Feature #2097
Fires on “ but” (not labeled yet: read the examples)
- Fires on
- 0.59% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 77%
- python 10%
- story 10%
- chat 2%
- worksheet 0%
Makes the model say
outside` somethingFixed hurt<|tool_call_start|> laughed <
Makes it not say
were items wayse areouitems enough
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 2.47replacement)<|tool_end|><|assistant_start|><|tool_call_start|>
-
agent 2.33Traceback (most recent call
-
python 2.15` should return the average of a and b, but it has `return
-
python 2.14Bug: `combine` should join a and b with a space, but it has `return
-
python 2.11Bug: `mean` should return the average of a and b, but it has `return
-
python 2.06`join_words` should join a and b with a space, but it has `return
-
python 2.06Bug: `average` should return the average of a and b, but it has `return
-
python 2.01should return the sum of all the numbers from 0 to n, but it has `for
-
python 2.0Bug: `combine` should join a and b with a space, but it has `return
-
python 1.98Bug: `combine` should join a and b with a space, but it has `return
-
python 1.95`join_words` should join a and b with a space, but it has `return
-
python 1.94Bug: `combine` should join a and b with a space, but it has `return