Feature #2071
Fires on “ to” (not labeled yet: read the examples)
- Fires on
- 0.37% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 33%
- worksheet 29%
- story 19%
- chat 13%
- python 6%
Makes the model say
9 is10 equals]) tree hill
Makes it not say
` " 'ideter "!"`. everyb
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.09in this project.<|assistant_end|><|bos|><|user_start|>thanks<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|user_start|>
-
chat 2.07to playing normally.<|bos|><|user_start|>thanks<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|user_start|>
-
chat 2.05for another story soon.<|assistant_end|><|user_start|>thanks<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|user_start|>
-
chat 2.04<|assistant_start|>You're welcome!<|assistant_end|><|user_start|>Thanks!<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|user_start|>
-
chat 2.025])⏎2⏎<|bos|><|user_start|>Thanks!<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|user_start|>
-
chat 1.96!<|assistant_end|><|bos|><|user_start|>Great, thanks!<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|user_start|>
-
chat 1.9assistant.<|system_end|><|user_start|>Great, thanks!<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|bos|>
-
chat 1.89.<|assistant_end|><|bos|><|user_start|>Great, thanks!<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|user_start|>
-
chat 1.88Have a nice day.<|assistant_end|><|user_start|>Thank you<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|user_start|>
-
chat 1.87or with a calculator.<|assistant_end|><|user_start|>thanks<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|bos|>
-
chat 1.86is 10.<|assistant_end|><|user_start|>thanks<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|user_start|>
-
chat 1.85<|bos|><|system_start|>tools: calculator<|system_end|><|user_start|>thanks<|user_end|><|assistant_start|>Happy to help!<|assistant_end|><|user_start|>