Feature #3056
Fires on “1” (not labeled yet: read the examples)
- Fires on
- 0.73% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 94%
- python 4%
- worksheet 1%
- story 1%
- chat 0%
Makes the model say
up for him)) is has warm play
Makes it not say
ButRenameterventfact did = And
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 6.4return a - b<|tool_call_end|><|assistant_end|><|tool_start|>Edited calc.py (1 replacement)<|tool_end|><|assistant_start|>
-
agent 6.27<|assistant_end|><|tool_start|>Edited main.py (1 replacement)<|tool_end|><|assistant_start|>
-
agent 6.27<|assistant_end|><|tool_start|>Edited main.py (1 replacement)<|tool_end|><|assistant_start|>
-
agent 6.24x + 1<|tool_call_end|><|assistant_end|><|tool_start|>Edited math_utils.py (1 replacement)<|tool_end|><|assistant_start|>
-
agent 6.24<|assistant_end|><|tool_start|>Edited number_utils.py (1 replacement)<|tool_end|><|assistant_start|>
-
agent 6.24<|assistant_end|><|tool_start|>Edited number_utils.py (1 replacement)<|tool_end|><|assistant_start|>
-
agent 6.17<|tool_start|>Edited run.py (1 replacement)<|tool_end|><|assistant_start|>
-
agent 6.17<|tool_start|>Edited run.py (1 replacement)<|tool_end|><|assistant_start|>
-
agent 6.16⏎ return a⏎ return b<|tool_call_end|><|assistant_end|><|tool_start|>Edited run.py (1 replacement)<|tool_end|><|assistant_start|>
-
agent 6.07) / 2<|tool_call_end|><|assistant_end|><|tool_start|>Edited math_utils.py (1 replacement)<|tool_end|><|assistant_start|>
-
agent 6.04% 2 == 1<|tool_call_end|><|assistant_end|><|tool_start|>Edited calc.py (1 replacement)<|tool_end|><|assistant_start|>
-
agent 6.02for n in nums]<|tool_call_end|><|assistant_end|><|tool_start|>Edited stats.py (1 replacement)<|tool_end|><|assistant_start|>