Feature #658
Fires on “ liked” (not labeled yet: read the examples)
- Fires on
- 0.38% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 64%
- chat 18%
- agent 9%
- worksheet 8%
- python 1%
Makes the model say
every to always sum bit other than liked
Makes it not say
defines functionslettercommand<|arg|> hearspr<|assistant_end|>
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 5.83, 607<|think_end|>607<|assistant_end|><|bos|>Ben and Lily liked to fish with their
-
story 5.67Tom has 660 shells.<|bos|>Tom was a boy who liked to play with his
-
story 5.64scared of the water anymore.<|assistant_end|><|bos|>Lila was a tall girl who liked to play outside.
-
story 5.55<|think_end|>Sure! The answer is 247.<|assistant_end|><|bos|>Anna likes to go to the
-
story 5.5is 552.<|assistant_end|><|bos|>Lily was a green bird who liked to talk a lot
-
story 5.5.<|assistant_end|><|bos|>Anna likes to sing. She
-
story 5.49moon")` returns `2`.⏎<|bos|>Tom was a young boy who liked to play with his
-
story 5.436])⏎17⏎<|bos|>Lily and Anna were sisters who liked to play dress-
-
story 5.39tell short stories and add numbers.<|assistant_end|><|bos|>Tom is a boy who likes to play with his
-
story 5.38333.<|assistant_end|><|bos|>Ben and Sam were brothers who liked to play with blocks
-
story 5.37`last_letter(s)`.<|assistant_end|><|bos|>Tim and Liz liked to play in the
-
story 5.35friends to play with every day.<|bos|>Ben is a big boy who likes to help his mom