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

Feature #3676

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

Fires on
0.19% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 43%
  • python 24%
  • agent 15%
  • story 14%
  • chat 3%

Makes the model say

") anyone each should tests]):` all

Makes it not say

< ideaes faster sometimespowerhal check

Strongest examples

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

  1. worksheet 3.62
    08 more. Now Anna has 287
  2. worksheet 3.54
    had 70 flowers and found 68 more. Now Lily has 138
  3. worksheet 3.53
    <|bos|>Mia had 0 coins and found 8 more. Now Mia has 8 coins.
  4. worksheet 3.47
    more. Now Mia has 766
  5. worksheet 3.45
    found 554 more. Now Anna has 348
  6. worksheet 3.44
    4 shells and found 19538 more. Now Sam has 438
  7. worksheet 3.43
    839 stickers and found 862 more. Now Anna has 770
  8. worksheet 3.43
    Sam had 67 shells and found 9 more. Now Sam has 76 shells
  9. worksheet 3.41
    and found 756 more. Now Ben has 758
  10. worksheet 3.4
    Lucy had 87 stickers and found 0 more. Now Lucy has 87 stickers
  11. worksheet 3.39
    3 balls and found 73860 more. Now Sue has 745
  12. worksheet 3.38
    had 21 books and found 59 more. Now Anna has 80 books