Skip to content
mini-lab
Features of prelude-2

Feature #1317

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

Fires on
0.32% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 54%
  • python 25%
  • chat 15%
  • story 6%
  • worksheet 0%

Makes the model say

ip toap /WORD anymoreber

Makes it not say

⏎ turned dirty hot3 Sheincludes not

Strongest examples

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

  1. python 3.5
    def mean(a, b):⏎ return (a + b) / 2⏎```
  2. agent 3.27
    ⏎8: return (a + b) / 2<|tool_end|><|assistant_start|>
  3. python 3.23
    def average(a, b):⏎ return (a + b) / 3⏎```
  4. python 3.22
    def mean(a, b):⏎ return (a + b) / 2⏎```
  5. python 3.19
    def mean(a, b):⏎ return (a + b) / 3⏎```
  6. agent 3.12
    def average(a, b):⏎ return (a + b) / 2⏎<|tool_call_end|>
  7. python 3.07
    def mean(a, b):⏎ return (a + b) / 2⏎```
  8. agent 3.07
    def mean(a, b):⏎ return (a + b) / 2⏎<|tool_call_end|>
  9. agent 3.05
    def mean(a, b):⏎ return (a + b) / 2⏎<|tool_call_end|>
  10. agent 3.05
    def mean(a, b):⏎ return (a + b) / 2<|tool_call_end|><|assistant_end|>
  11. agent 3.04
    def mean(a, b):⏎ return (a + b) / 2<|tool_call_end|><|assistant_end|>
  12. python 3.04
    def average(a, b):⏎ return (a + b) / 3⏎```