Feature #4079
Fires on “ defined” (not labeled yet: read the examples)
- Fires on
- 1.16% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 90%
- chat 8%
- story 2%
- python 0%
- worksheet 0%
Makes the model say
calculatorars calculator=**/g For <|assistant_start|>
Makes it not say
theIf0[ they5 proud p
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 2.0defined in `items.
-
agent 2.0defined in `text.
-
agent 1.92"⏎ ^^^^^^^^^^^^^^^^^^^^^^^^^^^⏎AssertionError⏎⏎
-
agent 1.84^^^⏎AssertionError⏎⏎
-
agent 1.83ulltiply(6, 1))⏎ ^^^^^^^^^⏎NameError: name
-
agent 1.820, 0]))⏎ ^^^^^^^^⏎NameError: name
-
agent 1.81me about yourself."<|user_end|><|assistant_start|>I'm prelude, a tiny language model trained from scratch on a
-
agent 1.81⏎ ^^^^^⏎NameError: name 'vowls' is not defined. Did you mean
-
agent 1.8>⏎ print(count_vwels("river"))⏎ ^^^^^^^^^^^⏎NameError: name
-
agent 1.79^^^^^^^^^^^^^^^^^^^^^^^^^⏎AssertionError⏎
-
agent 1.79⏎ ^^^^^^^^^^^^^^^^^^^^^^⏎AssertionError⏎⏎
-
agent 1.78<|user_end|><|assistant_start|><|tool_call_start|>grep<|arg|>pattern=LIMIT =<|tool_call_end|><|assistant_end|><|tool_start|>Found 1 matches⏎