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

Feature #433

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

Fires on
2.27% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 89%
  • python 4%
  • chat 3%
  • story 2%
  • worksheet 1%

Makes the model say

. it so I." is you but

Makes it not say

girl pythonfinalpython still +firstMIT

Strongest examples

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

  1. agent 2.85
    3, in <module>⏎ assert min_item([2, 6,
  2. agent 2.75
    3, in <module>⏎ assert keep_even([3, 0,
  3. agent 2.72
    5, in <module>⏎ assert double_all([3, 4,
  4. agent 2.72
    5, in <module>⏎ assert mean_list([1, 0,
  5. agent 2.71
    3, in <module>⏎ assert average_list([3, 1,
  6. agent 2.7
    3, in <module>⏎ assert keep_even([3, 0,
  7. agent 2.69
    ", line 3, in <module>⏎ assert doubled([4, 4,
  8. agent 2.64
    3, in <module>⏎ assert double_all([2, 0,
  9. agent 2.61
    3, in <module>⏎ assert min_item([3, 0,
  10. agent 2.6
    , in <module>⏎ print(max_itm([2, 3,
  11. agent 2.6
    ", line 3, in <module>⏎ assert biggest([3, 4,
  12. agent 2.6
    5, in <module>⏎ assert mean_list([8, 2,