Feature #3776
Fires on “ the” (not labeled yet: read the examples)
- Fires on
- 2.04% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 61%
- agent 19%
- chat 14%
- python 6%
- worksheet 0%
Makes the model say
big the toy l ball r c old
Makes it not say
<|tool_call_end|>.<|assistant_end|><|user_end|>estreplaceAll⏎?
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 3.83<|bos|><|system_start|>Answer with the number only.<|system_end|>
-
chat 3.83<|bos|><|system_start|>Answer with the number only.<|system_end|>
-
chat 3.761088<|think_end|>1088<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|>
-
python 3.61⏎print(size("cloud"))⏎⏎# README.md⏎# Tiny utils⏎⏎
-
python 3.582 for n in nums]⏎⏎# README.md⏎# Scratch⏎
-
story 3.51So Jack put the net back in the
-
chat 3.511 function: `tail(items)`.<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|>
-
chat 3.5⏎<|bos|><|system_start|>Answer with the number only.<|system_end|>
-
python 3.47` should return the average of a and
-
python 3.44⏎print(fact(5))⏎⏎# README.md⏎# Playground
-
chat 3.43return a ** b⏎```⏎<|bos|><|system_start|>Answer with the number only.<|system_end|>
-
python 3.43(count_vowels("robot"))⏎⏎# README.md⏎# Scratch⏎