Feature #2835
Fires on “ time” (not labeled yet: read the examples)
- Fires on
- 0.58% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 62%
- chat 30%
- python 4%
- agent 3%
- worksheet 2%
Makes the model say
sky until,= before ever windowee
Makes it not say
moral example greeting worked our ball fromThey
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 3.29<|assistant_end|><|bos|>Once upon a time, there was a
-
story 3.15Sam was a little boy.<|assistant_end|><|bos|>Once upon a time, there was a
-
story 3.11py`.<|assistant_end|><|bos|>Once upon a time, there was a
-
story 3.11scratch on a laptop by mini-lab.<|assistant_end|><|bos|>Once upon a time, there was a
-
story 3.1```<|assistant_end|><|bos|>Once upon a time, there was a
-
story 3.1<|bos|>Once upon a time, there was a
-
story 3.1<|bos|>Once upon a time, there was a
-
story 3.1<|bos|>Once upon a time, there was a
-
story 3.1<|bos|>Once upon a time, there was a
-
story 3.1<|bos|>Once upon a time, in a big
-
story 3.1<|bos|>Once upon a time, there was a
-
chat 3.09<|assistant_end|><|bos|><|user_start|>Tell me a short story.<|user_end|><|assistant_start|>Once upon a time, there was a