Skip to content
mini-lab
Features of prelude-2

Feature #3659

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

Fires on
0.29% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 89%
  • story 8%
  • python 2%
  • chat 1%
  • worksheet 0%

Makes the model say

e len <Sorry maybe enough miniCreated

Makes it not say

find saw, meet found+ friends always

Strongest examples

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

  1. agent 2.98
    webfetch, websearch, write, execute<|system_end|>
  2. agent 2.96
    webfetch, websearch, write, execute,
  3. agent 2.95
    webfetch, websearch, write, execute,
  4. agent 2.95
    webfetch, websearch, write, execute<|system_end|>
  5. agent 2.94
    webfetch, websearch, write, execute<|system_end|>
  6. agent 2.94
    webfetch, websearch, write, execute<|system_end|>
  7. agent 2.93
    webfetch, websearch, write, execute<|system_end|>
  8. agent 2.93
    webfetch, websearch, write, execute<|system_end|>
  9. agent 2.93
    webfetch, websearch, write, execute<|system_end|>
  10. agent 2.93
    webfetch, websearch, write, execute<|system_end|>
  11. agent 2.92
    webfetch, websearch, write, execute<|system_end|>
  12. agent 2.92
    webfetch, websearch, write, execute<|system_end|>