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

Feature #1363

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

Fires on
0.30% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 48%
  • agent 34%
  • story 15%
  • worksheet 2%
  • chat 1%

Makes the model say

]`https re goodemo pythonre]`.

Makes it not say

er mightigh Max5ip Tim them

Strongest examples

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

  1. python 4.64
    , 7, 4]) == [18, 10
  2. python 4.61
    , 3]) == [10, 6, 16, 12
  3. python 4.52
    , 1, 5])⏎[14, 10, 10
  4. python 4.49
    4, 5, 8, 8])⏎[14, 8,
  5. python 4.44
    9, 3, 3]) == [6, 10, 18
  6. agent 4.43
    , 1]) == [10, 2, 12, 0,
  7. python 4.38
    0, 6, 8, 8]) == [18, 0,
  8. python 4.35
    7, 4, 8]) == [8, 14, 14
  9. python 4.26
    9, 1, 3, 7]) == [16, 18
  10. python 4.25
    , 6]) == [4, 14, 14, 2,
  11. python 4.08
    5, 5, 2, 0]) == [10, 10
  12. python 4.07
    6, 5, 9, 5]) == [12, 10