ExternalVendorCode/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2024-12-19 04:37:51 +00:00
Code Issues Actions 4 Packages Projects Releases Wiki Activity

talk-llama : sync llama.cpp

Browse Source
This commit is contained in:
Georgi Gerganov 2024-05-12 20:12:46 +03:00
parent fe179ae0cc
commit 3fa7d29876
7 changed files with 4613 additions and 1485 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3178
examples/talk-llama/llama.cpp
View File

File diff suppressed because it is too large Load Diff

155
examples/talk-llama/llama.h
View File

@ -37,9 +37,13 @@
#define LLAMA_FILE_MAGIC_GGLA 0x67676c61u // 'ggla'
#define LLAMA_FILE_MAGIC_GGSN 0x6767736eu // 'ggsn'
#define LLAMA_FILE_MAGIC_GGSQ 0x67677371u // 'ggsq'
#define LLAMA_SESSION_MAGIC LLAMA_FILE_MAGIC_GGSN
#define LLAMA_SESSION_VERSION 5
#define LLAMA_SESSION_VERSION 6
#define LLAMA_STATE_SEQ_MAGIC LLAMA_FILE_MAGIC_GGSQ
#define LLAMA_STATE_SEQ_VERSION 1
#ifdef __cplusplus
extern "C" {
@ -65,6 +69,23 @@ extern "C" {
LLAMA_VOCAB_TYPE_WPM = 3, // BERT tokenizer based on WordPiece
};
// pre-tokenization types
enum llama_vocab_pre_type {
LLAMA_VOCAB_PRE_TYPE_DEFAULT = 0,
LLAMA_VOCAB_PRE_TYPE_LLAMA3 = 1,
LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_LLM = 2,
LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_CODER = 3,
LLAMA_VOCAB_PRE_TYPE_FALCON = 4,
LLAMA_VOCAB_PRE_TYPE_MPT = 5,
LLAMA_VOCAB_PRE_TYPE_STARCODER = 6,
LLAMA_VOCAB_PRE_TYPE_GPT2 = 7,
LLAMA_VOCAB_PRE_TYPE_REFACT = 8,
LLAMA_VOCAB_PRE_TYPE_COMMAND_R = 9,
LLAMA_VOCAB_PRE_TYPE_QWEN2 = 10,
LLAMA_VOCAB_PRE_TYPE_OLMO = 11,
LLAMA_VOCAB_PRE_TYPE_DBRX = 12,
};
// note: these values should be synchronized with ggml_rope
// TODO: maybe move this enum to ggml.h (ggml_rope_type)
enum llama_rope_type {
@ -118,6 +139,7 @@ extern "C" {
LLAMA_FTYPE_MOSTLY_IQ2_M = 29, // except 1d tensors
LLAMA_FTYPE_MOSTLY_IQ4_XS = 30, // except 1d tensors
LLAMA_FTYPE_MOSTLY_IQ1_M = 31, // except 1d tensors
LLAMA_FTYPE_MOSTLY_BF16 = 32, // except 1d tensors
LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file
};
@ -155,7 +177,7 @@ extern "C" {
bool sorted;
} llama_token_data_array;
typedef bool (*llama_progress_callback)(float progress, void *ctx);
typedef bool (*llama_progress_callback)(float progress, void * user_data);
// Input data for llama_decode
// A llama_batch object can contain input about one or many sequences
@ -191,15 +213,19 @@ extern "C" {
LLAMA_KV_OVERRIDE_TYPE_INT,
LLAMA_KV_OVERRIDE_TYPE_FLOAT,
LLAMA_KV_OVERRIDE_TYPE_BOOL,
LLAMA_KV_OVERRIDE_TYPE_STR,
};
struct llama_model_kv_override {
char key[128];
enum llama_model_kv_override_type tag;
char key[128];
union {
int64_t int_value;
double float_value;
bool bool_value;
int64_t val_i64;
double val_f64;
bool val_bool;
char val_str[128];
};
};
@ -231,6 +257,7 @@ extern "C" {
bool vocab_only; // only load the vocabulary, no weights
bool use_mmap; // use mmap if possible
bool use_mlock; // force system to keep model in RAM
bool check_tensors; // validate model tensor data
};
struct llama_context_params {
@ -266,6 +293,7 @@ extern "C" {
bool logits_all; // the llama_decode() call computes all logits, not just the last one (DEPRECATED - set llama_batch.logits instead)
bool embeddings; // if true, extract embeddings (together with logits)
bool offload_kqv; // whether to offload the KQV ops (including the KV cache) to GPU
bool flash_attn; // whether to use flash attention
// Abort callback
// if it returns true, execution of llama_decode() will be aborted
@ -284,6 +312,7 @@ extern "C" {
bool quantize_output_tensor; // quantize output.weight
bool only_copy; // only copy tensors - ftype, allow_requantize and quantize_output_tensor are ignored
bool pure; // quantize all tensors to the default type
bool keep_split; // quantize to the same number of shards
void * imatrix; // pointer to importance matrix data
void * kv_overrides; // pointer to vector containing overrides
} llama_model_quantize_params;
@ -386,7 +415,9 @@ extern "C" {
LLAMA_API uint32_t llama_n_ubatch (const struct llama_context * ctx);
LLAMA_API uint32_t llama_n_seq_max (const struct llama_context * ctx);
LLAMA_API enum llama_vocab_type llama_vocab_type(const struct llama_model * model);
LLAMA_API enum llama_pooling_type llama_pooling_type(const struct llama_context * ctx);
LLAMA_API enum llama_vocab_type llama_vocab_type (const struct llama_model * model);
LLAMA_API enum llama_rope_type llama_rope_type (const struct llama_model * model);
LLAMA_API int32_t llama_n_vocab (const struct llama_model * model);
@ -518,11 +549,12 @@ extern "C" {
// Returns the number of used KV cells (i.e. have at least one sequence assigned to them)
LLAMA_API int32_t llama_get_kv_cache_used_cells(const struct llama_context * ctx);
// Clear the KV cache
// Clear the KV cache - both cell info is erased and KV data is zeroed
LLAMA_API void llama_kv_cache_clear(
struct llama_context * ctx);
// Removes all tokens that belong to the specified sequence and have positions in [p0, p1)
// Returns false if a partial sequence cannot be removed. Removing a whole sequence never fails
// seq_id < 0 : match any sequence
// p0 < 0 : [0, p1]
// p1 < 0 : [p0, inf)
@ -594,34 +626,92 @@ extern "C" {
// Returns the maximum size in bytes of the state (rng, logits, embedding
// and kv_cache) - will often be smaller after compacting tokens
LLAMA_API size_t llama_get_state_size(const struct llama_context * ctx);
LLAMA_API size_t llama_state_get_size(const struct llama_context * ctx);
LLAMA_API DEPRECATED(size_t llama_get_state_size(const struct llama_context * ctx),
"use llama_state_get_size instead");
// Copies the state to the specified destination address.
// Destination needs to have allocated enough memory.
// Returns the number of bytes copied
LLAMA_API size_t llama_copy_state_data(
LLAMA_API size_t llama_state_get_data(
struct llama_context * ctx,
uint8_t * dst);
LLAMA_API DEPRECATED(size_t llama_copy_state_data(
struct llama_context * ctx,
uint8_t * dst),
"use llama_state_get_data instead");
// Set the state reading from the specified address
// Returns the number of bytes read
LLAMA_API size_t llama_set_state_data(
LLAMA_API size_t llama_state_set_data(
struct llama_context * ctx,
const uint8_t * src);
LLAMA_API DEPRECATED(size_t llama_set_state_data(
struct llama_context * ctx,
const uint8_t * src),
"use llama_state_set_data instead");
// Save/load session file
LLAMA_API bool llama_load_session_file(
LLAMA_API bool llama_state_load_file(
struct llama_context * ctx,
const char * path_session,
llama_token * tokens_out,
size_t n_token_capacity,
size_t * n_token_count_out);
LLAMA_API DEPRECATED(bool llama_load_session_file(
struct llama_context * ctx,
const char * path_session,
llama_token * tokens_out,
size_t n_token_capacity,
size_t * n_token_count_out),
"use llama_state_load_file instead");
LLAMA_API bool llama_save_session_file(
LLAMA_API bool llama_state_save_file(
