Skip to content
mini-lab
Features of prelude-2

Feature #1775

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

Fires on
0.48% of tokens
Mostly in
python
Layer
after block 4

Where it fires

  • python 65%
  • agent 24%
  • story 6%
  • chat 4%
  • worksheet 2%

Makes the model say

def_Nofrom⏎ (a`

Makes it not say

words surprise learned Now],Sara next brought

Strongest examples

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

  1. agent 3.97
    webfetch
  2. agent 3.97
    webfetch
  3. agent 3.97
    webfetch
  4. agent 3.97
    webfetch
  5. agent 3.97
    websearch
  6. agent 3.97
    websearch
  7. agent 3.97
    websearch
  8. agent 3.97
    websearch
  9. agent 3.97
    websearch
  10. agent 3.97
    webfetch
  11. python 2.86
    the Python files in this project.<|assistant_end|><|bos|># app.py⏎def half(x
  12. python 2.63
    The answer is 494.<|assistant_end|><|bos|># main.py⏎def how_many