tokenizer
family
Declared in clika_rt/clika_rt_core.h.
Members
| Name | Description |
|---|---|
clika_rt_tokenizer | |
tokenizer_apply_chat_template_messages_add_generation_prompt | ClikaRT::tokenizer::Tokenizer::apply_chat_template <- ClikaRT/tokenizer/tokenizer.h one form of the overloaded C++ apply_chat_template; the _messages_add_generation_prompt suffix names this form borrows self messages_json is a JSON document as text (not retained) sized-string protocol (contract §5) C++ defaults (dropped at this surface): add_generation_prompt=true |
tokenizer_bos_id | ClikaRT::tokenizer::Tokenizer::bos_id <- ClikaRT/tokenizer/tokenizer.h borrows self infallible |
tokenizer_decode | ClikaRT::tokenizer::Tokenizer::decode <- ClikaRT/tokenizer/tokenizer.h borrows self sized-string protocol (contract §5) C++ defaults (dropped at this surface): skip_special_tokens=true |
tokenizer_decode_batch_encoded_skip_special_tokens | ClikaRT::tokenizer::Tokenizer::decode_batch <- ClikaRT/tokenizer/tokenizer.h one form of the overloaded C++ decode_batch; the _encoded_skip_special_tokens suffix names this form borrows self borrows encoded single-owner list: index it with string_list_count / string_list_get_at; string_list_free frees it C++ defaults (dropped at this surface): skip_special_tokens=true |
tokenizer_decode_batch_ids_cu_seqlens | ClikaRT::tokenizer::Tokenizer::decode_batch <- ClikaRT/tokenizer/tokenizer.h one form of the overloaded C++ decode_batch; the _ids_cu_seqlens suffix names this form borrows self borrows ids borrows cu_seqlens single-owner list: index it with string_list_count / string_list_get_at; string_list_free frees it C++ defaults (dropped at this surface): skip_special_tokens=true |
tokenizer_decode_batch_ids_skip_special_tokens | ClikaRT::tokenizer::Tokenizer::decode_batch <- ClikaRT/tokenizer/tokenizer.h one form of the overloaded C++ decode_batch; the _ids_skip_special_tokens suffix names this form borrows self borrows ids single-owner list: index it with string_list_count / string_list_get_at; string_list_free frees it C++ defaults (dropped at this surface): skip_special_tokens=true |
tokenizer_encode | ClikaRT::tokenizer::Tokenizer::encode <- ClikaRT/tokenizer/tokenizer.h borrows self two-call sizing: buf == NULL sizes; count carries capacity in, required count out (contract §8) C++ defaults (dropped at this surface): add_special_tokens=true |
tokenizer_encode_batch | ClikaRT::tokenizer::Tokenizer::encode_batch <- ClikaRT/tokenizer/tokenizer.h borrows self texts[i] spans texts_lens[i] bytes (the string protocol's sized form, per element) options may be NULL; every option takes its default single-owner handle: the id rows + optional tensor columns ride its getters; encoded_free frees it C++ defaults (dropped at this surface): options={} |
tokenizer_encode_chat_messages | ClikaRT::tokenizer::Tokenizer::encode_chat <- ClikaRT/tokenizer/tokenizer.h one form of the overloaded C++ encode_chat; the _messages suffix names this form borrows self messages_json is a JSON document as text (not retained) two-call sizing: buf == NULL sizes; count carries capacity in, required count out (contract §8) C++ defaults (dropped at this surface): add_generation_prompt=true, add_special_tokens=true |
tokenizer_encode_text | ClikaRT::tokenizer::Tokenizer::encode_text <- ClikaRT/tokenizer/tokenizer.h borrows self options may be NULL; every option takes its default single-owner handle: the id rows + optional tensor columns ride its getters; encoded_free frees it C++ defaults (dropped at this surface): options={} |
tokenizer_eos_id | ClikaRT::tokenizer::Tokenizer::eos_id <- ClikaRT/tokenizer/tokenizer.h borrows self infallible |
tokenizer_free | |
tokenizer_from_bert_vocab | ClikaRT::tokenizer::Tokenizer::from_bert_vocab <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op) |
tokenizer_from_file | ClikaRT::tokenizer::Tokenizer::from_file <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op) |