struct llama_context * ctx,
const char * path_session,
const llama_token * tokens,
size_t n_token_count);
LLAMA_API DEPRECATED(bool llama_save_session_file(
struct llama_context * ctx,
const char * path_session,
const llama_token * tokens,
size_t n_token_count),
"use llama_state_save_file instead");
// Get the exact size needed to copy the KV cache of a single sequence
LLAMA_API size_t llama_state_seq_get_size(
struct llama_context * ctx,
llama_seq_id seq_id);
// Copy the KV cache of a single sequence into the specified buffer
LLAMA_API size_t llama_state_seq_get_data(
struct llama_context * ctx,
uint8_t * dst,
llama_seq_id seq_id);
// Copy the sequence data (originally copied with `llama_state_seq_get_data`) into the specified sequence
// Returns:
// - Positive: Ok
// - Zero: Failed to load
LLAMA_API size_t llama_state_seq_set_data(
struct llama_context * ctx,
const uint8_t * src,
llama_seq_id dest_seq_id);
LLAMA_API size_t llama_state_seq_save_file(
struct llama_context * ctx,
const char * filepath,
llama_seq_id seq_id,
const llama_token * tokens,
size_t n_token_count);
LLAMA_API size_t llama_state_seq_load_file(
struct llama_context * ctx,
const char * filepath,
llama_seq_id dest_seq_id,
llama_token * tokens_out,
size_t n_token_capacity,
size_t * n_token_count_out);
//
// Decoding
@ -684,8 +774,9 @@ extern "C" {
// Cols: n_vocab
LLAMA_API float * llama_get_logits(struct llama_context * ctx);
// Logits for the ith token. Equivalent to:
// Logits for the ith token. For positive indices, Equivalent to:
// llama_get_logits(ctx) + ctx->output_ids[i]*n_vocab
// Negative indicies can be used to access logits in reverse order, -1 is the last logit.
// returns NULL for invalid ids.
LLAMA_API float * llama_get_logits_ith(struct llama_context * ctx, int32_t i);
@ -697,8 +788,9 @@ extern "C" {
// Otherwise, returns NULL.
LLAMA_API float * llama_get_embeddings(struct llama_context * ctx);
// Get the embeddings for the ith token. Equivalent to:
// Get the embeddings for the ith token. For positive indices, Equivalent to:
// llama_get_embeddings(ctx) + ctx->output_ids[i]*n_embd
// Negative indicies can be used to access embeddings in reverse order, -1 is the last embedding.
// shape: [n_embd] (1-dimensional)
// returns NULL for invalid ids.
LLAMA_API float * llama_get_embeddings_ith(struct llama_context * ctx, int32_t i);
@ -718,9 +810,14 @@ extern "C" {
LLAMA_API enum llama_token_type llama_token_get_type(const struct llama_model * model, llama_token token);
// Check if the token is supposed to end generation (end-of-generation, eg. EOS, EOT, etc.)
LLAMA_API bool llama_token_is_eog(const struct llama_model * model, llama_token token);
// Special tokens
LLAMA_API llama_token llama_token_bos(const struct llama_model * model); // beginning-of-sentence
LLAMA_API llama_token llama_token_eos(const struct llama_model * model); // end-of-sentence
LLAMA_API llama_token llama_token_cls(const struct llama_model * model); // classification
LLAMA_API llama_token llama_token_sep(const struct llama_model * model); // sentence separator
LLAMA_API llama_token llama_token_nl (const struct llama_model * model); // next-line
// Returns -1 if unknown, 1 for true or 0 for false.
@ -729,7 +826,7 @@ extern "C" {
// Returns -1 if unknown, 1 for true or 0 for false.
LLAMA_API int32_t llama_add_eos_token(const struct llama_model * model);
// codellama infill tokens
// Codellama infill tokens
LLAMA_API llama_token llama_token_prefix(const struct llama_model * model); // Beginning of infill prefix
LLAMA_API llama_token llama_token_middle(const struct llama_model * model); // Beginning of infill middle
LLAMA_API llama_token llama_token_suffix(const struct llama_model * model); // Beginning of infill suffix
@ -743,26 +840,28 @@ extern "C" {
/// @param tokens The tokens pointer must be large enough to hold the resulting tokens.
/// @return Returns the number of tokens on success, no more than n_tokens_max
/// @return Returns a negative number on failure - the number of tokens that would have been returned
/// @param special Allow tokenizing special and/or control tokens which otherwise are not exposed and treated as plaintext.
/// Does not insert a leading space.
/// @param parse_special Allow tokenizing special and/or control tokens which otherwise are not exposed and treated
/// as plaintext. Does not insert a leading space.
LLAMA_API int32_t llama_tokenize(
const struct llama_model * model,
const char * text,
int32_t text_len,
llama_token * tokens,
int32_t n_tokens_max,
bool add_bos,
bool special);
bool add_special,
bool parse_special);
// Token Id -> Piece.
// Uses the vocabulary in the provided context.
// Does not write null terminator to the buffer.
// User code is responsible to remove the leading whitespace of the first non-BOS token when decoding multiple tokens.
// @param special If true, special tokens are rendered in the output.
LLAMA_API int32_t llama_token_to_piece(
const struct llama_model * model,
llama_token token,
char * buf,
int32_t length);
int32_t length,
bool special);
/// Apply chat template. Inspired by hf apply_chat_template() on python.
/// Both "model" and "custom_template" are optional, but at least one is required. "custom_template" has higher precedence than "model"
@ -915,7 +1014,7 @@ extern "C" {
struct llama_context * ctx,
llama_token_data_array * candidates);
/// @details Randomly selects a token from the candidates based on their probabilities.
/// @details Randomly selects a token from the candidates based on their probabilities using the RNG of ctx.
LLAMA_API llama_token llama_sample_token(
struct llama_context * ctx,
llama_token_data_array * candidates);
@ -1002,8 +1101,9 @@ extern "C" {
// Internal API to be implemented by llama.cpp and used by tests/benchmarks only
#ifdef LLAMA_API_INTERNAL
#include <vector>
#include <random>
#include <string>
#include <vector>
struct ggml_tensor;
@ -1030,15 +1130,20 @@ const std::vector<std::pair<std::string, struct ggml_tensor *>> & llama_internal
struct llama_context * ctx
);
std::vector<std::vector<const llama_grammar_element *>> llama_grammar_accept(
void llama_grammar_accept(
const std::vector<std::vector<llama_grammar_element>> & rules,
const std::vector<std::vector<const llama_grammar_element *>> & stacks,
const uint32_t chr);
const uint32_t chr,
std::vector<std::vector<const llama_grammar_element *>> & new_stacks);
std::pair<std::vector<uint32_t>, llama_partial_utf8> decode_utf8(
const std::string & src,
llama_partial_utf8 partial_start);
// Randomly selects a token from the candidates based on their probabilities using given std::mt19937.
// This is a temporary workaround in order to fix race conditions when sampling with multiple sequences.
llama_token llama_sample_token_with_rng(struct llama_context * ctx, llama_token_data_array * candidates, std::mt19937 & rng);
#endif // LLAMA_API_INTERNAL
#endif // LLAMA_H

4
examples/talk-llama/talk-llama.cpp
View File

@ -35,10 +35,10 @@ std::vector<llama_token> llama_tokenize(struct llama_context * ctx, const std::s
std::string llama_token_to_piece(const struct llama_context * ctx, llama_token token) {
std::vector<char> result(8, 0);
const int n_tokens = llama_token_to_piece(llama_get_model(ctx), token, result.data(), result.size());
const int n_tokens = llama_token_to_piece(llama_get_model(ctx), token, result.data(), result.size(), false);
if (n_tokens < 0) {
result.resize(-n_tokens);
int check = llama_token_to_piece(llama_get_model(ctx), token, result.data(), result.size());
int check = llama_token_to_piece(llama_get_model(ctx), token, result.data(), result.size(), false);
GGML_ASSERT(check == -n_tokens);
} else {
result.resize(n_tokens);

1844
examples/talk-llama/unicode-data.cpp
View File

File diff suppressed because it is too large Load Diff

3
examples/talk-llama/unicode-data.h
View File

@ -5,8 +5,9 @@
#include <utility>
#include <vector>
extern const std::vector<std::pair<uint32_t, uint32_t>> unicode_ranges_digit;
extern const std::vector<std::pair<uint32_t, uint32_t>> unicode_ranges_number;
extern const std::vector<std::pair<uint32_t, uint32_t>> unicode_ranges_letter;
extern const std::vector<std::pair<uint32_t, uint32_t>> unicode_ranges_separator;
extern const std::vector<std::pair<uint32_t, uint32_t>> unicode_ranges_whitespace;
extern const std::vector<std::pair<uint32_t, uint32_t>> unicode_ranges_accent_mark;
extern const std::vector<std::pair<uint32_t, uint32_t>> unicode_ranges_punctuation;

639
examples/talk-llama/unicode.cpp
View File

@ -5,11 +5,15 @@
#include <cstddef>
#include <cstdint>
#include <map>
#include <regex>
#include <stdexcept>
#include <string>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
#include <locale>
#include <codecvt>
static std::string unicode_cpts_to_utf8(const std::vector<uint32_t> & cps) {
std::string result;
@ -53,23 +57,22 @@ static uint32_t unicode_cpt_from_utf8(const std::string & utf8, size_t & offset)
offset += 4;
return result;
}
throw std::invalid_argument("invalid string");
throw std::invalid_argument("failed to convert utf8 to codepoint");
}
static std::vector<uint16_t> unicode_cpt_to_utf16(uint32_t cp) {
std::vector<uint16_t> result;
if (/* 0x0000 <= cp && */ cp <= 0xffff) {
result.emplace_back(cp);
}
else if (0x10000 <= cp && cp <= 0x10ffff) {
result.emplace_back(0xd800 | ((cp - 0x10000) >> 10));
result.emplace_back(0xdc00 | ((cp - 0x10000) & 0x03ff));
}
else {
throw std::invalid_argument("invalid cpt");
}
return result;
}
//static std::vector<uint16_t> unicode_cpt_to_utf16(uint32_t cp) {
// std::vector<uint16_t> result;
// if (/* 0x0000 <= cp && */ cp <= 0xffff) {
// result.emplace_back(cp);
// return result;
// }
// if (0x10000 <= cp && cp <= 0x10ffff) {
// result.emplace_back(0xd800 | ((cp - 0x10000) >> 10));
// result.emplace_back(0xdc00 | ((cp - 0x10000) & 0x03ff));
// return result;
// }
// throw std::invalid_argument("failed to convert codepoint to utf16");
//}
//static std::vector<uint16_t> unicode_cpts_to_utf16(const std::vector<uint32_t> & cps) {
// std::vector<uint16_t> result;
@ -80,56 +83,56 @@ static std::vector<uint16_t> unicode_cpt_to_utf16(uint32_t cp) {
// return result;
//}
static uint32_t cpt_from_utf16(const std::vector<uint16_t> & utf16, size_t & offset) {
assert(offset < utf16.size());
if (((utf16[0] >> 10) << 10) != 0xd800) {
auto result = utf16[offset + 0];
offset += 1;
return result;
}
if (offset + 1 >= utf16.size() || !((utf16[1] & 0xdc00) == 0xdc00)) {
throw std::invalid_argument("invalid character");
}
auto result = 0x10000 + (((utf16[0] & 0x03ff) << 10) | (utf16[1] & 0x03ff));
offset += 2;
return result;
}
//static uint32_t unicode_cpt_from_utf16(const std::vector<uint16_t> & utf16, size_t & offset) {
// assert(offset < utf16.size());
// if (((utf16[0] >> 10) << 10) != 0xd800) {
// auto result = utf16[offset + 0];
// offset += 1;
// return result;
// }
//
// if (offset + 1 >= utf16.size() || !((utf16[1] & 0xdc00) == 0xdc00)) {
// throw std::invalid_argument("invalid character");
// }
//
// auto result = 0x10000 + (((utf16[0] & 0x03ff) << 10) | (utf16[1] & 0x03ff));
// offset += 2;
// return result;
//}
//static std::vector<uint32_t> unicode_cpts_from_utf16(const std::vector<uint16_t> & utf16) {
// std::vector<uint32_t> result;
// size_t offset = 0;
// while (offset < utf16.size()) {
// result.push_back(cpt_from_utf16(utf16, offset));
// result.push_back(unicode_cpt_from_utf16(utf16, offset));
// }
// return result;
//}
static std::unordered_map<uint32_t, int> unicode_cpt_type_map() {
std::unordered_map<uint32_t, int> cpt_types;
for (auto p : unicode_ranges_digit) {
for (auto i = p.first; i <= p.second; ++ i) {
cpt_types[i] = CODEPOINT_TYPE_DIGIT;
for (auto p : unicode_ranges_number) {
for (auto i = p.first; i <= p.second; ++i) {
cpt_types[i] = CODEPOINT_TYPE_NUMBER;
}
}
for (auto p : unicode_ranges_letter) {
for (auto i = p.first; i <= p.second; ++ i) {
for (auto i = p.first; i <= p.second; ++i) {
cpt_types[i] = CODEPOINT_TYPE_LETTER;
}
}
for (auto p : unicode_ranges_whitespace) {
for (auto i = p.first; i <= p.second; ++ i) {
cpt_types[i] = CODEPOINT_TYPE_WHITESPACE;
for (auto p : unicode_ranges_separator) {
for (auto i = p.first; i <= p.second; ++i) {
cpt_types[i] = CODEPOINT_TYPE_SEPARATOR;
}
}
for (auto p : unicode_ranges_accent_mark) {
for (auto i = p.first; i <= p.second; ++ i) {
for (auto i = p.first; i <= p.second; ++i) {
cpt_types[i] = CODEPOINT_TYPE_ACCENT_MARK;
}
}
for (auto p : unicode_ranges_punctuation) {
for (auto i = p.first; i <= p.second; ++ i) {
for (auto i = p.first; i <= p.second; ++i) {
cpt_types[i] = CODEPOINT_TYPE_PUNCTUATION;
}
}
@ -139,7 +142,7 @@ static std::unordered_map<uint32_t, int> unicode_cpt_type_map() {
}
}
for (auto p : unicode_ranges_control) {
for (auto i = p.first; i <= p.second; ++ i) {
for (auto i = p.first; i <= p.second; ++i) {
cpt_types[i] = CODEPOINT_TYPE_CONTROL;
}
}
@ -194,34 +197,395 @@ static std::unordered_map<std::string, uint8_t> unicode_utf8_to_byte_map() {
return map;
}
static inline std::wstring unicode_wstring_from_utf8(const std::string & s) {
std::wstring_convert<std::codecvt_utf8<wchar_t>> conv;
return conv.from_bytes(s);
}
static std::vector<std::string> unicode_byte_encoding_process(const std::vector<std::string> & bpe_words) {
std::vector<std::string> bpe_encoded_words;
for (const auto & word : bpe_words) {
std::string text_utf;
auto utf_word = unicode_cpts_from_utf8(word);
for (size_t i = 0; i < utf_word.size(); ++i) {
text_utf += unicode_cpt_to_utf8(utf_word[i]);
}
std::string encoded_token;
for (char & c : text_utf) {
encoded_token += unicode_byte_to_utf8(c);
}
bpe_encoded_words.emplace_back(encoded_token);
}
return bpe_encoded_words;
}
// GPT2 system regex: 's|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+
static std::vector<size_t> unicode_regex_split_custom_gpt2(const std::string & text, const std::vector<size_t> & offsets) {
