diff options
| author | Georgi Gerganov <ggerganov@gmail.com> | 2023-08-21 23:07:43 +0300 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2023-08-21 23:07:43 +0300 |
| commit | 6381d4e110bd0ec02843a60bbeb8b6fc37a9ace9 (patch) | |
| tree | 15f5b726f864ad0913bc8dcf6ea08b90ecc7ada9 /examples | |
| parent | dadbed99e65252d79f81101a392d0d6497b86caa (diff) | |
gguf : new file format with flexible meta data (beta) (#2398)
* gguf : first API pass
* gguf : read header + meta data
* gguf : read tensor info
* gguf : initial model loading - not tested
* gguf : add gguf_get_tensor_name()
* gguf : do not support passing existing ggml_context to gguf_init
* gguf : simplify gguf_get_val
* gguf : gguf.c is now part of ggml.c
* gguf : read / write sample models
* gguf : add comments
* refactor : reduce code duplication and better API (#2415)
* gguf : expose the gguf_type enum through the API for now
* gguf : add array support
* gguf.py : some code style changes
* convert.py : start a new simplified implementation by removing old stuff
* convert.py : remove GGML vocab + other obsolete stuff
* GGUF : write tensor (#2426)
* WIP: Write tensor
* GGUF : Support writing tensors in Python
* refactor : rm unused import and upd todos
* fix : fix errors upd writing example
* rm example.gguf
* gitignore *.gguf
* undo formatting
* gguf : add gguf_find_key (#2438)
* gguf.cpp : find key example
* ggml.h : add gguf_find_key
* ggml.c : add gguf_find_key
* gguf : fix writing tensors
* gguf : do not hardcode tensor names to read
* gguf : write sample tensors to read
* gguf : add tokenization constants
* quick and dirty conversion example
* gguf : fix writing gguf arrays
* gguf : write tensors one by one and code reuse
* gguf : fix writing gguf arrays
* gguf : write tensors one by one
* gguf : write tensors one by one
* gguf : write tokenizer data
* gguf : upd gguf conversion script
* Update convert-llama-h5-to-gguf.py
* gguf : handle already encoded string
* ggml.h : get array str and f32
* ggml.c : get arr str and f32
* gguf.py : support any type
* Update convert-llama-h5-to-gguf.py
* gguf : fix set is not subscriptable
* gguf : update convert-llama-h5-to-gguf.py
* constants.py : add layer norm eps
* gguf.py : add layer norm eps and merges
* ggml.h : increase GGML_MAX_NAME to 64
* ggml.c : add gguf_get_arr_n
* Update convert-llama-h5-to-gguf.py
* add gptneox gguf example
* Makefile : add gptneox gguf example
* Update convert-llama-h5-to-gguf.py
* add gptneox gguf example
* Update convert-llama-h5-to-gguf.py
* Update convert-gptneox-h5-to-gguf.py
* Update convert-gptneox-h5-to-gguf.py
* Update convert-llama-h5-to-gguf.py
* gguf : support custom alignment value
* gguf : fix typo in function call
* gguf : mmap tensor data example
* fix : update convert-llama-h5-to-gguf.py
* Update convert-llama-h5-to-gguf.py
* convert-gptneox-h5-to-gguf.py : Special tokens
* gptneox-main.cpp : special tokens
* Update gptneox-main.cpp
* constants.py : special tokens
* gguf.py : accumulate kv and tensor info data + special tokens
* convert-gptneox-h5-to-gguf.py : accumulate kv and ti + special tokens
* gguf : gguf counterpart of llama-util.h
* gguf-util.h : update note
* convert-llama-h5-to-gguf.py : accumulate kv / ti + special tokens
* convert-llama-h5-to-gguf.py : special tokens
* Delete gptneox-common.cpp
* Delete gptneox-common.h
* convert-gptneox-h5-to-gguf.py : gpt2bpe tokenizer
* gptneox-main.cpp : gpt2 bpe tokenizer
* gpt2 bpe tokenizer (handles merges and unicode)
* Makefile : remove gptneox-common
* gguf.py : bytesarray for gpt2bpe tokenizer
* cmpnct_gpt2bpe.hpp : comments
* gguf.py : use custom alignment if present
* gguf : minor stuff
* Update gptneox-main.cpp
* map tensor names
* convert-gptneox-h5-to-gguf.py : map tensor names
* convert-llama-h5-to-gguf.py : map tensor names
* gptneox-main.cpp : map tensor names
* gguf : start implementing libllama in GGUF (WIP)
* gguf : start implementing libllama in GGUF (WIP)
* rm binary commited by mistake
* upd .gitignore
* gguf : calculate n_mult
* gguf : inference with 7B model working (WIP)
* gguf : rm deprecated function
* gguf : start implementing gguf_file_saver (WIP)
* gguf : start implementing gguf_file_saver (WIP)
* gguf : start implementing gguf_file_saver (WIP)
* gguf : add gguf_get_kv_type
* gguf : add gguf_get_kv_type
* gguf : write metadata in gguf_file_saver (WIP)
* gguf : write metadata in gguf_file_saver (WIP)
* gguf : write metadata in gguf_file_saver
* gguf : rm references to old file formats
* gguf : shorter name for member variable
* gguf : rm redundant method
* gguf : get rid of n_mult, read n_ff from file
* Update gguf_tensor_map.py
* Update gptneox-main.cpp
* gguf : rm references to old file magics
* gguf : start implementing quantization (WIP)
* gguf : start implementing quantization (WIP)
* gguf : start implementing quantization (WIP)
* gguf : start implementing quantization (WIP)
* gguf : start implementing quantization (WIP)
* gguf : start implementing quantization (WIP)
* gguf : quantization is working
* gguf : roper closing of file
* gguf.py : no need to convert tensors twice
* convert-gptneox-h5-to-gguf.py : no need to convert tensors twice
* convert-llama-h5-to-gguf.py : no need to convert tensors twice
* convert-gptneox-h5-to-gguf.py : simplify nbytes
* convert-llama-h5-to-gguf.py : simplify nbytes
* gptneox-main.cpp : n_layer --> n_block
* constants.py : n_layer --> n_block
* gguf.py : n_layer --> n_block
* convert-gptneox-h5-to-gguf.py : n_layer --> n_block
* convert-llama-h5-to-gguf.py : n_layer --> n_block
* gptneox-main.cpp : n_layer --> n_block
* Update gguf_tensor_map.py
* convert-gptneox-h5-to-gguf.py : load model in parts to save memory
* convert-llama-h5-to-gguf.py : load model in parts to save memory
* convert : write more metadata for LLaMA
* convert : rm quantization version
* convert-gptneox-h5-to-gguf.py : add file_type key
* gptneox-main.cpp : add file_type key
* fix conflicts
* gguf : add todos and comments
* convert-gptneox-h5-to-gguf.py : tensor name map changes
* Create gguf_namemap.py : tensor name map changes
* Delete gguf_tensor_map.py
* gptneox-main.cpp : tensor name map changes
* convert-llama-h5-to-gguf.py : fixes
* gguf.py : dont add empty strings
* simple : minor style changes
* gguf : use UNIX line ending
* Create convert-llama-7b-pth-to-gguf.py
* llama : sync gguf-llama.cpp with latest llama.cpp (#2608)
* llama : sync gguf-llama.cpp with latest llama.cpp
* minor : indentation + assert
* llama : refactor gguf_buffer and gguf_ctx_buffer
* llama : minor
* gitignore : add gptneox-main
* llama : tokenizer fixes (#2549)
* Merge tokenizer fixes into the gguf branch.
* Add test vocabularies
* convert : update convert-new.py with tokenizer fixes (#2614)
* Merge tokenizer fixes into the gguf branch.
