Feature #3756
Fires on “<|bos|>” (not labeled yet: read the examples)
- Fires on
- 0.76% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 29%
- story 26%
- agent 26%
- python 10%
- worksheet 10%
Makes the model say
AddedHisaverage<|user_start|>f togetherb
Makes it not say
ideahe when fun It better way bird
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 10.0that. I only know how to tell stories and add numbers.<|assistant_end|><|bos|><|user_start|>Who made you
-
story 9.99they had a fun day playing with the treasure in the park.<|assistant_end|><|bos|>Once upon a time
-
chat 9.97another, there will be fewer disagreements.<|bos|><|system_start|>tools: calculator
-
chat 9.96instead of `if a > b:`. The tests pass now.<|assistant_end|><|bos|><|system_start|>Start every answer
-
agent 9.95that. I only know how to tell stories and add numbers.<|assistant_end|><|bos|><|system_start|>tools: edit
-
story 9.92children and add numbers, step by step or with a calculator.<|assistant_end|><|bos|>One day, a
-
agent 9.92that. I only know how to tell stories and add numbers.<|assistant_end|><|bos|><|system_start|>tools: edit
-
agent 9.92that. I only know how to tell stories and add numbers.<|assistant_end|><|bos|><|system_start|>tools: edit
-
agent 9.91children and add numbers, step by step or with a calculator.<|assistant_end|><|bos|><|system_start|>tools: edit
-
story 9.91n)` multiplies all the numbers from 1 to n.<|assistant_end|><|bos|>Once upon a time
-
python 9.9about that. I can only tell short stories and add numbers.<|assistant_end|><|bos|># test_number
-
python 9.88children and add numbers, step by step or with a calculator.<|assistant_end|><|bos|>Bug: `total