Global Token Co-occurrence Weight Matrix ()
Weak
Strong
哥哥
→
?
Candidate Token Probability Distribution
The first step in LLM learning: reading the massive text data provided by humans.