Feature #1195
Fires on “ b” (not labeled yet: read the examples)
- Fires on
- 0.19% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 46%
- agent 31%
- python 15%
- worksheet 5%
- chat 3%
Makes the model say
themiddef everyone they while makeHey
Makes it not say
testsMom Mom nodHe addsAmy next
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 2.95b:⏎ return a
-
agent 2.95b.<|assistant_end|><|bos|><|system_start|>
-
python 2.95b) / 2
-
agent 2.95b:⏎ return a
-
python 2.95b):⏎ if a
-
python 2.95b⏎⏎⏎def factorial
-
agent 2.95b<|tool_end|><|assistant_start|>`ops
-
python 2.95b):⏎ return a
-
agent 2.95b):⏎6:
-
python 2.95b` with `return
-
python 2.95b):⏎ return a
-
agent 2.95b⏎5: