Skip to content
mini-lab
Features of prelude-2

Feature #1221

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

Fires on
0.62% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 40%
  • python 36%
  • story 19%
  • chat 3%
  • worksheet 2%

Makes the model say

_s2(read to dogness

Makes it not say

her sheig Her made set their subtract

Strongest examples

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

  1. python 3.5
    nums)` returns the average of nums.⏎⏎```python⏎def mean_list(nums
  2. python 2.91
    333333333333⏎assert mean_list([5
  3. python 2.88
    nums)` returns the average of nums.⏎⏎```python⏎def average_list(nums
  4. python 2.8
    sum(nums) / len(nums)⏎```⏎⏎`mean_list` returns
  5. python 2.8
    sum(nums) / len(nums)⏎```⏎⏎`mean_list` returns
  6. python 2.79
    _lists.py⏎from lists import average_list⏎⏎assert average_list([6
  7. python 2.79
    sum(nums) / len(nums)⏎```⏎⏎`mean_list` returns
  8. python 2.77
    sum(nums) / 2`.⏎⏎```python⏎def mean_list(nums
  9. agent 2.76
    , a coding agent harness.<|system_end|><|user_start|>Add mean_list to stats
  10. python 2.76
    sum(nums) / len(nums)⏎```⏎⏎>>> mean_list([5
  11. story 2.75
    . She was so sad and started to cry. The monster didn't mean to cause her
  12. agent 2.75
    agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Explain mean_list from lists