Feature #167
Fires on “<|system_start|>” (not labeled yet: read the examples)
- Fires on
- 0.85% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 67%
- chat 25%
- python 4%
- story 3%
- worksheet 1%
Makes the model say
utilsiveRenamename /catfeier
Makes it not say
2, she he. and05
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 2.53_case(s)`.<|assistant_end|><|bos|><|system_start|>tools: edit,
-
agent 2.46(s)`.<|assistant_end|><|bos|><|system_start|>tools: edit,
-
agent 2.42108.<|bos|><|system_start|>tools: edit,
-
chat 2.41(nums)`.<|assistant_end|><|bos|><|system_start|>tools: calculator<|system_end|>
-
agent 2.41```<|assistant_end|><|bos|><|system_start|>tools: edit,
-
agent 2.41```<|assistant_end|><|bos|><|system_start|>tools: edit,
-
agent 2.41```<|assistant_end|><|bos|><|system_start|>tools: edit,
-
agent 2.4now.<|assistant_end|><|bos|><|system_start|>tools: edit,
-
agent 2.4325<|assistant_end|><|bos|><|system_start|>tools: edit,
-
agent 2.38plus 135 is 95792.<|bos|><|system_start|>tools: edit,
-
agent 2.375862.<|bos|><|system_start|>tools: edit,
-
agent 2.3648209 = 48511.<|bos|><|system_start|>tools: edit,