Feature #317
Fires on “t” (not labeled yet: read the examples)
- Fires on
- 0.82% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 87%
- story 4%
- chat 4%
- worksheet 3%
- python 3%
Makes the model say
mom tree `" everhalve `[case
Makes it not say
lot mostThis there= toysly h
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 5.39<|arg|>oldString=welcome<|arg|>newString=greet<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|>
-
agent 5.33py<|arg|>oldString=raise_to<|arg|>newString=power<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|>
-
agent 5.12<|arg|>oldString=square<|arg|>newString=squared<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|>
-
agent 5.01main.py<|arg|>oldString=first<|arg|>newString=head<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|>
-
agent 5.01<|arg|>oldString=double<|arg|>newString=double_v2<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|>
-
agent 5.01<|arg|>oldString=head<|arg|>newString=head_v2<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|>
-
agent 4.99=sum_list<|arg|>newString=total<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|>
-
agent 4.96_item<|arg|>newString=max_item_v2<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|>
-
agent 4.95replaceAll=true<|tool_call_end|><|assistant_end|>
-
agent 4.94newString=my_tail<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|>
-
agent 4.93text.py<|arg|>oldString=greet<|arg|>newString=welcome<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|>
-
agent 4.91main.py<|arg|>oldString=hello<|arg|>newString=greet<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|>