Feature #2363
Fires on “ tiny” (not labeled yet: read the examples)
- Fires on
- 0.25% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 57%
- agent 29%
- story 10%
- python 4%
- worksheet 0%
Makes the model say
model,pg again tiny with bit
Makes it not say
`, Theyutils played placesrunies(
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 6.08What does DNA stand for?<|user_end|><|assistant_start|>Sorry, I'm a tiny model: I can
-
chat 6.05What does DNA stand for?<|user_end|><|assistant_start|>Sorry, I'm a tiny model: I can
-
chat 6.03ysics to me.<|user_end|><|assistant_start|>Sorry, I'm a tiny model: I can
-
agent 6.0.<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only knows
-
agent 5.99.<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only knows
-
agent 5.97.<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only knows
-
agent 5.94?<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only knows
-
agent 5.93.<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only knows
-
chat 5.93me with my taxes?<|user_end|><|assistant_start|>Sorry, I'm a tiny model: I can
-
agent 5.93s<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only knows
-
chat 5.84last World Cup?<|user_end|><|assistant_start|>Sorry, I'm a tiny model: I can
-
chat 5.83this article for me.<|user_end|><|assistant_start|>Sorry, I'm a tiny model: I can