Feature #884
Fires on “p” (not labeled yet: read the examples)
- Fires on
- 1.16% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 89%
- story 5%
- chat 3%
- python 2%
- worksheet 0%
Makes the model say
x<|tool_call_end|>if examplecaseassert<|think_end|>()
Makes it not say
i more( they kind other again average
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 2.31pretty river.<|assistant_end|><|bos|><|system_start|>tools: edit, glob, grep, question, read
-
agent 2.31a walk.<|assistant_end|><|bos|><|system_start|>tools: edit, glob, grep, question, read
-
agent 2.28proud of her balance.<|bos|><|system_start|>tools: edit, glob, grep, question,
-
agent 2.26the yard.<|assistant_end|><|bos|><|system_start|>tools: edit, glob, grep, question, read
-
agent 2.25greetings.py<|assistant_end|><|bos|><|system_start|>tools: edit, glob, grep, question, read
-
agent 2.24big boat.<|assistant_end|><|bos|><|system_start|>tools: edit, glob, grep, question, read
-
agent 2.24a sailor.<|assistant_end|><|bos|><|system_start|>tools: edit, glob, grep, question, read
-
agent 2.24.py with bigger<|assistant_end|><|bos|><|system_start|>tools: edit, glob, grep, question,
-
agent 2.23and add numbers.<|assistant_end|><|bos|><|system_start|>tools: edit, glob, grep, question,
-
agent 2.23and add numbers.<|assistant_end|><|bos|><|system_start|>tools: edit, glob, grep, question,
-
agent 2.23named Tim.<|assistant_end|><|bos|><|system_start|>tools: edit, glob, grep, question, read
-
agent 2.21a walk.<|assistant_end|><|bos|><|system_start|>tools: edit, glob, grep, question, read