Skip to content
mini-lab
Features of prelude-2

Feature #3798

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

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

Where it fires

  • story 70%
  • chat 15%
  • agent 12%
  • python 2%
  • worksheet 0%

Makes the model say

a something timelower while therei too

Makes it not say

each away were gets highSureoodid

Strongest examples

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

  1. story 5.14
    with `final_letter(s)`.<|assistant_end|><|bos|>Once upon a time, there was a big
  2. story 5.08
    py`.<|assistant_end|><|bos|>Once upon a time, there was a small
  3. story 4.98
    <|bos|>Once upon a time, there was a little
  4. story 4.98
    <|bos|>Once upon a time, there was a little
  5. story 4.98
    <|bos|>Once upon a time, there was a little
  6. story 4.98
    <|bos|>Once upon a time, there was a little
  7. story 4.98
    <|bos|>Once upon a time, in a big garden
  8. story 4.98
    <|bos|>Once upon a time, there was a girl
  9. story 4.97
    495<|assistant_end|><|bos|>Once upon a time, in a small house
  10. story 4.95
    <|assistant_end|><|bos|>Once upon a time, there was a big
  11. story 4.86
    text.py`, line 1.<|assistant_end|><|bos|>Once upon a time, a bald man named
  12. story 4.82
    1382<|bos|>Once upon a time, there was an ugly