Feature #441
Fires on “=” (not labeled yet: read the examples)
- Fires on
- 1.17% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 94%
- story 5%
- python 1%
- chat 1%
- worksheet 0%
Makes the model say
newStringoldStringcontent`:defpe<|system_end|>
Makes it not say
., it⏎ years⏎⏎ project there
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.07path=lists.py<|arg|>oldString=head<|arg|>newString=
-
agent 3.07))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=hello.py<|arg|>oldString=NUMBER = 7
-
agent 3.07))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=script.py<|arg|>oldString=SIZE = 7
-
agent 3.05]<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=lists.py<|arg|>oldString= return [n
-
agent 3.01"))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=hello.py<|arg|>oldString=print(numchars
-
agent 2.99))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=hello.py<|arg|>oldString= SIZE =
-
agent 2.98x<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=lib.py<|arg|>oldString= return x *
-
agent 2.98))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=app.py<|arg|>oldString=print(next_
-
agent 2.982<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=listlib.py<|arg|>oldString= return sum(
-
agent 2.97))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=script.py<|arg|>oldString=TXT =
-
agent 2.97]))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=script.py<|arg|>oldString=print(evenn
-
agent 2.97))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=main.py<|arg|>oldString=LIMIT = -4