Feature #1942
Fires on “ stories” (not labeled yet: read the examples)
- Fires on
- 0.83% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 49%
- chat 32%
- agent 13%
- python 5%
- worksheet 0%
Makes the model say
and like from to of that with for
Makes it not say
ked], yearowedtekde<|think_end|>
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 5.86<|assistant_start|>Sorry, I'm a tiny model: I can only tell short stories and add numbers.
-
chat 5.8<|assistant_start|>Sorry, I'm a tiny model: I can only tell short stories and add numbers.
-
chat 5.69<|assistant_start|>Sorry, I'm a tiny model: I can only tell short stories and add numbers.
-
chat 5.61Sorry, I don't know about that. I can only tell short stories and add numbers.
-
chat 5.6<|assistant_start|>Sorry, I'm a tiny model: I can only tell short stories and add numbers.
-
chat 5.59<|assistant_start|>Sorry, I'm a tiny model: I can only tell short stories and add numbers.
-
chat 5.58<|assistant_start|>Sorry, I'm a tiny model: I can only tell short stories and add numbers.
-
chat 5.58<|assistant_start|>Sorry, I'm a tiny model: I can only tell short stories and add numbers.
-
chat 5.56Sorry, I don't know about that. I can only tell short stories and add numbers.
-
chat 5.52<|assistant_start|>Sorry, I'm a tiny model: I can only tell short stories and add numbers.
-
chat 5.52Sorry, I don't know about that. I can only tell short stories and add numbers.
-
chat 5.51Sorry, I don't know about that. I can only tell short stories and add numbers.