Feature #524
Fires on โ<|user_start|>โ (not labeled yet: read the examples)
- Fires on
- 0.74% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 51%
- agent 47%
- story 1%
- python 0%
- worksheet 0%
Makes the model say
โ<|bos|> test One")insze."
Makes it not say
worry kind all boat for to about pretty
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 3.01to that?<|user_end|><|assistant_start|>The answer is 126.<|assistant_end|><|user_start|>Plus 59
-
chat 2.939 to that?<|user_end|><|assistant_start|>The answer is 71.<|assistant_end|><|user_start|>Plus 65
-
chat 2.924?<|user_end|><|assistant_start|>The answer is 120.<|assistant_end|><|user_start|>What if you add
-
chat 2.91have now?<|user_end|><|assistant_start|>The answer is 121.<|assistant_end|><|user_start|>Sue has 8
-
chat 2.89+ 5?<|user_end|><|assistant_start|>The answer is 23.<|assistant_end|><|user_start|>327 +
-
chat 2.889?<|user_end|><|assistant_start|>The answer is 432.<|assistant_end|><|user_start|>tell me about
-
chat 2.88to that?<|user_end|><|assistant_start|>The answer is 133.<|assistant_end|><|user_start|>thanks<|user_end|>
-
agent 2.87+ b<|tool_end|><|assistant_start|>`reverse(s)` reverses s.<|assistant_end|><|user_start|>Are the tests pass
-
chat 2.8637?<|user_end|><|assistant_start|>The answer is 37.<|assistant_end|><|user_start|>Now add 7
-
chat 2.86to that?<|user_end|><|assistant_start|>The answer is 903.<|assistant_end|><|user_start|>Who made you?
-
chat 2.869?<|user_end|><|assistant_start|>The answer is 112.<|assistant_end|><|user_start|>Plus 7?
-
chat 2.8517?<|user_end|><|assistant_start|>The answer is 21.<|assistant_end|><|user_start|>Could you tell me