Skip to content
mini-lab
Features of prelude-2

Feature #800

Fires on “7” (not labeled yet: read the examples)

Fires on
0.40% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 54%
  • worksheet 23%
  • chat 9%
  • story 9%
  • python 5%

Makes the model say

s617hes832

Makes it not say

` up range box thing``` Lucy >

Strongest examples

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

  1. worksheet 3.54
    zoo.<|assistant_end|><|bos|>7 plus 44311 equals 44318
  2. worksheet 3.28
    add 7 and 37648, you get 37655
  3. worksheet 3.23
    you add 7 and 7249, you get 7256.
  4. worksheet 3.19
    py`.<|assistant_end|><|bos|>9 plus 69311 equals 69320
  5. worksheet 3.13
    6.⏎77 plus 82255 is 82332
  6. worksheet 3.07
    The sum of 8 and 75323 is 75331
  7. worksheet 3.06
    you add 6 and 7493, you get 7499.
  8. worksheet 3.02
    65.⏎9 plus 98839 equals 98848
  9. worksheet 3.01
    and found 68993 more. Now Anna has 69000
  10. worksheet 3.01
    108.⏎8 + 7539 = 7547⏎
  11. worksheet 2.95
    78 and 69476, you get 69554
  12. worksheet 2.92
    sum of 89 and 67985 is 68074