Feature #1525
Fires on “ if” (not labeled yet: read the examples)
- Fires on
- 0.15% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 40%
- story 31%
- agent 11%
- python 11%
- worksheet 7%
Makes the model say
Python yourver testsOnceexcite numbersfrom
Makes it not say
friend = same fones Now so!".
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 3.66answer is 7.<|assistant_end|><|user_start|>What if you add 9
-
chat 3.37145.<|assistant_end|><|user_start|>What if you add 6
-
chat 3.2657.<|assistant_end|><|user_start|>What if you add 8
-
chat 3.237.<|user_end|><|assistant_start|>The answer is 91.<|assistant_end|><|user_start|>What if you add 8
-
chat 3.064<|tool_end|><|assistant_start|>The answer is 144.<|assistant_end|><|user_start|>What if you add 2
-
chat 3.035?<|user_end|><|assistant_start|>The answer is 76.<|assistant_end|><|user_start|>What if you add 4
-
chat 3.00?<|user_end|><|assistant_start|>The answer is 61.<|assistant_end|><|user_start|>What if you add 2
-
chat 2.929<|tool_end|><|assistant_start|>The answer is 29.<|assistant_end|><|user_start|>What if you add 1
-
chat 2.89.<|user_end|><|assistant_start|>The answer is 132.<|assistant_end|><|user_start|>What if you add 6
-
chat 2.86more?<|user_end|><|assistant_start|>The answer is 24.<|assistant_end|><|user_start|>What if you add 6
-
chat 2.82.<|user_end|><|assistant_start|>The answer is 962.<|assistant_end|><|user_start|>What if you add 2
-
chat 2.787.<|user_end|><|assistant_start|>The answer is 13.<|assistant_end|><|user_start|>What if you add 7