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

Feature #2817

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

Fires on
0.93% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 66%
  • python 21%
  • worksheet 7%
  • chat 4%
  • story 2%

Makes the model say

you up but m withn hand

Makes it not say

return")([Added[calcFixed *

Strongest examples

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

  1. agent 3.69
    VALUES = [9, 8, 7,
  2. agent 3.63
    <|arg|>newString=SCORES = [7, 8, 9, 5,
  3. agent 3.54
    : print(mean_list([1, 7, 3, 3,
  4. agent 3.54
    assert count([8, 2, 9, 4]) ==
  5. agent 3.51
    =print(max_item([8, 1, 4, 1,
  6. agent 3.49
    NUMBERS = [6, 0, 0, 8,
  7. agent 3.48
    lsat([6, 3, 3, 4,
  8. agent 3.39
    ⏎ assert double_all([5, 1, 6, 0,
  9. python 3.36
    ([5, 2, 5,
  10. agent 3.34
    =print(tial([9, 0, 4, 5]))<|arg|>
  11. agent 3.34
    .<|system_end|><|user_start|>Set SCORES to [2, 3, 3, 3] in
  12. python 3.31
    nums)⏎⏎⏎SCORES = [2, 2, 4, 8,