Feature #3336
Fires on “ model” (not labeled yet: read the examples)
- Fires on
- 0.23% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 41%
- chat 30%
- agent 15%
- worksheet 9%
- python 5%
Makes the model say
because that're anymore thank about were all
Makes it not say
x wind cloud leaves b?"kesitems
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.24rench: good morning.<|user_end|><|assistant_start|>Sorry, I'm a tiny model: I can only
-
story 2.18thanked the little boy for bringing all the things. He was so proud of himself and smiled
-
agent 2.18<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only knows small
-
agent 2.16<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only knows small
-
agent 2.12<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only knows small
-
agent 2.1<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only knows small
-
agent 2.1<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only knows small
-
story 2.1lots of interesting things from his magazine, and he was so proud of his knowled
-
chat 2.06sics to me.<|user_end|><|assistant_start|>Sorry, I'm a tiny model: I can only
-
agent 2.06<|user_end|><|assistant_start|>Sorry, I can't do that: I'm a tiny model that only knows small
-
chat 2.05does DNA stand for?<|user_end|><|assistant_start|>Sorry, I'm a tiny model: I can only
-
chat 2.05in Berlin?<|user_end|><|assistant_start|>Sorry, I'm a tiny model: I can only