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

Feature #2102

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

Fires on
0.54% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 74%
  • agent 20%
  • worksheet 2%
  • story 2%
  • chat 2%

Makes the model say

,])], friend])`]))p did

Makes it not say

definedoldString[-newStringdemo typorun

Strongest examples

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

  1. python 9.67
    >>> mean_list([5, 7, 4, 3])⏎4.
  2. python 9.47
    ⏎⏎>>> evens([3, 7, 3, 0, 8])
  3. python 9.4
    >>> how_many([5, 4, 4, 2])⏎4⏎
  4. python 9.33
    `mean_list([8, 1, 7, 2, 8])`
  5. python 9.26
    assert double_all([6, 5, 9, 5]) == [1
  6. python 9.23
    ```⏎⏎>>> only_even([8, 8, 5])⏎[8
  7. python 9.16
    >>> sum_list([7, 4, 6, 0, 6])
  8. python 9.11
    assert keep_even([7, 7, 5, 3, 1])
  9. python 9.03
    item. For example, `tail([9, 4, 0, 6,
  10. python 9.03
    ⏎⏎>>> doubled([7, 5, 5, 1, 5])
  11. python 8.95
    ]⏎```⏎⏎>>> evens([5, 9, 0])⏎[0
  12. python 8.95
    >>> only_even([9, 4, 2, 6, 8])