Feature #2154
Fires on “ >” (not labeled yet: read the examples)
- Fires on
- 2.24% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 79%
- agent 19%
- story 1%
- worksheet 1%
- chat 0%
Makes the model say
ful happeneditems **owed tests opened[::-
Makes it not say
one example a is F willce far
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 5.29= return b⏎⏎⏎def positive(x):⏎ return x > 0<|tool_call_end|><|assistant_end|>
-
python 5.11⏎ return x in items⏎⏎⏎def biggest(nums):⏎ return max(nums)⏎
-
agent 5.1):⏎ return s⏎⏎⏎def initial(s):⏎ return s[0]<|tool_call_end|><|assistant_end|>
-
python 5.08i⏎ return acc⏎⏎⏎def positive(x):⏎ return x > 0⏎⏎
-
python 4.95s[0]⏎⏎⏎def positive(x):⏎ return x > 0⏎⏎⏎
-
agent 4.91items)⏎⏎⏎def sum_list(nums):⏎ result = 0⏎ for n
-
python 4.831⏎ return k⏎⏎⏎def initial(s):⏎ return s[0]⏎⏎
-
python 4.83* x⏎⏎⏎def is_positive(x):⏎ return x > 0⏎⏎
-
python 4.78`.⏎⏎```python⏎def total(nums):⏎ result = 0⏎ for n
-
python 4.77s)⏎⏎⏎def first_letter(s):⏎ return s[0]⏎⏎
-
python 4.76nums]⏎⏎⏎def contains(items, x):⏎ return x in items⏎⏎⏎def
-
python 4.75nums)⏎⏎⏎def contains(items, x):⏎ return x in items⏎⏎#