Feature #51
Fires on “.” (not labeled yet: read the examples)
- Fires on
- 1.26% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 68%
- python 22%
- chat 5%
- story 4%
- worksheet 1%
Makes the model say
pyutilsmdjsonlistsite list(
Makes it not say
idea") / z magic```aeiou#
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 5.88project⏎⏎A small Python project.⏎⏎# test_listlib.py⏎from listlib
-
python 5.83<|think_end|>The answer is 88.<|assistant_end|><|bos|># test_lists.py⏎from lists
-
python 5.68md` and `textlib.py`.<|assistant_end|><|bos|># test_core.py⏎from core
-
python 5.65, 8], 5))⏎⏎# test_listlib.py⏎from listlib
-
python 5.55⏎{"debug": false}⏎⏎# test_textlib.py⏎from textlib
-
python 5.542]` in `main.py`.<|assistant_end|><|bos|># test_listlib.py⏎from listlib
-
python 5.53(echo("tree", 2))⏎⏎# test_textlib.py⏎from textlib
-
python 5.53_words("apple", "cloud"))⏎⏎# test_textlib.py⏎from textlib
-
python 5.49How can I help with your code?<|assistant_end|><|bos|># test_textlib.py⏎from textlib
-
python 5.47910<|think_end|>910<|assistant_end|><|bos|># test_textlib.py⏎from textlib
-
python 5.47print(final_letter("tree"))⏎⏎# test_strings.py⏎from strings
-
python 5.464, 3, 7]))⏎⏎# test_listlib.py⏎from listlib