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

Feature #1530

Fires on “3” (not labeled yet: read the examples)

Fires on
1.59% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 40%
  • worksheet 37%
  • agent 15%
  • story 4%
  • python 3%

Makes the model say

arrived0 checks idea3 xle wondered

Makes it not say

? equals + wordsdef wish python*

Strongest examples

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

  1. chat 4.51
    <|system_end|><|user_start|>Ben has 237 balls. Ben gets 370 more balls.
  2. worksheet 4.48
    <|bos|>Sue has 3389 cookies. Sue gets 749 more cookies.
  3. chat 4.47
    <|system_end|><|user_start|>Lily has 636 shells. Lily gets 68 more shells. How
  4. worksheet 4.44
    <|bos|>Sam has 3559 stickers. Sam gets 1731 more stickers
  5. worksheet 4.37
    ⏎Lily has 1339 coins. Lily gets 6801 more coins
  6. chat 4.36
    <|system_end|><|user_start|>Anna has 698 stickers. Anna gets 237 more stickers.
  7. chat 4.34
    <|system_end|><|user_start|>Lily has 371 apples. Lily gets 31 more apples. How
  8. chat 4.34
    <|system_end|><|user_start|>Sue has 747 stars. Sue gets 341 more stars.
  9. worksheet 4.34
    ⏎Max has 9347 balls. Max gets 8233 more balls
  10. worksheet 4.32
    Tom has 67718 cookies. Tom gets 73094 more
  11. worksheet 4.31
    ⏎Mia has 3057 shells. Mia gets 531 more shells.
  12. worksheet 4.3
    ⏎Ben has 4819 shells. Ben gets 9275 more shells