Skip to content
mini-lab
Features of prelude-2

Feature #3273

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

Fires on
0.35% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 50%
  • agent 45%
  • chat 4%
  • story 1%
  • worksheet 0%

Makes the model say

⏎<|think_end|><|bos|><|assistant_end|><|tool_call_end|>]`<|tool_start|>()`

Makes it not say

uobprinth with smallfor together

Strongest examples

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

  1. python 3.78
    ` with `sub_one(x)`.<|assistant_end|><|bos|># words.py⏎def whisper(
  2. python 3.76
    ⏎ return x % 2 == 1⏎⏎# stats.py⏎def contains(
  3. python 3.75
    sum(nums) / len(nums)⏎⏎# words.py⏎def combine(
  4. python 3.73
    ):⏎ return a % b == 0⏎⏎# text.py⏎def greet(
  5. python 3.73
    n * 2 for n in nums]⏎⏎# text.py⏎def hello(
  6. python 3.71
    <|bos|># requirements.txt⏎pytest⏎⏎# number_utils.py⏎def is_
  7. python 3.7
    ):⏎ return a % b == 0⏎⏎# listlib.py⏎def head(
  8. python 3.67
    (x):⏎ return x > 0⏎⏎# stats.py⏎def tail(
  9. python 3.64
    6<|think_end|>The answer is 76.<|assistant_end|><|bos|># strings.py⏎def hello(
  10. python 3.62
    )` and `average(a, b)`.<|assistant_end|><|bos|># stats.py⏎def has_
  11. python 3.62
    6, 8]⏎22⏎<|bos|># utils.py⏎def sub_
  12. python 3.6
    friend, and they played together all day long.<|bos|># textlib.py⏎def join_