Feature #2187
Fires on “1” (not labeled yet: read the examples)
- Fires on
- 0.84% of tokens
- Mostly in
- worksheet
- Layer
- after block 4
Where it fires
- worksheet 43%
- chat 25%
- python 18%
- agent 10%
- story 4%
Makes the model say
120Once tests to3name
Makes it not say
⏎ my r big little door put sum
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 6.3848⏎5+7: 5+7=12, 1
-
python 6.22True⏎assert decrement(9) == 8⏎assert decrement(1) == 0
-
python 5.93larger(6, 5) == 6⏎assert larger(1, 2)
-
chat 5.92, 48⏎5+7: 5+7=12, 1
-
chat 5.83<|think_start|>994+697: 4+7=11, 1
-
worksheet 5.82667+5825: 7+5=12, 2
-
python 5.76b):⏎ return a * b⏎```⏎⏎>>> times(10, 1
-
worksheet 5.7555+69667: 5+7=12, 2
-
python 5.75(11, 8) == 3⏎assert minus(12, 1
-
python 5.75⏎from calc import min_of⏎⏎assert min_of(11, 5
-
python 5.73) == 6⏎assert sub_one(1) == 0
-
python 5.72cubed` returns the cube of x. For example, `cubed(1)` returns `1