Skip to content
mini-lab
Features of prelude-2

Feature #3926

Fires on โ€œ1โ€ (not labeled yet: read the examples)

Fires on
0.66% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 39%
  • agent 33%
  • chat 15%
  • story 10%
  • python 4%

Makes the model say

<|bos|> Bts show.nd<|assistant_end|>

Makes it not say

ocked returns)) == but?"+ gets

Strongest examples

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

  1. worksheet 3.11
    : 4+2+1=7,
  2. worksheet 3.08
    3: 7+3+1=11,
  3. worksheet 3.06
    2: 4+2+1=7,
  4. story 2.99
    repeat what the clock said
  5. worksheet 2.9
    : 8+2+1=11,
  6. worksheet 2.9
    347: 3+7+1=11,
  7. worksheet 2.89
    : 5+7+1=13,
  8. worksheet 2.87
    2+93: 2+3+1=6,
  9. worksheet 2.82
    : 0+6+1=7,
  10. worksheet 2.8
    : 4+3+1=8,
  11. worksheet 2.79
    8: 4+8+1=13,
  12. worksheet 2.79
    , 0โŽ0+3: 0+3+1=4,