* Add test vocabularies
* Adapt convert-new.py (and fix a clang-cl compiler error on windows)
* llama : sync gguf-llama with llama (#2613)
* llama : sync gguf-llama with llama
* tests : fix build + warnings (test-tokenizer-1 still fails)
* tests : fix wstring_convert
* convert : fix layer names
* llama : sync gguf-llama.cpp
* convert : update HF converter to new tokenizer voodoo magics
* llama : update tokenizer style
* convert-llama-h5-to-gguf.py : add token types
* constants.py : add token types
* gguf.py : add token types
* convert-llama-7b-pth-to-gguf.py : add token types
* gguf-llama.cpp : fix n_head_kv
* convert-llama-h5-to-gguf.py : add 70b gqa support
* gguf.py : add tensor data layout
* convert-llama-h5-to-gguf.py : add tensor data layout
* convert-llama-7b-pth-to-gguf.py : add tensor data layout
* gptneox-main.cpp : add tensor data layout
* convert-llama-h5-to-gguf.py : clarify the reverse permute
* llama : refactor model loading code (#2620)
* llama : style formatting + remove helper methods
* llama : fix quantization using gguf tool
* llama : simplify gguf_file_saver
* llama : fix method names
* llama : simplify write_header()
* llama : no need to pass full file loader to the file saver
just gguf_ctx
* llama : gguf_file_saver write I32
* llama : refactor tensor names (#2622)
* gguf: update tensor names searched in quantization
* gguf : define tensor names as constants
* gguf : initial write API (not tested yet)
* gguf : write to file API (not tested)
* gguf : initial write API ready + example
* gguf : fix header write
* gguf : fixes + simplify example + add ggml_nbytes_pad()
* gguf : minor
* llama : replace gguf_file_saver with new gguf write API
* gguf : streaming support when writing files
* gguf : remove oboslete write methods
* gguf : remove obosolete gguf_get_arr_xxx API
* llama : simplify gguf_file_loader
* llama : move hparams and vocab from gguf_file_loader to llama_model_loader
* llama : merge gguf-util.h in llama.cpp
* llama : reorder definitions in .cpp to match .h
* llama : minor simplifications
* llama : refactor llama_model_loader (WIP)
wip : remove ggml_ctx from llama_model_loader
wip : merge gguf_file_loader in llama_model_loader
* llama : fix shape prints
* llama : fix Windows build + fix norm_rms_eps key
* llama : throw error on missing KV paris in model meta data
* llama : improve printing + log meta data
* llama : switch print order of meta data
---------
Co-authored-by: M. Yusuf Sarıgöz <yusufsarigoz@gmail.com>
* gguf : deduplicate (#2629)
* gguf : better type names
* dedup : CPU + Metal is working
* ggml : fix warnings about unused results
* llama.cpp : fix line feed and compiler warning
* llama : fix strncpy warning + note token_to_str does not write null
* llama : restore the original load/save session implementation
Will migrate this to GGUF in the future
* convert-llama-h5-to-gguf.py : support alt ctx param name
* ggml : assert when using ggml_mul with non-F32 src1
* examples : dedup simple
---------
Co-authored-by: klosax <131523366+klosax@users.noreply.github.com>
* gguf.py : merge all files in gguf.py
* convert-new.py : pick #2427 for HF 70B support
* examples/gguf : no need to keep q option for quantization any more
* llama.cpp : print actual model size
* llama.cpp : use ggml_elements()
* convert-new.py : output gguf (#2635)
* convert-new.py : output gguf (WIP)
* convert-new.py : add gguf key-value pairs
* llama : add hparams.ctx_train + no longer print ftype
* convert-new.py : minor fixes
* convert-new.py : vocab-only option should work now
* llama : fix tokenizer to use llama_char_to_byte
* tests : add new ggml-vocab-llama.gguf
* convert-new.py : tensor name mapping
* convert-new.py : add map for skipping tensor serialization
* convert-new.py : convert script now works
* gguf.py : pick some of the refactoring from #2644
* convert-new.py : minor fixes
* convert.py : update to support GGUF output
* Revert "ci : disable CI temporary to not waste energy"
This reverts commit 7e82d25f40386540c2c15226300ad998ecd871ea.
