Feature #1032
Fires on “_” (not labeled yet: read the examples)
- Fires on
- 0.91% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 63%
- python 34%
- story 2%
- chat 1%
- worksheet 0%
Makes the model say
_letterjsonitem{utilsockedof
Makes it not say
In. him One⏎ not each and
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 7.68.<|assistant_end|><|bos|># test_words.py⏎from words import final_letter, exclaim⏎
-
python 7.66⏎ return len(s)⏎```⏎⏎Bug: `final_letter` should return
-
python 7.6⏎ return s[0]⏎```⏎⏎Bug: `last_letter` should return
-
python 7.53the tests.⏎⏎# hello.py⏎from strings import final_letter, count_
-
python 7.5292679.<|bos|># lib.py⏎def final_letter(s):
-
python 7.46⏎ return s[::-1]⏎```⏎⏎Bug: `final_letter` should return
-
python 7.31only tell short stories and add numbers.<|assistant_end|><|bos|>Bug: `last_letter` should return
-
python 7.29always helping each other when they needed it.<|bos|>Bug: `final_letter` should return
-
python 7.24The answer is 756.<|assistant_end|><|bos|>Bug: `max_item` should return
-
python 7.24⏎ return "Hello, " + name + "!"⏎⏎⏎def max_item(nums):
-
python 7.23<|bos|>Bug: `max_of` should return
-
python 7.18):⏎ return x * x⏎```⏎⏎Bug: `last_letter` should return