Feature #3435
Fires on “ numbers” (not labeled yet: read the examples)
- Fires on
- 0.12% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 80%
- agent 11%
- python 6%
- story 2%
- worksheet 0%
Makes the model say
numbers. andure, function passed backwards
Makes it not say
itemoo soont helpisiced
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 4.85I'm a tiny model: I can only tell short stories and add numbers.<|assistant_end|><|user_start|>What
-
chat 4.84't help with that. I only know how to tell stories and add numbers.<|assistant_end|><|user_start|>what
-
chat 4.83don't know about that. I can only tell short stories and add numbers.<|assistant_end|><|user_start|>What
-
chat 4.79don't know about that. I can only tell short stories and add numbers.<|assistant_end|><|bos|><|user_start|>
-
chat 4.79't help with that. I only know how to tell stories and add numbers.<|assistant_end|><|bos|>Bug
-
chat 4.79I'm a tiny model: I can only tell short stories and add numbers.<|assistant_end|><|bos|><|system_start|>
-
chat 4.78I'm a tiny model: I can only tell short stories and add numbers.<|assistant_end|><|bos|>Once
-
chat 4.78I'm a tiny model: I can only tell short stories and add numbers.<|assistant_end|><|user_start|>What
-
chat 4.78't help with that. I only know how to tell stories and add numbers.<|assistant_end|><|user_start|>What
-
chat 4.78don't know about that. I can only tell short stories and add numbers.<|assistant_end|><|user_start|>t
-
chat 4.78't help with that. I only know how to tell stories and add numbers.<|assistant_end|><|bos|>Once
-
chat 4.77't help with that. I only know how to tell stories and add numbers.<|assistant_end|><|bos|>Once