Skip to content
mini-lab
Log in Sign up
Features of prelude-2

Feature #1258

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

Fires on
0.56% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 51%
  • python 38%
  • story 6%
  • chat 3%
  • worksheet 2%

Makes the model say

careful Python numsall together min lower function

Makes it not say

import made unexpected")))?"!") what

Strongest examples

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

  1. agent 6.53
    return [n * 2 for n in nums]<|tool_end|><|assistant_start|>
  2. agent 6.47
    return [n * 2 for n in nums]`. The tests
  3. python 6.23
    for n in nums]⏎```
  4. agent 6.23
    for n in nums]<|tool_call_end|><|assistant_end|>
  5. python 6.23
    for n in nums:⏎ result
  6. agent 6.14
    = return [n * 2 for n in nums]<|tool_call_end|><|assistant_end|>
  7. agent 6.09
    def doubled(nums):⏎ return [n * 2 for n in nums]<|tool_call_end|><|assistant_end|>
  8. python 6.08
    def doubled(nums):⏎ return [n * 2 for n in nums]⏎⏎
  9. agent 6.02
    * 2 for n in nums]`. The tests
  10. agent 6.02
    2: return [n * 3 for n in nums]<|tool_end|><|assistant_start|>
  11. python 6.01
    in nums]` with `return [n * 2 for n in nums]`.⏎⏎
  12. agent 5.94
    in nums]` instead of `return [n * 2 for n in nums]`. The tests