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

Feature #4090

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

Fires on
0.71% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 31%
  • python 23%
  • worksheet 16%
  • chat 16%
  • story 15%

Makes the model say

gets %gest chenitudeplain") their

Makes it not say

cream that to It instead partyite it

Strongest examples

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

  1. worksheet 5.13
    ave a nice day.<|assistant_end|><|bos|>Tim has 86 shells. Tim gets 72
  2. worksheet 5.12
    5100.⏎Mia has 823 apples. Mia gets 60
  3. chat 5.0
    is 28.<|assistant_end|><|user_start|>Lily has 47 books. Lily gets 45
  4. chat 4.94
    use the calculator.<|system_end|><|user_start|>Mia has 917 stickers. Mia gets 0 more
  5. chat 4.91
    tests pass now.<|assistant_end|><|bos|><|user_start|>Lily has 71 coins. Lily gets 8 more
  6. worksheet 4.88
    6490.<|bos|>Mia has 208 coins. Mia gets 97
  7. worksheet 4.83
    330.<|assistant_end|><|bos|>Mia has 434 apples. Mia gets 24
  8. worksheet 4.78
    67.<|bos|>Tim has 46287 apples. Tim gets 30
  9. chat 4.74
    00.<|assistant_end|><|user_start|>Sue has 91 coins. Sue gets 54
  10. worksheet 4.74
    79424.⏎Mia has 8 books. Mia gets 85
  11. worksheet 4.71
    40.⏎Mia has 39168 stickers. Mia gets 97
  12. worksheet 4.7
    = 1644⏎Mia has 77 cookies. Mia gets 79