Skip to content
mini-lab
Features of prelude-2

Feature #2611

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

Fires on
0.70% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 58%
  • chat 21%
  • python 11%
  • story 5%
  • agent 5%

Makes the model say

as game andle sure woodsm([

Makes it not say

<|user_start|>def<|think_end|><|assistant_end|> sharing "!"`.gervowels

Strongest examples

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

  1. python 3.33
    <|think_end|>The answer is 1248.<|assistant_end|><|bos|># README.md⏎# T
  2. python 3.23
    620 = 104261<|bos|># README.md⏎# P
  3. python 3.16
    first(items)` to `main.py`.<|assistant_end|><|bos|># README.md⏎# Tiny
  4. python 3.14
    (s, n):⏎ return s * n⏎⏎# README.md⏎# D
  5. python 3.13
    79<|think_end|>The answer is 79.<|assistant_end|><|bos|># README.md⏎# P
  6. python 3.11
    65 = 37234.<|bos|># README.md⏎# P
  7. python 3.09
    popular.<|assistant_end|><|bos|># notes.txt⏎Remember
  8. python 3.03
    2+0=2, 2<|think_end|>2<|assistant_end|><|bos|># README.md⏎# S
  9. python 3.02
    93<|think_end|>The answer is 93.<|assistant_end|><|bos|># README.md⏎# S
  10. python 2.99
    return s[::-1]`. The tests pass now.<|assistant_end|><|bos|># README.md⏎# Tiny
  11. python 2.94
    175 equals 79961.<|bos|># README.md⏎# Tiny
  12. python 2.91
    def square(x):⏎ return x * x⏎⏎# README.md⏎# P