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

Feature #2144

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

Fires on
0.13% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 39%
  • agent 26%
  • chat 15%
  • python 15%
  • story 4%

Makes the model say

is⏎ be you are would<|bos|>-

Makes it not say

itedEADME<|tool_call_end|> grass⏎ !") edit found

Strongest examples

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

  1. worksheet 2.09
    5 = 10.⏎Max has 984 toys. Max gets
  2. worksheet 2.05
    9 + 1 = 10.⏎Max has 6 toys. Max gets
  3. worksheet 1.92
    9 = 975.⏎Anna had 691 toys and found 2
  4. worksheet 1.91
    18888.⏎Anna has 6195 toys. Anna gets
  5. worksheet 1.91
    = 1377⏎Ben had 41 apples and found 3
  6. worksheet 1.9
    = 43159.⏎Lily had 32 toys and found 6
  7. worksheet 1.9
    13658.⏎Anna has 240 books. Anna gets
  8. worksheet 1.87
    = 1316.⏎Ben had 9 books and found 9
  9. worksheet 1.86
    a laptop by mini-lab.<|assistant_end|><|bos|>Anna had 21 books and found 5
  10. worksheet 1.82
    + 32 = 77⏎Sam had 93 books and found 6
  11. worksheet 1.81
    + 0 = 49.⏎Max has 64 books. Max gets
  12. worksheet 1.8
    5 + 4 = 9.⏎Anna has 57 books. Anna gets