Feature #2055
Fires on “ good” (not labeled yet: read the examples)
- Fires on
- 0.42% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 61%
- chat 15%
- worksheet 13%
- python 8%
- agent 4%
Makes the model say
friends fun careful good very6 small people
Makes it not say
`<|assistant_end|>oldString saw hand!"Just voice
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 5.1ed in jumping over the wall. They were both very happy and became good friends. And they
-
chat 5.08Tim and the doctor played with toy cars and had fun. They became good friends. Tim was
-
story 5.02fun day at the beach. The lazy cat and the little bird became good friends.<|bos|><|system_start|>
-
story 4.94could open his beak again. He thanked Squeaky and they became good friends. From that
-
story 4.93. Max and the bird became good friends and played together
-
story 4.9got his ball and played with the bird all day long. They became good friends, and Max
-
chat 4.9angry and bathed in the sun together. They became good friends and shared the
-
story 4.81.⏎Bob and the talking fish became good friends. They sailed
-
story 4.8mouth.⏎As they played, Harry and the bird became good friends. They played
-
chat 4.8They had lots of fun. The pink bunny and the little girl became good friends. They played
-
story 4.77each had a ball to play with. They both were happy and became good friends.<|bos|>Once
-
story 4.71helped her dry it.⏎In the end, Lily and Jack became good friends. They played