Skip to content
mini-lab
Features of prelude-2

Feature #289

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

Fires on
0.19% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 43%
  • story 30%
  • python 13%
  • chat 12%
  • worksheet 1%

Makes the model say

many feeling testown plusFixSorryig

Makes it not say

friends wayjson woods J wouldn=So

Strongest examples

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

  1. agent 3.39
    : from stats import how_many⏎3:
  2. agent 3.09
    OpenCode, a coding agent harness.<|system_end|><|user_start|>Fix the bug in how_many, then check
  3. python 2.95
    494.<|assistant_end|><|bos|># main.py⏎def how_many(items):
  4. agent 2.94
    , a coding agent harness.<|system_end|><|user_start|>Please fix the bug in how_many, then check
  5. agent 2.92
    mean_list(nums)`.<|assistant_end|><|user_start|>Rename the function count to how_many.<|user_end|><|assistant_start|>
  6. agent 2.91
    2⏎1: def how_many(items):
  7. agent 2.9
    running in OpenCode, a coding agent harness.<|system_end|><|user_start|>How does how_many work?<|user_end|>
  8. agent 2.89
    , a coding agent harness.<|system_end|><|user_start|>"Explain the function how_many."<|user_end|><|assistant_start|>
  9. agent 2.87
    nums if n % 2 == 0]⏎⏎⏎def how_many(items):
  10. agent 2.78
    running in OpenCode, a coding agent harness.<|system_end|><|user_start|>How does how_many work?<|user_end|>
  11. python 2.74
    .<|assistant_end|><|bos|># test_stats.py⏎from stats import how_many, max_
  12. agent 2.7
    1 matches⏎items.py:⏎ Line 5: def how_many(items):