Feature #1219
Fires on “ only” (not labeled yet: read the examples)
- Fires on
- 0.16% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 49%
- agent 27%
- story 18%
- python 4%
- worksheet 1%
Makes the model say
knows: one *_ heard two idea
Makes it not say
wasas "!"pped Tom's newThe
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 3.27together, and they became good friends.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>What
-
chat 3.26pass now.<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>What
-
chat 3.26and not shaken the tree.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>What
-
chat 3.21they can turn out to be good.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>Anna
-
chat 3.19ready for another day of fun together.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>what
-
chat 3.17do them in the right order.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>Calculate
-
chat 3.16198<|think_end|>198<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>what
-
chat 3.13she learned something new with her mom.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>Could
-
chat 3.1)` returns a greeting for name.<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>Add
-
chat 3.1is 1706.<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>What
-
chat 3.1answer is 706.<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>Please
-
chat 3.09, or some numbers to add?<|assistant_end|><|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>Please