std::vector<size_t> bpe_offsets; // store the offset of each word
bpe_offsets.reserve(offsets.size()); // Reserve memory for the approximate size
const auto cpts = unicode_cpts_from_utf8(text);
size_t start = 0;
for (auto offset : offsets) {
const size_t offset_ini = start;
const size_t offset_end = start + offset;
assert(offset_end <= cpts.size());
start = offset_end;
auto _get_cpt = [&] (const size_t pos) -> char32_t {
return (offset_ini <= pos && pos < offset_end) ? cpts[pos] : 0;
};
auto _get_cpt_type = [&] (const size_t pos) -> int {
return (offset_ini <= pos && pos < offset_end) ? unicode_cpt_type(cpts[pos]) : CODEPOINT_TYPE_UNIDENTIFIED;
};
size_t _prev_end = offset_ini;
auto _add_token = [&] (const size_t end) -> size_t {
assert(_prev_end <= end && end <= offset_end);
size_t len = end - _prev_end;
if (len > 0) {
bpe_offsets.push_back(len);
}
_prev_end = end;
//if (len > 0) {
// std::string s = "";
// for(size_t p = end-len; p < end; p++)
// s += unicode_cpt_to_utf8(cpts[p]);
// printf(">>> '%s'\n", s.c_str());
//}
return len;
};
for (size_t pos = offset_ini; pos < offset_end; /*pos++*/ ) {
const char32_t cpt = _get_cpt(pos);
const int cpt_type = _get_cpt_type(pos);
// regex: 's|'t|'re|'ve|'m|'ll|'d
if (cpt == '\'' && pos+1 < offset_end) {
char32_t cpt_next = _get_cpt(pos+1);
if (cpt_next == 's' || cpt_next == 't' || cpt_next == 'm' || cpt_next == 'd') {
pos += _add_token(pos+2);
continue;
}
if (pos+2 < offset_end) {
char32_t cpt_next_next = _get_cpt(pos+2);
if ((cpt_next == 'r' && cpt_next_next == 'e') ||
(cpt_next == 'v' && cpt_next_next == 'e') ||
(cpt_next == 'l' && cpt_next_next == 'l')) {
pos += _add_token(pos+3);
continue;
}
}
}
char32_t cpt2 = (cpt == ' ' ? _get_cpt(pos+1) : cpt);
int cpt2_type = (cpt == ' ' ? _get_cpt_type(pos+1) : cpt_type);
// regex: <space>?\p{L}+
if (cpt2_type == CODEPOINT_TYPE_LETTER) {
pos += (cpt == ' ');
while (cpt2_type == CODEPOINT_TYPE_LETTER) {
cpt2_type = _get_cpt_type(++pos);
}
_add_token(pos);
continue;
}
// regex: <space>?\p{N}+
if (cpt2_type == CODEPOINT_TYPE_NUMBER) {
pos += (cpt == ' ');
while (cpt2_type == CODEPOINT_TYPE_NUMBER) {
cpt2_type = _get_cpt_type(++pos);
}
_add_token(pos);
continue;
}
// regex: <space>?[^\s\p{L}\p{N}]+
if (!unicode_cpt_is_whitespace(cpt2) && cpt2_type != CODEPOINT_TYPE_LETTER && cpt2_type != CODEPOINT_TYPE_NUMBER && cpt2_type != CODEPOINT_TYPE_UNIDENTIFIED) {
pos += (cpt == ' ');
while (!unicode_cpt_is_whitespace(cpt2) && cpt2_type != CODEPOINT_TYPE_LETTER && cpt2_type != CODEPOINT_TYPE_NUMBER && cpt2_type != CODEPOINT_TYPE_UNIDENTIFIED) {
cpt2_type = _get_cpt_type(++pos);
cpt2 = _get_cpt(pos);
}
_add_token(pos);
continue;
}
size_t num_whitespaces = 0;
while (unicode_cpt_is_whitespace(_get_cpt(pos+num_whitespaces))) {
num_whitespaces++;
}
// regex: \s+(?!\S)
if (num_whitespaces > 1 && _get_cpt(pos+num_whitespaces) != 0) {
pos += num_whitespaces - 1;
_add_token(pos);
continue;
}
// regex: \s+
if (num_whitespaces > 0) {
pos += num_whitespaces;
_add_token(pos);
continue;
}
// no matches
_add_token(++pos);
}
}
return bpe_offsets;
}
// LLAMA3 system regex: "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}{1,3}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+"
static std::vector<size_t> unicode_regex_split_custom_llama3(const std::string & text, const std::vector<size_t> & offsets) {
std::vector<size_t> bpe_offsets; // store the offset of each word
bpe_offsets.reserve(offsets.size()); // Reserve memory for the approximate size
const auto cpts = unicode_cpts_from_utf8(text);
size_t start = 0;
for (auto offset : offsets) {
const size_t offset_ini = start;
const size_t offset_end = start + offset;
assert(offset_end <= cpts.size());
start = offset_end;
auto _get_cpt = [&] (const size_t pos) -> char32_t {
return (offset_ini <= pos && pos < offset_end) ? cpts[pos] : 0;
};
auto _get_cpt_type = [&] (const size_t pos) -> int {
return (offset_ini <= pos && pos < offset_end) ? unicode_cpt_type(cpts[pos]) : CODEPOINT_TYPE_UNIDENTIFIED;
};
size_t _prev_end = offset_ini;
auto _add_token = [&] (const size_t end) -> size_t {
assert(_prev_end <= end && end <= offset_end);
size_t len = end - _prev_end;
if (len > 0) {
bpe_offsets.push_back(len);
}
_prev_end = end;
//if (len > 0) {
// std::string s = "";
// for(size_t p = end-len; p < end; p++)
// s += unicode_cpt_to_utf8(cpts[p]);
// printf(">>> '%s'\n", s.c_str());
//}
return len;
};
for (size_t pos = offset_ini; pos < offset_end; /*pos++*/ ) {
const char32_t cpt = _get_cpt(pos);
const int cpt_type = _get_cpt_type(pos);
// regex: (?i:'s|'t|'re|'ve|'m|'ll|'d) // case insensitive
if (cpt == '\'' && pos+1 < offset_end) {
char32_t cpt_next = unicode_tolower(_get_cpt(pos+1));
if (cpt_next == 's' || cpt_next == 't' || cpt_next == 'm' || cpt_next == 'd') {
pos += _add_token(pos+2);
continue;
}
if (pos+2 < offset_end) {
char32_t cpt_next_next = unicode_tolower(_get_cpt(pos+2));
if ((cpt_next == 'r' && cpt_next_next == 'e') ||
(cpt_next == 'v' && cpt_next_next == 'e') ||
(cpt_next == 'l' && cpt_next_next == 'l')) {
pos += _add_token(pos+3);
continue;
}
}
}
// regex: [^\r\n\p{L}\p{N}]?\p{L}+ //####FIXME: the first \p{L} is correct?
if (cpt != '\r' && cpt != '\n' && /*cpt_type != CODEPOINT_TYPE_LETTER &&*/ cpt_type != CODEPOINT_TYPE_NUMBER) {
if (cpt_type == CODEPOINT_TYPE_LETTER || _get_cpt_type(pos+1) == CODEPOINT_TYPE_LETTER) { // one or more letters
pos++;