* convert.py : n_head_kv optional and .gguf file extension
* convert.py : better always have n_head_kv and default it to n_head
* llama : sync with recent PRs on master
* editorconfig : ignore models folder
ggml-ci
* ci : update ".bin" to ".gguf" extension
ggml-ci
* llama : fix llama_model_loader memory leak
* gptneox : move as a WIP example
* llama : fix lambda capture
ggml-ci
* ggml : fix bug in gguf_set_kv
ggml-ci
* common.h : .bin --> .gguf
* quantize-stats.cpp : .bin --> .gguf
* convert.py : fix HF tensor permuting / unpacking
ggml-ci
* llama.cpp : typo
* llama : throw error if gguf fails to init from file
ggml-ci
* llama : fix tensor name grepping during quantization
ggml-ci
* gguf.py : write tensors in a single pass (#2644)
* gguf : single pass for writing tensors + refactoring writer
* gguf : single pass for writing tensors + refactoring writer
* gguf : single pass for writing tensors + refactoring writer
* gguf : style fixes in simple conversion script
* gguf : refactor gptneox conversion script
* gguf : rename h5 to hf (for HuggingFace)
* gguf : refactor pth to gguf conversion script
* gguf : rm file_type key and method
* gguf.py : fix vertical alignment
* gguf.py : indentation
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* convert-gptneox-hf-to-gguf.py : fixes
* gguf.py : gptneox mapping
* convert-llama-hf-to-gguf.py : fixes
* convert-llama-7b-pth-to-gguf.py : fixes
* ggml.h : reverse GGUF_MAGIC
* gguf.py : reverse GGUF_MAGIC
* test-tokenizer-0.cpp : fix warning
* llama.cpp : print kv general.name
* llama.cpp : get special token kv and linefeed token id
* llama : print number of tensors per type + print arch + style
* tests : update vocab file with new magic
* editorconfig : fix whitespaces
* llama : re-order functions
* llama : remove C++ API + reorganize common source in /common dir
* llama : minor API updates
* llama : avoid hardcoded special tokens
* llama : fix MPI build
ggml-ci
* llama : introduce enum llama_vocab_type + remove hardcoded string constants
* convert-falcon-hf-to-gguf.py : falcon HF --> gguf conversion, not tested
* falcon-main.cpp : falcon inference example
* convert-falcon-hf-to-gguf.py : remove extra kv
* convert-gptneox-hf-to-gguf.py : remove extra kv
* convert-llama-7b-pth-to-gguf.py : remove extra kv
* convert-llama-hf-to-gguf.py : remove extra kv
* gguf.py : fix for falcon 40b
* falcon-main.cpp : fix for falcon 40b
* convert-falcon-hf-to-gguf.py : update ref
* convert-falcon-hf-to-gguf.py : add tensor data layout
* cmpnct_gpt2bpe.hpp : fixes
* falcon-main.cpp : fixes
* gptneox-main.cpp : fixes
* cmpnct_gpt2bpe.hpp : remove non-general stuff
* Update examples/server/README.md
Co-authored-by: slaren <slarengh@gmail.com>
* cmpnct_gpt2bpe.hpp : cleanup
* convert-llama-hf-to-gguf.py : special tokens
* convert-llama-7b-pth-to-gguf.py : special tokens
* convert-permute-debug.py : permute debug print
* convert-permute-debug-master.py : permute debug for master
* convert-permute-debug.py : change permute type of attn_q
* convert.py : 70b model working (change attn_q permute)
* Delete convert-permute-debug-master.py
* Delete convert-permute-debug.py
* convert-llama-hf-to-gguf.py : fix attn_q permute
* gguf.py : fix rope scale kv
* convert-llama-hf-to-gguf.py : rope scale and added tokens
* convert-llama-7b-pth-to-gguf.py : rope scale and added tokens
* llama.cpp : use rope scale kv
* convert-llama-7b-pth-to-gguf.py : rope scale fix
* convert-llama-hf-to-gguf.py : rope scale fix
* py : fix whitespace
* gguf : add Python script to convert GGMLv3 LLaMA models to GGUF (#2682)
* First pass at converting GGMLv3 LLaMA models to GGUF
* Cleanups, better output during conversion
* Fix vocab space conversion logic
* More vocab conversion fixes
* Add description to converted GGUF files
* Improve help text, expand warning
* Allow specifying name and description for output GGUF
* Allow overriding vocab and hyperparams from original model metadata
* Use correct params override var name
* Fix wrong type size for Q8_K
Better handling of original style metadata
* Set default value for gguf add_tensor raw_shape KW arg
* llama : improve token type support (#2668)
* Merge tokenizer fixes into the gguf branch.
* Add test vocabularies
* Adapt convert-new.py (and fix a clang-cl compiler error on windows)
* Improved tokenizer test
But does it work on MacOS?
* Improve token type support
- Added @klosax code to convert.py
- Improved token type support in vocabulary
* Exclude platform dependent tests
* More sentencepiece compatibility by eliminating magic numbers
* Restored accidentally removed comment
* llama : add API for token type
ggml-ci
* tests : use new tokenizer type API (#2692)
* Merge tokenizer fixes into the gguf branch.
* Add test vocabularies
* Adapt convert-new.py (and fix a clang-cl compiler error on windows)
* Improved tokenizer test
But does it work on MacOS?
