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

Feature #1729

Fires on “2” (not labeled yet: read the examples)

Fires on
1.33% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 83%
  • python 8%
  • worksheet 5%
  • chat 3%
  • story 0%

Makes the model say

2347506 =

Makes it not say

told one landed then when tell way

Strongest examples

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

  1. agent 3.3
    min(nums)⏎3: ⏎4
  2. agent 3.28
    def mean_list(nums):⏎2: return sum
  3. agent 3.27
    def average(a, b):⏎2: return a
  4. agent 3.14
    lines 1-6⏎1: def cubed(x):⏎2: return x
  5. agent 3.06
    _item(nums):⏎2: return min
  6. agent 3.04
    ⏎1: def num_chars(s):⏎2: return len
  7. agent 2.99
    2: return sum(nums) / len(nums)⏎3: ⏎4
  8. agent 2.98
    _one(x):⏎2: return x
  9. agent 2.98
    2: return sum(nums) / len(nums)⏎3: ⏎4
  10. agent 2.97
    1-2⏎1: def average_list(nums):⏎2: return sum
  11. agent 2.94
    1-8⏎1: def add_one(x):⏎2: return x
  12. agent 2.94
    lines 1-6⏎1: def reverse(s):⏎2: return s