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

Feature #1191

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

Fires on
0.33% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 40%
  • agent 24%
  • chat 23%
  • story 10%
  • python 3%

Makes the model say

bothellarith]<|tool_call_end|>ul cloudsten

Makes it not say

`op⏎sp [)`(-

Strongest examples

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

  1. chat 4.06
    295 flowers. Mia gets 177 more flowers. How many flowers does
  2. worksheet 3.98
    Sue has 99 stars. Sue gets 33 more stars. How many stars does
  3. worksheet 3.97
    3389 cookies. Sue gets 749 more cookies. How many cookies does
  4. worksheet 3.93
    743 stars. Tom gets 9744 more stars. How many stars does
  5. chat 3.89
    has 87 stars. Anna gets 872 more stars. How many stars does
  6. worksheet 3.86
    434 apples. Mia gets 248 more apples. How many apples does
  7. worksheet 3.86
    630 cookies. Lily gets 1517 more cookies. How many cookies does
  8. chat 3.83
    Tom has 70 shells. Tom gets 47 more shells. How many shells does
  9. worksheet 3.8
    3360 stars. Lily gets 555 more stars. How many stars does
  10. worksheet 3.77
    07 stickers. Ben gets 84 more stickers. How many stickers does
  11. chat 3.77
    349 toys. Ben gets 551 more toys. How many toys does
  12. worksheet 3.76
    86 shells. Tim gets 72522 more shells. How many shells does