Feature #234
Fires on “py” (not labeled yet: read the examples)
- Fires on
- 0.44% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 62%
- story 13%
- worksheet 13%
- python 6%
- chat 6%
Makes the model say
<|assistant_end|>ThankHi xMe:` live like
Makes it not say
-Shepython! herself him SheFinally
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 5.36py:⏎ Line
-
agent 5.36py with a function smallest
-
agent 5.36py`.<|assistant_end|><|bos|>#
-
agent 5.36py` and `script
-
agent 5.36py (1 replacement)
-
agent 5.36py", line 3
-
agent 5.36py<|tool_end|><|assistant_start|>Created `
-
agent 5.36py:⏎ Line
-
agent 5.36py`: it used `
-
agent 5.36py, lines 1
-
agent 5.36py<|tool_call_end|><|assistant_end|><|tool_start|>all
-
agent 5.36py (1 replacement)