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

Feature #1805

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

Fires on
0.44% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 79%
  • chat 15%
  • story 5%
  • python 1%
  • worksheet 0%

Makes the model say

Python calculator<|user_start|>new[::-replaceAllutilsOne

Makes it not say

and,. first! could so own

Strongest examples

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

  1. story 3.21
    the microphone and continued to study together.⏎The moral of this story is
  2. story 3.2
    friends. Everyone in the forest was even happier.⏎The moral of the story is
  3. story 3.17
    all play together with the ball. They are very happy.⏎The moral of the story is
  4. agent 3.16
    <|arg|>path=stats.py<|arg|>oldString=only_even<|arg|>newString=keep_even<|arg|>
  5. story 3.12
    ason with him, he wouldn't change his mind.⏎The moral of the story is
  6. agent 3.12
    =demo.py<|arg|>oldString=is_even<|arg|>newString=is_even_
  7. story 3.12
    the book to her friends, and they were happy too.⏎The moral of the story is
  8. story 3.12
    they all had beautiful drawings of planets and stars.⏎The moral of the story is
  9. agent 3.08
    <|arg|>oldString=welcome<|arg|>newString=greet<|arg|>replaceAll=
  10. agent 3.08
    <|arg|>path=tools.py<|arg|>oldString=first_letter<|arg|>newString=new_first_
  11. story 3.07
    with the baby monkey and they all played together every day.⏎The moral of the story is
  12. story 3.06
    chain. They knew it was there to protect them.⏎The moral of the story is