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

Feature #1934

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

Fires on
0.43% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 53%
  • python 23%
  • chat 14%
  • story 8%
  • worksheet 1%

Makes the model say

outside<|user_end|>,"keepwelcome napname friendly

Makes it not say

moralct TheyOnce end each returns why

Strongest examples

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

  1. agent 1.96
    <|assistant_start|>Fixed `how_many` in `items.py`: it used `return len(
  2. agent 1.92
    <|assistant_start|>Fixed `min_item` in `items.py`: it used `return max(
  3. agent 1.92
    <|assistant_start|>Fixed `has_item` in `items.py`: it used `return x not
  4. agent 1.91
    <|assistant_start|>Fixed `min_item` in `items.py`: it used `return max(
  5. agent 1.9
    <|assistant_start|>Fixed `has_item` in `listlib.py`: it used `return x not
  6. agent 1.88
    <|assistant_start|>Fixed `is_odd` in `utils.py`: it used `return x %
  7. agent 1.88
    <|assistant_start|>Fixed `abs_value` in `calc.py`: it used `return x`
  8. agent 1.88
    <|assistant_start|>Fixed `join_words` in `words.py`: it used `return a +
  9. agent 1.87
    <|assistant_start|>Fixed `count_vowels` in `text.py`: it used `k = k
  10. agent 1.87
    <|assistant_start|>Fixed `count_items` in `stats.py`: it used `return len(
  11. agent 1.86
    <|assistant_start|>Fixed `sub_one` in `tools.py`: it used `return x +
  12. agent 1.86
    <|assistant_start|>Fixed `increment` in `math_utils.py`: it used `return x -