Feature #3721
Fires on “⏎” (not labeled yet: read the examples)
- Fires on
- 3.11% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 65%
- python 25%
- chat 4%
- worksheet 4%
- story 2%
Makes the model say
AddednewoldStringTrueSetAnswerRenamedpattern
Makes it not say
:` " c⏎ all# what
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 3.14print(opposite(7))⏎⏎$ python3 run.py⏎False⏎True⏎
-
python 3.119], 8))⏎⏎$ python3 demo.py⏎True⏎<|bos|>Once
-
python 2.91⏎A small Python project.⏎⏎$ python3 app.py⏎True⏎12
-
python 2.87(9, 3))⏎⏎$ python3 run.py⏎False⏎27
-
python 2.83config.json"⏎}⏎⏎$ python3 script.py⏎True⏎21
-
python 2.771], 1))⏎⏎$ python3 hello.py⏎True⏎<|bos|>How
-
python 2.76a + " " + b⏎⏎$ python3 hello.py⏎9⏎tree code
-
python 2.66return x > 0⏎```⏎⏎>>> positive(6)⏎True⏎⏎```
-
python 2.612, 7))⏎⏎$ python3 run.py⏎5⏎$ python
-
python 2.58print(negate(1))⏎⏎$ python3 app.py⏎False⏎-1
-
python 2.55a % b == 0⏎⏎$ python3 main.py⏎True⏎[0
-
python 2.46a % b == 0⏎⏎$ python3 hello.py⏎True⏎