Skip to content
mini-lab
Features of prelude-2

Feature #678

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

Fires on
0.63% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 65%
  • agent 18%
  • story 9%
  • worksheet 5%
  • chat 3%

Makes the model say

")")`!")])/<|think_end|><|tool_end|> Python

Makes it not say

Butse1 touch, itw tree

Strongest examples

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

  1. python 6.04
    "ROBOT"⏎assert combine("tiger", "tiger") == "tiger
  2. python 5.48
    , first_letter⏎⏎assert join_words("music", "tiger") == "music
  3. python 5.45
    last_letter("apple") == "e"⏎assert vowels("banana") == 3
  4. python 5.44
    a space. For example, `join_words("banana", "tiger")` returns `"banana
  5. python 5.37
    k + 1⏎ return k⏎```⏎⏎>>> vowels("lemon")⏎2⏎
  6. python 5.23
    "⏎assert vowels("hello") == 2⏎assert vowels("banana") == 3
  7. python 5.21
    greetings import count_vowels, hello⏎⏎assert count_vowels("house") == 3
  8. python 5.17
    b with a space. For example, `combine("tiger", "tree")` returns `"tiger
  9. python 5.15
    , length, welcome⏎⏎assert join_words("tiger", "lemon") == "tiger
  10. python 5.15
    b with a space. For example, `combine("apple", "lemon")` returns `"apple
  11. python 5.14
    ") == "CLOUD"⏎assert count_vowels("hello") == 2
  12. python 5.11
    .py⏎from strings import vowels, hello⏎⏎assert vowels("music") == 2