Feature #3298
Fires on “ model” (not labeled yet: read the examples)
- Fires on
- 0.38% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 46%
- chat 37%
- python 11%
- story 5%
- worksheet 0%
Makes the model say
trained model mini]`])) divisible best's
Makes it not say
returns"052 import day4
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 4.85name?<|user_end|><|assistant_start|>My name is prelude. I'm a tiny language model: I can tell
-
chat 4.56name?<|user_end|><|assistant_start|>My name is prelude. I'm a tiny language model: I can tell
-
chat 4.55name?<|user_end|><|assistant_start|>My name is prelude. I'm a tiny language model: I can tell
-
chat 4.53you?<|user_end|><|assistant_start|>My name is prelude. I'm a tiny language model: I can tell
-
chat 4.52your name<|user_end|><|assistant_start|>My name is prelude. I'm a tiny language model: I can tell
-
chat 4.51you?<|user_end|><|assistant_start|>My name is prelude. I'm a tiny language model: I can tell
-
chat 4.5you?<|user_end|><|assistant_start|>My name is prelude. I'm a tiny language model: I can tell
-
chat 4.49you?<|user_end|><|assistant_start|>My name is prelude. I'm a tiny language model: I can tell
-
chat 4.41you?<|user_end|><|assistant_start|>My name is prelude. I'm a tiny language model: I can tell
-
chat 4.4Who made you?<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on
-
chat 4.39name?<|user_end|><|assistant_start|>My name is prelude. I'm a tiny language model: I can tell
-
chat 4.38name?<|user_end|><|assistant_start|>My name is prelude. I'm a tiny language model: I can tell