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

Feature #1869

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

Fires on
0.23% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 47%
  • python 35%
  • story 15%
  • chat 3%
  • worksheet 0%

Makes the model say

name sorrymainopsesug=cked

Makes it not say

am TheIatThe cur there

Strongest examples

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

  1. python 4.39
    py⏎def includes(items, x):⏎ return x in items⏎⏎⏎def first(
  2. python 4.38
    (nums):⏎ return sum(nums) / len(nums)⏎⏎⏎def magnitude(
  3. python 4.35
    combine(a, b):⏎ return a + " " + b⏎⏎⏎def excite(
  4. python 4.32
    ⏎def final_letter(s):⏎ return s[-1]⏎⏎⏎print(final
  5. python 4.29
    greet(name):⏎ return "Hello, " + name + "!"⏎⏎⏎def squared(
  6. python 4.28
    def has_item(items, x):⏎ return x in items⏎⏎⏎def only_
  7. python 4.28
    .py⏎def tail(items):⏎ return items[-1]⏎⏎⏎def double_
  8. python 4.28
    hello.py⏎def excite(s):⏎ return s + "!"⏎⏎⏎def last(
  9. python 4.26
    ⏎def how_many(items):⏎ return len(items)⏎⏎⏎def tail(
  10. python 4.22
    ⏎def last_letter(s):⏎ return s[-1]⏎⏎⏎def first_
  11. python 4.2
    words(a, b):⏎ return a + " " + b⏎⏎⏎def loud(
  12. python 4.19
    ⏎def max_item(nums):⏎ return max(nums)⏎⏎⏎def double_