Feature #2597
Fires on “ only” (not labeled yet: read the examples)
- Fires on
- 0.41% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 38%
- story 29%
- agent 20%
- python 10%
- worksheet 2%
Makes the model say
wanted way her other wants returns want
Makes it not say
unexpectedpe<|tool_call_end|>`B thatOnceletter
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 4.78together, and they became good friends.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>What
-
chat 4.62pass now.<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>What
-
chat 4.62ready for another day of fun together.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>what
-
chat 4.57she learned something new with her mom.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>Could
-
chat 4.54and not shaken the tree.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>What
-
chat 4.52is 64886.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>Hey
-
chat 4.574593 balls.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>What
-
chat 4.49, or some numbers to add?<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>Please
-
chat 4.472 is 1534.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>what
-
chat 4.47return a ** b⏎```⏎<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>Could
-
chat 4.46= 98926.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>Could
-
chat 4.45so much fun with their new toy.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>I