Feature #2239
Fires on “And” (not labeled yet: read the examples)
- Fires on
- 1.00% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 44%
- agent 41%
- story 9%
- python 3%
- worksheet 3%
Makes the model say
pythonayisndw editfinal saf
Makes it not say
command namedmean came Now I function while
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 3.8The answer is 101.<|assistant_end|><|user_start|>And add 10
-
chat 3.79and 0.<|user_end|><|assistant_start|>The answer is 0.<|assistant_end|><|user_start|>And if I add
-
chat 3.634?<|user_end|><|assistant_start|>The answer is 87.<|assistant_end|><|user_start|>And add 7 to
-
chat 3.5252.<|user_end|><|assistant_start|>The answer is 59.<|assistant_end|><|user_start|>And if I add
-
chat 3.514?<|user_end|><|assistant_start|>The answer is 121.<|assistant_end|><|user_start|>And add 51
-
chat 3.516 = ?<|user_end|><|assistant_start|>The answer is 62.<|assistant_end|><|user_start|>And add 9 to
-
chat 3.489+8<|user_end|><|assistant_start|>The answer is 17.<|assistant_end|><|user_start|>And if I add
-
chat 3.4635.<|user_end|><|assistant_start|>The answer is 92.<|assistant_end|><|user_start|>And add 34
-
chat 3.451 = ?<|user_end|><|assistant_start|>The answer is 89.<|assistant_end|><|user_start|>And if I add
-
chat 3.39.<|user_end|><|assistant_start|>The answer is 49.<|assistant_end|><|user_start|>And if I add
-
chat 3.381 + 8<|user_end|><|assistant_start|>The answer is 9.<|assistant_end|><|user_start|>And if I add
-
chat 3.374 is?<|user_end|><|assistant_start|>The answer is 22.<|assistant_end|><|user_start|>And if I add