Feature #773
Fires on “Once” (not labeled yet: read the examples)
- Fires on
- 0.29% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 74%
- chat 21%
- agent 4%
- python 0%
- worksheet 0%
Makes the model say
day playing fast happy nextbye equals README
Makes it not say
only divisibletools tests himself": wse
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 4.26: hello.py<|tool_end|><|assistant_start|>Created `hello.py`.<|assistant_end|><|bos|>Once upon a time,
-
story 4.13<|assistant_start|>The answer is 86.<|assistant_end|><|bos|>Once upon a time,
-
story 4.09greeting.py<|tool_end|><|assistant_start|>Created `greeting.py`.<|assistant_end|><|bos|>Once upon a time,
-
story 4.04stats.py` with `keep_even(nums)`.<|assistant_end|><|bos|>Once upon a time there
-
story 4.03stats.py` with `max_item(nums)`.<|assistant_end|><|bos|>Once upon a time,
-
story 4.02Added `total(nums)` to `utils.py`.<|assistant_end|><|bos|>Once upon a time,
-
story 3.99py`.<|assistant_end|><|bos|>Once upon a time,
-
story 3.97They played all day and enjoyed being friends in the sea.<|assistant_end|><|bos|>Once upon a time,
-
story 3.96<|think_end|>The answer is 32.<|assistant_end|><|bos|>Once upon a time,
-
story 3.96model: I can only tell short stories and add numbers.<|assistant_end|><|bos|>Once upon a time,
-
story 3.94Created `utils.py` with `largest(nums)`.<|assistant_end|><|bos|>Once upon a time,
-
story 3.9461<|tool_end|><|assistant_start|>The answer is 261.<|assistant_end|><|bos|>Once upon a time there