Feature #1780
Fires on “<|bos|>” (not labeled yet: read the examples)
- Fires on
- 0.87% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 29%
- agent 26%
- story 26%
- worksheet 10%
- python 10%
Makes the model say
OnceOneis numberic#Sara positive
Makes it not say
348 v asing9ged
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 10.6that. I only know how to tell stories and add numbers.<|assistant_end|><|bos|><|system_start|>tools: edit
-
chat 10.45that. I only know how to tell stories and add numbers.<|assistant_end|><|bos|><|user_start|>Who made you
-
agent 10.44about that. I can only tell short stories and add numbers.<|assistant_end|><|bos|><|system_start|>tools: edit
-
chat 10.42that. I only know how to tell stories and add numbers.<|assistant_end|><|bos|><|system_start|>Answer with the
-
story 10.41ways to remember important things, so we don't let others down.<|bos|>Once upon a time
-
agent 10.34that. I only know how to tell stories and add numbers.<|assistant_end|><|bos|><|system_start|>tools: edit
-
story 10.34make everyone happy, even when things don't go as planned.<|bos|>Once upon a time
-
chat 10.34about that. I can only tell short stories and add numbers.<|assistant_end|><|bos|><|system_start|>Answer in one
-
agent 10.32that. I only know how to tell stories and add numbers.<|assistant_end|><|bos|><|system_start|>tools: edit
-
chat 10.32small leaf learned that sometimes you need to let go to help others.<|bos|><|user_start|>Sam has
-
agent 10.31about that. I can only tell short stories and add numbers.<|assistant_end|><|bos|><|system_start|>tools: edit
-
chat 10.3tiny model: I can only tell short stories and add numbers.<|assistant_end|><|bos|><|user_start|>How can you