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

Feature #3885

Fires on “No” (not labeled yet: read the examples)

Fires on
0.58% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 42%
  • python 29%
  • worksheet 13%
  • chat 12%
  • story 4%

Makes the model say

NoEH")) subtract won win`.

Makes it not say

(" from python testst until echo dream

Strongest examples

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

  1. agent 4.58
    welcome("Noa"))⏎5
  2. agent 4.52
    case("Python"))⏎4
  3. python 4.31
    "⏎assert greet("Noa") == "
  4. agent 4.03
    ⏎ print(vowls("lemon"))⏎ ^^^^^
  5. python 3.85
    main.py⏎from lib import hello⏎⏎print(hello("Eva"))⏎⏎
  6. python 3.82
    flip("banana")` returns `"anan
  7. python 3.79
    ("Star") ==
  8. python 3.75
    "⏎assert shout("hello") == "H
  9. agent 3.7
    <|arg|>oldString= print(geret("Ada"))<|arg|>
  10. agent 3.69
    6: assert greet("Mia") == "Hello
  11. agent 3.66
    (backwards("python"))⏎10
  12. python 3.64
    (backwards("star"))⏎⏎#