ClikaRT::tokenizer::Model
class
Header: ClikaRT/tokenizer/builder.h
The core tokenization stage (required): BPE / Unigram / WordLevel / WordPiece, built from an explicit vocabulary; no file needed. Consumed by TokenizerBuilder's constructor; move-only handle.
Static member functions
bpe()
static Model bpe(
Span<const VocabEntry> vocab,
Span<const MergeRule> merges,
const BpeOptions& options = {}
)
Byte-pair encoding from a vocabulary and an ordered merge list (rank = position). Raises if a merge references a token missing from the vocab.
Declared in ClikaRT/tokenizer/builder.h, line 85
unigram()
static Model unigram(
Span<const UnigramEntry> vocab,
std::int64_t unk_id = -1,
bool has_unk_id = false,
bool byte_fallback = false
)
Unigram language model. unk_id (if set) names the unknown token's id; byte_fallback decomposes an unknown piece into byte tokens.
Declared in ClikaRT/tokenizer/builder.h, line 94
word_level()
static Model word_level(Span<const VocabEntry> vocab, std::string_view unk_token)
Whole-word vocabulary lookup; an out-of-vocabulary word maps to unk_token.
Declared in ClikaRT/tokenizer/builder.h, line 102
word_piece()
static Model word_piece(
Span<const VocabEntry> vocab,
std::string_view unk_token,
std::string_view continuing_subword_prefix = "##",
std::int64_t max_input_chars_per_word = 100
)
Greedy longest-match WordPiece. continuing_subword_prefix (e.g. ##) marks a non-initial subword; a word longer than max_input_chars_per_word maps to unk_token.
Declared in ClikaRT/tokenizer/builder.h, line 113
Member functions
Model(void)
explicit Model(void* impl) noexcept
Declared in ClikaRT/tokenizer/builder.h, line 121
Model(Model)
Model(Model&& other) noexcept
Move-only handle; the moved-from object is empty.
Declared in ClikaRT/tokenizer/builder.h, line 123
operator=()
Move-only handle; the moved-from object is empty.
Declared in ClikaRT/tokenizer/builder.h, line 125
~Model()
~Model()
Releases the model.
Declared in ClikaRT/tokenizer/builder.h, line 127