Feature #742
Fires on “⏎ ” (not labeled yet: read the examples)
- Fires on
- 0.39% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 90%
- story 8%
- python 1%
- worksheet 1%
- chat 1%
Makes the model say
oldString<|assistant_end|><|system_end|><|arg|> laptop("ont defines
Makes it not say
The, and. theCan more returns
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.05quiet<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎utils.py:⏎ Line 1:
-
agent 2.98_positive<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎utils.py:⏎ Line 5:
-
agent 2.87=initial<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎words.py:⏎ Line 5:
-
agent 2.82=first<|tool_call_end|><|assistant_end|><|tool_start|>Found 4 matches⏎listlib.py:⏎ Line 1:
-
agent 2.8<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎app.py:⏎ Line 3:
-
agent 2.8<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎app.py:⏎ Line 9:
-
agent 2.78<|tool_call_end|><|assistant_end|><|tool_start|>Found 5 matches⏎mathlib.py:⏎ Line 8:
-
agent 2.71_even<|tool_call_end|><|assistant_end|><|tool_start|>Found 5 matches⏎main.py:⏎ Line 1:
-
agent 2.71def initial<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎textlib.py:⏎ Line 1:
-
agent 2.69<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎stats.py:⏎ Line 9:
-
agent 2.69def backwards<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎greetings.py:⏎ Line 5:
-
agent 2.69_to<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎utils.py:⏎ Line 5: