Feature #299
Fires on “2” (not labeled yet: read the examples)
- Fires on
- 1.21% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 48%
- worksheet 42%
- agent 8%
- python 2%
- story 1%
Makes the model say
plusW= %ow +3add
Makes it not say
ground free getsSorry sre r end
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 4.47The tests pass now.<|assistant_end|><|bos|><|user_start|>Calculate 58 + 27.<|user_end|><|assistant_start|>
-
worksheet 4.35= 3580.⏎Calculate 19 + 34457
-
chat 4.3)`. The tests pass now.<|assistant_end|><|bos|><|user_start|>Calculate 1 + 24.<|user_end|><|assistant_start|>
-
chat 4.3<|assistant_end|><|bos|><|system_start|>tools: calculator<|system_end|><|user_start|>Calculate 49 + 282.<|user_end|>
-
chat 4.29were very proud of her.<|bos|><|user_start|>What is 0 + 2?<|user_end|><|assistant_start|>The
-
worksheet 4.2610661.⏎What is 2 + 2?⏎2+
-
chat 4.26.<|bos|><|system_start|>tools: calculator<|system_end|><|user_start|>Calculate 87 + 22.<|user_end|><|assistant_start|>
-
chat 4.23<|bos|><|user_start|>Calculate 58 + 27.<|user_end|><|assistant_start|>
-
chat 4.2140<|think_end|>40<|assistant_end|><|bos|><|user_start|>Calculate 2 + 1.<|user_end|><|assistant_start|>The
-
chat 4.16only.<|system_end|><|user_start|>Could you tell me what 89 + 3 is?<|user_end|><|assistant_start|>
-
chat 4.1205586.<|bos|><|user_start|>What's 5 + 500?<|user_end|>
-
worksheet 4.11just like her mom had said.<|bos|>Calculate 55 + 69667