+0.189
reviewed
Tokenreviewed
Feature activation+0.311
literature
Token literature
Feature activation+0.252
âĢĵ
Token âĢĵ
Feature activation+0.059
".
Token".
Feature activation+0.021
The
Token The
Feature activation+0.000
energy
Token energy
Feature activation+0.084
monopol
Token monopol
Feature activation+0.251
ies
Tokenies
Feature activation+0.037
how
Token how
Feature activation+0.160
led
Tokenled
Feature activation+0.320
that
Token that
Feature activation+0.000
investment
Token investment
Feature activation+0.000
would
Token would
Feature activation+0.000
collapse
Token collapse
Feature activation+0.000
afternoon
Token afternoon
Feature activation+0.000
with
Token with
Feature activation+0.000
five
Token five
Feature activation+0.000
fellow
Token fellow
Feature activation+0.167
migrants
Token migrants
Feature activation+0.387
from
Token from
Feature activation+0.171
Bangladesh
Token Bangladesh
Feature activation+0.269
,
Token,
Feature activation+0.006
India
Token India
Feature activation+0.080
and
Token and
Feature activation+0.000
Nepal
Token Nepal
Feature activation+0.000

INTERVAL 0.000 - 0.127
CONTAINS 97.003%

experiences
Token experiences
Feature activation+0.000
.
Token.
Feature activation+0.000
For
Token For
Feature activation+0.000
instance
Token instance
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
traveler
Token traveler
Feature activation+0.000
planning
Token planning
Feature activation+0.000
a
Token a
Feature activation+0.000
trip
Token trip
Feature activation+0.000
might
Token might
Feature activation+0.000
West
Token West
Feature activation+0.000
Germany
Token Germany
Feature activation+0.000
when
Token when
Feature activation+0.000
he
Token he
Feature activation+0.000
set
Token set
Feature activation+0.000
the
Token the
Feature activation+0.000
previous
Token previous
Feature activation+0.000
benchmark
Token benchmark
Feature activation+0.000
in
Token in
Feature activation+0.000
1972
Token 1972
Feature activation+0.000
and
Token and
Feature activation+0.000
's
Token's
Feature activation+0.000
fate
Token fate
Feature activation+0.000
.
Token.
Feature activation+0.000
The
Token The
Feature activation+0.000
formal
Token formal
Feature activation+0.000
minutes
Token minutes
Feature activation+0.000
reveal
Token reveal
Feature activation+0.000
that
Token that
Feature activation+0.000
her
Token her
Feature activation+0.000
application
Token application
Feature activation+0.000
was
Token was
Feature activation+0.000
was
Token was
Feature activation+0.000
born
Token born
Feature activation+0.000
outside
Token outside
Feature activation+0.000
the
Token the
Feature activation+0.000
United
Token United
Feature activation+0.000
States
Token States
Feature activation+0.000
and
Token and
Feature activation+0.000
is
Token is
Feature activation+0.000
thus
Token thus
Feature activation+0.000
Constitution
Token Constitution
Feature activation+0.000
ally
Tokenally
Feature activation+0.000
7
Token 7
Feature activation+0.000
Prediction
Token Prediction
Feature activation+0.000
7
Token 7
Feature activation+0.000
.
Token.
Feature activation+0.000
1
Token1
Feature activation+0.000
The
Token The
Feature activation+0.000
Boston
Token Boston
Feature activation+0.000
Bruins
Token Bruins
Feature activation+0.000
problem
Token problem
Feature activation+0.000
In
Token In
Feature activation+0.000
the
Token the
Feature activation+0.000

BOTTOM ACTIVATIONS
MIN = 0.000

former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
Organisation
Token Organisation
Feature activation+0.000
,
Token,
Feature activation+0.000
which
Token which
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
Organisation
Token Organisation
Feature activation+0.000
,
Token,
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
Organisation
Token Organisation
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.079
Saturday
Token Saturday
Feature activation+0.464
.
Token.
Feature activation+0.131
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.014
headed
Token headed
Feature activation+0.196
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.079
Saturday
Token Saturday
Feature activation+0.464
.
Token.
Feature activation+0.131
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.014
said
Token said
Feature activation+0.079
Saturday
Token Saturday
Feature activation+0.464
.
Token.
Feature activation+0.131
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.014
headed
Token headed
Feature activation+0.196
by
Token by
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.464
.
Token.
Feature activation+0.131
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.014
headed
Token headed
Feature activation+0.196
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
committee
Token committee
Feature activation+0.014
headed
Token headed
Feature activation+0.196
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
headed
Token headed
Feature activation+0.196
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.014
headed
Token headed
Feature activation+0.196
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.014
headed
Token headed
Feature activation+0.196
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
.
Token.
Feature activation+0.131
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.014
headed
Token headed
Feature activation+0.196
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.014
headed
Token headed
Feature activation+0.196
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000

Top feature 5 in H1.3: (feature 23223)

