Feature #307
Fires on “ me” (not labeled yet: read the examples)
- Fires on
- 0.65% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 57%
- agent 19%
- worksheet 18%
- story 6%
- python 1%
Makes the model say
ant`:! tell what()` > in
Makes it not say
pe5doubledHow3]`.ote
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 5.81<|user_end|><|assistant_start|>The answer is 21.<|assistant_end|><|user_start|>Could you tell me what 1 +
-
chat 5.74⏎<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>Could you tell me what 89
-
chat 5.724544 = 4550<|bos|><|user_start|>Tell me about yourself.<|user_end|>
-
chat 5.62was a sailor.<|assistant_end|><|bos|><|system_start|>tools: calculator<|system_end|><|user_start|>Tell me a story.<|user_end|>
-
chat 5.61(nums)`.<|assistant_end|><|bos|><|system_start|>tools: calculator<|system_end|><|user_start|>Tell me a story about a
-
chat 5.6<|assistant_end|><|bos|><|user_start|>Tell me a story about the
-
chat 5.6<|user_end|><|assistant_start|>The answer is 21.<|assistant_end|><|user_start|>Could you tell me what 1 +
-
chat 5.58their friends. And they all lived happily ever after.<|bos|><|user_start|>Tell me a story.<|user_end|>
-
chat 5.57new friend and went home to get warm.<|bos|><|user_start|>Could you tell me what 43
-
chat 5.52.<|bos|><|system_start|>Answer with the number only.<|system_end|><|user_start|>Could you tell me what 12
-
chat 5.52⏎lemon⏎house!⏎```<|assistant_end|><|bos|><|user_start|>Could you tell me what 96
-
chat 5.49+ 815 = 819.<|bos|><|user_start|>Tell me about the hist