Feature #4068
Fires on “What” (not labeled yet: read the examples)
- Fires on
- 0.24% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 31%
- python 30%
- agent 19%
- worksheet 14%
- story 6%
Makes the model say
<|assistant_end|> lastbug =]`⏎ sum ==
Makes it not say
to backg downthamM off
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.41, edit and fix the Python files in this project.<|assistant_end|><|bos|><|user_start|>What can you do?
-
python 2.25: I can only tell short stories and add numbers.<|assistant_end|><|bos|>Bug: `lower_case
-
chat 2.18and fix the Python files in this project.<|assistant_end|><|bos|><|user_start|>What can you do?
-
chat 2.17friend and had a lot of fun finding the wheat.<|bos|><|user_start|>What is 4 plus
-
chat 2.14and catch the ball, happy to be with his friends.<|bos|><|user_start|>What is 82
-
chat 2.14and add numbers, step by step or with a calculator.<|assistant_end|><|user_start|>What should I invest
-
chat 2.12for a story!<|assistant_end|><|user_start|>What's your name?
-
chat 2.12Try asking me for a story!<|assistant_end|><|bos|><|user_start|>What's your name?
-
chat 2.12. I can only tell short stories and add numbers.<|assistant_end|><|bos|><|user_start|>What is 36
-
python 2.1`how_many(items)` counts the items.<|assistant_end|><|bos|>Bug: `combine` should
-
chat 2.1spraying the lemons and having a great time.<|bos|><|user_start|>What is 36
-
chat 2.1I only know how to tell stories and add numbers.<|assistant_end|><|bos|><|user_start|>What is the sum of