Feature #63
Fires on “ numbers” (not labeled yet: read the examples)
- Fires on
- 0.15% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 72%
- story 12%
- worksheet 8%
- agent 6%
- python 1%
Makes the model say
? or?", little.``` to
Makes it not say
Thankice range thanked Thank biggerg<|arg|>
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 4.13<|user_start|>Hi!<|user_end|><|assistant_start|>Hi there! Would you like a story, or some numbers to
-
chat 3.87<|user_end|><|assistant_start|>Hi there! Would you like a story, or some numbers to add?<|assistant_end|>
-
chat 3.82Hi!<|user_end|><|assistant_start|>Hi there! Would you like a story, or some numbers to
-
chat 3.78<|user_end|><|assistant_start|>Hi there! Would you like a story, or some numbers to add?<|assistant_end|>
-
chat 3.78! Hi there! Would you like a story, or some numbers to add?<|assistant_end|>
-
chat 3.76<|user_end|><|assistant_start|>Hi there! Would you like a story, or some numbers to add?<|assistant_end|>
-
chat 3.76! Hi there! Would you like a story, or some numbers to add?<|assistant_end|>
-
chat 3.74! Hi there! Would you like a story, or some numbers to add?<|assistant_end|>
-
chat 3.69<|user_end|><|assistant_start|>Hi there! Would you like a story, or some numbers to add?<|assistant_end|>
-
chat 3.67! Hi there! Would you like a story, or some numbers to add?<|assistant_end|>
-
chat 3.67<|user_end|><|assistant_start|>Hi there! Would you like a story, or some numbers to add?<|assistant_end|>
-
chat 3.67<|user_end|><|assistant_start|>Hi there! Would you like a story, or some numbers to add?<|assistant_end|>