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

Feature #3976

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

Fires on
1.25% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 46%
  • agent 24%
  • worksheet 17%
  • chat 11%
  • story 2%

Makes the model say

2ay thought mouse store lookedful

Makes it not say

trained +`,=**/*.<|think_end|> Pythonpositivedef

Strongest examples

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

  1. python 9.95
    result = 1⏎ for i in range(1, n + 1):⏎ result
  2. python 9.92
    result = 1⏎ for i in range(1, n + 1):⏎ result
  3. python 9.51
    result = 1⏎ for i in range(1, n + 1):⏎ result
  4. python 9.38
    result = 1⏎ for i in range(1, n + 1):⏎ result
  5. agent 9.24
    result = 1⏎ for i in range(1, n + 1):⏎ result
  6. python 9.07
    result = 1⏎ for i in range(1, n + 1):⏎ result
  7. agent 9.07
    result = 1⏎ for i in range(1, n + 1):⏎ result
  8. agent 8.99
    result = 1⏎ for i in range(1, n + 1):⏎ result
  9. agent 8.91
    result = 1⏎ for i in range(1, n + 1):⏎ result
  10. python 8.86
    result = 1⏎ for i in range(1, n + 1):⏎ result
  11. python 8.84
    result = 1⏎ for i in range(1, n + 1):⏎ result
  12. python 8.81
    result = 1⏎ for i in range(1, n + 1):⏎ result