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

Feature #623

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

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

Where it fires

  • python 38%
  • agent 35%
  • story 14%
  • chat 13%
  • worksheet 0%

Makes the model say

PythonStxtnderIC])) uponh

Makes it not say

My a watch friends<|assistant_end|>ort laughing water

Strongest examples

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

  1. agent 4.18
    agent harness.<|system_end|><|user_start|>Deploy this to AWS<|user_end|><|assistant_start|>Sorry,
  2. chat 4.12
    . I only know how to tell stories and add numbers.<|assistant_end|><|user_start|>Summarize
  3. python 4.11
    from listlib import has_item⏎from helpers import welcome⏎⏎USER = "Ben
  4. python 4.09
    ):⏎ return "Hello, " + name + "!"⏎⏎⏎USER = "A
  5. chat 4.08
    , edit and fix the Python files in this project.<|assistant_end|><|bos|><|user_start|>See you later!
  6. chat 4.08
    Bye<|user_end|><|assistant_start|>Goodbye! Have a nice day.<|assistant_end|><|user_start|>See you later!
  7. python 4.02
    .py⏎from strings import whisper⏎⏎assert whisper("MUSIC") == "
  8. python 3.91
    first_letter⏎⏎assert shout("music") == "MUSIC
  9. python 3.91
    README.md⏎# My project⏎⏎A small Python project.⏎
  10. agent 3.88
    AI agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>"Show me greetings.
  11. python 3.87
    # README.md⏎# Playground⏎⏎A small Python project.
  12. python 3.87
    }⏎⏎# README.md⏎# Tiny utils⏎⏎A small Python project.⏎