tokenizer_from_gpt2 | ClikaRT::tokenizer::Tokenizer::from_gpt2 <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op) |
tokenizer_from_huggingface | ClikaRT::tokenizer::Tokenizer::from_huggingface <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op) |
tokenizer_from_sentencepiece | ClikaRT::tokenizer::Tokenizer::from_sentencepiece <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op) |
tokenizer_from_tekken | ClikaRT::tokenizer::Tokenizer::from_tekken <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op) |
tokenizer_from_tiktoken | ClikaRT::tokenizer::Tokenizer::from_tiktoken <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op) |
tokenizer_has_chat_template | ClikaRT::tokenizer::Tokenizer::has_chat_template <- ClikaRT/tokenizer/tokenizer.h borrows self infallible |
tokenizer_id_to_token | ClikaRT::tokenizer::Tokenizer::id_to_token <- ClikaRT/tokenizer/tokenizer.h borrows self sized-string protocol (contract §5) |
tokenizer_pad_id | ClikaRT::tokenizer::Tokenizer::pad_id <- ClikaRT/tokenizer/tokenizer.h borrows self infallible |
tokenizer_streaming_decoder | ClikaRT::tokenizer::Tokenizer::streaming_decoder <- ClikaRT/tokenizer/tokenizer.h borrows self infallible; single-owner handle: streaming_decoder_free frees it (NULL is a no-op) C++ defaults (dropped at this surface): skip_special_tokens=true |
tokenizer_token_to_id | ClikaRT::tokenizer::Tokenizer::token_to_id <- ClikaRT/tokenizer/tokenizer.h borrows self |
tokenizer_tokenize | ClikaRT::tokenizer::Tokenizer::tokenize <- ClikaRT/tokenizer/tokenizer.h borrows self single-owner list: token_list_count + the per-index field getters; token_list_free frees it C++ defaults (dropped at this surface): add_special_tokens=true |
tokenizer_vocab_size | ClikaRT::tokenizer::Tokenizer::vocab_size <- ClikaRT/tokenizer/tokenizer.h borrows self infallible |
Type aliases
using clika_rt_tokenizer
using clika_rt_tokenizer = struct clika_rt_tokenizer
Declared in clika_rt/clika_rt_core.h, line 302
Functions
tokenizer_apply_chat_template_messages_add_generation_prompt
clika_rt_error*(* tokenizer_apply_chat_template_messages_add_generation_prompt)(clika_rt_tokenizer *self, const char *messages_json, size_t messages_json_len, uint8_t add_generation_prompt, char *buf, size_t *len)
ClikaRT::tokenizer::Tokenizer::apply_chat_template <- ClikaRT/tokenizer/tokenizer.h one form of the overloaded C++ apply_chat_template; the _messages_add_generation_prompt suffix names this form borrows self messages_json is a JSON document as text (not retained) sized-string protocol (contract §5) C++ defaults (dropped at this surface): add_generation_prompt=true
Declared in clika_rt/clika_rt_core.h, line 9374
tokenizer_bos_id
int64_t(* tokenizer_bos_id)(clika_rt_tokenizer *self)
ClikaRT::tokenizer::Tokenizer::bos_id <- ClikaRT/tokenizer/tokenizer.h borrows self infallible
Declared in clika_rt/clika_rt_core.h, line 9331
tokenizer_decode
clika_rt_error*(* tokenizer_decode)(clika_rt_tokenizer *self, const int32_t *ids, size_t ids_count, uint8_t skip_special_tokens, char *buf, size_t *len)
ClikaRT::tokenizer::Tokenizer::decode <- ClikaRT/tokenizer/tokenizer.h borrows self sized-string protocol (contract §5) C++ defaults (dropped at this surface): skip_special_tokens=true
Declared in clika_rt/clika_rt_core.h, line 9252
tokenizer_decode_batch_encoded_skip_special_tokens
clika_rt_error*(* tokenizer_decode_batch_encoded_skip_special_tokens)(clika_rt_tokenizer *self, const clika_rt_encoded *encoded, uint8_t skip_special_tokens, clika_rt_string_list **out)
ClikaRT::tokenizer::Tokenizer::decode_batch <- ClikaRT/tokenizer/tokenizer.h one form of the overloaded C++ decode_batch; the _encoded_skip_special_tokens suffix names this form borrows self borrows encoded single-owner list: index it with string_list_count / string_list_get_at; string_list_free frees it C++ defaults (dropped at this surface): skip_special_tokens=true
Declared in clika_rt/clika_rt_core.h, line 9299
tokenizer_decode_batch_ids_cu_seqlens
