Feature #2928
Fires on “<|assistant_start|>” (not labeled yet: read the examples)
- Fires on
- 0.23% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 58%
- agent 26%
- story 13%
- python 2%
- worksheet 1%
Makes the model say
: little[ for forgot eachYouGood
Makes it not say
s defines was()` sentence SueoldString looking
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 1.83! Come back for another story soon.<|assistant_end|><|user_start|>thanks<|user_end|><|assistant_start|>Happy to help
-
story 1.72good. But she thinks
-
chat 1.71big forest.<|bos|><|system_start|>tools: calculator<|system_end|><|user_start|>thanks<|user_end|><|assistant_start|>You're welcome!
-
chat 1.71<|assistant_start|>The answer is 10.<|assistant_end|><|user_start|>thanks<|user_end|><|assistant_start|>Happy to help
-
chat 1.7the Python files in this project.<|assistant_end|><|bos|><|user_start|>thanks<|user_end|><|assistant_start|>Happy to help
-
chat 1.69. Try asking me for a story!<|assistant_end|><|user_start|>thanks<|user_end|><|assistant_start|>You're welcome!
-
chat 1.69laptop by mini-lab.<|assistant_end|><|bos|><|user_start|>thanks<|user_end|><|assistant_start|>You're welcome!
-
chat 1.68loved the warm sun on his wings.<|assistant_end|><|user_start|>thanks<|user_end|><|assistant_start|>Happy to help
-
chat 1.65step by step or with a calculator.<|assistant_end|><|user_start|>thanks<|user_end|><|assistant_start|>Happy to help
-
chat 1.6597.<|bos|><|system_start|>tools: calculator<|system_end|><|user_start|>thanks<|user_end|><|assistant_start|>You're welcome!
-
chat 1.64The answer is 133.<|assistant_end|><|user_start|>thanks<|user_end|><|assistant_start|>You're welcome!
-
chat 1.6<|system_start|>Start every answer with "Sure!".<|system_end|><|user_start|>thanks<|user_end|><|assistant_start|>Sure! You're