TOP ACTIVATIONS
MAX = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000

Top DFA by src position
MAX = 0.556

atre
Tokenatre
Feature activation+0.030
Top resid features:
,
Token,
Feature activation+0.005
Top resid features:
a
Token a
Feature activation+0.074
Top resid features:
former
Token former
Feature activation+0.028
Top resid features:
boss
Token boss
Feature activation+0.091
Top resid features:
of
Token of
Feature activation+0.267
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
&
Token &
Feature activation+0.000
Top resid features:
Development
Token Development
Feature activation+0.000
Top resid features:
A
Token A
Feature activation-0.013
Top resid features:
atre
Tokenatre
Feature activation+0.027
Top resid features:
,
Token,
Feature activation-0.004
Top resid features:
a
Token a
Feature activation+0.077
Top resid features:
former
Token former
Feature activation+0.036
Top resid features:
boss
Token boss
Feature activation+0.309
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
&
Token &
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.026
Top resid features:
.
Token.
Feature activation+0.045
Top resid features:
A
Token A
Feature activation+0.138
Top resid features:
atre
Tokenatre
Feature activation+0.055
Top resid features:
,
Token,
Feature activation+0.056
Top resid features:
a
Token a
Feature activation+0.266
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.215
Top resid features:
rik
Tokenrik
Feature activation+0.022
Top resid features:
ar
Tokenar
Feature activation+0.046
Top resid features:
said
Token said
Feature activation+0.051
Top resid features:
Saturday
Token Saturday
Feature activation-0.017
Top resid features:
.
Token.
Feature activation+0.049
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.525
Top resid features:
rik
Tokenrik
Feature activation+0.051
Top resid features:
ar
Tokenar
Feature activation+0.106
Top resid features:
said
Token said
Feature activation+0.271
Top resid features:
Saturday
Token Saturday
Feature activation-0.049
Top resid features:
.
Token.
Feature activation+0.203
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.556
Top resid features:
rik
Tokenrik
Feature activation+0.006
Top resid features:
ar
Tokenar
Feature activation+0.110
Top resid features:
said
Token said
Feature activation+0.311
Top resid features:
Saturday
Token Saturday
Feature activation-0.081
Top resid features:
.
Token.
Feature activation+0.314
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.449
Top resid features:
rik
Tokenrik
Feature activation+0.056
Top resid features:
ar
Tokenar
Feature activation+0.093
Top resid features:
said
Token said
Feature activation+0.225
Top resid features:
Saturday
Token Saturday
Feature activation-0.042
Top resid features:
.
Token.
Feature activation+0.168
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.377
Top resid features:
rik
Tokenrik
Feature activation+0.028
Top resid features:
ar
Tokenar
Feature activation+0.096
Top resid features:
said
Token said
Feature activation+0.190
Top resid features:
Saturday
Token Saturday
Feature activation-0.042
Top resid features:
.
Token.
Feature activation+0.133
Top resid features:
defence
Token defence
Feature activation+0.021
Top resid features:
ministry
Token ministry
Feature activation+0.061
Top resid features:
committee
Token committee
Feature activation+0.076
Top resid features:
headed
Token headed
Feature activation+0.048
Top resid features:
by
Token by
Feature activation+0.070
Top resid features:
V
Token V
Feature activation+0.243
Top resid features:
.
Token.
Feature activation+0.074
Top resid features:
K
TokenK
Feature activation+0.194
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.270
Top resid features:
rik
Tokenrik
Feature activation+0.021
Top resid features:
ar
Tokenar
Feature activation+0.050
Top resid features:
said
Token said
Feature activation+0.077
Top resid features:
Saturday
Token Saturday
Feature activation-0.034
Top resid features:
.
Token.
Feature activation+0.106
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.281
Top resid features:
rik
Tokenrik
Feature activation+0.024
Top resid features:
ar
Tokenar
Feature activation+0.053
Top resid features:
said
Token said
Feature activation+0.103
Top resid features:
Saturday
Token Saturday
Feature activation-0.023
Top resid features:
.
Token.
Feature activation+0.097
Top resid features:
defence
Token defence
Feature activation+0.031
Top resid features:
ministry
Token ministry
Feature activation+0.079
Top resid features:
committee
Token committee
Feature activation+0.095
Top resid features:
headed
Token headed
Feature activation+0.024
Top resid features:
by
Token by
Feature activation+0.114
Top resid features:
V
Token V
Feature activation+0.285
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.066
Top resid features:
V
Token V
Feature activation+0.080
Top resid features:
.
Token.
Feature activation+0.051
Top resid features:
K
TokenK
Feature activation+0.051
Top resid features:
.
Token.
Feature activation+0.055
Top resid features:
A
Token A
Feature activation+0.316
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.253
Top resid features:
rik
Tokenrik
Feature activation+0.022
Top resid features:
ar
Tokenar
Feature activation+0.042
Top resid features:
said
Token said
Feature activation+0.068
Top resid features:
Saturday
Token Saturday
Feature activation-0.032
Top resid features:
.
Token.
Feature activation+0.093
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.265
Top resid features:
rik
Tokenrik
Feature activation+0.020
Top resid features:
ar
Tokenar
Feature activation+0.020
Top resid features:
said
Token said
Feature activation+0.059
Top resid features:
Saturday
Token Saturday
Feature activation-0.017
Top resid features:
.
Token.
Feature activation+0.073
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.252
Top resid features:
rik
Tokenrik
Feature activation+0.026
Top resid features:
ar
Tokenar
Feature activation+0.043
Top resid features:
said
Token said
Feature activation+0.065
Top resid features:
Saturday
Token Saturday
Feature activation-0.020
Top resid features:
.
Token.
Feature activation+0.076
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.296
Top resid features:
rik
Tokenrik
Feature activation+0.024
Top resid features:
ar
Tokenar
Feature activation+0.067
Top resid features:
said
Token said
Feature activation+0.111
Top resid features:
Saturday
Token Saturday
Feature activation-0.024
Top resid features:
.
Token.
Feature activation+0.120
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.276
Top resid features:
rik
Tokenrik
Feature activation+0.038
Top resid features:
ar
Tokenar
Feature activation+0.050
Top resid features:
said
Token said
Feature activation+0.115
Top resid features:
Saturday
Token Saturday
Feature activation-0.025
Top resid features:
.
Token.
Feature activation+0.097
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.328
Top resid features:
rik
Tokenrik
Feature activation+0.033
Top resid features:
ar
Tokenar
Feature activation+0.058
Top resid features:
said
Token said
Feature activation+0.165
Top resid features:
Saturday
Token Saturday
Feature activation-0.040
Top resid features:
.
Token.
Feature activation+0.132
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.312
Top resid features:
rik
Tokenrik
Feature activation+0.057
Top resid features:
ar
Tokenar
Feature activation+0.016
Top resid features:
said
Token said
Feature activation+0.140
Top resid features:
Saturday
Token Saturday
Feature activation-0.035
Top resid features:
.
Token.
Feature activation+0.140
Top resid features:

Decoder Weights Distribution

Head 0: 0.10

Head 1: 0.09

Head 2: 0.08

Head 3: 0.10

Head 4: 0.07

Head 5: 0.08

Head 6: 0.10

Head 7: 0.09

Head 8: 0.06

Head 9: 0.06

Head 10: 0.07

Head 11: 0.09

Positive logits

incorpor1.63

loved1.60

adena1.45

reads1.43

enhagen1.42

LOVE1.39

Reyes1.39

certific1.38

accus1.38

innocence1.35

morals1.34

polite1.33

avorite1.33

experien1.28

xon1.27

pronouns1.26

loves1.26

referen1.26

narrator1.25

Gohan1.24

Negative logits

aspx-1.54

min-1.48

nit-1.44

mini-1.44

Bund-1.43

rh-1.39

bull-1.37

prem-1.37

ember-1.37

Bangladesh-1.34

Kraken-1.33

course-1.32

TPS-1.30

python-1.30

ummies-1.29

lam-1.29

zar-1.26

rovers-1.26

balance-1.25

China-1.24

INTERVAL 0.000 - 0.000
CONTAINS 100.000%

such
Token such
Feature activation+0.000
detail
Token detail
Feature activation+0.000
and
Token and
Feature activation+0.000
were
Token were
Feature activation+0.000
likely
Token likely
Feature activation+0.000
telling
Token telling
Feature activation+0.000
the
Token the
Feature activation+0.000
truth
Token truth
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
to
Token to
Feature activation+0.000
the
Token the
Feature activation+0.000
S
Token S
Feature activation+0.000
AST
TokenAST
Feature activation+0.000
IND
TokenIND
Feature activation+0.000
media
Token media
Feature activation+0.000
center
Token center
Feature activation+0.000
this
Token this
Feature activation+0.000
morning
Token morning
Feature activation+0.000
--
Token --
Feature activation+0.000
and
Token and
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Through
TokenThrough
Feature activation+0.000
the
Token the
Feature activation+0.000
end
Token end
Feature activation+0.000
of
Token of
Feature activation+0.000
July
Token July
Feature activation+0.000
,
Token,
Feature activation+0.000
interest
Token interest
Feature activation+0.000
groups
Token groups
Feature activation+0.000
had
Token had
Feature activation+0.000
than
Token than
Feature activation+0.000
her
Token her
Feature activation+0.000
debut
Token debut
Feature activation+0.000
in
Token in
Feature activation+0.000
New
Token New
Feature activation+0.000
York
Token York
Feature activation+0.000
last
Token last
Feature activation+0.000
year
Token year
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Index
Token Index
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
East
TokenEast
Feature activation+0.000
Antarctic
Token Antarctic
Feature activation+0.000
Ice
Token Ice
Feature activation+0.000
Sheet
Token Sheet
Feature activation+0.000
More
Token More
Feature activation+0.000
V
Token V
Feature activation+0.000
ulnerable
Tokenulnerable
Feature activation+0.000
to
Token to
Feature activation+0.000

BOTTOM ACTIVATIONS
MIN = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000

Top feature 6 in H1.3: (feature 9206)

