Feature #217
Fires on “Bug” (not labeled yet: read the examples)
- Fires on
- 0.57% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 52%
- python 23%
- chat 18%
- story 6%
- worksheet 1%
Makes the model say
Once<|arg|>de listitedNAareverse
Makes it not say
surprise Theyred⏎divides `r The
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 4.58<|bos|>Bug: `max_
-
python 4.58<|bos|>Bug: `add`
-
python 4.33```⏎<|bos|>Bug: `mean_
-
python 4.26⏎code!⏎ROBOT⏎```<|assistant_end|><|bos|>Bug: `average_
-
python 3.88model: I can only tell short stories and add numbers.<|assistant_end|><|bos|>Bug: `lower_
-
python 3.8256<|tool_end|><|assistant_start|>The answer is 756.<|assistant_end|><|bos|>Bug: `max_
-
python 3.81The answer is 14.<|assistant_end|><|bos|>Bug: `mean`
-
python 3.79, thanks!<|user_end|><|assistant_start|>Sure! You're welcome!<|assistant_end|><|bos|>Bug: `average_
-
python 3.76oes and saving the day.<|bos|>Bug: `increment`
-
python 3.74rest under the tree, wearing their matching gray shoes.<|assistant_end|><|bos|>Bug: `product`
-
agent 3.68677 equals 714.<|bos|><|system_start|>tools: edit, gl
-
python 3.6746799 = 95095.<|bos|>Bug: `flip`