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

Feature #76

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

Fires on
0.44% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 56%
  • story 25%
  • python 15%
  • chat 4%
  • worksheet 1%

Makes the model say

all both<|assistant_end|> I All are me /

Makes it not say

=ive after time her OneThey they

Strongest examples

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

  1. python 3.01
    sum(nums) / 2` with `return sum(nums) / len(nums
  2. python 2.97
    ⏎⏎⏎def average_list(nums):⏎ return sum(nums) / len(nums
  3. agent 2.95
    (nums) / 2` instead of `return sum(nums) / len(nums
  4. python 2.92
    sum(nums) / 2` with `return sum(nums) / len(nums
  5. python 2.89
    sum(nums) / 2` with `return sum(nums) / len(nums
  6. agent 2.89
    (nums) / 2` instead of `return sum(nums) / len(nums
  7. agent 2.88
    (nums) / 2` instead of `return sum(nums) / len(nums
  8. python 2.87
    sum(nums) / 2` with `return sum(nums) / len(nums
  9. python 2.79
    sum(nums) / 2` with `return sum(nums) / len(nums
  10. agent 2.77
    (nums) / 2` instead of `return sum(nums) / len(nums
  11. python 2.77
    py⏎def average_list(nums):⏎ return sum(nums) / len(
  12. python 2.76
    sum(nums) / 2` with `return sum(nums) / len(nums