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

Feature #3351

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

Fires on
0.20% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 46%
  • agent 31%
  • story 18%
  • chat 3%
  • worksheet 1%

Makes the model say

knowto back eat upen reachvalue

Makes it not say

]))includesob Comeited friend vowels alone

Strongest examples

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

  1. python 4.45
    `.⏎⏎```python⏎def raise_to(a,
  2. python 4.34
    with `return a ** b`.⏎⏎```python⏎def raise_to(a,
  3. python 4.13
    ` in `main.py`.<|assistant_end|><|bos|>```python⏎def raise_to(a,
  4. python 4.12
    7], 1)⏎False⏎⏎```python⏎def raise_to(a,
  5. python 4.12
    5])` returns `[2]`.⏎⏎```python⏎def raise_to(a,
  6. python 3.93
    6⏎True⏎```<|assistant_end|><|bos|># helpers.py⏎def raise_to(a,
  7. agent 3.84
    newString=print(abs_value(2))⏎⏎⏎def raise_to(a,
  8. agent 3.81
    rite<|arg|>path=math_utils.py<|arg|>content=def raise_to(a,
  9. python 3.77
    (a, b):⏎ return a * b⏎⏎⏎def raise_to(a,
  10. agent 3.76
    lines 1-6⏎1: from number_utils import raise_to, average⏎
  11. python 3.73
    again.<|bos|># number_utils.py⏎def sum_up_to(n):
  12. python 3.7
    _to(2, 0) == 1⏎assert raise_to(4,