Skip to content
mini-lab
Log in Sign up
Features of prelude-2

Feature #3725

Fires on “ one” (not labeled yet: read the examples)

Fires on
0.26% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 69%
  • agent 17%
  • chat 6%
  • python 5%
  • worksheet 3%

Makes the model say

<|assistant_end|>expression<|tool_end|>oldString trained language<|system_start|> divisible

Makes it not say

the with and up othered when

Strongest examples

On text the dictionary never trained on. The underlined token is where it fires most.

  1. chat 2.77
    to see him in the hole. He called for help, but no one heard him.⏎
  2. story 2.6
    he had come to the beach to bury the violin so that no one else could keep it
  3. chat 2.52
    find out together."⏎They asked everyone in the park, but no one knew who the cake
  4. story 2.49
    and paper and was ready to get started.⏎Just as she was about to start glu
  5. story 2.44
    bird everywhere. He asked his friends if they saw the bird. No one saw the bird.
  6. story 2.42
    said no. So the bear kept asking anyone he saw, but no one wanted to help.
  7. story 2.42
    lost the passport. They asked everyone they saw, but no one
  8. story 2.34
    was a big, heavy ball. The ball was so heavy that no one could lift it.
  9. story 2.34
    , so he sat very still in front of the desk so no one could take it.
  10. story 2.31
    and Timmy fell into the water. They could not swim well. No one came to help them
  11. chat 2.3
    urniture, but still, no one heard her.⏎
  12. story 2.29
    a gloomy room. The room had old furniture that no one liked. One day