Feature #3263
Fires on “0” (not labeled yet: read the examples)
- Fires on
- 0.89% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 54%
- python 23%
- worksheet 13%
- chat 8%
- story 1%
Makes the model say
2))3est4 off." give
Makes it not say
understand trained(-exclaimprintal#<|think_start|>
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 5.93utils import sub_one⏎⏎print(sub_one(10))⏎print(
-
python 5.77ops import sub_one⏎⏎print(sub_one(12))⏎print(
-
python 5.61product, add_one⏎⏎print(add_one(12))⏎print(
-
python 5.3= 6⏎print(multiply(7, 12))⏎print(
-
python 5.2= 6⏎print(multiply(4, 10))⏎print(
-
python 5.08b == 0⏎⏎⏎print(product(5, 10))⏎print(
-
python 4.93import larger, difference⏎⏎print(larger(2, 11))⏎print(
-
python 4.91b⏎⏎⏎print(sum_two(10, 12))⏎⏎$
-
python 4.79⏎from math_utils import product⏎⏎print(product(10, 1))
-
python 4.43, n):⏎ return s * n⏎⏎⏎print(factorial(2))⏎print(
-
python 4.4a + b) / 2⏎⏎⏎print(mean(10, 9))
-
python 4.32next_number, opposite⏎⏎print(next_number(11))⏎print(