TOP ACTIVATIONS
MAX = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000

Top DFA by src position
MAX = 0.358

atre
Tokenatre
Feature activation+0.010
Top resid features:
,
Token,
Feature activation-0.027
Top resid features:
a
Token a
Feature activation+0.011
Top resid features:
former
Token former
Feature activation-0.005
Top resid features:
boss
Token boss
Feature activation-0.023
Top resid features:
of
Token of
Feature activation+0.176
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
&
Token &
Feature activation+0.000
Top resid features:
Development
Token Development
Feature activation+0.000
Top resid features:
A
Token A
Feature activation-0.031
Top resid features:
atre
Tokenatre
Feature activation+0.010
Top resid features:
,
Token,
Feature activation-0.033
Top resid features:
a
Token a
Feature activation+0.021
Top resid features:
former
Token former
Feature activation-0.028
Top resid features:
boss
Token boss
Feature activation+0.221
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
&
Token &
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.031
Top resid features:
.
Token.
Feature activation+0.011
Top resid features:
A
Token A
Feature activation+0.076
Top resid features:
atre
Tokenatre
Feature activation+0.017
Top resid features:
,
Token,
Feature activation-0.010
Top resid features:
a
Token a
Feature activation+0.148
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.004
Top resid features:
A
Token A
Feature activation-0.028
Top resid features:
atre
Tokenatre
Feature activation+0.010
Top resid features:
,
Token,
Feature activation-0.041
Top resid features:
a
Token a
Feature activation+0.047
Top resid features:
former
Token former
Feature activation+0.160
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.166
Top resid features:
rik
Tokenrik
Feature activation+0.034
Top resid features:
ar
Tokenar
Feature activation+0.067
Top resid features:
said
Token said
Feature activation+0.263
Top resid features:
Saturday
Token Saturday
Feature activation-0.005
Top resid features:
.
Token.
Feature activation+0.079
Top resid features:
Ċ
TokenĊ
Feature activation+0.173
Top resid features:
Ċ
TokenĊ
Feature activation+0.000
Top resid features:
The
TokenThe
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.138
Top resid features:
rik
Tokenrik
Feature activation+0.011
Top resid features:
ar
Tokenar
Feature activation+0.069
Top resid features:
said
Token said
Feature activation+0.338
Top resid features:
Saturday
Token Saturday
Feature activation-0.001
Top resid features:
.
Token.
Feature activation+0.154
Top resid features:
Ċ
TokenĊ
Feature activation+0.000
Top resid features:
Ċ
TokenĊ
Feature activation+0.000
Top resid features:
The
TokenThe
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.140
Top resid features:
rik
Tokenrik
Feature activation+0.039
Top resid features:
ar
Tokenar
Feature activation+0.060
Top resid features:
said
Token said
Feature activation+0.217
Top resid features:
Saturday
Token Saturday
Feature activation-0.001
Top resid features:
.
Token.
Feature activation+0.061
Top resid features:
Ċ
TokenĊ
Feature activation+0.137
Top resid features:
Ċ
TokenĊ
Feature activation+0.139
Top resid features:
The
TokenThe
Feature activation+0.000
Top resid features:
said
Token said
Feature activation+0.155
Top resid features:
Saturday
Token Saturday
Feature activation-0.025
Top resid features:
.
Token.
Feature activation+0.058
Top resid features:
Ċ
TokenĊ
Feature activation+0.046
Top resid features:
Ċ
TokenĊ
Feature activation+0.031
Top resid features:
The
TokenThe
Feature activation+0.275
Top resid features:
defence
Token defence
Feature activation+0.000
Top resid features:
ministry
Token ministry
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation+0.000
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation+0.054
Top resid features:
headed
Token headed
Feature activation-0.005
Top resid features:
by
Token by
Feature activation+0.049
Top resid features:
V
Token V
Feature activation+0.173
Top resid features:
.
Token.
Feature activation+0.001
Top resid features:
K
TokenK
Feature activation+0.247
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
defence
Token defence
Feature activation-0.004
Top resid features:
ministry
Token ministry
Feature activation+0.021
Top resid features:
committee
Token committee
Feature activation+0.064
Top resid features:
headed
Token headed
Feature activation-0.038
Top resid features:
by
Token by
Feature activation+0.049
Top resid features:
V
Token V
Feature activation+0.157
Top resid features:
.
Token.
Feature activation+0.074
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
The
TokenThe
Feature activation+0.050
Top resid features:
defence
Token defence
Feature activation+0.006
Top resid features:
ministry
Token ministry
Feature activation+0.045
Top resid features:
committee
Token committee
Feature activation+0.059
Top resid features:
headed
Token headed
Feature activation-0.184
Top resid features:
by
Token by
Feature activation+0.165
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
defence
Token defence
Feature activation-0.010
Top resid features:
ministry
Token ministry
Feature activation+0.008
Top resid features:
committee
Token committee
Feature activation+0.055
Top resid features:
headed
Token headed
Feature activation-0.047
Top resid features:
by
Token by
Feature activation+0.049
Top resid features:
V
Token V
Feature activation+0.358
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.030
Top resid features:
V
Token V
Feature activation+0.060
Top resid features:
.
Token.
Feature activation+0.012
Top resid features:
K
TokenK
Feature activation+0.064
Top resid features:
.
Token.
Feature activation-0.006
Top resid features:
A
Token A
Feature activation+0.205
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.119
Top resid features:
rik
Tokenrik
Feature activation+0.012
Top resid features:
ar
Tokenar
Feature activation+0.034
Top resid features:
said
Token said
Feature activation+0.052
Top resid features:
Saturday
Token Saturday
Feature activation-0.005
Top resid features:
.
Token.
Feature activation+0.053
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.157
Top resid features:
rik
Tokenrik
Feature activation+0.028
Top resid features:
ar
Tokenar
Feature activation+0.044
Top resid features:
said
Token said
Feature activation+0.045
Top resid features:
Saturday
Token Saturday
Feature activation-0.007
Top resid features:
.
Token.
Feature activation+0.032
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.132
Top resid features:
rik
Tokenrik
Feature activation+0.015
Top resid features:
ar
Tokenar
Feature activation+0.028
Top resid features:
said
Token said
Feature activation+0.050
Top resid features:
Saturday
Token Saturday
Feature activation-0.007
Top resid features:
.
Token.
Feature activation+0.038
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.132
Top resid features:
rik
Tokenrik
Feature activation+0.032
Top resid features:
ar
Tokenar
Feature activation+0.041
Top resid features:
said
Token said
Feature activation+0.093
Top resid features:
Saturday
Token Saturday
Feature activation-0.011
Top resid features:
.
Token.
Feature activation+0.045
Top resid features:
Ċ
TokenĊ
Feature activation+0.041
Top resid features:
Ċ
TokenĊ
Feature activation+0.033
Top resid features:
The
TokenThe
Feature activation+0.075
Top resid features:
defence
Token defence
Feature activation+0.024
Top resid features:
ministry
Token ministry
Feature activation-0.008
Top resid features:
committee
Token committee
Feature activation+0.200
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.130
Top resid features:
rik
Tokenrik
Feature activation+0.028
Top resid features:
ar
Tokenar
Feature activation+0.060
Top resid features:
said
Token said
Feature activation+0.119
Top resid features:
Saturday
Token Saturday
Feature activation-0.029
Top resid features:
.
Token.
Feature activation+0.045
Top resid features:
Saturday
Token Saturday
Feature activation-0.026
Top resid features:
.
Token.
Feature activation+0.055
Top resid features:
Ċ
TokenĊ
Feature activation+0.049
Top resid features:
Ċ
TokenĊ
Feature activation+0.040
Top resid features:
The
TokenThe
Feature activation+0.094
Top resid features:
defence
Token defence
Feature activation+0.127
Top resid features:
ministry
Token ministry
Feature activation+0.106
Top resid features:
committee
Token committee
Feature activation+0.000
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:

