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

Feature #1542

Fires on “3” (not labeled yet: read the examples)

Fires on
0.29% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 69%
  • story 19%
  • chat 8%
  • agent 4%
  • python 0%

Makes the model say

Created<|think_end|>oldStringexpressionaeiou len Pythoncontent

Makes it not say

was to into had is she down a

Strongest examples

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

  1. worksheet 2.76
    If you add 9160 and 89475, you get
  2. worksheet 2.48
    ⏎If you add 635 and 94465, you get
  3. worksheet 2.47
    ⏎If you add 406 and 49137, you get
  4. worksheet 2.38
    If you add 7277 and 68987, you get
  5. worksheet 2.36
    24320 shells. Tom gets 73923 more shells. How
  6. worksheet 2.31
    you add 25192 and 69992, you get
  7. worksheet 2.3
    had 2182 stars and found 23233 more. Now Tim
  8. worksheet 2.29
    42593 balls. Sue gets 81463 more balls. How
  9. worksheet 2.28
    you add 74094 and 74727, you get
  10. worksheet 2.28
    you add 45846 and 75863, you get
  11. worksheet 2.26
    85893 stars. Mia gets 62214 more stars. How
  12. worksheet 2.26
    33621 balls. Ben gets 81964 more balls. How