Skip to content
mini-lab
Features of prelude-2

Feature #202

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

Fires on
0.67% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 76%
  • python 8%
  • story 6%
  • chat 6%
  • worksheet 4%

Makes the model say

ce<|tool_call_end|>Once pythonor +v unexpected

Makes it not say

over thirstytxt asked of run up outside

Strongest examples

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

  1. agent 2.3
    webfetch, websearch, write, execute<|system_end|>
  2. agent 2.3
    webfetch, websearch, write, execute<|system_end|>
  3. agent 2.26
    webfetch, websearch, write, execute<|system_end|>
  4. agent 2.26
    webfetch, websearch, write, execute<|system_end|>
  5. agent 2.26
    webfetch, websearch, write, execute<|system_end|>
  6. agent 2.26
    webfetch, websearch, write, execute<|system_end|>
  7. story 2.26
    with a big smile on her face. As she laid down, exhausted
  8. agent 2.25
    webfetch, websearch, write, execute<|system_end|>
  9. agent 2.24
    webfetch, websearch, write, execute,
  10. agent 2.24
    webfetch, websearch, write, execute<|system_end|>
  11. agent 2.24
    webfetch, websearch, write, execute<|system_end|>
  12. agent 2.24
    webfetch, websearch, write, execute<|system_end|>