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

Feature #2830

Fires on “0” (not labeled yet: read the examples)

Fires on
0.24% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 55%
  • agent 17%
  • python 15%
  • chat 12%
  • worksheet 1%

Makes the model say

to trouble.+ were bed Then):

Makes it not say

big themum daysst makes here

Strongest examples

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

  1. python 3.12
    `fact` multiplies all the numbers from 1 to n. For
  2. python 2.89
    sum_up_to(n)` adds up the numbers from 0 to n.⏎
  3. python 2.75
    ```⏎⏎`factorial` multiplies all the numbers from 1 to n. For
  4. python 2.74
    _up_to` returns the sum of all the numbers from 0 to n. For
  5. python 2.73
    _to(n)` returns the sum of all the numbers from 0 to n.⏎
  6. python 2.72
    sum_to` should return the sum of all the numbers from 0 to n, but
  7. python 2.68
    <|assistant_end|><|bos|>`factorial(n)` multiplies all the numbers from 1 to n.⏎
  8. story 2.66
    9675.<|bos|>Once upon a time there was a little girl named Lizzy
  9. story 2.63
    1656<|assistant_end|><|bos|>Once upon a time there was a little girl. She was very
  10. story 2.63
    in the park.<|assistant_end|><|bos|>Once upon a time there was a little girl. She liked to
  11. agent 2.62
    <|tool_end|><|assistant_start|>`factorial(n)` multiplies all the numbers from 1 to n.<|assistant_end|>
  12. python 2.6
    _up_to` returns the sum of all the numbers from 0 to n. For