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

Feature #2203

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

Fires on
2.92% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 65%
  • python 17%
  • chat 12%
  • story 4%
  • worksheet 2%

Makes the model say

pyjson Hemdlowerupper She It

Makes it not say

rement wished red help park helped pleaseMia

Strongest examples

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

  1. python 9.08
    s):⏎ return s[0]⏎⏎# test_textlib.py⏎from textlib
  2. python 9.04
    910<|think_end|>910<|assistant_end|><|bos|># test_textlib.py⏎from textlib
  3. python 8.73
    items)` and `total(nums)`.<|assistant_end|><|bos|># test_words.py⏎from words
  4. python 8.72
    add numbers.<|assistant_end|><|bos|># listlib.py⏎def biggest
  5. python 8.7
    <|think_end|>The answer is 88.<|assistant_end|><|bos|># test_lists.py⏎from lists
  6. python 8.67
    (nums) / len(nums)⏎⏎# test_listlib.py⏎from listlib
  7. python 8.66
    short stories and add numbers.<|assistant_end|><|bos|># test_number_utils.py⏎from number
  8. python 8.61
    n % 2 == 0]⏎⏎# test_strings.py⏎from strings
  9. python 8.56
    ⏎{"debug": false}⏎⏎# test_textlib.py⏎from textlib
  10. python 8.55
    4, 0]⏎5⏎$ python3 test_listlib.py⏎all tests
  11. python 8.52
    = k + 1⏎ return k⏎⏎# test_words.py⏎from words
  12. python 8.48
    .py`.<|assistant_end|><|bos|># test_words.py⏎from words