Feature #2940
Fires on “.” (not labeled yet: read the examples)
- Fires on
- 0.64% of tokens
- Mostly in
- worksheet
- Layer
- after block 4
Where it fires
- worksheet 34%
- story 18%
- python 17%
- chat 16%
- agent 16%
Makes the model say
4 `53 main off cream(
Makes it not say
Fixeddefemoopencodeconfig# happened man
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.7there was a little girl named Lily. She had a toy named Mr. Bear. They liked
-
story 2.68He wanted to talk to the crab. Tim asked, "Mr. Crab,
-
chat 2.67wanted to play with it.⏎Tim said, "Hello, Mr. Crab!
-
chat 2.65wanted to play with it.⏎Tim said, "Hello, Mr. Crab!
-
story 2.64there was a little girl named Lily. Lily had a toy named Mr. Bear. Mr.
-
story 2.6bear and flew near him. The bird said, "Hello, Mr. Bear! What are
-
python 2.59python⏎def only_even(nums):⏎ return [n for n in nums if
-
story 2.59a small, fast car. The car said, "Hello, Mr. Wagon!
-
python 2.58(nums):⏎ return [n for n in nums if
-
python 2.52has `return [n for n in nums if
-
chat 2.5wanted to play with it.⏎Tim said, "Hello, Mr. Crab!
-
python 2.48⏎⏎```python⏎def evens(nums):⏎ return [n for n in nums if