Feature #3257
Fires on “S” (not labeled yet: read the examples)
- Fires on
- 0.61% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 36%
- agent 27%
- chat 20%
- python 16%
- worksheet 0%
Makes the model say
w=**/ H <UleangA
Makes it not say
words knew allIf said sayot che
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 4.31, edit and fix the Python files in this project.<|assistant_end|><|bos|><|user_start|>See you later!
-
story 4.19<|bos|>Once upon a time, there was a little car named Speedy.
-
chat 4.09. I only know how to tell stories and add numbers.<|assistant_end|><|user_start|>Summarize
-
story 4.08= 402.<|bos|>One day, a fast cat named Speedy saw
-
chat 3.99the kitchen, and they made many yummy meals together.<|bos|><|user_start|>Summarize
-
chat 3.96Once upon a time, there was a dog. His name was Scruffy
-
chat 3.85_strings.py` and `strings.py`.<|assistant_end|><|bos|><|user_start|>See you later!
-
chat 3.78for rescuing her.<|bos|><|system_start|>tools: calculator<|system_end|><|user_start|>See you later!
-
chat 3.76Bye<|user_end|><|assistant_start|>Goodbye! Have a nice day.<|assistant_end|><|user_start|>See you later!
-
chat 3.763.<|user_end|><|assistant_start|>The answer is 154.<|assistant_end|><|user_start|>See you later!
-
chat 3.7122<|think_end|>The answer is 22.<|assistant_end|><|bos|><|user_start|>See you later!
-
chat 3.71Once upon a time, there was a dog. His name was Scruffy