Feature #371
Fires on “most” (not labeled yet: read the examples)
- Fires on
- 0.34% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 93%
- story 3%
- chat 2%
- python 1%
- worksheet 0%
Makes the model say
accant too himutils a anyHi
Makes it not say
of stop If named stopped shouldIf towards
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.12=python3 test_calc.py<|tool_call_end|><|assistant_end|><|tool_start|>Traceback (most recent call last):
-
agent 3.12=python3 test_calc.py<|tool_call_end|><|assistant_end|><|tool_start|>Traceback (most recent call last):
-
agent 3.12python3 test_tools.py<|tool_call_end|><|assistant_end|><|tool_start|>Traceback (most recent call last):
-
agent 3.09test_text.py<|tool_call_end|><|assistant_end|><|tool_start|>Traceback (most recent call last):
-
agent 3.03py<|tool_call_end|><|assistant_end|><|tool_start|>Traceback (most recent call last):
-
agent 2.94_strings.py<|tool_call_end|><|assistant_end|><|tool_start|>Traceback (most recent call last):
-
agent 2.81command=python3 test_tools.py<|tool_call_end|><|assistant_end|><|tool_start|>Traceback (most recent call last):
-
agent 2.79<|tool_call_end|><|assistant_end|><|tool_start|>Traceback (most recent call last):
-
agent 2.79<|tool_call_end|><|assistant_end|><|tool_start|>Traceback (most recent call last):
-
agent 2.79command=python3 test_tools.py<|tool_call_end|><|assistant_end|><|tool_start|>Traceback (most recent call last):
-
agent 2.78.py<|tool_call_end|><|assistant_end|><|tool_start|>Traceback (most recent call last):
-
agent 2.78.py<|tool_call_end|><|assistant_end|><|tool_start|>Traceback (most recent call last):