clika_rt_error*(* tokenizer_decode_batch_ids_cu_seqlens)(clika_rt_tokenizer *self, clika_rt_tensor *ids, clika_rt_tensor *cu_seqlens, uint8_t skip_special_tokens, clika_rt_string_list **out)
ClikaRT::tokenizer::Tokenizer::decode_batch <- ClikaRT/tokenizer/tokenizer.h one form of the overloaded C++ decode_batch; the _ids_cu_seqlens suffix names this form borrows self borrows ids borrows cu_seqlens single-owner list: index it with string_list_count / string_list_get_at; string_list_free frees it C++ defaults (dropped at this surface): skip_special_tokens=true
Declared in clika_rt/clika_rt_core.h, line 9322
tokenizer_decode_batch_ids_skip_special_tokens
clika_rt_error*(* tokenizer_decode_batch_ids_skip_special_tokens)(clika_rt_tokenizer *self, clika_rt_tensor *ids, uint8_t skip_special_tokens, clika_rt_string_list **out)
ClikaRT::tokenizer::Tokenizer::decode_batch <- ClikaRT/tokenizer/tokenizer.h one form of the overloaded C++ decode_batch; the _ids_skip_special_tokens suffix names this form borrows self borrows ids single-owner list: index it with string_list_count / string_list_get_at; string_list_free frees it C++ defaults (dropped at this surface): skip_special_tokens=true
Declared in clika_rt/clika_rt_core.h, line 9310
tokenizer_encode
clika_rt_error*(* tokenizer_encode)(clika_rt_tokenizer *self, const char *text, size_t text_len, uint8_t add_special_tokens, int32_t *buf, size_t *count)
ClikaRT::tokenizer::Tokenizer::encode <- ClikaRT/tokenizer/tokenizer.h borrows self two-call sizing: buf == NULL sizes; count carries capacity in, required count out (contract §8) C++ defaults (dropped at this surface): add_special_tokens=true
Declared in clika_rt/clika_rt_core.h, line 9241
tokenizer_encode_batch
clika_rt_error*(* tokenizer_encode_batch)(clika_rt_tokenizer *self, const char *const *texts, const size_t *texts_lens, size_t texts_count, const clika_rt_encode_options *options, clika_rt_encoded **out)
ClikaRT::tokenizer::Tokenizer::encode_batch <- ClikaRT/tokenizer/tokenizer.h borrows self texts[i] spans texts_lens[i] bytes (the string protocol's sized form, per element) options may be NULL; every option takes its default single-owner handle: the id rows + optional tensor columns ride its getters; encoded_free frees it C++ defaults (dropped at this surface): options={}
Declared in clika_rt/clika_rt_core.h, line 9286
tokenizer_encode_chat_messages
clika_rt_error*(* tokenizer_encode_chat_messages)(clika_rt_tokenizer *self, const char *messages_json, size_t messages_json_len, uint8_t add_generation_prompt, uint8_t add_special_tokens, int32_t *buf, size_t *count)
ClikaRT::tokenizer::Tokenizer::encode_chat <- ClikaRT/tokenizer/tokenizer.h one form of the overloaded C++ encode_chat; the _messages suffix names this form borrows self messages_json is a JSON document as text (not retained) two-call sizing: buf == NULL sizes; count carries capacity in, required count out (contract §8) C++ defaults (dropped at this surface): add_generation_prompt=true, add_special_tokens=true
Declared in clika_rt/clika_rt_core.h, line 9387
tokenizer_encode_text
clika_rt_error*(* tokenizer_encode_text)(clika_rt_tokenizer *self, const char *text, size_t text_len, const clika_rt_encode_options *options, clika_rt_encoded **out)
ClikaRT::tokenizer::Tokenizer::encode_text <- ClikaRT/tokenizer/tokenizer.h borrows self options may be NULL; every option takes its default single-owner handle: the id rows + optional tensor columns ride its getters; encoded_free frees it C++ defaults (dropped at this surface): options={}
Declared in clika_rt/clika_rt_core.h, line 9274
tokenizer_eos_id
int64_t(* tokenizer_eos_id)(clika_rt_tokenizer *self)
ClikaRT::tokenizer::Tokenizer::eos_id <- ClikaRT/tokenizer/tokenizer.h borrows self infallible
Declared in clika_rt/clika_rt_core.h, line 9336
tokenizer_free
void(* tokenizer_free)(clika_rt_tokenizer *h)
Declared in clika_rt/clika_rt_core.h, line 653
tokenizer_from_bert_vocab
clika_rt_error*(* tokenizer_from_bert_vocab)(const char *dir, size_t dir_len, clika_rt_tokenizer **out)
