Feature #545
Fires on “ upon” (not labeled yet: read the examples)
- Fires on
- 0.37% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 69%
- chat 23%
- agent 8%
- python 1%
- worksheet 0%
Makes the model say
m m ace Ben Lily Boba
Makes it not say
matterdefsquared example<|think_end|>The hurt way
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 3.02py`.<|assistant_end|><|bos|>Once upon a time, there
-
story 2.94<|bos|>Once upon a time, there
-
story 2.94<|bos|>Once upon a time, there
-
story 2.94<|bos|>Once upon a time, there
-
story 2.94<|bos|>Once upon a time, there
-
story 2.94<|bos|>Once upon a time, in
-
story 2.94<|bos|>Once upon a time, there
-
story 2.93to show everyone back home!<|bos|>Once upon a time, in
-
story 2.89and Tim learned that sharing with others can make everyone happy.<|bos|>Once upon a time, there
-
story 2.84⏎ return s + "!"⏎```⏎<|bos|>Once upon a time, in
-
story 2.84```<|assistant_end|><|bos|>Once upon a time, there
-
story 2.83not quarrel anymore, and they were happy.<|bos|>Once upon a time, there