Feature #3139
Fires on “ trained” (not labeled yet: read the examples)
- Fires on
- 0.66% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 72%
- story 12%
- chat 11%
- python 4%
- worksheet 0%
Makes the model say
typo(-ighedAssertionError `[ than⏎ File
Makes it not say
you = big6 sted c3
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.61me about yourself<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
story 2.55mom just shrugged, and said she'd have to take care of it forever.
-
chat 2.47about yourself.<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
story 2.44.⏎Mia enjoyed her cube so much and always made sure to take care of it. She
-
chat 2.39are you?<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
story 2.39As he rubbed the belt, a kind and compassionate genie came out. The genie
-
story 2.38in his mind. One day, Jack's mom asked him to take care of the house while
-
chat 2.38me about yourself<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
chat 2.37are you?<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
chat 2.36are you?<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
agent 2.35's your name?<|user_end|><|assistant_start|>I'm prelude, a tiny language model trained from scratch on a
-
chat 2.35your name?<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a