Feature #654
Fires on “ double” (not labeled yet: read the examples)
- Fires on
- 2.83% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 68%
- agent 23%
- story 6%
- worksheet 3%
- chat 1%
Makes the model say
itemsnumslistitem([read cannot tail
Makes it not say
woreving wasoneis on specialf
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 8.38_number(11))⏎⏎# lists.py⏎def double_all(nums
-
python 8.23(backwards("star"))⏎⏎# items.py⏎def double_all(nums
-
python 8.22_many(items):⏎ return len(items)⏎⏎⏎def double_all(nums
-
python 8.18add numbers.<|assistant_end|><|bos|># listlib.py⏎def biggest(nums):⏎
-
python 8.18`. The tests pass now.<|assistant_end|><|bos|># items.py⏎def double_all(nums
-
python 8.11tests passed")⏎⏎# listlib.py⏎def add_all(nums
-
python 8.11_item(nums):⏎ return max(nums)⏎⏎⏎def double_all(nums
-
python 8.05print("all tests passed")⏎⏎# stats.py⏎def mean_list(nums
-
python 8.02:`. The tests pass now.<|assistant_end|><|bos|># lists.py⏎def double_all(nums
-
python 8.0print("all tests passed")⏎⏎# listlib.py⏎def double_all(nums
-
python 7.91def tail(items):⏎ return items[-1]⏎⏎⏎def double_all(nums
-
python 7.85tests passed⏎<|bos|># run.py⏎from items import head, double_all⏎⏎