Feature #1230
Fires on “<|tool_call_start|>” (not labeled yet: read the examples)
- Fires on
- 0.32% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 78%
- chat 12%
- story 9%
- python 0%
- worksheet 0%
Makes the model say
is was a noise such herDone
Makes it not say
filesAddedincludesopleed rever wishpe
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 1.5<|assistant_start|><|tool_call_start|>shell<|arg|>
-
agent 1.5<|assistant_start|><|tool_call_start|>write<|arg|>path
-
agent 1.5<|assistant_start|><|tool_call_start|>read<|arg|>path=
-
agent 1.5<|assistant_start|><|tool_call_start|>edit<|arg|>path
-
agent 1.5<|assistant_start|><|tool_call_start|>edit<|arg|>path
-
chat 1.5<|assistant_start|><|tool_call_start|>calculator<|arg|>expression=
-
agent 1.5<|assistant_start|><|tool_call_start|>read<|arg|>path=
-
agent 1.5<|assistant_start|><|tool_call_start|>shell<|arg|>
-
agent 1.5<|assistant_start|><|tool_call_start|>read<|arg|>path=
-
agent 1.5<|assistant_start|><|tool_call_start|>edit<|arg|>path
-
agent 1.5<|assistant_start|><|tool_call_start|>read<|arg|>path=
-
agent 1.5<|assistant_start|><|tool_call_start|>read<|arg|>path=