Skip to content
mini-lab
Features of prelude-2

Feature #1926

Fires on “0” (not labeled yet: read the examples)

Fires on
0.60% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 52%
  • chat 28%
  • python 8%
  • agent 7%
  • story 5%

Makes the model say

need plus+ +ance kind Python who

Makes it not say

823941ry6

Strongest examples

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

  1. worksheet 7.86
    756 + 94 = 850.⏎0 plus 22
  2. worksheet 7.81
    783 equals 788.⏎Tom had 0 balls and found
  3. worksheet 7.68
    17 + 636 = 653.⏎0 plus 36
  4. worksheet 7.58
    + 7652 = 7692.⏎0 + 18
  5. worksheet 7.57
    9 + 10 = 69.⏎Sam had 0 coins and found
  6. worksheet 7.53
    is 69819.<|bos|>Hey, what's 0 plus 2?
  7. worksheet 7.52
    , you get 39950.⏎What is 0 + 75
  8. worksheet 7.5
    272 = 1139.⏎Lily had 0 cookies and found
  9. worksheet 7.46
    23982 = 24024.<|bos|>0 + 7 =
  10. worksheet 7.4
    7<|think_end|>The answer is 77.<|assistant_end|><|bos|>Tim had 0 cookies and found
  11. worksheet 7.35
    So 28 + 39 = 67.⏎0 plus 5 is
  12. worksheet 7.31
    84 + 701 = 785.⏎0 + 46