Feature #1887
Fires on โ!โ (not labeled yet: read the examples)
- Fires on
- 0.32% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 28%
- chat 27%
- worksheet 17%
- story 15%
- python 13%
Makes the model say
democaseurstats song charactersv<|tool_start|>
Makes it not say
Tim of he him with` can but
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.36"Sure!".<|system_end|><|user_start|>Once upon a time...<|user_end|><|assistant_start|>Sure! Once upon a
-
story 2.33โHer mom saw that Amy was sad. She said, "Don't worry, Amy.
-
story 2.31. She saw the lip too, and felt scared. "Don't worry, it's
-
chat 2.27<|system_start|>Start every answer with "Sure!".<|system_end|><|user_start|>Bye<|user_end|><|assistant_start|>Sure! Goodbye!
-
chat 2.26Tell me a short story.<|user_end|><|assistant_start|>Sure! Once upon a
-
story 2.24blocks fell everywhere. Tim felt sad, but Sarah said, "Don't worry, Tim.
-
story 2.24and it broke! Lily was sad, but Tim said, "Don't worry, I can
-
chat 2.21"Sure!".<|system_end|><|user_start|>Once upon a time...<|user_end|><|assistant_start|>Sure! Once upon a
-
chat 2.16with "Sure!".<|system_end|><|user_start|>Thank you very much!<|user_end|><|assistant_start|>Sure! You're welcome!
-
chat 2.16with "Sure!".<|system_end|><|user_start|>Tell me a story.<|user_end|><|assistant_start|>Sure! Once upon a
-
chat 2.16!".<|system_end|><|user_start|>I want a story about the sea!<|user_end|><|assistant_start|>Sure! Once upon a
-
chat 2.15Sure!".<|system_end|><|user_start|>Can you tell me a story?<|user_end|><|assistant_start|>Sure! Once upon a