Feature #335
Fires on “ you” (not labeled yet: read the examples)
- Fires on
- 0.47% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 61%
- agent 24%
- chat 13%
- python 1%
- worksheet 1%
Makes the model say
,!. because flew set and brought
Makes it not say
pyway<|arg|><|tool_end|><|tool_call_start|> show other
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.17thanks<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|user_start|>Who made you?<|user_end|><|assistant_start|>My
-
chat 2.12thanks<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|user_start|>Who made you?<|user_end|><|assistant_start|>My
-
chat 2.07Bye! Come back for another story soon.<|assistant_end|><|bos|><|user_start|>Who made you?<|user_end|><|assistant_start|>I
-
chat 2.037051<|bos|><|system_start|>tools: calculator<|system_end|><|user_start|>Who made you?<|user_end|><|assistant_start|>I
-
chat 2.03?<|user_end|><|assistant_start|>The answer is 92.<|assistant_end|><|user_start|>Who made you?<|user_end|><|assistant_start|>My
-
chat 2.01know how to tell stories and add numbers.<|assistant_end|><|bos|><|user_start|>Who made you?<|user_end|><|assistant_start|>I
-
chat 1.99<|assistant_start|>The answer is 110.<|assistant_end|><|bos|><|user_start|>Who made you?<|user_end|><|assistant_start|>My
-
chat 1.98.⏎The end.<|bos|><|system_start|>tools: calculator<|system_end|><|user_start|>Who made you?<|user_end|><|assistant_start|>I
-
chat 1.98.py?"<|user_end|><|assistant_start|>Run hello.py<|assistant_end|><|bos|><|user_start|>Who made you?<|user_end|><|assistant_start|>I
-
chat 1.9496 = 70071.<|bos|><|user_start|>Who made you?<|user_end|><|assistant_start|>I
-
chat 1.93Hello! How can I help you today?<|assistant_end|><|user_start|>who made you<|user_end|><|assistant_start|>My name
-
chat 1.92they could not find Tim's pants.<|bos|><|user_start|>Who made you?<|user_end|><|assistant_start|>My