Skip to content
mini-lab
Features of prelude-2

Feature #4078

Fires on ā€œlā€ (not labeled yet: read the examples)

Fires on
0.10% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 49%
  • story 31%
  • worksheet 12%
  • chat 7%
  • python 1%

Makes the model say

their his a her smiled unique them the

Makes it not say

function hurtsscript codeWrite checksICENSEāŽ

Strongest examples

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

  1. agent 1.79
    Fix the failing test.<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/
  2. agent 1.69
    Do the tests pass?"<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/
  3. agent 1.66
    are in this project?<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/*.
  4. agent 1.5
    agent harness.<|system_end|><|user_start|>Fix the failing test.<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/
  5. agent 1.48
    files<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/*.
  6. agent 1.46
    .<|system_end|><|user_start|>Find all .py files.<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/*.
  7. agent 1.46
    <|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/
  8. agent 1.46
    <|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/
  9. agent 1.45
    harness.<|system_end|><|user_start|>Check the tests.<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/
  10. agent 1.44
    harness.<|system_end|><|user_start|>Check the tests.<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/
  11. agent 1.44
    harness.<|system_end|><|user_start|>fix the failing test.<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/
  12. agent 1.43
    name.<|assistant_end|><|user_start|>Check the tests.<|user_end|><|assistant_start|><|tool_call_start|>glob<|arg|>pattern=**/