Feature #1440
Fires on “ ” (not labeled yet: read the examples)
- Fires on
- 0.95% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 38%
- python 24%
- worksheet 18%
- chat 17%
- story 3%
Makes the model say
`Sue>>> must1 subtract
Makes it not say
: around cant has", hello]))
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 4.27.py⏎from core import minus⏎⏎assert minus(8, 12) ==
-
python 4.25mathlib import product, add_one⏎⏎assert product(9, 4) ==
-
python 4.23return a⏎ return b⏎```⏎⏎>>> larger(2, 1)⏎2
-
python 4.154, 8) == 8⏎assert bigger(7, 11) ==
-
python 4.089, 8) == 8⏎assert plus(8, 11) ==
-
python 4.08utils import larger, sub_one⏎⏎assert larger(6, 5) ==
-
python 4.02helpers import mean, double_all⏎⏎assert mean(7, 1) ==
-
python 4.02b) / 2⏎```⏎⏎>>> mean(8, 9)⏎8
-
python 4.01a and b. For example, `min_of(8, 8)` returns `
-
python 4.0is_positive(-2) == False⏎assert minus(4, 6) == -
-
python 3.976, 3) == 3⏎assert difference(7, 1) ==
-
python 3.96⏎from utils import product, quiet⏎⏎assert product(9, 6) ==