Skip to content
mini-lab
Features of prelude-2

Feature #582

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

Fires on
0.42% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 55%
  • chat 25%
  • worksheet 8%
  • python 8%
  • agent 2%

Makes the model say

upon there Suddenly today")) Insteadreturn element

Makes it not say

of` played that who>>>1 be

Strongest examples

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

  1. story 5.14
    : 7+2=9, 9<|think_end|>9<|assistant_end|><|bos|>Once upon a time,
  2. story 5.03
    stats.py` with `max_item(nums)`.<|assistant_end|><|bos|>Once upon a time,
  3. story 4.98
    4+3=7, 78<|think_end|>78<|assistant_end|><|bos|>Once upon a time,
  4. story 4.96
    Added `positive(x)` to `ops.py`.<|assistant_end|><|bos|>Once upon a time,
  5. story 4.95
    : `vowels(s)` and `reverse(s)`.<|assistant_end|><|bos|>Once upon a time,
  6. story 4.95
    stats.py` with `keep_even(nums)`.<|assistant_end|><|bos|>Once upon a time there
  7. story 4.94
    1=6, 627<|think_end|>627<|assistant_end|><|bos|>Once there was a little
  8. story 4.93
    6+0=6, 68<|think_end|>68<|assistant_end|><|bos|>Once upon a time,
  9. story 4.93
    0=5, 572<|think_end|>572<|assistant_end|><|bos|>Once upon a time,
  10. story 4.93
    0=7, 753<|think_end|>753<|assistant_end|><|bos|>Once upon a time there
  11. story 4.93
    ce(a, b)` to `arith.py`.<|assistant_end|><|bos|>Once upon a time,
  12. story 4.9
    2=2, 283<|think_end|>283<|assistant_end|><|bos|>Once upon a time there