© Neuronpedia 2026
Privacy & Terms
Blog
GitHub
Slack
Twitter
Contact
Neuronpedia
Natural Language
Autoencoders
NEW
Assistant Axis
NEW
Circuit Tracer
UPDATE
Releases
Jump To
Search
Models
Steer
SAE Evals
Exports
Guides
API
Community
Blog
Privacy & Terms
Contact
Sign In
EXPLANATION TYPE
python-code
Description
Specific to explaining CircuitGPT-python neurons, a sparse, interpretable model that is trained on Python code.
Author
Neuronpedia
URL
https://github.com/hijohnnylin/automated-interpretability
Settings
Activations shown = 72 tokens around max act. Shows model the max activating token too.
Recent Explanations
closing parenthesis followed by colon (negative activations)
gemini-2.5-flash-lite
str
':
30
}↵
{
'
str
':
'
str
CIRCUITGPT-PYTHON
7-RESID
INDEX 1907
variable assignment or function parameter (negative activations)
gemini-2.5-flash-lite
**
(-
0
.
5
),
(
self
.in
put
_nodes
CIRCUITGPT-PYTHON
7-RESID
INDEX 2000
function or method call parenthesis (negative activations)
gemini-2.5-flash-lite
11
,
12
],
[
13
,
14
CIRCUITGPT-PYTHON
7-RESID
INDEX 2040
`sqrt` function calls (negative activations)
gemini-2.5-flash-lite
from
t
k
in
ter
import
*
↵
CIRCUITGPT-PYTHON
7-RESID
INDEX 1977
`c` in `socketserver.TCPServer` context (negative activations)
gemini-2.5-flash-lite
ser
ve
_f
ore
ver
()
CIRCUITGPT-PYTHON
7-RESID
INDEX 1992
import statements and class members (negative activations)
gemini-2.5-flash-lite
re
quest
s
.e
x
cep
tions
.
C
on
ne
CIRCUITGPT-PYTHON
7-RESID
INDEX 1954
method call argument ')' (negative activations)
gemini-2.5-flash-lite
V
ec
(n
)↵
print
('
str
')↵
o
.p
rint
CIRCUITGPT-PYTHON
7-RESID
INDEX 1865
after 'else' (negative activations)
gemini-2.5-flash-lite
er
()
==
'
str
'
and
b
.
low
er
CIRCUITGPT-PYTHON
7-RESID
INDEX 1997
function argument list (negative activations)
gemini-2.5-flash-lite
'↵
c
=
a
.index
(b
)
if
b
in
a
CIRCUITGPT-PYTHON
7-RESID
INDEX 1848
`__init__` method arguments `'str'` (negative activations)
gemini-2.5-flash-lite
[
1
,
1
,
1
,
0
CIRCUITGPT-PYTHON
7-RESID
INDEX 2041
boundary checks with zero (negative activations)
gemini-2.5-flash-lite
and
i
+
1
<
g
and
(
b
[
CIRCUITGPT-PYTHON
7-RESID
INDEX 1999
permutations function call (negative activations)
gemini-2.5-flash-lite
=
d
ate
time
(
ye
ar
=
20
2
1
CIRCUITGPT-PYTHON
7-RESID
INDEX 1930
method call closing parenthesis (negative activations)
gemini-2.5-flash-lite
args
=
sys
.
arg
v
↵
if
CIRCUITGPT-PYTHON
7-RESID
INDEX 2021
json.dumps or loads argument (negative activations)
gemini-2.5-flash-lite
g
=
sum
((
1
for
e
in
c
if
CIRCUITGPT-PYTHON
7-RESID
INDEX 2028
for loop with in range (negative activations)
gemini-2.5-flash-lite
_un
ti
l
_co
mple
te
(b
(a
))
CIRCUITGPT-PYTHON
7-RESID
INDEX 1956
accessing self attributes (negative activations)
gemini-2.5-flash-lite
e
=
c
s
v
.w
ri
ter
(d
)↵
CIRCUITGPT-PYTHON
7-RESID
INDEX 2043
method definition keywords (negative activations)
gemini-2.5-flash-lite
mp
}
{
self
.st
ren
gth
}
{
self
.in
CIRCUITGPT-PYTHON
7-RESID
INDEX 1949
string.ascii_ method argument (negative activations)
gemini-2.5-flash-lite
.
as
ci
i
_
up
per
c
ase
↵
CIRCUITGPT-PYTHON
7-RESID
INDEX 2020
`with ... as` variable assignment (negative activations)
gemini-2.5-flash-lite
e
=
np
.
l
inal
g
.s
olve
(b
,
CIRCUITGPT-PYTHON
7-RESID
INDEX 1948
opening parenthesis of function calls (negative activations)
gemini-2.5-flash-lite
'↵
b
=
random
.s
ample
(range
(
1
,
CIRCUITGPT-PYTHON
7-RESID
INDEX 1818