ClikaRT::tokenizer::Tokenizer::from_bert_vocab <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op)
Declared in clika_rt/clika_rt_core.h, line 9227
tokenizer_from_file
clika_rt_error*(* tokenizer_from_file)(const char *path, size_t path_len, clika_rt_tokenizer **out)
ClikaRT::tokenizer::Tokenizer::from_file <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op)
Declared in clika_rt/clika_rt_core.h, line 9233
tokenizer_from_gpt2
clika_rt_error*(* tokenizer_from_gpt2)(const char *dir, size_t dir_len, clika_rt_tokenizer **out)
ClikaRT::tokenizer::Tokenizer::from_gpt2 <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op)
Declared in clika_rt/clika_rt_core.h, line 9215
tokenizer_from_huggingface
clika_rt_error*(* tokenizer_from_huggingface)(const char *path, size_t path_len, clika_rt_tokenizer **out)
ClikaRT::tokenizer::Tokenizer::from_huggingface <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op)
Declared in clika_rt/clika_rt_core.h, line 9197
tokenizer_from_sentencepiece
clika_rt_error*(* tokenizer_from_sentencepiece)(const char *path, size_t path_len, clika_rt_tokenizer **out)
ClikaRT::tokenizer::Tokenizer::from_sentencepiece <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op)
Declared in clika_rt/clika_rt_core.h, line 9203
tokenizer_from_tekken
clika_rt_error*(* tokenizer_from_tekken)(const char *path, size_t path_len, clika_rt_tokenizer **out)
ClikaRT::tokenizer::Tokenizer::from_tekken <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op)
Declared in clika_rt/clika_rt_core.h, line 9209
tokenizer_from_tiktoken
clika_rt_error*(* tokenizer_from_tiktoken)(const char *path, size_t path_len, clika_rt_tokenizer **out)
ClikaRT::tokenizer::Tokenizer::from_tiktoken <- ClikaRT/tokenizer/tokenizer.h single-owner handle: tokenizer_free frees it (NULL is a no-op)
Declared in clika_rt/clika_rt_core.h, line 9221
tokenizer_has_chat_template
uint8_t(* tokenizer_has_chat_template)(clika_rt_tokenizer *self)
ClikaRT::tokenizer::Tokenizer::has_chat_template <- ClikaRT/tokenizer/tokenizer.h borrows self infallible
Declared in clika_rt/clika_rt_core.h, line 9366
tokenizer_id_to_token
clika_rt_error*(* tokenizer_id_to_token)(clika_rt_tokenizer *self, int32_t id, char *buf, size_t *len)
ClikaRT::tokenizer::Tokenizer::id_to_token <- ClikaRT/tokenizer/tokenizer.h borrows self sized-string protocol (contract §5)
Declared in clika_rt/clika_rt_core.h, line 9358
tokenizer_pad_id
int64_t(* tokenizer_pad_id)(clika_rt_tokenizer *self)
ClikaRT::tokenizer::Tokenizer::pad_id <- ClikaRT/tokenizer/tokenizer.h borrows self infallible
Declared in clika_rt/clika_rt_core.h, line 9341
tokenizer_streaming_decoder
clika_rt_streaming_decoder*(* tokenizer_streaming_decoder)(clika_rt_tokenizer *self, uint8_t skip_special_tokens)
ClikaRT::tokenizer::Tokenizer::streaming_decoder <- ClikaRT/tokenizer/tokenizer.h borrows self infallible; single-owner handle: streaming_decoder_free frees it (NULL is a no-op) C++ defaults (dropped at this surface): skip_special_tokens=true
Declared in clika_rt/clika_rt_core.h, line 9399
tokenizer_token_to_id
clika_rt_error*(* tokenizer_token_to_id)(clika_rt_tokenizer *self, const char *token, size_t token_len, int32_t *out)
ClikaRT::tokenizer::Tokenizer::token_to_id <- ClikaRT/tokenizer/tokenizer.h borrows self
Declared in clika_rt/clika_rt_core.h, line 9350
tokenizer_tokenize
clika_rt_error*(* tokenizer_tokenize)(clika_rt_tokenizer *self, const char *text, size_t text_len, uint8_t add_special_tokens, clika_rt_token_list **out)
ClikaRT::tokenizer::Tokenizer::tokenize <- ClikaRT/tokenizer/tokenizer.h borrows self single-owner list: token_list_count + the per-index field getters; token_list_free frees it C++ defaults (dropped at this surface): add_special_tokens=true
Declared in clika_rt/clika_rt_core.h, line 9263
tokenizer_vocab_size
int64_t(* tokenizer_vocab_size)(clika_rt_tokenizer *self)
ClikaRT::tokenizer::Tokenizer::vocab_size <- ClikaRT/tokenizer/tokenizer.h borrows self infallible
Declared in clika_rt/clika_rt_core.h, line 9346