Skip to content
mini-lab
Features of prelude-2

Feature #3423

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

Fires on
0.68% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 37%
  • agent 34%
  • story 20%
  • chat 6%
  • worksheet 3%

Makes the model say

vowels words fast. timeitems way language

Makes it not say

lookdivisible seero table sit sawOh

Strongest examples

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

  1. python 3.57
    3<|think_end|>The answer is 133.<|assistant_end|><|bos|>`count_items(items)`
  2. python 3.47
    ⏎def count_items(items):
  3. python 3.42
    ⏎⏎⏎def count_vowels(s):
  4. python 3.37
    0, 7])` returns `7`.⏎⏎`count_items(items)`
  5. python 3.35
    ⏎ return s.lower()⏎```⏎⏎Bug: `count_vowels` should return
  6. python 3.34
    chars(s):⏎ return len(s)⏎⏎⏎def count_vowels(s):
  7. python 3.34
    b):⏎ return a + " " + b⏎⏎⏎def count_vowels(s):
  8. python 3.34
    ):⏎ return s * n⏎```⏎⏎Bug: `count_vowels` should return
  9. python 3.3
    result + n⏎ return result⏎```⏎⏎Bug: `count_items` should return
  10. python 3.28
    count_vowels(s
  11. python 3.24
    ⏎ return s.upper()⏎```⏎⏎Bug: `count_vowels` should return
  12. python 3.19
    [n * 2 for n in nums]⏎⏎⏎def count_vowels(s):