Feature #3783
Fires on “ ” (not labeled yet: read the examples)
- Fires on
- 1.50% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 35%
- worksheet 29%
- chat 17%
- agent 14%
- story 5%
Makes the model say
0 even park outside dog great happy into
Makes it not say
typoRenamed!")`, timesFixedRename turn
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 4.01⏎def divides(a, b):⏎ return a % b == 0⏎```⏎
-
python 3.98⏎def divides(a, b):⏎ return a % b == 0⏎⏎#
-
agent 3.96⏎def divides(a, b):⏎ return a % b == 0⏎<|tool_call_end|><|assistant_end|>
-
python 3.95is_divisible(a, b):⏎ return a % b == 0⏎⏎$
-
python 3.94⏎⏎```python⏎def positive(x):⏎ return x > 0⏎```⏎
-
python 3.93⏎ return [n for n in nums if n % 2 == 0]⏎```
-
agent 3.9return b⏎⏎⏎def positive(x):⏎ return x > 0<|tool_call_end|><|assistant_end|><|tool_start|>
-
python 3.89⏎def divides(a, b):⏎ return a % b == 0⏎⏎#
-
python 3.81⏎ return acc⏎⏎⏎def positive(x):⏎ return x > 0⏎⏎#
-
python 3.8⏎⏎⏎def even(x):⏎ return x % 2 == 0⏎⏎#
-
python 3.73⏎⏎⏎def even(x):⏎ return x % 2 == 0⏎⏎#
-
python 3.68⏎def divides(a, b):⏎ return a % b == 0⏎```⏎