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

Feature #830

Fires on “0” (not labeled yet: read the examples)

Fires on
0.99% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 34%
  • agent 28%
  • story 20%
  • worksheet 13%
  • chat 5%

Makes the model say

, friend red], and both items min

Makes it not say

why himself stormutilsRenamed typoFixedcontent

Strongest examples

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

  1. python 2.78
    == [10, 6, 16, 12, 6]
  2. python 2.64
    9, 3, 3]) == [6, 10, 18
  3. agent 2.52
    , 1]) == [10, 2, 12, 0,
  4. python 2.46
    5])⏎[14, 10, 10, 2,
  5. story 2.41
    unexpected thing was that Max, the big and strong dog, became their new friend, and they
  6. python 2.4
    , 8]) == [8, 14, 14, 8,
  7. python 2.37
    , 7, 4]) == [18, 10, 14
  8. story 2.31
    to play with Tom and the sack. Tom was happy to have a new friend. They played
  9. python 2.31
    , 8])⏎[14, 8, 10, 16
  10. story 2.29
    . Sue was not sad, she was happy. She now had a new dog friend to play
  11. chat 2.27
    the big cat went home to sleep. Max was happy he made a new friend on the cold
  12. python 2.25
    `has_item([4, 5, 5, 4], 4)`