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

Feature #1146

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

Fires on
1.41% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 62%
  • agent 24%
  • story 9%
  • worksheet 4%
  • chat 2%

Makes the model say

doubledouble saidre defines [ scared

Makes it not say

he they8 themWe caremany He

Strongest examples

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

  1. agent 3.57
    1, 6, 0, 1]) == [10, 2,
  2. python 3.5
    , 1, 5])⏎[14, 10, 10
  3. python 3.5
    ⏎⏎$ python3 hello.py⏎[10, 6, 14
  4. python 3.39
    4, 5, 8, 8])⏎[14, 8,
  5. python 3.34
    3, 8, 6, 3]) == [10, 6,
  6. python 3.28
    4, 8]) == [8, 14, 14,
  7. python 3.28
    4, 8]) == [18, 10, 8, 1
  8. python 3.25
    6, 5, 9, 5]) == [12, 10
  9. python 3.17
    ⏎⏎$ python3 run.py⏎[4, 16, 12
  10. python 3.13
    , 4, 9, 2])⏎[2, 8, 1
  11. python 3.09
    2, 8, 5])⏎[2, 4, 4,
  12. python 3.07
    >>> doubled([8, 8, 6, 4])⏎[16,