Feature #3870
Fires on “⏎” (not labeled yet: read the examples)
- Fires on
- 1.39% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 45%
- worksheet 30%
- agent 14%
- story 8%
- chat 2%
Makes the model say
Thank⏎ topationceoc<|bos|> Thank
Makes it not say
him things he rememberock wished then make
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 2.84. For example, `exclaim("python")` returns `"python!"`.⏎⏎```python⏎
-
python 2.82, `combine("tiger", "robot")` returns `"tiger robot"`.⏎⏎`includes(
-
python 2.81-x⏎```⏎⏎>>> opposite(4)⏎-4⏎⏎`min_
-
python 2.813, 6, 5], 1)⏎False⏎<|bos|>Once upon a
-
python 2.78. For example, `exclaim("river")` returns `"river!"`.⏎⏎```python⏎
-
python 2.78```⏎⏎>>> loud("cat")⏎"CAT"⏎⏎```python⏎
-
python 2.73, `combine("apple", "lemon")` returns `"apple lemon"`.⏎<|bos|><|user_start|>I need
-
python 2.73, `combine("tiger", "tree")` returns `"tiger tree"`.⏎<|bos|><|system_start|>tools:
-
python 2.72`welcome("Zoe")` returns `"Hello, Zoe!"`.⏎<|bos|><|system_start|>tools:
-
python 2.7example, `hello("Mia")` returns `"Hello, Mia!"`.⏎⏎```python⏎
-
python 2.68. For example, `excite("hello")` returns `"hello!"`.⏎<|bos|>8 plus
-
python 2.678, 7, 7], 1)⏎False⏎⏎```python⏎