Decoder Weights Distribution

Head 0: 0.08

Head 1: 0.08

Head 2: 0.07

Head 3: 0.10

Head 4: 0.08

Head 5: 0.07

Head 6: 0.09

Head 7: 0.08

Head 8: 0.07

Head 9: 0.09

Head 10: 0.08

Head 11: 0.10

Positive logits

BuyableInstoreAndOnline1.71

fav1.42

��1.38

Carnage1.37

Newsletter1.33

brill1.31

neighbourhood1.27

unrecogn1.25

api1.24

Ont1.23

GTA1.22

fertilizer1.21

Trafford1.21

mathemat1.21

upstream1.19

explor1.19

awa1.19

Els1.18

ecosystem1.18

dads1.17

Negative logits

rule-1.74

BILITY-1.51

odan-1.50

ibles-1.48

perjury-1.48

waivers-1.47

abo-1.46

srfAttach-1.44

arag-1.43

ayson-1.43

oath-1.42

ISSION-1.42

igham-1.40

shall-1.40

eton-1.39

anamo-1.39

arently-1.37

ittee-1.36

abbage-1.35

astical-1.33

INTERVAL 0.000 - 0.000
CONTAINS 100.000%

05
Token05
Feature activation+0.000
-
Token-
Feature activation+0.000
22
Token22
Feature activation+0.000
%
Token%
Feature activation+0.000
2
Token2
Feature activation+0.000
B
TokenB
Feature activation+0.000
photo
Tokenphoto
Feature activation+0.000
%
Token%
Feature activation+0.000
2
Token2
Feature activation+0.000
B
TokenB
Feature activation+0.000
crop
Tokencrop
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
S
TokenS
Feature activation+0.000
owell
Tokenowell
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
There
TokenThere
Feature activation+0.000
are
Token are
Feature activation+0.000
a
Token a
Feature activation+0.000
lot
Token lot
Feature activation+0.000
of
Token of
Feature activation+0.000
recipes
Token recipes
Feature activation+0.000
volcan
Token volcan
Feature activation+0.000
oes
Tokenoes
Feature activation+0.000
,
Token,
Feature activation+0.000
there
Token there
Feature activation+0.000
are
Token are
Feature activation+0.000
no
Token no
Feature activation+0.000
cameras
Token cameras
Feature activation+0.000
or
Token or
Feature activation+0.000
monitoring
Token monitoring
Feature activation+0.000
stations
Token stations
Feature activation+0.000
on
Token on
Feature activation+0.000
18
Token 18
Feature activation+0.000
%
Token%
Feature activation+0.000
11
Token 11
Feature activation+0.000
%
Token%
Feature activation+0.000
Romney
Token Romney
Feature activation+0.000
31
Token 31
Feature activation+0.000
%
Token%
Feature activation+0.000
40
Token 40
Feature activation+0.000
%
Token%
Feature activation+0.000
Santorum
Token Santorum
Feature activation+0.000
38
Token 38
Feature activation+0.000
aldo
Tokenaldo
Feature activation+0.000
Alonso
Token Alonso
Feature activation+0.000
(
Token (
Feature activation+0.000
ca
Tokenca
Feature activation+0.000
ution
Tokenution
Feature activation+0.000
;
Token;
Feature activation+0.000
dissent
Token dissent
Feature activation+0.000
)
Token)
Feature activation+0.000
4
Token 4
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000

BOTTOM ACTIVATIONS
MIN = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000

Top feature 7 in H1.3: (feature 8195)

TOP ACTIVATIONS
MAX = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000

Top DFA by src position
MAX = 0.388

<|endoftext|>
Token<|endoftext|>
Feature activation+0.214
Top resid features:
rik
Tokenrik
Feature activation+0.022
Top resid features:
ar
Tokenar
Feature activation+0.040
Top resid features:
said
Token said
Feature activation+0.037
Top resid features:
Saturday
Token Saturday
Feature activation+0.006
Top resid features:
.
Token.
Feature activation+0.040
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.182
Top resid features:
rik
Tokenrik
Feature activation+0.022
Top resid features:
ar
Tokenar
Feature activation+0.042
Top resid features:
said
Token said
Feature activation+0.041
Top resid features:
Saturday
Token Saturday
Feature activation+0.008
Top resid features:
.
Token.
Feature activation+0.041
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.194
Top resid features:
rik
Tokenrik
Feature activation+0.028
Top resid features:
ar
Tokenar
Feature activation+0.040
Top resid features:
said
Token said
Feature activation+0.047
Top resid features:
Saturday
Token Saturday
Feature activation+0.007
Top resid features:
.
Token.
Feature activation+0.053
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.176
Top resid features:
rik
Tokenrik
Feature activation+0.025
Top resid features:
ar
Tokenar
Feature activation+0.042
Top resid features:
said
Token said
Feature activation+0.045
Top resid features:
Saturday
Token Saturday
Feature activation+0.004
Top resid features:
.
Token.
Feature activation+0.047
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.388
Top resid features:
rik
Tokenrik
Feature activation+0.065
Top resid features:
ar
Tokenar
Feature activation+0.107
Top resid features:
said
Token said
Feature activation+0.204
Top resid features:
Saturday
Token Saturday
Feature activation+0.037
Top resid features:
.
Token.
Feature activation+0.160
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.385
Top resid features:
rik
Tokenrik
Feature activation+0.070
Top resid features:
ar
Tokenar
Feature activation+0.119
Top resid features:
said
Token said
Feature activation+0.262
Top resid features:
Saturday
Token Saturday
Feature activation+0.051
Top resid features:
.
Token.
Feature activation+0.251
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.329
Top resid features:
rik
Tokenrik
Feature activation+0.059
Top resid features:
ar
Tokenar
Feature activation+0.097
Top resid features:
said
Token said
Feature activation+0.171
Top resid features:
Saturday
Token Saturday
Feature activation+0.029
Top resid features:
.
Token.
Feature activation+0.135
Top resid features:
said
Token said
Feature activation+0.150
Top resid features:
Saturday
Token Saturday
Feature activation+0.020
Top resid features:
.
Token.
Feature activation+0.130
Top resid features:
Ċ
TokenĊ
Feature activation+0.051
Top resid features:
Ċ
TokenĊ
Feature activation+0.042
Top resid features:
The
TokenThe
Feature activation+0.359
Top resid features:
defence
Token defence
Feature activation+0.000
Top resid features:
ministry
Token ministry
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation+0.000
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation+0.128
Top resid features:
headed
Token headed
Feature activation+0.048
Top resid features:
by
Token by
Feature activation+0.043
Top resid features:
V
Token V
Feature activation+0.125
Top resid features:
.
Token.
Feature activation+0.096
Top resid features:
K
TokenK
Feature activation+0.199
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.213
Top resid features:
rik
Tokenrik
Feature activation+0.038
Top resid features:
ar
Tokenar
Feature activation+0.050
Top resid features:
said
Token said
Feature activation+0.072
Top resid features:
Saturday
Token Saturday
Feature activation+0.012
Top resid features:
.
Token.
Feature activation+0.096
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.223
Top resid features:
rik
Tokenrik
Feature activation+0.036
Top resid features:
ar
Tokenar
Feature activation+0.064
Top resid features:
said
Token said
Feature activation+0.082
Top resid features:
Saturday
Token Saturday
Feature activation+0.017
Top resid features:
.
Token.
Feature activation+0.085
Top resid features:
defence
Token defence
Feature activation+0.049
Top resid features:
ministry
Token ministry
Feature activation+0.105
Top resid features:
committee
Token committee
Feature activation+0.160
Top resid features:
headed
Token headed
Feature activation+0.014
Top resid features:
by
Token by
Feature activation+0.098
Top resid features:
V
Token V
Feature activation+0.269
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.045
Top resid features:
V
Token V
Feature activation+0.077
Top resid features:
.
Token.
Feature activation+0.072
Top resid features:
K
TokenK
Feature activation+0.022
Top resid features:
.
Token.
Feature activation+0.078
Top resid features:
A
Token A
Feature activation+0.215
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.201
Top resid features:
rik
Tokenrik
Feature activation+0.036
Top resid features:
ar
Tokenar
Feature activation+0.042
Top resid features:
said
Token said
Feature activation+0.062
Top resid features:
Saturday
Token Saturday
Feature activation+0.009
Top resid features:
.
Token.
Feature activation+0.086
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.211
Top resid features:
rik
Tokenrik
Feature activation+0.037
Top resid features:
ar
Tokenar
Feature activation+0.038
Top resid features:
said
Token said
Feature activation+0.054
Top resid features:
Saturday
Token Saturday
Feature activation+0.011
Top resid features:
.
Token.
Feature activation+0.061
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.201
Top resid features:
rik
Tokenrik
Feature activation+0.028
Top resid features:
ar
Tokenar
Feature activation+0.043
Top resid features:
said
Token said
Feature activation+0.048
Top resid features:
Saturday
Token Saturday
Feature activation+0.006
Top resid features:
.
Token.
Feature activation+0.065
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.236
Top resid features:
rik
Tokenrik
Feature activation+0.045
Top resid features:
ar
Tokenar
Feature activation+0.062
Top resid features:
said
Token said
Feature activation+0.105
Top resid features:
Saturday
Token Saturday
Feature activation+0.019
Top resid features:
.
Token.
Feature activation+0.101
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.212
Top resid features:
rik
Tokenrik
Feature activation+0.043
Top resid features:
ar
Tokenar
Feature activation+0.067
Top resid features:
said
Token said
Feature activation+0.121
Top resid features:
Saturday
Token Saturday
Feature activation+0.020
Top resid features:
.
Token.
Feature activation+0.088
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.260
Top resid features:
rik
Tokenrik
Feature activation+0.050
Top resid features:
ar
Tokenar
Feature activation+0.076
Top resid features:
said
Token said
Feature activation+0.147
Top resid features:
Saturday
Token Saturday
Feature activation-0.001
Top resid features:
.
Token.
Feature activation+0.112
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.251
Top resid features:
rik
Tokenrik
Feature activation+0.058
Top resid features:
ar
Tokenar
Feature activation+0.062
Top resid features:
said
Token said
Feature activation+0.135
Top resid features:
Saturday
Token Saturday
Feature activation+0.017
Top resid features:
.
Token.
Feature activation+0.119
Top resid features:

