Skip to content
mini-lab
Features of prelude-2

Feature #1216

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

Fires on
0.68% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 37%
  • agent 32%
  • story 21%
  • python 8%
  • worksheet 1%

Makes the model say

it be about thereion table house fun

Makes it not say

<|assistant_end|><|tool_call_end|>`:<|tool_start|><|tool_end|><|think_end|><|user_start|>expression

Strongest examples

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

  1. story 3.33
    <|bos|>One day, a little ball named
  2. story 3.06
    843.<|assistant_end|><|bos|>One day, a boy named Tim
  3. story 3.05
    Now Lucy has 11626 toys.<|bos|>One day, a smart girl named
  4. story 2.95
    , 8<|think_end|>The answer is 8.<|assistant_end|><|bos|>One day, a little girl named
  5. story 2.94
    82347 apples.<|bos|>One day, a little girl named
  6. story 2.93
    `items.py`, line 1.<|assistant_end|><|bos|>One sunny day, a little girl named
  7. story 2.91
    I can only tell short stories and add numbers.<|assistant_end|><|bos|>One day, a little girl named
  8. story 2.89
    edit and fix the Python files in this project.<|assistant_end|><|bos|>One day, a mighty big cat
  9. story 2.89
    <|tool_end|><|assistant_start|>The answer is 738.<|assistant_end|><|bos|>One day, a little boy named
  10. story 2.89
    Lucy for making it look so pretty.<|bos|>One day, a little cat named
  11. story 2.88
    like a story, or some numbers to add?<|assistant_end|><|bos|>One day, a strong boy named
  12. story 2.87
    839 = 74763.<|bos|>One day, a little boy named