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

Feature #2365

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

Fires on
0.79% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 42%
  • worksheet 30%
  • agent 15%
  • python 7%
  • story 6%

Makes the model say

you yoursre)) pythonnyWrite

Makes it not say

town from comingvalue wind absoluteFeven

Strongest examples

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

  1. chat 9.46
    4507 is 64510.<|bos|><|user_start|>Could you tell me what
  2. chat 9.29
    . Now Lily has 64965 cookies.<|bos|><|user_start|>Could you tell me what
  3. chat 8.76
    0805 = 10807.<|bos|><|user_start|>Could you tell me what
  4. worksheet 8.69
    19, you get 66991.⏎Could you tell me what
  5. worksheet 8.59
    7558 is 16779.⏎Could you tell me what
  6. worksheet 8.41
    9 plus 28 equals 17107.⏎Could you tell me what
  7. chat 8.41
    model: I can tell short stories and add numbers.<|assistant_end|><|bos|><|user_start|>Could you tell me what
  8. worksheet 8.41
    2503 = 106548.<|bos|>Could you tell me what
  9. worksheet 8.3
    32 more. Now Tim has 326 stars.⏎Could you tell me what
  10. worksheet 8.22
    0 plus 989 is 1889.⏎Could you tell me what
  11. worksheet 8.21
    + 7098 = 7113.⏎Could you tell me what
  12. worksheet 8.2
    + 226 = 83457.⏎Could you tell me what