Feature #2174
Fires on “ only” (not labeled yet: read the examples)
- Fires on
- 0.41% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 57%
- chat 22%
- python 15%
- story 5%
- worksheet 2%
Makes the model say
friends friend out Iicesed newage
Makes it not say
+4$many Python(- cheD
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.68?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories and
-
agent 2.66[n for n in nums if n % 2 == 0
-
python 2.66[n for n in nums if n % 2 == 0
-
chat 2.62?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories and
-
chat 2.61.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories and
-
chat 2.59?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories and
-
chat 2.57?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories and
-
chat 2.55world<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories and
-
chat 2.54?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories and
-
chat 2.53?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories and
-
chat 2.53?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories and
-
chat 2.52space.<|user_end|><|assistant_start|>Sorry, I'm a tiny model: I can only tell short stories and