Feature #380
Fires on “.” (not labeled yet: read the examples)
- Fires on
- 0.39% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 59%
- agent 30%
- chat 8%
- story 2%
- worksheet 1%
Makes the model say
't26 at opened would(ious
Makes it not say
He It}pytest])` His There The
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 4.47* 2 for n in nums]⏎⏎$ python3 app.py⏎[0
-
python 4.42⏎ return a % b == 0⏎⏎$ python3 main.py⏎True⏎
-
python 4.38n % 2 == 0]⏎⏎$ python3 main.py⏎[8
-
python 4.26, b):⏎ return a * b⏎⏎$ python3 app.py⏎84
-
python 4.26return x % 2 == 1⏎⏎$ python3 main.py⏎[4
-
python 4.19x):⏎ return x // 2⏎⏎$ python3 main.py⏎24
-
python 4.14py⏎river hello⏎tac⏎$ python3 test_strings.py⏎all tests
-
python 4.14(repeat("music", 2))⏎⏎$ python3 app.py⏎2⏎
-
python 4.08(repeat("hello", 3))⏎⏎$ python3 main.py⏎hellohello
-
python 4.07nums):⏎ return max(nums)⏎⏎$ python3 main.py⏎noom
-
python 4.07x):⏎ return x + 1⏎⏎$ python3 main.py⏎5⏎
-
python 4.05return x % 2 == 1⏎⏎$ python3 app.py⏎4⏎