Skip to content
mini-lab
Features of prelude-2

Feature #1432

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

Fires on
0.52% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 64%
  • python 33%
  • story 2%
  • chat 1%
  • worksheet 1%

Makes the model say

] elementlet trained⏎⏎ "!"<|tool_call_end|><|arg|>

Makes it not say

of toyopencodee ask pretty known

Strongest examples

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

  1. agent 4.42
    )<|tool_end|><|assistant_start|>Added `total(nums)` to `
  2. python 4.35
    (s)`. The tests pass now.<|assistant_end|><|bos|>`double_all(nums)` doubles
  3. agent 4.2
    <|tool_end|><|assistant_start|>Added `evens(nums)` to `
  4. python 4.12
    0, you get 135.<|bos|>`double_all(nums)` doubles
  5. agent 3.89
    functions: `average_list(nums)` and `doubled(nums)`.<|assistant_end|><|user_start|>
  6. agent 3.84
    function double_all(nums) in core
  7. agent 3.82
    * 2 for n in nums]<|tool_end|><|assistant_start|>`double_all(nums)` doubles
  8. python 3.77
    _list(nums)` returns the
  9. agent 3.72
    return "Hello, " + name + "!"<|tool_end|><|assistant_start|>`welcome(name)` returns a
  10. python 3.7
    , 2])` returns `0`.⏎⏎`double_all(nums)` doubles
  11. python 3.6
    his strength to help others.<|bos|>`mean_list(nums)` returns the
  12. agent 3.59
    upper()<|tool_end|><|assistant_start|>`shout(s)` makes s