Feature #129
Fires on “1” (not labeled yet: read the examples)
- Fires on
- 0.42% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 92%
- chat 4%
- story 2%
- worksheet 2%
- python 1%
Makes the model say
multipl she it<|tool_call_end|> nod replace defined glad
Makes it not say
manyhero count first other + own
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 2.99<|tool_call_start|>grep<|arg|>pattern=def hello<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎lib.
-
agent 2.97<|tool_call_start|>grep<|arg|>pattern=def includes<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎lists.
-
agent 2.96<|tool_call_start|>grep<|arg|>pattern=def exclaim<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎hello.
-
agent 2.92<|tool_call_start|>grep<|arg|>pattern=def doubled<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎helpers.
-
agent 2.9rep<|arg|>pattern=def lower_of<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎calc.
-
agent 2.89<|tool_call_start|>grep<|arg|>pattern=def cube<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎math_
-
agent 2.89<|tool_call_start|>grep<|arg|>pattern=def biggest<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎lists.
-
agent 2.88<|tool_call_start|>grep<|arg|>pattern=def larger<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎script.
-
agent 2.88<|tool_call_start|>grep<|arg|>pattern=def fact<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎number_
-
agent 2.87'lenggth' is not defined. Did you mean: 'length'?
-
agent 2.84rep<|arg|>pattern=def has_item<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎items.
-
agent 2.84<|tool_call_start|>grep<|arg|>pattern=def biggest<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎listlib.