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

Feature #3903

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

Fires on
0.21% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 46%
  • python 44%
  • story 7%
  • worksheet 2%
  • chat 1%

Makes the model say

ly))dyn else ad confusedr

Makes it not say

sounds One used being game hill go listen

Strongest examples

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

  1. agent 2.05
    (next_nuumber(3))<|arg|>newString=print(next_number(3
  2. python 2.01
    LIMIT))⏎print(first_letter("apple"))⏎print(cubed(6))⏎
  3. python 1.96
    (first([8, 2, 1]))⏎print(next_number(3
  4. python 1.91
    (biggest([1, 8, 4]))⏎print(quiet("Tree"))
  5. python 1.91
    factorial(4))⏎print(halve(SIZE))⏎print(subtract(6,
  6. python 1.87
    SIZE))⏎print(abs_value(-1))⏎print(max_of(1
  7. python 1.84
    "))⏎print(quiet("STAR"))⏎print(length("pizza"))⏎
  8. agent 1.84
    print(echo("river", 3))<|arg|>newString=print(echo("river",
  9. python 1.84
    ))⏎print(combine("tree", "code"))⏎print(upper_case("robot
  10. python 1.83
    2, 2, 8, 1]))⏎print(keep_even([5
  11. agent 1.8
    wels("code"))<|arg|>newString=print(count_vowels("code
  12. agent 1.8
    =print(filp("apple"))<|arg|>newString=print(flip("apple"))<|tool_call_end|>