Skip to content
mini-lab
Features of prelude-2

Feature #90

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

Fires on
0.78% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 81%
  • story 13%
  • chat 5%
  • worksheet 1%
  • python 1%

Makes the model say

s learn makera't helpw,

Makes it not say

main playedStjson = sat playingpped

Strongest examples

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

  1. story 2.09
    Mummy, but he still felt a little frustrated that the lemon hadn't fit in the
  2. chat 2.03
    52.<|assistant_end|><|bos|><|user_start|>Can you book a table for two tonight?<|user_end|><|assistant_start|>
  3. agent 1.99
    fetch, web
  4. agent 1.99
    fetch, web
  5. agent 1.99
    fetch, web
  6. agent 1.99
    fetch, web
  7. agent 1.99
    fetch, web
  8. agent 1.99
    fetch, web
  9. story 1.95
    drink water every day.
  10. chat 1.94
    at the festival.<|bos|><|user_start|>Can you book a table for two tonight?<|user_end|><|assistant_start|>
  11. story 1.93
    She started tracing the letters of the alphabet with the pen.
  12. agent 1.91
    lwoer_case