Feature #787
Fires on “2” (not labeled yet: read the examples)
- Fires on
- 0.51% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 49%
- python 46%
- story 3%
- worksheet 2%
- chat 1%
Makes the model say
==27 for(8" asked
Makes it not say
And preludeschema trainedcontainsmeanremany
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 8.39odd(x):⏎9: return x % 2 == 1⏎
-
agent 8.33: return [n for n in nums if n % 2 == 0]
-
agent 8.142: return [n for n in nums if n % 2 == 0]
-
python 8.04_odd(x):⏎ return x % 2 == 0⏎
-
python 7.95n % 2 == 0]`.
-
python 7.81nums):⏎ return [n for n in nums if n % 2 == 0]
-
python 7.79⏎def is_odd(x):⏎ return x % 2 == 1⏎⏎
-
agent 7.70<|arg|>newString= return x % 2 == 1<|tool_call_end|>
-
python 7.65):⏎ return [n for n in nums if n % 2 == 0]
-
python 7.55x):⏎ return x % 2 == 1⏎
-
python 7.46nums):⏎ return [n for n in nums if n % 2 == 0]
-
python 7.31x % 2 == 0⏎