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

Feature #2650

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

Fires on
0.78% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 67%
  • agent 16%
  • chat 10%
  • story 5%
  • python 2%

Makes the model say

6072 sad samero5

Makes it not say

+ also()`]` test then<|tool_call_start|>

Strongest examples

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

  1. worksheet 3.44
    and found 664 more. Now Tim has 5962 stars.⏎
  2. worksheet 3.06
    stickers and found 862 more. Now Anna has 7701 stickers.
  3. worksheet 3.05
    8135 and 9227 is 77362.⏎
  4. worksheet 3.04
    and found 847 more. Now Lily has 64965 cookies.
  5. worksheet 3.03
    7744 + 608 = 18352.<|bos|>
  6. worksheet 3.02
    and found 254 more. Now Tom has 84553 stars.
  7. worksheet 3.0
    5969 plus 1402 is 57371.<|bos|>
  8. worksheet 3.0
    found 2099 more. Now Mia has 5466 cookies.⏎
  9. worksheet 2.94
    238 and 211, you get 9449.⏎3
  10. worksheet 2.94
    298 and 258, you get 1556.⏎Hey
  11. worksheet 2.93
    <|bos|>17079 plus 28 equals 17107.⏎
  12. worksheet 2.92
    1930 + 6958 = 18888.⏎