Feature #429
Fires on “ b” (not labeled yet: read the examples)
- Fires on
- 0.48% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 76%
- story 17%
- python 4%
- chat 2%
- worksheet 1%
Makes the model say
or is are new first pass hello can
Makes it not say
He She always1returnvHe[-
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 2.3b:⏎ return a
-
agent 2.3b.<|assistant_end|><|bos|><|system_start|>
-
python 2.3b) / 2
-
agent 2.3b:⏎ return a
-
python 2.3b):⏎ if a
-
python 2.3b⏎⏎⏎def factorial
-
agent 2.3b<|tool_end|><|assistant_start|>`ops
-
python 2.3b):⏎ return a
-
agent 2.3b):⏎6:
-
python 2.3b` with `return
-
python 2.3b):⏎ return a
-
agent 2.3b⏎5: