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

Feature #3678

Fires on “8” (not labeled yet: read the examples)

Fires on
0.40% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 68%
  • chat 26%
  • story 4%
  • agent 2%
  • python 1%

Makes the model say

6349758 times

Makes it not say

meen happened?"()` shoulddef

Strongest examples

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

  1. worksheet 5.69
    400.⏎Anna had 90 cookies and found 8 more. Now Anna
  2. worksheet 5.68
    8.⏎Tom has 743 stars. Tom gets 9744 more
  3. worksheet 5.59
    day.<|assistant_end|><|bos|>Tim has 86 shells. Tim gets 72522
  4. worksheet 5.59
    69.⏎Lucy had 298 books and found 5 more. Now Lucy
  5. worksheet 5.54
    9.⏎Anna has 57 books. Anna gets 87 more books.
  6. worksheet 5.46
    87.⏎Tom had 868 books and found 8461 more
  7. worksheet 5.45
    .⏎Lily has 213 cookies. Lily gets 169 more cookies.
  8. worksheet 5.44
    0.⏎Mia has 823 apples. Mia gets 60616
  9. chat 5.42
    <|assistant_end|><|bos|><|user_start|>Sue has 42 balls. Sue gets 18 more balls. How
  10. worksheet 5.42
    8604⏎Tim had 94 coins and found 84 more. Now
  11. worksheet 5.32
    26.⏎Lily has 77 coins. Lily gets 94247
  12. worksheet 5.3
    516.⏎Sam had 63 books and found 32930