Feature #1909
Fires on “ divides” (not labeled yet: read the examples)
- Fires on
- 0.52% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 53%
- agent 33%
- chat 9%
- story 4%
- worksheet 0%
Makes the model say
, firstashLion out idea!
Makes it not say
$ Python<|user_start|> mini#lab tests://
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 5.250]⏎```⏎<|bos|># demo.py⏎from calc import divides, decrement⏎⏎
-
python 5.2all tests passed")⏎⏎# main.py⏎from text import reverse⏎⏎TEXT =
-
python 5.11all tests passed")⏎⏎# hello.py⏎from ops import next_number, opposite
-
python 5.1core.py`.<|assistant_end|><|bos|># demo.py⏎from words import flip, final_letter
-
python 5.070⏎⏎# run.py⏎from math_utils import opposite, is_odd
-
python 5.01small Python project.⏎⏎# main.py⏎from calc import positive, product⏎⏎
-
python 5.041.<|assistant_end|><|bos|># main.py⏎from text import echo⏎⏎print(
-
python 4.991⏎⏎# main.py⏎from math_utils import larger, sub_one
-
python 4.99- b⏎⏎# script.py⏎from math_utils import product⏎⏎print(
-
python 4.97s[-1]⏎⏎# main.py⏎from textlib import loud⏎⏎print(
-
python 4.86# demo.py⏎from core import vowels⏎⏎TEXT =
-
python 4.84all tests passed")⏎⏎# demo.py⏎from words import size⏎from lists import