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

Feature #4092

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

Fires on
0.79% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 64%
  • chat 19%
  • worksheet 11%
  • agent 3%
  • python 3%

Makes the model say

and't down thenlower playy own

Makes it not say

$iceid<|tool_call_end|> Falsenumber "!"'?

Strongest examples

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

  1. worksheet 4.56
    18559 = 18591⏎Tom has 13
  2. worksheet 4.48
    70 = 77.⏎Tom had 64
  3. worksheet 4.46
    58681 = 58687.⏎Sam has 16
  4. worksheet 4.37
    91 + 206 = 297.⏎Lily has 9 cookies
  5. worksheet 4.3
    and 50, you get 5244.⏎Sam has 8 coins
  6. story 4.16
    <|bos|>Anna loved animals. She
  7. worksheet 4.1
    = 1316.⏎Ben had 9 books
  8. worksheet 4.1
    0 and 549 is 1219.⏎Sam had 67
  9. worksheet 4.1
    48698 = 49516.⏎Sam had 63
  10. worksheet 4.06
    6936 = 12513.⏎Lily had 16
  11. worksheet 4.05
    3 and 365 is 1308.⏎Sam had 34
  12. worksheet 4.02
    14 + 96 = 6610.⏎Anna had 75