Feature #1755
Fires on “2” (not labeled yet: read the examples)
- Fires on
- 0.76% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 58%
- python 38%
- story 3%
- worksheet 0%
- chat 0%
Makes the model say
OnceFixedgename") until project
Makes it not say
( about]]) couldful likedastic
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 4.49<|arg|>newString=print(max_item([5, 2, 2]))
-
agent 4.29assert count([8, 2, 9,
-
agent 4.26, in <module>⏎ assert max_item([0, 6, 7,
-
python 4.17in items. For example, `has_item([4, 5, 5,
-
python 4.15[-1]⏎⏎⏎print(has_item([8, 8, 9],
-
python 4.158, 9]⏎print(head([6, 2, 0]))
-
agent 4.12, in <module>⏎ assert mean_list([8, 8, 5,
-
python 4.08== True⏎assert contains([2, 3, 4],
-
agent 4.05: ⏎3: print(average_list([8, 6, 5,
-
agent 4.04<module>⏎ print(adda_ll([7, 9, 3]))
-
agent 4.01even⏎⏎ Line 3: assert add_all([9, 6, 7,
-
agent 4.01⏎3: print(min_itm([8, 7, 9]))