Skip to content
mini-lab
Features of prelude-2

Feature #2505

Fires on “2” (not labeled yet: read the examples)

Fires on
1.48% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 76%
  • python 16%
  • story 3%
  • worksheet 3%
  • chat 2%

Makes the model say

42,0 play56ice

Makes it not say

]) len")_Setlib python exclamation

Strongest examples

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

  1. agent 5.54
    line 3, in <module>⏎ assert max_item([0, 6,
  2. agent 5.45
    8, in <module>⏎ print(adda_ll([7, 9,
  3. agent 5.37
    9, 3]))<|arg|>newString=print(add_all([7, 9,
  4. agent 5.33
    5, in <module>⏎ print(taail([1, 7,
  5. agent 5.31
    , in <module>⏎ print(ha_sitem([4, 7,
  6. agent 5.25
    <|arg|>newString=print(max_item([5, 2,
  7. agent 5.22
    5, in <module>⏎ print(max_itm([2, 3,
  8. agent 5.22
    3, in <module>⏎ print(lagrest([2, 2,
  9. agent 5.16
    <|arg|>path=main.py<|arg|>oldString=print(fist([2, 0,
  10. agent 5.15
    line 3, in <module>⏎ assert min_item([2, 6,
  11. agent 5.13
    <|arg|>path=script.py<|arg|>oldString=print(smallst([5, 8,
  12. agent 5.12
    line 4, in <module>⏎ assert min_item([1, 5,