* Improve token type support
- Added @klosax code to convert.py
- Improved token type support in vocabulary
* Exclude platform dependent tests
* More sentencepiece compatibility by eliminating magic numbers
* Restored accidentally removed comment
* Improve commentary
* Use token type API in test-tokenizer-1.cpp
* py : cosmetics
* readme : add notice about new file format
ggml-ci
---------
Co-authored-by: M. Yusuf Sarıgöz <yusufsarigoz@gmail.com>
Co-authored-by: klosax <131523366+klosax@users.noreply.github.com>
Co-authored-by: goerch <jhr.walter@t-online.de>
Co-authored-by: slaren <slarengh@gmail.com>
Co-authored-by: Kerfuffle <44031344+KerfuffleV2@users.noreply.github.com>
Diffstat (limited to 'examples')
25 files changed, 3825 insertions, 2157 deletions
diff --git a/examples/CMakeLists.txt b/examples/CMakeLists.txt index d5365281..d2176c91 100644 --- a/examples/CMakeLists.txt +++ b/examples/CMakeLists.txt @@ -6,27 +6,6 @@ find_package(Threads REQUIRED) # ... -# common - -set(TARGET common) - -add_library(${TARGET} OBJECT - common.h - common.cpp - console.h - console.cpp - grammar-parser.h - grammar-parser.cpp - ) - -if (BUILD_SHARED_LIBS) - set_target_properties(${TARGET} PROPERTIES POSITION_INDEPENDENT_CODE ON) -endif() - -target_include_directories(${TARGET} PUBLIC .) -target_compile_features(${TARGET} PUBLIC cxx_std_11) -target_link_libraries(${TARGET} PRIVATE llama) - # examples include_directories(${CMAKE_CURRENT_SOURCE_DIR}) diff --git a/examples/common.cpp b/examples/common.cpp deleted file mode 100644 index bd39d922..00000000 --- a/examples/common.cpp +++ /dev/null @@ -1,718 +0,0 @@ -#include "common.h" - -#include <cassert> -#include <iostream> -#include <cstring> -#include <fstream> -#include <string> -#include <iterator> -#include <algorithm> -#include <sstream> -#include <unordered_set> -#include <regex> - -#if defined(__APPLE__) && defined(__MACH__) -#include <sys/types.h> -#include <sys/sysctl.h> -#endif - -#if defined(_WIN32) -#define WIN32_LEAN_AND_MEAN -#define NOMINMAX -#include <windows.h> -#include <fcntl.h> -#include <io.h> -#else -#include <sys/ioctl.h> -#include <unistd.h> -#endif - -#if defined(_MSC_VER) -#pragma warning(disable: 4244 4267) // possible loss of data -#endif - -int32_t get_num_physical_cores() { -#ifdef __linux__ - // enumerate the set of thread siblings, num entries is num cores - std::unordered_set<std::string> siblings; - for (uint32_t cpu=0; cpu < UINT32_MAX; ++cpu) { - std::ifstream thread_siblings("/sys/devices/system/cpu" - + std::to_string(cpu) + "/topology/thread_siblings"); - if (!thread_siblings.is_open()) { - break; // no more cpus - } - std::string line; - if (std::getline(thread_siblings, line)) { - siblings.insert(line); - } - } - if (siblings.size() > 0) { - return static_cast<int32_t>(siblings.size()); - } -#elif defined(__APPLE__) && defined(__MACH__) - int32_t num_physical_cores; - size_t len = sizeof(num_physical_cores); - int result = sysctlbyname("hw.perflevel0.physicalcpu", &num_physical_cores, &len, NULL, 0); - if (result == 0) { - return num_physical_cores; - } - result = sysctlbyname("hw.physicalcpu", &num_physical_cores, &len, NULL, 0); - if (result == 0) { - return num_physical_cores; - } -#elif defined(_WIN32) - //TODO: Implement -#endif - unsigned int n_threads = std::thread::hardware_concurrency(); - return n_threads > 0 ? (n_threads <= 4 ? n_threads : n_threads / 2) : 4; -} - -void process_escapes(std::string& input) { - std::size_t input_len = input.length(); - std::size_t output_idx = 0; - - for (std::size_t input_idx = 0; input_idx < input_len; ++input_idx) { - if (input[input_idx] == '\\' && input_idx + 1 < input_len) { - switch (input[++input_idx]) { - case 'n': input[output_idx++] = '\n'; break; - case 'r': input[output_idx++] = '\r'; break; - case 't': input[output_idx++] = '\t'; break; - case '\'': input[output_idx++] = '\''; break; - case '\"': input[output_idx++] = '\"'; break; - case '\\': input[output_idx++] = '\\'; break; - default: input[output_idx++] = '\\'; - input[output_idx++] = input[input_idx]; break; - } - } else { - input[output_idx++] = input[input_idx]; - } - } - - input.resize(output_idx); -} - -bool gpt_params_parse(int argc, char ** argv, gpt_params & params) { - bool invalid_param = false; - bool escape_prompt = false; - std::string arg; - gpt_params default_params; - const std::string arg_prefix = "--"; - - for (int i = 1; i < argc; i++) { - arg = argv[i]; - if (arg.compare(0, arg_prefix.size(), arg_prefix) == 0) { - std::replace(arg.begin(), arg.end(), '_', '-'); - } - - if (arg == "-s" || arg == "--seed") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.seed = std::stoul(argv[i]); - } else if (arg == "-t" || arg == "--threads") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.n_threads = std::stoi(argv[i]); - if (params.n_threads <= 0) { - params.n_threads = std::thread::hardware_concurrency(); - } - } else if (arg == "-p" || arg == "--prompt") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.prompt = argv[i]; - } else if (arg == "-e") { - escape_prompt = true; - } else if (arg == "--prompt-cache") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.path_prompt_cache = argv[i]; - } else if (arg == "--prompt-cache-all") { - params.prompt_cache_all = true; - } else if (arg == "--prompt-cache-ro") { - params.prompt_cache_ro = true; - } else if (arg == "-f" || arg == "--file") { - if (++i >= argc) { - invalid_param = true; - break; - } - std::ifstream file(argv[i]); - if (!file) { - fprintf(stderr, "error: failed to open file '%s'\n", argv[i]); - invalid_param = true; - break; - } - std::copy(std::istreambuf_iterator<char>(file), std::istreambuf_iterator<char>(), back_inserter(params.prompt)); - if (params.prompt.back() == '\n') { - params.prompt.pop_back(); - } - } else if (arg == "-n" || arg == "--n-predict") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.n_predict = std::stoi(argv[i]); - } else if (arg == "--top-k") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.top_k = std::stoi(argv[i]); - } else if (arg == "-c" || arg == "--ctx-size") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.n_ctx = std::stoi(argv[i]); - } else if (arg == "-gqa" || arg == "--gqa") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.n_gqa = std::stoi(argv[i]); - } else if (arg == "-eps" || arg == "--rms-norm-eps") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.rms_norm_eps = std::stof(argv[i]); - } else if (arg == "--rope-freq-base") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.rope_freq_base = std::stof(argv[i]); - } else if (arg == "--rope-freq-scale") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.rope_freq_scale = std::stof(argv[i]); - } else if (arg == "--rope-scale") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.rope_freq_scale = 1.0f/std::stof(argv[i]); - } else if (arg == "--memory-f32") { - params.memory_f16 = false; - } else if (arg == "--top-p") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.top_p = std::stof(argv[i]); - } else if (arg == "--temp") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.temp = std::stof(argv[i]); - } else if (arg == "--tfs") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.tfs_z = std::stof(argv[i]); - } else if (arg == "--typical") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.typical_p = std::stof(argv[i]); - } else if (arg == "--repeat-last-n") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.repeat_last_n = std::stoi(argv[i]); - } else if (arg == "--repeat-penalty") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.repeat_penalty = std::stof(argv[i]); - } else if (arg == "--frequency-penalty") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.frequency_penalty = std::stof(argv[i]); - } else if (arg == "--presence-penalty") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.presence_penalty = std::stof(argv[i]); - } else if (arg == "--mirostat") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.mirostat = std::stoi(argv[i]); - } else if (arg == "--mirostat-lr") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.mirostat_eta = std::stof(argv[i]); - } else if (arg == "--mirostat-ent") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.mirostat_tau = std::stof(argv[i]); - } else if (arg == "--cfg-negative-prompt") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.cfg_negative_prompt = argv[i]; - } else if (arg == "--cfg-negative-prompt-file") { - if (++i >= argc) { - invalid_param = true; - break; - } - std::ifstream file(argv[i]); - if (!file) { - fprintf(stderr, "error: failed to open