Feature #1430
Fires on “ and” (not labeled yet: read the examples)
- Fires on
- 0.08% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 33%
- story 33%
- python 20%
- agent 11%
- worksheet 2%
Makes the model say
L kind hard smallROncehowtools
Makes it not say
⏎(, theyAtIn(" They
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.4773865.<|bos|><|system_start|>You are mini, a small and friendly assistant.<|system_end|>
-
chat 2.47awe.<|bos|><|system_start|>You are mini, a small and friendly assistant.<|system_end|>
-
chat 2.33`listlib.py`.<|assistant_end|><|bos|><|system_start|>You are mini, a small and friendly assistant.<|system_end|>
-
chat 2.28<|assistant_end|><|bos|><|system_start|>tools: calculator<|system_end|><|system_start|>You are mini, a small and friendly assistant.<|system_end|>
-
chat 2.28<|assistant_end|><|bos|><|system_start|>tools: calculator<|system_end|><|system_start|>You are mini, a small and friendly assistant.<|system_end|>
-
chat 2.27⏎<|bos|><|system_start|>tools: calculator<|system_end|><|system_start|>You are mini, a small and friendly assistant.<|system_end|>
-
chat 2.26.<|bos|><|system_start|>tools: calculator<|system_end|><|system_start|>You are mini, a small and friendly assistant.<|system_end|>
-
chat 2.24<|system_start|>You are mini, a small and friendly assistant.<|system_end|>
-
chat 2.2and playing in the sand.<|bos|><|system_start|>You are mini, a small and friendly assistant.<|system_end|>
-
chat 2.18723.<|assistant_end|><|bos|><|system_start|>You are mini, a small and friendly assistant.<|system_end|>
-
chat 2.18py⏎all tests passed⏎<|bos|><|system_start|>You are mini, a small and friendly assistant.<|system_end|>
-
chat 2.15<|assistant_end|><|bos|><|system_start|>tools: calculator<|system_end|><|system_start|>You are mini, a small and friendly assistant.<|system_end|>