Feature #2155
Fires on “ trained” (not labeled yet: read the examples)
- Fires on
- 0.25% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 72%
- chat 18%
- story 4%
- worksheet 4%
- python 1%
Makes the model say
numsememberAddedMITLet checks divisible **
Makes it not say
print found met"` pretty dirty have
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.31me about yourself<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
chat 2.2about yourself.<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
chat 2.17your name?<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
chat 2.17made you?<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
chat 2.16your name?<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
agent 2.14What are you?<|user_end|><|assistant_start|>I'm prelude, a tiny language model trained from scratch on a
-
chat 2.14your name?<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
chat 2.13's your name<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
chat 2.12your name?<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
chat 2.12your name?<|user_end|><|assistant_start|>I'm prelude, a very small language model trained from scratch on a
-
agent 2.12Who are you?<|user_end|><|assistant_start|>I'm prelude, a tiny language model trained from scratch on a
-
agent 2.11's your name?<|user_end|><|assistant_start|>I'm prelude, a tiny language model trained from scratch on a