Skip to content
mini-lab
Features of prelude-2

Feature #3060

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

Fires on
0.27% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 77%
  • worksheet 9%
  • story 9%
  • agent 2%
  • chat 2%

Makes the model say

def best```<|bos|>⏎ biggest chean

Makes it not say

<|think_end|> main books answer<|think_start|>] definedexpression

Strongest examples

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

  1. python 7.31
    , `combine("tiger", "robot")` returns `"tiger robot"`.⏎⏎`includes(
  2. python 7.25
    3, 6, 5], 1)⏎False⏎<|bos|>Once upon a
  3. python 7.22
    , `combine("tiger", "tree")` returns `"tiger tree"`.⏎<|bos|><|system_start|>tools:
  4. python 7.18
    , `combine("apple", "lemon")` returns `"apple lemon"`.⏎<|bos|><|user_start|>I need
  5. python 7.06
    ```⏎⏎>>> loud("cat")⏎"CAT"⏎⏎```python⏎
  6. python 7.03
    `welcome("Zoe")` returns `"Hello, Zoe!"`.⏎<|bos|><|system_start|>tools:
  7. python 7.02
    , 5, 1, 5])` returns `2`.⏎<|bos|>One day,
  8. python 7.01
    8, 7, 7], 1)⏎False⏎⏎```python⏎
  9. python 6.97
    For example, `first_letter("tiger")` returns `"t"`.⏎⏎`repeat(
  10. python 6.95
    . For example, `exclaim("python")` returns `"python!"`.⏎⏎```python⏎
  11. python 6.93
    , 9, 9, 9])` returns `9`.⏎<|bos|><|user_start|>What can
  12. python 6.93
    example, `hello("Mia")` returns `"Hello, Mia!"`.⏎⏎```python⏎