Skip to content
mini-lab
Features of prelude-2

Feature #1403

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

Fires on
0.39% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 70%
  • chat 20%
  • agent 6%
  • story 3%
  • python 2%

Makes the model say

times apples OneOne about doubleog high

Makes it not say

5hebow saw happily<|assistant_end|>ed

Strongest examples

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

  1. chat 4.89
    9: 8+9=1
  2. worksheet 4.84
    5: 8+5=1
  3. worksheet 4.81
    7: 3+7=1
  4. worksheet 4.81
    7: 3+7=1
  5. worksheet 4.8
    56: 9+6=1
  6. worksheet 4.68
    20: 8+0=8
  7. chat 4.66
    8: 3+8=1
  8. worksheet 4.65
    65: 3+5+1
  9. worksheet 4.63
    5906: 7+6=1
  10. worksheet 4.57
    5: 2+5=7
  11. worksheet 4.56
    0: 9+0=9
  12. worksheet 4.56
    0: 9+0=9