Decoder Weights Distribution

Head 0: 0.09

Head 1: 0.08

Head 2: 0.07

Head 3: 0.10

Head 4: 0.08

Head 5: 0.09

Head 6: 0.07

Head 7: 0.09

Head 8: 0.09

Head 9: 0.08

Head 10: 0.07

Head 11: 0.08

Positive logits

charms1.51

scept1.44

emot1.42

lazy1.41

welcome1.36

gestures1.34

delicate1.31

quick1.30

weary1.30

moaning1.29

mindless1.27

patriotic1.27

ginger1.26

gems1.25

mindful1.24

��1.24

ignorant1.22

hats1.22

cheerful1.21

hysterical1.21

Negative logits

oya-1.59

arthed-1.58

SPONSORED-1.56

hai-1.47

arten-1.46

GOODMAN-1.45

istics-1.45

identified-1.44

fleet-1.41

opsis-1.41

ederation-1.39

foreseen-1.39

tower-1.37

ensional-1.36

eki-1.36

OIL-1.36

been-1.36

nance-1.35

DAQ-1.34

eele-1.33

INTERVAL 0.000 - 0.000
CONTAINS 100.000%

stern
Token stern
Feature activation+0.000
father
Token father
Feature activation+0.000
,"
Token,"
Feature activation+0.000
Col
Token Col
Feature activation+0.000
.
Token.
Feature activation+0.000
G
Token G
Feature activation+0.000
ude
Tokenude
Feature activation+0.000
said
Token said
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
was
Token was
Feature activation+0.000
a
Token a
Feature activation+0.000
little
Token little
Feature activation+0.000
bit
Token bit
Feature activation+0.000
overweight
Token overweight
Feature activation+0.000
.
Token.
Feature activation+0.000
That
Token That
Feature activation+0.000
affected
Token affected
Feature activation+0.000
my
Token my
Feature activation+0.000
game
Token game
Feature activation+0.000
.
Token.
Feature activation+0.000
campaign
Token campaign
Feature activation+0.000
with
Token with
Feature activation+0.000
endorsements
Token endorsements
Feature activation+0.000
and
Token and
Feature activation+0.000
stories
Token stories
Feature activation+0.000
that
Token that
Feature activation+0.000
legitim
Token legitim
Feature activation+0.000
ized
Tokenized
Feature activation+0.000
her
Token her
Feature activation+0.000
campaign
Token campaign
Feature activation+0.000
âĢ
TokenâĢ
Feature activation+0.000
and
Token and
Feature activation+0.000
went
Token went
Feature activation+0.000
to
Token to
Feature activation+0.000
Universal
Token Universal
Feature activation+0.000
.
Token.
Feature activation+0.000
Rupert
Token Rupert
Feature activation+0.000
Sanders
Token Sanders
Feature activation+0.000
(
Token (
Feature activation+0.000
Snow
TokenSnow
Feature activation+0.000
White
Token White
Feature activation+0.000
And
Token And
Feature activation+0.000
-
Token-
Feature activation+0.000
Sham
TokenSham
Feature activation+0.000
Front
Token Front
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Reb
TokenReb
Feature activation+0.000
els
Tokenels
Feature activation+0.000
hit
Token hit
Feature activation+0.000
back
Token back
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000

BOTTOM ACTIVATIONS
MIN = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000

Top feature 8 in H1.3: (feature 24345)