while (_get_cpt_type(pos) == CODEPOINT_TYPE_LETTER) {
pos++;
}
_add_token(pos);
continue;
}
}
// regex: \p{N}{1,3}
if (cpt_type == CODEPOINT_TYPE_NUMBER) {
size_t ini = pos;
while (_get_cpt_type(pos) == CODEPOINT_TYPE_NUMBER) {
if (++pos - ini >= 3 ) {
_add_token(pos);
ini = pos;
}
}
_add_token(pos);
continue;
}
// regex: <space>?[^\s\p{L}\p{N}]+[\r\n]*
char32_t cpt2 = (cpt == ' ' ? _get_cpt(pos+1) : cpt);
int cpt2_type = (cpt == ' ' ? _get_cpt_type(pos+1) : cpt_type);
if (!unicode_cpt_is_whitespace(cpt2) && cpt2_type != CODEPOINT_TYPE_LETTER && cpt2_type != CODEPOINT_TYPE_NUMBER && cpt2_type != CODEPOINT_TYPE_UNIDENTIFIED) {
pos += (cpt == ' ');
while (!unicode_cpt_is_whitespace(cpt2) && cpt2_type != CODEPOINT_TYPE_LETTER && cpt2_type != CODEPOINT_TYPE_NUMBER && cpt2_type != CODEPOINT_TYPE_UNIDENTIFIED) {
cpt2_type = _get_cpt_type(++pos);
cpt2 = _get_cpt(pos);
}
while (cpt2 == '\r' || cpt2 == '\n') {
cpt2 = _get_cpt(++pos);
}
_add_token(pos);
continue;
}
size_t num_whitespaces = 0;
size_t last_end_r_or_n = 0;
while (unicode_cpt_is_whitespace(_get_cpt(pos+num_whitespaces))) {
char32_t cpt2 = _get_cpt(pos+num_whitespaces);
if (cpt2 == '\r' || cpt2 == '\n') {
last_end_r_or_n = pos + num_whitespaces + 1;
}
num_whitespaces++;
}
// regex: \s*[\r\n]+
if (last_end_r_or_n > 0) {
pos = last_end_r_or_n;
_add_token(pos);
continue;
}
// regex: \s+(?!\S)
if (num_whitespaces > 1 && _get_cpt(pos+num_whitespaces) != 0) {
pos += num_whitespaces - 1;
_add_token(pos);
continue;
}
// regex: \s+
if (num_whitespaces > 0) {
pos += num_whitespaces;
_add_token(pos);
continue;
}
// no matches
_add_token(++pos);
}
}
return bpe_offsets;
}
// use std::wregex to split the text
static std::vector<size_t> unicode_regex_split_stl(const std::wstring & wtext, const std::wstring & regex_expr, const std::vector<size_t> & offsets) {
std::wregex expr(regex_expr);
std::vector<size_t> bpe_offsets; // store the offset of each word
bpe_offsets.reserve(offsets.size()); // Reserve memory for the approximate size
size_t start = 0;
for (auto offset : offsets) {
std::wcregex_iterator it(wtext.data() + start, wtext.data() + start + offset, expr);
std::wcregex_iterator end;
int64_t start_idx = 0;
while (it != end) {
std::wcmatch match = *it;
if (match.position() > start_idx) {
bpe_offsets.emplace_back(match.position() - start_idx);
}
bpe_offsets.emplace_back(match.length());
start_idx = match.position() + match.length();
++it;
}
if (start_idx < (int64_t) offset) {
bpe_offsets.emplace_back(offset - start_idx);
}
start += offset;
}
return bpe_offsets;
}
// use std::regex to split the text
static std::vector<size_t> unicode_regex_split_stl(const std::string & text, const std::string & regex_expr, const std::vector<size_t> & offsets) {
std::regex expr(regex_expr);
std::vector<size_t> bpe_offsets; // store the offset of each word
bpe_offsets.reserve(offsets.size()); // Reserve memory for the approximate size
size_t start = 0;
for (auto offset : offsets) {
std::cregex_iterator it(text.data() + start, text.data() + start + offset, expr);
std::cregex_iterator end;
int64_t start_idx = 0;
while (it != end) {
std::cmatch match = *it;
if (match.position() > start_idx) {
bpe_offsets.emplace_back(match.position() - start_idx);
}
bpe_offsets.emplace_back(match.length());
start_idx = match.position() + match.length();
++it;
}
if (start_idx < (int64_t) offset) {
bpe_offsets.emplace_back(offset - start_idx);
}
start += offset;
}
return bpe_offsets;
}
static std::vector<size_t> unicode_regex_split_custom(const std::string & text, const std::string & regex_expr, const std::vector<size_t> & offsets) {
std::vector<size_t> bpe_offsets;
if (regex_expr == "'s|'t|'re|'ve|'m|'ll|'d| ?\\p{L}+| ?\\p{N}+| ?[^\\s\\p{L}\\p{N}]+|\\s+(?!\\S)") {
bpe_offsets = unicode_regex_split_custom_gpt2(text, offsets);
} else if (
regex_expr == "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+" ||
regex_expr == "(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+") {
bpe_offsets = unicode_regex_split_custom_llama3(text, offsets);
}
return bpe_offsets;
}
//
// interface
//
std::string unicode_cpt_to_utf8(uint32_t cp) {
std::string result;
if (/* 0x00 <= cp && */ cp <= 0x7f) {
result.push_back(cp);
return result;
}
else if (0x80 <= cp && cp <= 0x7ff) {
if (0x80 <= cp && cp <= 0x7ff) {
result.push_back(0xc0 | ((cp >> 6) & 0x1f));
result.push_back(0x80 | (cp & 0x3f));
return result;
}
else if (0x800 <= cp && cp <= 0xffff) {
if (0x800 <= cp && cp <= 0xffff) {
result.push_back(0xe0 | ((cp >> 12) & 0x0f));
result.push_back(0x80 | ((cp >> 6) & 0x3f));
result.push_back(0x80 | (cp & 0x3f));
return result;
}
else if (0x10000 <= cp && cp <= 0x10ffff) {
if (0x10000 <= cp && cp <= 0x10ffff) {
result.push_back(0xf0 | ((cp >> 18) & 0x07));
result.push_back(0x80 | ((cp >> 12) & 0x3f));
result.push_back(0x80 | ((cp >> 6) & 0x3f));
result.push_back(0x80 | (cp & 0x3f));
}
else {
throw std::invalid_argument("invalid codepoint");
}
return result;
}
throw std::invalid_argument("invalid codepoint");
}
std::vector<uint32_t> unicode_cpts_normalize_nfd(const std::vector<uint32_t> & cpts) {
@ -261,6 +625,19 @@ int unicode_cpt_type(const std::string & utf8) {
return unicode_cpt_type(unicode_cpt_from_utf8(utf8, offset));
}
bool unicode_cpt_is_whitespace(uint32_t cp) {
static const std::unordered_set<uint32_t> is_whitespace = [] {
std::unordered_set<uint32_t> is_whitespace;
for (auto p : unicode_ranges_whitespace) {
