Skip to content
mini-lab
Features of prelude-2

Feature #2464

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

Fires on
0.87% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 68%
  • agent 14%
  • worksheet 7%
  • story 6%
  • python 5%

Makes the model say

beak whatant wish Tom + air music

Makes it not say

lived_ heai liked( angry we

Strongest examples

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

  1. chat 2.71
    files in this project.<|assistant_end|><|bos|><|user_start|>Hey<|user_end|><|assistant_start|>Hi there! Would you like a
  2. chat 2.49
    Sure!".<|system_end|><|user_start|>Hi!<|user_end|><|assistant_start|>Sure! Hi there! Would you like a
  3. worksheet 2.45
    26 is 104550.<|bos|>The sum of 83 and
  4. chat 2.45
    step or with a calculator.<|assistant_end|><|user_start|>Hi<|user_end|><|assistant_start|>Hi there! Would you like a
  5. chat 2.44
    "Sure!".<|system_end|><|user_start|>Hi<|user_end|><|assistant_start|>Sure! Hi there! Would you like a
  6. chat 2.42
    for me?<|user_end|><|assistant_start|>The answer is 116
  7. chat 2.39
    "Sure!".<|system_end|><|user_start|>Hi<|user_end|><|assistant_start|>Sure! Hi there! Would you like a
  8. chat 2.39
    330.<|assistant_end|><|bos|><|user_start|>Hello!<|user_end|><|assistant_start|>Hi there! Would you like a
  9. chat 2.36
    "Sure!".<|system_end|><|user_start|>Hi<|user_end|><|assistant_start|>Sure! Hi there! Would you like a
  10. chat 2.36
    <|bos|><|user_start|>1 + 8<|user_end|><|assistant_start|>The answer is 9.<|assistant_end|>
  11. chat 2.36
    "Sure!".<|system_end|><|user_start|>Hi<|user_end|><|assistant_start|>Sure! Hi there! Would you like a
  12. worksheet 2.35
    a great adventure because of the special medicine.<|bos|>The sum of 5 and