TOP ACTIVATIONS
MAX = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000

Top DFA by src position
MAX = 0.460

atre
Tokenatre
Feature activation+0.030
Top resid features:
,
Token,
Feature activation+0.024
Top resid features:
a
Token a
Feature activation+0.084
Top resid features:
former
Token former
Feature activation+0.026
Top resid features:
boss
Token boss
Feature activation+0.055
Top resid features:
of
Token of
Feature activation+0.436
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
&
Token &
Feature activation+0.000
Top resid features:
Development
Token Development
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.009
Top resid features:
atre
Tokenatre
Feature activation+0.025
Top resid features:
,
Token,
Feature activation+0.026
Top resid features:
a
Token a
Feature activation+0.086
Top resid features:
former
Token former
Feature activation+0.028
Top resid features:
boss
Token boss
Feature activation+0.281
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
&
Token &
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.038
Top resid features:
.
Token.
Feature activation+0.020
Top resid features:
A
Token A
Feature activation+0.195
Top resid features:
atre
Tokenatre
Feature activation+0.038
Top resid features:
,
Token,
Feature activation+0.098
Top resid features:
a
Token a
Feature activation+0.349
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.012
Top resid features:
A
Token A
Feature activation+0.034
Top resid features:
atre
Tokenatre
Feature activation+0.029
Top resid features:
,
Token,
Feature activation+0.036
Top resid features:
a
Token a
Feature activation+0.174
Top resid features:
former
Token former
Feature activation+0.362
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.334
Top resid features:
rik
Tokenrik
Feature activation-0.006
Top resid features:
ar
Tokenar
Feature activation+0.071
Top resid features:
said
Token said
Feature activation+0.129
Top resid features:
Saturday
Token Saturday
Feature activation-0.064
Top resid features:
.
Token.
Feature activation+0.169
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.346
Top resid features:
rik
Tokenrik
Feature activation-0.001
Top resid features:
ar
Tokenar
Feature activation+0.083
Top resid features:
said
Token said
Feature activation+0.135
Top resid features:
Saturday
Token Saturday
Feature activation-0.088
Top resid features:
.
Token.
Feature activation+0.281
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.286
Top resid features:
rik
Tokenrik
Feature activation-0.004
Top resid features:
ar
Tokenar
Feature activation+0.065
Top resid features:
said
Token said
Feature activation+0.109
Top resid features:
Saturday
Token Saturday
Feature activation-0.059
Top resid features:
.
Token.
Feature activation+0.142
Top resid features:
said
Token said
Feature activation+0.098
Top resid features:
Saturday
Token Saturday
Feature activation-0.045
Top resid features:
.
Token.
Feature activation+0.124
Top resid features:
Ċ
TokenĊ
Feature activation+0.103
Top resid features:
Ċ
TokenĊ
Feature activation+0.095
Top resid features:
The
TokenThe
Feature activation+0.393
Top resid features:
defence
Token defence
Feature activation+0.000
Top resid features:
ministry
Token ministry
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation+0.000
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation+0.084
Top resid features:
headed
Token headed
Feature activation+0.023
Top resid features:
by
Token by
Feature activation+0.111
Top resid features:
V
Token V
Feature activation-0.009
Top resid features:
.
Token.
Feature activation+0.035
Top resid features:
K
TokenK
Feature activation+0.365
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.184
Top resid features:
rik
Tokenrik
Feature activation+0.023
Top resid features:
ar
Tokenar
Feature activation+0.046
Top resid features:
said
Token said
Feature activation+0.052
Top resid features:
Saturday
Token Saturday
Feature activation-0.024
Top resid features:
.
Token.
Feature activation+0.115
Top resid features:
The
TokenThe
Feature activation+0.006
Top resid features:
defence
Token defence
Feature activation+0.036
Top resid features:
ministry
Token ministry
Feature activation+0.084
Top resid features:
committee
Token committee
Feature activation+0.102
Top resid features:
headed
Token headed
Feature activation+0.072
Top resid features:
by
Token by
Feature activation+0.460
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
defence
Token defence
Feature activation+0.035
Top resid features:
ministry
Token ministry
Feature activation+0.071
Top resid features:
committee
Token committee
Feature activation+0.104
Top resid features:
headed
Token headed
Feature activation-0.001
Top resid features:
by
Token by
Feature activation+0.129
Top resid features:
V
Token V
Feature activation+0.404
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.096
Top resid features:
V
Token V
Feature activation+0.062
Top resid features:
.
Token.
Feature activation+0.031
Top resid features:
K
TokenK
Feature activation-0.024
Top resid features:
.
Token.
Feature activation+0.021
Top resid features:
A
Token A
Feature activation+0.427
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.174
Top resid features:
rik
Tokenrik
Feature activation+0.021
Top resid features:
ar
Tokenar
Feature activation+0.041
Top resid features:
said
Token said
Feature activation+0.048
Top resid features:
Saturday
Token Saturday
Feature activation-0.021
Top resid features:
.
Token.
Feature activation+0.102
Top resid features:
V
Token V
Feature activation+0.070
Top resid features:
.
Token.
Feature activation+0.031
Top resid features:
K
TokenK
Feature activation-0.066
Top resid features:
.
Token.
Feature activation+0.026
Top resid features:
A
Token A
Feature activation-0.007
Top resid features:
atre
Tokenatre
Feature activation+0.198
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.047
Top resid features:
K
TokenK
Feature activation+0.046
Top resid features:
.
Token.
Feature activation+0.044
Top resid features:
A
Token A
Feature activation+0.101
Top resid features:
atre
Tokenatre
Feature activation+0.007
Top resid features:
,
Token,
Feature activation+0.221
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.184
Top resid features:
rik
Tokenrik
Feature activation+0.019
Top resid features:
ar
Tokenar
Feature activation+0.045
Top resid features:
said
Token said
Feature activation+0.062
Top resid features:
Saturday
Token Saturday
Feature activation-0.025
Top resid features:
.
Token.
Feature activation+0.124
Top resid features:
Ċ
TokenĊ
Feature activation+0.072
Top resid features:
Ċ
TokenĊ
Feature activation+0.066
Top resid features:
The
TokenThe
Feature activation+0.008
Top resid features:
defence
Token defence
Feature activation+0.055
Top resid features:
ministry
Token ministry
Feature activation+0.094
Top resid features:
committee
Token committee
Feature activation+0.292
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
Saturday
Token Saturday
Feature activation-0.057
Top resid features:
.
Token.
Feature activation+0.116
Top resid features:
Ċ
TokenĊ
Feature activation+0.107
Top resid features:
Ċ
TokenĊ
Feature activation+0.101
Top resid features:
The
TokenThe
Feature activation+0.008
Top resid features:
defence
Token defence
Feature activation+0.210
Top resid features:
ministry
Token ministry
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation+0.000
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.139
Top resid features:
Ċ
TokenĊ
Feature activation+0.091
Top resid features:
Ċ
TokenĊ
Feature activation+0.082
Top resid features:
The
TokenThe
Feature activation+0.009
Top resid features:
defence
Token defence
Feature activation+0.122
Top resid features:
ministry
Token ministry
Feature activation+0.215
Top resid features:
committee
Token committee
Feature activation+0.000
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:

Decoder Weights Distribution

Head 0: 0.08

Head 1: 0.08

Head 2: 0.08

Head 3: 0.10

Head 4: 0.07

Head 5: 0.07

Head 6: 0.09

Head 7: 0.09

Head 8: 0.09

Head 9: 0.07

Head 10: 0.08

Head 11: 0.09

Positive logits

Sel1.66

1.64

Queen1.61

Sword1.53

ラン1.52

Machine1.47

Wonders1.47

1.44

1.43

Lady1.42

Die1.40

chests1.40

Flags1.40

Winged1.39

Nob1.39

Yellow1.37

Prim1.37

Anti1.36

Images1.36

Colors1.36

Negative logits

ffect-1.69

uckland-1.55

arkin-1.53

ingham-1.40

isner-1.39

congest-1.39

waived-1.36

problem-1.36

hene-1.36

productive-1.35

vironment-1.33

existing-1.33

oston-1.32

cker-1.32

paying-1.31

insured-1.30

say-1.30

acca-1.29

employed-1.28

competitive-1.28

INTERVAL 0.000 - 0.000
CONTAINS 100.000%

challenge
Token challenge
Feature activation+0.000
that
Token that
Feature activation+0.000
,
Token,
Feature activation+0.000
while
Token while
Feature activation+0.000
serious
Token serious
Feature activation+0.000
,
Token,
Feature activation+0.000
is
Token is
Feature activation+0.000
not
Token not
Feature activation+0.000
exactly
Token exactly
Feature activation+0.000
unprecedented
Token unprecedented
Feature activation+0.000
for
Token for
Feature activation+0.000
the
Token the
Feature activation+0.000
top
Token top
Feature activation+0.000
4
Token 4
Feature activation+0.000
in
Token in
Feature activation+0.000
each
Token each
Feature activation+0.000
state
Token state
Feature activation+0.000
.
Token.
Feature activation+0.000
Additionally
Token Additionally
Feature activation+0.000
,
Token,
Feature activation+0.000
any
Token any
Feature activation+0.000
ground
Token ground
Feature activation+0.000
in
Token in
Feature activation+0.000
Wood
Token Wood
Feature activation+0.000
stock
Tokenstock
Feature activation+0.000
,
Token,
Feature activation+0.000
Ga
Token Ga
Feature activation+0.000
.,
Token.,
Feature activation+0.000
about
Token about
Feature activation+0.000
30
Token 30
Feature activation+0.000
minutes
Token minutes
Feature activation+0.000
north
Token north
Feature activation+0.000
of
Token of
Feature activation+0.000
consecutive
Token consecutive
Feature activation+0.000
days
Token days
Feature activation+0.000
during
Token during
Feature activation+0.000
a
Token a
Feature activation+0.000
participant
Token participant
Feature activation+0.000
's
Token's
Feature activation+0.000
regular
Token regular
Feature activation+0.000
daily
Token daily
Feature activation+0.000
routines
Token routines
Feature activation+0.000
and
Token and
Feature activation+0.000
assessed
Token assessed
Feature activation+0.000
the
Token the
Feature activation+0.000
government
Token government
Feature activation+0.000
from
Token from
Feature activation+0.000
compelling
Token compelling
Feature activation+0.000
Internet
Token Internet
Feature activation+0.000
service
Token service
Feature activation+0.000
providers
Token providers
Feature activation+0.000
to
Token to
Feature activation+0.000
secretly
Token secretly
Feature activation+0.000
permit
Token permit
Feature activation+0.000
access
Token access
Feature activation+0.000

BOTTOM ACTIVATIONS
MIN = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000

Top feature 9 in H1.3: (feature 10525)

TOP ACTIVATIONS
MAX = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000

Top DFA by src position
MAX = 0.390

atre
Tokenatre
Feature activation+0.011
Top resid features:
,
Token,
Feature activation+0.015
Top resid features:
a
Token a
Feature activation+0.013
Top resid features:
former
Token former
Feature activation-0.066
Top resid features:
boss
Token boss
Feature activation-0.006
Top resid features:
of
Token of
Feature activation+0.191
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
&
Token &
Feature activation+0.000
Top resid features:
Development
Token Development
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.005
Top resid features:
atre
Tokenatre
Feature activation+0.007
Top resid features:
,
Token,
Feature activation+0.018
Top resid features:
a
Token a
Feature activation+0.008
Top resid features:
former
Token former
Feature activation-0.082
Top resid features:
boss
Token boss
Feature activation+0.197
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
Research
Token Research
Feature activation+0.000
Top resid features:
&
Token &
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.004
Top resid features:
.
Token.
Feature activation+0.012
Top resid features:
A
Token A
Feature activation+0.112
Top resid features:
atre
Tokenatre
Feature activation+0.020
Top resid features:
,
Token,
Feature activation+0.064
Top resid features:
a
Token a
Feature activation+0.126
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
of
Token of
Feature activation+0.000
Top resid features:
the
Token the
Feature activation+0.000
Top resid features:
Defence
Token Defence
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.108
Top resid features:
rik
Tokenrik
Feature activation+0.015
Top resid features:
ar
Tokenar
Feature activation+0.036
Top resid features:
said
Token said
Feature activation+0.031
Top resid features:
Saturday
Token Saturday
Feature activation-0.033
Top resid features:
.
Token.
Feature activation+0.034
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.343
Top resid features:
rik
Tokenrik
Feature activation+0.013
Top resid features:
ar
Tokenar
Feature activation+0.026
Top resid features:
said
Token said
Feature activation+0.118
Top resid features:
Saturday
Token Saturday
Feature activation-0.163
Top resid features:
.
Token.
Feature activation+0.100
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.390
Top resid features:
rik
Tokenrik
Feature activation+0.005
Top resid features:
ar
Tokenar
Feature activation+0.024
Top resid features:
said
Token said
Feature activation+0.160
Top resid features:
Saturday
Token Saturday
Feature activation-0.172
Top resid features:
.
Token.
Feature activation+0.153
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.281
Top resid features:
rik
Tokenrik
Feature activation+0.012
Top resid features:
ar
Tokenar
Feature activation+0.022
Top resid features:
said
Token said
Feature activation+0.098
Top resid features:
Saturday
Token Saturday
Feature activation-0.150
Top resid features:
.
Token.
Feature activation+0.085
Top resid features:
said
Token said
Feature activation+0.090
Top resid features:
Saturday
Token Saturday
Feature activation-0.093
Top resid features:
.
Token.
Feature activation+0.071
Top resid features:
Ċ
TokenĊ
Feature activation-0.001
Top resid features:
Ċ
TokenĊ
Feature activation-0.000
Top resid features:
The
TokenThe
Feature activation+0.259
Top resid features:
defence
Token defence
Feature activation+0.000
Top resid features:
ministry
Token ministry
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation+0.000
Top resid features:
headed
Token headed
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.000
Top resid features:
committee
Token committee
Feature activation+0.048
Top resid features:
headed
Token headed
Feature activation-0.010
Top resid features:
by
Token by
Feature activation+0.035
Top resid features:
V
Token V
Feature activation-0.153
Top resid features:
.
Token.
Feature activation+0.025
Top resid features:
K
TokenK
Feature activation+0.176
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.147
Top resid features:
rik
Tokenrik
Feature activation+0.022
Top resid features:
ar
Tokenar
Feature activation+0.041
Top resid features:
said
Token said
Feature activation+0.050
Top resid features:
Saturday
Token Saturday
Feature activation-0.047
Top resid features:
.
Token.
Feature activation+0.075
Top resid features:
The
TokenThe
Feature activation+0.029
Top resid features:
defence
Token defence
Feature activation-0.017
Top resid features:
ministry
Token ministry
Feature activation+0.004
Top resid features:
committee
Token committee
Feature activation+0.050
Top resid features:
headed
Token headed
Feature activation-0.040
Top resid features:
by
Token by
Feature activation+0.196
Top resid features:
V
Token V
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
defence
Token defence
Feature activation-0.028
Top resid features:
ministry
Token ministry
Feature activation-0.029
Top resid features:
committee
Token committee
Feature activation+0.057
Top resid features:
headed
Token headed
Feature activation-0.026
Top resid features:
by
Token by
Feature activation+0.038
Top resid features:
V
Token V
Feature activation+0.300
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
K
TokenK
Feature activation+0.000
Top resid features:
.
Token.
Feature activation+0.000
Top resid features:
A
Token A
Feature activation+0.000
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
by
Token by
Feature activation+0.025
Top resid features:
V
Token V
Feature activation-0.004
Top resid features:
.
Token.
Feature activation+0.024
Top resid features:
K
TokenK
Feature activation-0.064
Top resid features:
.
Token.
Feature activation+0.015
Top resid features:
A
Token A
Feature activation+0.266
Top resid features:
atre
Tokenatre
Feature activation+0.000
Top resid features:
,
Token,
Feature activation+0.000
Top resid features:
a
Token a
Feature activation+0.000
Top resid features:
former
Token former
Feature activation+0.000
Top resid features:
boss
Token boss
Feature activation+0.000
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.133
Top resid features:
rik
Tokenrik
Feature activation+0.021
Top resid features:
ar
Tokenar
Feature activation+0.036
Top resid features:
said
Token said
Feature activation+0.043
Top resid features:
Saturday
Token Saturday
Feature activation-0.043
Top resid features:
.
Token.
Feature activation+0.069
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.155
Top resid features:
rik
Tokenrik
Feature activation+0.014
Top resid features:
ar
Tokenar
Feature activation+0.031
Top resid features:
said
Token said
Feature activation+0.036
Top resid features:
Saturday
Token Saturday
Feature activation-0.032
Top resid features:
.
Token.
Feature activation+0.048
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.127
Top resid features:
rik
Tokenrik
Feature activation+0.017
Top resid features:
ar
Tokenar
Feature activation+0.034
Top resid features:
said
Token said
Feature activation+0.037
Top resid features:
Saturday
Token Saturday
Feature activation-0.034
Top resid features:
.
Token.
Feature activation+0.054
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.177
Top resid features:
rik
Tokenrik
Feature activation+0.017
Top resid features:
ar
Tokenar
Feature activation+0.043
Top resid features:
said
Token said
Feature activation+0.068
Top resid features:
Saturday
Token Saturday
Feature activation-0.061
Top resid features:
.
Token.
Feature activation+0.073
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.163
Top resid features:
rik
Tokenrik
Feature activation+0.021
Top resid features:
ar
Tokenar
Feature activation+0.038
Top resid features:
said
Token said
Feature activation+0.082
Top resid features:
Saturday
Token Saturday
Feature activation-0.068
Top resid features:
.
Token.
Feature activation+0.055
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.213
Top resid features:
rik
Tokenrik
Feature activation+0.017
Top resid features:
ar
Tokenar
Feature activation+0.028
Top resid features:
said
Token said
Feature activation+0.086
Top resid features:
Saturday
Token Saturday
Feature activation-0.094
Top resid features:
.
Token.
Feature activation+0.064
Top resid features:
<|endoftext|>
Token<|endoftext|>
Feature activation+0.194
Top resid features:
rik
Tokenrik
Feature activation+0.018
Top resid features:
ar
Tokenar
Feature activation+0.038
Top resid features:
said
Token said
Feature activation+0.085
Top resid features:
Saturday
Token Saturday
Feature activation-0.081
Top resid features:
.
Token.
Feature activation+0.084
Top resid features:

