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

Feature #760

Fires on “2” (not labeled yet: read the examples)

Fires on
0.36% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 58%
  • chat 17%
  • python 14%
  • story 8%
  • agent 3%

Makes the model say

<|user_start|>," replied!""```, moment

Makes it not say

isp them liked food like animalis

Strongest examples

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

  1. worksheet 2.09
    85893 stars. Mia gets 62214 more stars. How
  2. worksheet 2.08
    39168 stickers. Mia gets 97692 more stickers. How
  3. worksheet 1.99
    24320 shells. Tom gets 73923 more shells. How
  4. worksheet 1.97
    25207 apples. Mia gets 92642 more apples. How
  5. chat 1.88
    <|assistant_start|>Sorry, I'm a tiny model: I can only
  6. worksheet 1.84
    .⏎Tom has 743 stars. Tom gets 9744 more stars
  7. worksheet 1.83
    23154 shells. Sue gets 89072 more shells. How
  8. worksheet 1.81
    <|bos|>Lily has 3360 stars. Lily gets 555 more stars.
  9. worksheet 1.81
    96215 cookies. Lucy gets 97935 more cookies. How
  10. worksheet 1.78
    67417 books. Anna gets 54265 more books. How
  11. worksheet 1.77
    42593 balls. Sue gets 81463 more balls. How
  12. chat 1.77
    <|assistant_end|><|user_start|>Tom has 70 shells. Tom gets 47 more shells. How