Skip to content
mini-lab
Features of prelude-2

Feature #3283

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

Fires on
0.38% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 42%
  • story 26%
  • agent 19%
  • worksheet 9%
  • python 3%

Makes the model say

path tail=**/script were defines generator whether

Makes it not say

⏎. H(" you ini went

Strongest examples

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

  1. chat 5.38
    and fix the Python files in this project.<|assistant_end|><|bos|><|user_start|>What can you do?<|user_end|><|assistant_start|>
  2. chat 5.14
    and fix the Python files in this project.<|assistant_end|><|bos|><|user_start|>What can you do?<|user_end|><|assistant_start|>
  3. chat 5.13
    tell short stories and add numbers.<|assistant_end|><|user_start|>What can you do?<|user_end|><|assistant_start|>
  4. chat 5.12
    88.<|assistant_end|><|user_start|>Can you add 55
  5. chat 5.04
    a helpful assistant.<|system_end|><|user_start|>Can you add 53
  6. chat 4.88
    story!<|assistant_end|><|user_start|>Can you book a table for
  7. chat 4.83
    <|user_start|>Can you tell me a story
  8. agent 4.79
    agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>"Can you run the tests?"
  9. agent 4.72
    agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>"Can you run main.py
  10. agent 4.69
    agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>"Can you run demo.py
  11. agent 4.68
    agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>"Can you run hello.py
  12. agent 4.68
    agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>"Can you run the tests?"