pub fn propose(ctx: &[u32], max_draft: usize) -> DraftExpand description
Propose up to max_draft tokens by matching the longest available suffix of ctx against an
earlier occurrence in ctx, and returning the tokens that followed that occurrence.
- Tries suffix lengths
MAX_NGRAM..=1(longest match first — higher-order n-grams are more selective, so their continuation is likelier to be accepted). - For a given n-gram, uses the MOST RECENT earlier occurrence (locality: recent context is the best predictor of the immediate continuation).
- Never proposes past the end of
ctx(only real, already-seen tokens are drafted). - Returns
Draft::emptywhen nothing recurs (the common case for novel text).