Module shared 

SharedDictionary: pre-parsed trie + raw data that can be shared across multiple tokenizer instances via Arc, avoiding ~150 MB per-tokenizer trie duplication.

Structs

SharedDictionary

Shared dictionary state that can be cloned cheaply across tokenizers.

Enums

DictData

Dictionary data that can be either an owned Vec<u8> (when mutation was needed for connection inhibitions) or a memory-mapped file (zero-copy, OS-managed pages).