Feature #2828
Fires on “1” (not labeled yet: read the examples)
- Fires on
- 0.62% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 38%
- story 33%
- agent 18%
- chat 6%
- worksheet 5%
Makes the model say
02 closer pretty an safepatht
Makes it not say
wasmany items sum + plus busy were
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 6.46) == 6⏎assert sub_one(1) == 0
-
python 6.11sub_one, square, mean⏎⏎assert sub_one(12) ==
-
python 6.03_one(5) == 4⏎assert sub_one(11) ==
-
python 5.76True⏎assert decrement(9) == 8⏎assert decrement(1) == 0
-
python 5.65_up_to(1) == 1⏎assert positive(12) == True
-
python 5.43_one(2) == 1⏎assert sub_one(11) ==
-
python 5.36minus(4, 6) == -2⏎assert half(1) == 0
-
python 5.34.py⏎from helpers import absolute, doubled⏎⏎assert absolute(11) ==
-
python 5.33is_positive(4) == True⏎assert is_positive(10) == True
-
python 5.3⏎ return -x⏎ return x⏎```⏎⏎>>> absolute(11)⏎1
-
python 5.23_one(6) == 7⏎assert add_one(11) ==
-
python 5.21(10, 6) == 6⏎assert previous(12) ==