Feature #1450
Fires on “ tools” (not labeled yet: read the examples)
- Fires on
- 1.07% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 53%
- python 29%
- chat 12%
- story 3%
- worksheet 3%
Makes the model say
Python mini laptoppytest:// "$Fixedalse
Makes it not say
1 are⏎ were<|bos|> count home am
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 4.14_list(SCORES))⏎⏎$ python3 app.py⏎6.0⏎<|bos|>
-
python 3.74py`.<|assistant_end|><|bos|># tools.py⏎def
-
python 3.63<|tool_end|><|assistant_start|>The answer is 1656.<|assistant_end|><|bos|># listlib.py⏎def
-
python 3.61, 8, 3, 8, 2])⏎4.8⏎<|bos|>
-
python 3.57home sad.<|bos|># utils.py⏎def
-
python 3.56shadows on the wall. They were happy and safe.<|bos|># tools.py⏎def
-
python 3.53add numbers.<|assistant_end|><|bos|># listlib.py⏎def
-
python 3.52513<|think_end|>513<|assistant_end|><|bos|># tools.py⏎def
-
python 3.58, 4, 6, 8])` returns `6.2`.⏎
-
python 3.498, 6, 8]⏎22⏎<|bos|># utils.py⏎def
-
python 3.492670 = 92679.<|bos|># lib.py⏎def
-
python 3.47list([0, 7, 5, 7])⏎4.75⏎