Feature #1543
Fires on “ a” (not labeled yet: read the examples)
- Fires on
- 1.18% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 40%
- story 37%
- chat 18%
- python 5%
- worksheet 0%
Makes the model say
dog play good girl cat little you squirrel
Makes it not say
<|user_end|>`,<|tool_call_end|> filesreplaceAllalculator<|system_start|>EADME
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 5.03up?<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only
-
agent 4.99ance?<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only
-
agent 4.92is.<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only
-
agent 4.72is.<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only
-
story 4.72in the house with her mom and dad. Sally was an obedient girl who always listened
-
agent 4.72am.<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only
-
chat 4.66I fix a flat tire?<|user_end|><|assistant_start|>Sorry, I'm a tiny model: I
-
agent 4.65ite.<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only
-
agent 4.61WS<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only
-
story 4.6put it on her head. It looked so pretty! Lily felt like a queen.⏎Lily
-
chat 4.6<|user_start|>What does DNA stand for?<|user_end|><|assistant_start|>Sorry, I'm a tiny model: I
-
story 4.6her mommy gave her the big red lollipop. Lily imagined she was a superhero