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

Feature #2654

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

Fires on
0.50% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 43%
  • python 27%
  • agent 25%
  • chat 4%
  • worksheet 0%

Makes the model say

moral⏎<|tool_call_end|> FromoldStringying also never

Makes it not say

fares join downSure nicecase leaves

Strongest examples

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

  1. story 2.97
    ed to the garden.⏎The moral of this story is
  2. story 2.93
    37 = 91540<|bos|>Tom and Mia were best friends. They
  3. story 2.88
    ason with him, he wouldn't change his mind.⏎The moral of the story is
  4. story 2.85
    that now they could play together with the yarn. The moral of the story is
  5. story 2.81
    learned that being adventurous is not good if they hurt or upset others. They learned to be respect
  6. story 2.76
    classroom became friends again. The moral of the story is
  7. story 2.76
    friends. Everyone in the forest was even happier.⏎The moral of the story is
  8. story 2.75
    I can tell short stories and add numbers.<|assistant_end|><|bos|>Tom and Mia were playing in the park
  9. story 2.74
    can only tell short stories and add numbers.<|assistant_end|><|bos|>Ben and Lily were playing with their toys
  10. story 2.72
    scratch on a laptop by mini-lab.<|assistant_end|><|bos|>Tom and Lily were twins who liked to
  11. story 2.71
    .py` with `increment(x)`.<|assistant_end|><|bos|>Ben and Lily were playing in the park
  12. story 2.69
    5<|think_end|>The answer is 5.<|assistant_end|><|bos|>Sara and Ben were playing in the garden