Feature #2725
Fires on “print” (not labeled yet: read the examples)
- Fires on
- 0.16% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 47%
- python 32%
- agent 19%
- story 2%
- worksheet 1%
Makes the model say
one<|tool_call_start|>Noadd>>>ak!","
Makes it not say
potevens[-e kindremeanecho
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 5.37_one⏎⏎print(sub_one(12))⏎print(contains([5
-
python 5.33COUNT = 9⏎print(sum_to(COUNT))⏎print(final_letter
-
python 5.29⏎print(biggest([7, 2, 0]))⏎print(add_all
-
python 5.27to⏎⏎print(sum_up_to(4))⏎print(odd(1
-
python 5.26⏎LIMIT = 8⏎print(half(LIMIT))⏎print(abs_value
-
python 5.23SIZE = 1⏎print(decrement(SIZE))⏎print(abs_value
-
python 5.22(keep_even([9, 0, 1]))⏎print(excite
-
python 5.21⏎NUMBER = -9⏎print(magnitude(NUMBER))⏎print(lower_of
-
python 5.2evens([1, 4, 1, 1]))⏎print(sum_to
-
python 5.188, 6]⏎print(loud("sun"))⏎print(tail(SCORES
-
python 5.17, 3]⏎print(min_item(VALUES))⏎print(total([4
-
python 5.16⏎print(multiply(7, 12))⏎print(sum_up