for (auto i = p.first; i <= p.second; ++i) {
is_whitespace.insert(i);
}
}
return is_whitespace;
}();
return (bool)is_whitespace.count(cp);
}
std::string unicode_byte_to_utf8(uint8_t byte) {
static std::unordered_map<uint8_t, std::string> map = unicode_byte_to_utf8_map();
return map.at(byte);
@ -275,3 +652,167 @@ char32_t unicode_tolower(char32_t cp) {
auto it = unicode_map_lowercase.find(cp);
return it == unicode_map_lowercase.end() ? cp : it->second;
}
std::vector<std::string> unicode_regex_split(const std::string & text, const std::vector<std::string> & regex_exprs) {
// unicode categories
static const std::map<std::string, int> k_ucat_enum = {
{ "\\p{N}", CODEPOINT_TYPE_NUMBER },
{ "\\p{L}", CODEPOINT_TYPE_LETTER },
{ "\\p{P}", CODEPOINT_TYPE_PUNCTUATION },
};
static const std::map<int, int> k_ucat_cpt = {
{ CODEPOINT_TYPE_NUMBER, 0xD1 },
{ CODEPOINT_TYPE_LETTER, 0xD2 },
{ CODEPOINT_TYPE_PUNCTUATION, 0xD3 },
};
static const std::map<int, std::string> k_ucat_map = {
{ CODEPOINT_TYPE_NUMBER, "\x30-\x39" }, // 0-9
{ CODEPOINT_TYPE_LETTER, "\x41-\x5A\x61-\x7A" }, // A-Za-z
{ CODEPOINT_TYPE_PUNCTUATION, "\x21-\x23\x25-\x2A\x2C-\x2F\x3A-\x3B\x3F-\x40\\\x5B-\\\x5D\x5F\\\x7B\\\x7D" }, // !-#%-*,-/:-;?-@\[-\]_\{\}
};
// compute collapsed codepoints only if needed by at least one regex
bool need_collapse = false;
for (auto & regex_expr : regex_exprs) {
// search for unicode categories
for (const auto & ucat : k_ucat_enum) {
if (std::string::npos != regex_expr.find(ucat.first)) {
need_collapse = true;
break;
}
}
}
const auto cpts = unicode_cpts_from_utf8(text);
// generate a "collapsed" representation of the text, where all codepoints are replaced by a single byte
// ref: https://github.com/ggerganov/llama.cpp/pull/6920#issuecomment-2081479935
std::string text_collapsed;
if (need_collapse) {
// collapse all unicode categories
text_collapsed.resize(cpts.size());
for (size_t i = 0; i < cpts.size(); ++i) {
// keep single-byte codepoints as is
if (cpts[i] < 128) {
text_collapsed[i] = cpts[i];
continue;
}
const int cpt_type = unicode_cpt_type(cpts[i]);
if (k_ucat_cpt.find(cpt_type) != k_ucat_cpt.end()) {
text_collapsed[i] = k_ucat_cpt.at(cpt_type);
} else {
text_collapsed[i] = (char) 0xD0; // fallback
}
}
}
std::vector<size_t> bpe_offsets = { cpts.size() };
for (auto & regex_expr : regex_exprs) {
// first, see if we have an efficient custom regex implementation
auto tmp = unicode_regex_split_custom(text, regex_expr, bpe_offsets);
if (!tmp.empty()) {
bpe_offsets = std::move(tmp);
continue;
}
// fallback to general-purpose std::regex / std::wregex
try {
// if a unicode category is used in the regex, we use the collapsed text and replace the unicode category
// with the corresponding collapsed representation
bool use_collapsed = false;
for (auto & ucat : k_ucat_enum) {
if (std::string::npos != regex_expr.find(ucat.first)) {
use_collapsed = true;
break;
}
}
if (use_collapsed) {
// sanity-check that the original regex does not contain any non-ASCII characters
const auto cpts_regex = unicode_cpts_from_utf8(regex_expr);
for (size_t i = 0; i < cpts_regex.size(); ++i) {
if (cpts_regex[i] >= 128) {
throw std::runtime_error("Regex includes both unicode categories and non-ASCII characters - not supported");
}
}
// generate a collapsed representation of the regex
std::string regex_expr_collapsed;
// track if we are inside [], because nested [] are not allowed
bool inside = false;
for (size_t i = 0; i < regex_expr.size(); ++i) {
if (regex_expr[i] == '[' && (i == 0 || regex_expr[i - 1] != '\\')) {
regex_expr_collapsed += '[';
inside = true;
continue;
}
if (inside && regex_expr[i] == ']' && regex_expr[i - 1] != '\\') {
regex_expr_collapsed += ']';
inside = false;
continue;
}
if (regex_expr[i + 0] == '\\' && i + 4 < regex_expr.size() &&
regex_expr[i + 1] == 'p' &&
regex_expr[i + 2] == '{' &&
regex_expr[i + 4] == '}') {
const std::string pat = regex_expr.substr(i, 5);
if (k_ucat_enum.find(pat) != k_ucat_enum.end()) {
if (!inside) {
regex_expr_collapsed += '[';
}
regex_expr_collapsed += k_ucat_cpt.at(k_ucat_enum.at(pat));
regex_expr_collapsed += k_ucat_map.at(k_ucat_enum.at(pat));
if (!inside) {
regex_expr_collapsed += ']';
}
i += 4;
continue;
}
}
regex_expr_collapsed += regex_expr[i];
}
//printf("text_collapsed: %s\n", text_collapsed.c_str());
//printf("regex_expr_collapsed: %s\n", regex_expr_collapsed.c_str());
bpe_offsets = unicode_regex_split_stl(text_collapsed, regex_expr_collapsed, bpe_offsets);
} else {
// no unicode category used, we can use std::wregex directly
const std::wstring wtext = unicode_wstring_from_utf8(text);
const std::wstring wregex_expr = unicode_wstring_from_utf8(regex_expr);
//printf("text: %s\n", text.c_str());
//printf("regex_expr: %s\n", regex_expr.c_str());
bpe_offsets = unicode_regex_split_stl(wtext, wregex_expr, bpe_offsets);
}
} catch (std::regex_error & e) {
fprintf(stderr, "Failed to process regex: '%s'\n", regex_expr.c_str());
fprintf(stderr, "Regex error: %s\n", e.what());
throw std::runtime_error("Failed to process regex");
}
}
std::vector<std::string> bpe_words;
bpe_words.reserve(bpe_offsets.size()); // reserve memory for the approximate size
size_t start = 0;
for (size_t & offset : bpe_offsets) {
bpe_words.emplace_back();
for (size_t i = start; i < start + offset; ++i) {
bpe_words.back() += unicode_cpt_to_utf8(cpts[i]);
}
start += offset;
}
return unicode_byte_encoding_process(bpe_words);
}

