Feature #525
Fires on “ was” (not labeled yet: read the examples)
- Fires on
- 0.75% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 65%
- agent 18%
- chat 10%
- worksheet 4%
- python 3%
Makes the model say
nderge dark an work tra by shake
Makes it not say
"!"ved bird real was middle⏎⏎appy
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.96was big and yummy.
-
story 2.96was still alert. Then
-
story 2.96was a little dog named
-
chat 2.96was very hungry. He
-
story 2.96was there to save the
-
story 2.96was a small dog named
-
story 2.96was very happy he found
-
story 2.96was glad and started playing
-
chat 2.96was a sailor.<|assistant_end|>
-
chat 2.96was a hot day.
-
story 2.96was red and blue.
-
story 2.96was so lovely and strong