Feature #2610
Fires on “ one” (not labeled yet: read the examples)
- Fires on
- 0.72% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 46%
- worksheet 36%
- chat 11%
- story 5%
- python 2%
Makes the model say
txt returnO green twistrement")?
Makes it not say
Lily well`: princessup So fell nest
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.57best treasure hunters in the world.<|bos|><|system_start|>Answer in one short sentence.<|system_end|>
-
chat 2.54can help you solve problems and help others.<|bos|><|system_start|>Answer in one short sentence.<|system_end|>
-
chat 2.4830<|think_end|>1230<|assistant_end|><|bos|><|system_start|>Answer in one short sentence.<|system_end|>
-
chat 2.48their day. They were the best of friends.<|bos|><|system_start|>Answer in one short sentence.<|system_end|>
-
chat 2.44very happy he found the box with the toys.<|bos|><|system_start|>Answer in one short sentence.<|system_end|>
-
chat 2.44how to pay. They were happy friends again.<|bos|><|system_start|>Answer in one short sentence.<|system_end|>
-
chat 2.44, they all played with the toy car together.<|bos|><|system_start|>Answer in one short sentence.<|system_end|>
-
chat 2.42friends and never take candy from strangers.<|bos|><|system_start|>Answer in one short sentence.<|system_end|>
-
chat 2.38. And they all lived happily ever after.<|assistant_end|><|bos|><|system_start|>Answer in one short sentence.<|system_end|>
-
chat 2.35know how to tell stories and add numbers.<|assistant_end|><|bos|><|system_start|>Answer in one short sentence.<|system_end|>
-
chat 2.35iating what you already have.<|bos|><|system_start|>Answer in one short sentence.<|system_end|>
-
chat 2.35, 659<|think_end|>659<|assistant_end|><|bos|><|system_start|>Answer in one short sentence.<|system_end|>