Skip to content
mini-lab
Features of prelude-2

Feature #1353

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

Fires on
0.55% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 50%
  • agent 36%
  • chat 9%
  • story 4%
  • worksheet 1%

Makes the model say

Createdth numsaeiouexpressioncontent<|think_start|>=**/

Makes it not say

they run.ier not cat happy the

Strongest examples

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

  1. python 3.62
    , 4, 2])⏎4⏎<|bos|>```python⏎def odd(x):
  2. python 3.47
    <|assistant_end|><|bos|>```python⏎def double_all(
  3. python 3.38
    ```python⏎def echo(s,
  4. python 3.38
    ```python⏎def mean(a,
  5. python 3.38
    ```python⏎def count_items(
  6. python 3.38
    ```python⏎def repeat(s,
  7. python 3.36
    ⏎⏎>>> positive(6)⏎True⏎⏎```python⏎def factorial(n):
  8. python 3.35
    TIGER")` returns `"tiger"`.⏎⏎```python⏎def sum_up_
  9. python 3.34
    answer is 494.<|assistant_end|><|bos|># main.py⏎def how_many(
  10. python 3.26
    the Python files in this project.<|assistant_end|><|bos|># app.py⏎def half(x):
  11. python 3.23
    answer is 61.<|assistant_end|><|bos|># script.py⏎def first_letter(
  12. python 3.22
    5])⏎[8, 8]⏎⏎```python⏎def last(items):