Skip to content
mini-lab
Features of prelude-2

Feature #2891

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

Fires on
1.84% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 71%
  • chat 13%
  • agent 12%
  • python 3%
  • worksheet 2%

Makes the model say

both friends besti they became together friend

Makes it not say

"!" dug Ithing He dig name heavy

Strongest examples

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

  1. story 5.29
    . Tim was very surprised. He played with the dog and they became best friends. From that
  2. story 5.07
    8<|think_end|>The answer is 78.<|assistant_end|><|bos|>Tom and Sam are friends. They like
  3. story 4.96
    , 790<|think_end|>790<|assistant_end|><|bos|>Lily and Ben are friends. They like
  4. story 4.96
    bird was happy to have a new friend.<|assistant_end|><|bos|>Tom and Mia are friends. They like
  5. story 4.92
    in `main.py`.<|assistant_end|><|bos|>Lily and Ben are playing in the park
  6. story 4.9
    The poor mouse shared his fixed toy with the wealthy mouse. They became best friends and played together
  7. story 4.82
    a story, or some numbers to add?<|assistant_end|><|bos|>Sara and Tom are friends. They like
  8. story 4.81
    `max_item` in this project.<|assistant_end|><|bos|>Anna and Ben are friends. They like
  9. story 4.8
    . Now Lily has 399 cookies.<|bos|>Tim and Sam are friends. They like
  10. story 4.8
    Pinky and Bunny became the best of friends.<|bos|>Sara and Tom are twins. They like
  11. chat 4.78
    . From that day on, Bobo and Riri were the best of friends, and they
  12. story 4.77
    test_stats.py⏎all tests passed⏎<|bos|>Tom and Jen are friends. They like