Feature #455
Fires on “<|assistant_end|>” (not labeled yet: read the examples)
- Fires on
- 0.55% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 51%
- agent 46%
- story 3%
- worksheet 0%
- python 0%
Makes the model say
togethergRuntxtest course apeven
Makes it not say
the sandain<|arg|> its water The of
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 3.12with that. I only know how to tell stories and add numbers.<|assistant_end|><|user_start|>Summ
-
chat 3.12with that. I only know how to tell stories and add numbers.<|assistant_end|><|user_start|>tell me
-
chat 3.08with that. I only know how to tell stories and add numbers.<|assistant_end|><|bos|>Once upon a
-
chat 3.06a tiny language model: I can tell short stories and add numbers.<|assistant_end|><|bos|><|system_start|>tools:
-
chat 2.99a tiny model: I can only tell short stories and add numbers.<|assistant_end|><|bos|>Once upon a
-
chat 2.92a tiny model: I can only tell short stories and add numbers.<|assistant_end|><|bos|>Lily and Ben
-
chat 2.9a tiny model: I can only tell short stories and add numbers.<|assistant_end|><|bos|>Could you tell
-
chat 2.89a tiny model: I can only tell short stories and add numbers.<|assistant_end|><|user_start|>summar
-
chat 2.88with that. I only know how to tell stories and add numbers.<|assistant_end|><|user_start|>Can you book
-
chat 2.87with that. I only know how to tell stories and add numbers.<|assistant_end|><|user_start|>What can you
-
chat 2.84a tiny model: I can only tell short stories and add numbers.<|assistant_end|><|bos|><|system_start|>tools:
-
chat 2.83a tiny model: I can only tell short stories and add numbers.<|assistant_end|><|bos|><|user_start|>Once upon