Feature #403
Fires on โ:โ (not labeled yet: read the examples)
- Fires on
- 0.29% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 43%
- python 36%
- story 11%
- chat 10%
- worksheet 0%
Makes the model say
definedโ <|tool_call_start|><|system_start|> multiplcalculator!")Renamed
Makes it not say
,. off won to out?
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 1.95Goodbye!<|user_end|><|assistant_start|>Sure! Bye!
-
python 1.9: I can only tell short stories and add numbers.<|assistant_end|><|bos|>Bug: `lower_case
-
python 1.88<|assistant_start|>Hey! How can I help with your code?<|assistant_end|><|bos|>Bug: `max_item
-
python 1.86<|user_end|><|assistant_start|>Hello! How can I help you today?<|assistant_end|><|bos|>Bug: `quiet` should
-
chat 1.86Tell me a short story.<|user_end|><|assistant_start|>Sure! Once upon
-
python 1.83: I can only tell short stories and add numbers.<|assistant_end|><|bos|>Bug: `last_letter
-
python 1.8250<|tool_end|><|assistant_start|>The answer is 50.<|assistant_end|><|bos|>Bug: `vowels` should
-
python 1.8259<|think_end|>The answer is 59.<|assistant_end|><|bos|>Bug: `triangle` should
-
python 1.82model: I can tell short stories and add numbers.<|assistant_end|><|bos|>Bug: `half`
-
python 1.81I only know how to tell stories and add numbers.<|assistant_end|><|bos|>Bug: `square` should
-
python 1.790+6=6, 6<|think_end|>6<|assistant_end|><|bos|>Bug: `positive` should
-
python 1.79thanks!<|user_end|><|assistant_start|>Sure! You're welcome!<|assistant_end|><|bos|>Bug: `average_list