9
examples/talk-llama/unicode.h
View File

@ -5,9 +5,9 @@
#include <vector>
#define CODEPOINT_TYPE_UNIDENTIFIED 0
#define CODEPOINT_TYPE_DIGIT 1
#define CODEPOINT_TYPE_NUMBER 1
#define CODEPOINT_TYPE_LETTER 2
#define CODEPOINT_TYPE_WHITESPACE 3
#define CODEPOINT_TYPE_SEPARATOR 3
#define CODEPOINT_TYPE_ACCENT_MARK 4
#define CODEPOINT_TYPE_PUNCTUATION 5
#define CODEPOINT_TYPE_SYMBOL 6
@ -21,8 +21,11 @@ std::vector<uint32_t> unicode_cpts_normalize_nfd(const std::vector<uint32_t> & c
int unicode_cpt_type(uint32_t cp);
int unicode_cpt_type(const std::string & utf8);
bool unicode_cpt_is_whitespace(uint32_t cp);
std::string unicode_byte_to_utf8(uint8_t byte);
uint8_t unicode_utf8_to_byte(const std::string & utf8);
// simple tolower that only implements one-to-one mapping, not one-to-many
char32_t unicode_tolower(char32_t cp);
std::vector<std::string> unicode_regex_split(const std::string & text, const std::vector<std::string> & regex_exprs);