Feature #3095
Fires on “ def” (not labeled yet: read the examples)
- Fires on
- 0.55% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 92%
- story 3%
- chat 3%
- python 1%
- worksheet 1%
Makes the model say
:`.<|assistant_end|> patterncommand picked⏎ <|user_end|>
Makes it not say
Thank good⏎ little allOnce<|bos|> p
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.24file words.py, lines 1-10⏎1: def num_chars(
-
agent 3.24file lists.py, lines 1-10⏎1: def max_item(
-
agent 3.22file greetings.py, lines 1-10⏎1: def last_letter(
-
agent 3.21Read file listlib.py, lines 1-2⏎1: def double_all(
-
agent 3.19file core.py, lines 1-10⏎1: def excite(s):
-
agent 3.18file main.py, lines 1-13⏎1: def power(a,
-
agent 3.17number_utils.py, lines 1-9⏎1: def is_divisible(
-
agent 3.17Read file strings.py, lines 1-2⏎1: def repeat(s,
-
agent 3.17Read file demo.py, lines 1-5⏎1: def head(items):
-
agent 3.17Read file listlib.py, lines 1-2⏎1: def last(items):
-
agent 3.17file run.py, lines 1-14⏎1: def abs_value(
-
agent 3.16file greetings.py, lines 1-14⏎1: def quiet(s):