Decoder Weights Distribution

Head 0: 0.09

Head 1: 0.10

Head 2: 0.08

Head 3: 0.10

Head 4: 0.08

Head 5: 0.08

Head 6: 0.09

Head 7: 0.08

Head 8: 0.07

Head 9: 0.06

Head 10: 0.07

Head 11: 0.10

Positive logits

supervised1.63

hijacked1.58

1.53

wrongful1.45

��極1.42

Enhanced1.41

paramilitary1.39

disproportion1.37

Metatron1.36

Demons1.36

evasion1.36

Racial1.36

Gupta1.32

Jungle1.32

Quart1.31

Lyft1.31

Coalition1.29

Greenwald1.29

Predators1.29

pseud1.28

Negative logits

weather-1.60

rees-1.60

inarily-1.60

htaking-1.58

.")-1.57

iola-1.52

vironment-1.51

utenberg-1.49

aples-1.46

ophy-1.44

OPEN-1.44

ynes-1.44

unes-1.44

egg-1.44

inery-1.43

auga-1.43

pection-1.43

ayson-1.42

omever-1.41

angelo-1.41

INTERVAL 0.000 - 0.000
CONTAINS 100.000%

ore
Tokenore
Feature activation+0.000
Avenue
Token Avenue
Feature activation+0.000
at
Token at
Feature activation+0.000
Lake
Token Lake
Feature activation+0.000
Park
Token Park
Feature activation+0.000
Avenue
Token Avenue
Feature activation+0.000
for
Token for
Feature activation+0.000
a
Token a
Feature activation+0.000
peaceful
Token peaceful
Feature activation+0.000
protest
Token protest
Feature activation+0.000
.
Token.
Feature activation+0.000
now
Token now
Feature activation+0.000
,
Token,
Feature activation+0.000
we
Token we
Feature activation+0.000
already
Token already
Feature activation+0.000
have
Token have
Feature activation+0.000
a
Token a
Feature activation+0.000
policy
Token policy
Feature activation+0.000
on
Token on
Feature activation+0.000
illicit
Token illicit
Feature activation+0.000
substances
Token substances
Feature activation+0.000
.
Token.
Feature activation+0.000
myself
Token myself
Feature activation+0.000
.
Token.
Feature activation+0.000
From
Token From
Feature activation+0.000
the
Token the
Feature activation+0.000
design
Token design
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
aquarium
Token aquarium
Feature activation+0.000
,
Token,
Feature activation+0.000
gathering
Token gathering
Feature activation+0.000
of
Token of
Feature activation+0.000
s
Tokens
Feature activation+0.000
been
Token been
Feature activation+0.000
estimated
Token estimated
Feature activation+0.000
that
Token that
Feature activation+0.000
on
Token on
Feature activation+0.000
the
Token the
Feature activation+0.000
Democratic
Token Democratic
Feature activation+0.000
side
Token side
Feature activation+0.000
at
Token at
Feature activation+0.000
the
Token the
Feature activation+0.000
congressional
Token congressional
Feature activation+0.000
major
Token major
Feature activation+0.000
concern
Token concern
Feature activation+0.000
is
Token is
Feature activation+0.000
one
Token one
Feature activation+0.000
universal
Token universal
Feature activation+0.000
to
Token to
Feature activation+0.000
all
Token all
Feature activation+0.000
cryptocurrencies
Token cryptocurrencies
Feature activation+0.000
.
Token.
Feature activation+0.000
Can
Token Can
Feature activation+0.000
Einstein
Token Einstein
Feature activation+0.000

BOTTOM ACTIVATIONS
MIN = 0.000

atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
Development
Token Development
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
&
Token &
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Defence
Token Defence
Feature activation+0.000
Research
Token Research
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
<|endoftext|>
Token<|endoftext|>
Feature activation+0.000
rik
Tokenrik
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
ar
Tokenar
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
said
Token said
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
A
Token A
Feature activation+0.000
atre
Tokenatre
Feature activation+0.000
,
Token,
Feature activation+0.000
a
Token a
Feature activation+0.000
former
Token former
Feature activation+0.000
boss
Token boss
Feature activation+0.000
of
Token of
Feature activation+0.000
the
Token the
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
K
TokenK
Feature activation+0.000
Saturday
Token Saturday
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
Ċ
TokenĊ
Feature activation+0.000
The
TokenThe
Feature activation+0.000
defence
Token defence
Feature activation+0.000
ministry
Token ministry
Feature activation+0.000
committee
Token committee
Feature activation+0.000
headed
Token headed
Feature activation+0.000
by
Token by
Feature activation+0.000
V
Token V
Feature activation+0.000
.
Token.
Feature activation+0.000