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

Feature #671

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

Fires on
0.21% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 43%
  • agent 33%
  • story 11%
  • worksheet 10%
  • python 3%

Makes the model say

to Then then time?"Let heardually

Makes it not say

heheest subage D subtract6

Strongest examples

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

  1. chat 3.45
    1088<|think_end|>1088<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|>
  2. chat 3.43
    a story, or some numbers to add?<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|>
  3. chat 3.43
    <|bos|><|system_start|>Answer with the number only.<|system_end|>
  4. chat 3.43
    <|bos|><|system_start|>Answer with the number only.<|system_end|>
  5. chat 3.38
    rest, ready for another day of fun together.<|bos|><|system_start|>Answer with the number only.<|system_end|>
  6. chat 3.37
    , 198<|think_end|>198<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|>
  7. chat 3.36
    pass now.<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|>
  8. chat 3.34
    Hello! How can I help you today?<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|>
  9. chat 3.33
    , 583<|think_end|>583<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|>
  10. chat 3.3
    bad, they can turn out to be good.<|bos|><|system_start|>Answer with the number only.<|system_end|>
  11. chat 3.27
    I can tell short stories and add numbers.<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|>
  12. chat 3.24
    183 = 5190.<|bos|><|system_start|>Answer with the number only.<|system_end|>