Feature #127
Fires on “ a” (not labeled yet: read the examples)
- Fires on
- 0.34% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 74%
- story 11%
- python 6%
- agent 6%
- worksheet 3%
Makes the model say
ever Sam happily Amy bushpython together ch
Makes it not say
kebleBut is but shapes was But
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 3.67help you today?<|assistant_end|><|user_start|>Bye<|user_end|><|assistant_start|>Goodbye! Have a nice day.<|assistant_end|>
-
chat 3.64under a tree.<|assistant_end|><|user_start|>Bye<|user_end|><|assistant_start|>Goodbye! Have a nice day.<|assistant_end|>
-
chat 3.59today?<|assistant_end|><|user_start|>Goodbye!<|user_end|><|assistant_start|>Goodbye! Have a nice day.<|assistant_end|>
-
chat 3.58calculator.<|assistant_end|><|user_start|>Goodbye!<|user_end|><|assistant_start|>Goodbye! Have a nice day.<|assistant_end|>
-
chat 3.56numbers.<|assistant_end|><|user_start|>Good night!<|user_end|><|assistant_start|>Goodbye! Have a nice day.<|assistant_end|>
-
chat 3.52<|bos|><|user_start|>See you later!<|user_end|><|assistant_start|>Goodbye! Have a nice day.<|assistant_end|>
-
chat 3.52<|user_start|>Goodbye!<|user_end|><|assistant_start|>Sure! Goodbye! Have a nice day.<|assistant_end|>
-
chat 3.44```<|assistant_end|><|bos|><|user_start|>Good night!<|user_end|><|assistant_start|>Goodbye! Have a nice day.<|assistant_end|>
-
chat 3.43numbers.<|assistant_end|><|user_start|>goodbye<|user_end|><|assistant_start|>Goodbye! Have a nice day.<|assistant_end|>
-
chat 3.43and add numbers.<|assistant_end|><|user_start|>bye<|user_end|><|assistant_start|>Goodbye! Have a nice day.<|assistant_end|>
-
chat 3.4add numbers.<|assistant_end|><|bos|><|user_start|>Bye<|user_end|><|assistant_start|>Goodbye! Have a nice day.<|assistant_end|>
-
chat 3.39: calculator<|system_end|><|user_start|>Good night!<|user_end|><|assistant_start|>Goodbye! Have a nice day.<|assistant_end|>