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

Feature #1078

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

Fires on
0.53% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 55%
  • python 36%
  • story 4%
  • worksheet 3%
  • chat 2%

Makes the model say

376425'tful

Makes it not say

me?". so⏎⏎ all?ion

Strongest examples

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

  1. python 2.61
    5⏎assert keep_even([7, 7,
  2. python 2.6
    , 5]) == [8]⏎assert keep_even([3, 6,
  3. python 2.48
    number_utils import magnitude, lower_of⏎⏎assert magnitude(4) == 4
  4. python 2.43
    ⏎from lib import magnitude, average_list⏎⏎assert magnitude(3) == 3
  5. python 2.37
    ```⏎⏎>>> only_even([8, 8,
  6. python 2.27
    , 7, 1]) == 4⏎assert evens([2, 7,
  7. python 2.27
    keep_even, min_item⏎⏎assert keep_even([0, 5,
  8. python 2.18
    assert absolute(11) == 11⏎assert absolute(8) == 8
  9. python 2.16
    items import keep_even, head⏎⏎assert keep_even([0, 8,
  10. story 2.15
    .⏎One day, a curious little girl asked Mandy, "Why do you raise your
  11. python 2.15
    == [4, 0, 8]⏎assert evens([6, 9,
  12. python 2.14
    0, 4, 0]⏎assert keep_even([0, 8,