Feature #2305
Fires on “What” (not labeled yet: read the examples)
- Fires on
- 2.00% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 43%
- agent 28%
- worksheet 15%
- python 12%
- story 2%
Makes the model say
> < % calculator Python + ** prelude
Makes it not say
⏎ one86719 waited
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 5.61400 = 39589.<|bos|><|user_start|>What is the sum of
-
chat 5.579 + 844 = 853.<|bos|><|user_start|>What's the weather
-
chat 5.552288 = 2353.<|bos|><|user_start|>What is 1 +
-
chat 5.546<|think_end|>The answer is 1006.<|assistant_end|><|bos|><|user_start|>What's your name?
-
chat 5.5205<|think_end|>The answer is 505.<|assistant_end|><|bos|><|user_start|>What's your name?
-
chat 5.41917 = 105586.<|bos|><|user_start|>What's 5 +
-
chat 5.37, edit and fix the Python files in this project.<|assistant_end|><|bos|><|user_start|>What can you do?
-
chat 5.18` with `max_of(a, b)`.<|assistant_end|><|bos|><|user_start|>What is the sum of
-
chat 5.17and fix the Python files in this project.<|assistant_end|><|bos|><|user_start|>What can you do?
-
python 5.17result⏎```⏎⏎Bug: `includes` should return True if x is in items,
-
python 5.17```⏎⏎Bug: `is_positive` should return True if x is positive, but
-
chat 5.16)`.<|assistant_end|><|bos|><|user_start|>What's the weather