file '%s'\n", argv[i]); - invalid_param = true; - break; - } - std::copy(std::istreambuf_iterator<char>(file), std::istreambuf_iterator<char>(), back_inserter(params.cfg_negative_prompt)); - if (params.cfg_negative_prompt.back() == '\n') { - params.cfg_negative_prompt.pop_back(); - } - } else if (arg == "--cfg-scale") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.cfg_scale = std::stof(argv[i]); - } else if (arg == "-b" || arg == "--batch-size") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.n_batch = std::stoi(argv[i]); - params.n_batch = std::min(512, params.n_batch); - } else if (arg == "--keep") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.n_keep = std::stoi(argv[i]); - } else if (arg == "--chunks") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.n_chunks = std::stoi(argv[i]); - } else if (arg == "-m" || arg == "--model") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.model = argv[i]; - } else if (arg == "-a" || arg == "--alias") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.model_alias = argv[i]; - } else if (arg == "--lora") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.lora_adapter = argv[i]; - params.use_mmap = false; - } else if (arg == "--lora-base") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.lora_base = argv[i]; - } else if (arg == "-i" || arg == "--interactive") { - params.interactive = true; - } else if (arg == "--embedding") { - params.embedding = true; - } else if (arg == "--interactive-first") { - params.interactive_first = true; - } else if (arg == "-ins" || arg == "--instruct") { - params.instruct = true; - } else if (arg == "--multiline-input") { - params.multiline_input = true; - } else if (arg == "--simple-io") { - params.simple_io = true; - } else if (arg == "--color") { - params.use_color = true; - } else if (arg == "--mlock") { - params.use_mlock = true; - } else if (arg == "--gpu-layers" || arg == "-ngl" || arg == "--n-gpu-layers") { - if (++i >= argc) { - invalid_param = true; - break; - } -#ifdef LLAMA_SUPPORTS_GPU_OFFLOAD - params.n_gpu_layers = std::stoi(argv[i]); -#else - fprintf(stderr, "warning: not compiled with GPU offload support, --n-gpu-layers option will be ignored\n"); - fprintf(stderr, "warning: see main README.md for information on enabling GPU BLAS support\n"); -#endif - } else if (arg == "--main-gpu" || arg == "-mg") { - if (++i >= argc) { - invalid_param = true; - break; - } -#ifdef GGML_USE_CUBLAS - params.main_gpu = std::stoi(argv[i]); -#else - fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS. It is not possible to set a main GPU.\n"); -#endif - } else if (arg == "--tensor-split" || arg == "-ts") { - if (++i >= argc) { - invalid_param = true; - break; - } -#ifdef GGML_USE_CUBLAS - std::string arg_next = argv[i]; - - // split string by , and / - const std::regex regex{R"([,/]+)"}; - std::sregex_token_iterator it{arg_next.begin(), arg_next.end(), regex, -1}; - std::vector<std::string> split_arg{it, {}}; - GGML_ASSERT(split_arg.size() <= LLAMA_MAX_DEVICES); - - for (size_t i = 0; i < LLAMA_MAX_DEVICES; ++i) { - if (i < split_arg.size()) { - params.tensor_split[i] = std::stof(split_arg[i]); - } else { - params.tensor_split[i] = 0.0f; - } - } -#else - fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS. It is not possible to set a tensor split.\n"); -#endif // GGML_USE_CUBLAS - } else if (arg == "--mul-mat-q" || arg == "-mmq") { -#ifdef GGML_USE_CUBLAS - params.mul_mat_q = true; -#else - fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS. It is not possible to use mul_mat_q kernels.\n"); -#endif // GGML_USE_CUBLAS - } else if (arg == "--low-vram" || arg == "-lv") { -#ifdef GGML_USE_CUBLAS - params.low_vram = true; -#else - fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS. It is not possible to set lower vram usage.\n"); -#endif // GGML_USE_CUBLAS - } else if (arg == "--no-mmap") { - params.use_mmap = false; - } else if (arg == "--mtest") { - params.mem_test = true; - } else if (arg == "--numa") { - params.numa = true; - } else if (arg == "--export") { - params.export_cgraph = true; - } else if (arg == "--verbose-prompt") { - params.verbose_prompt = true; - } else if (arg == "-r" || arg == "--reverse-prompt") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.antiprompt.push_back(argv[i]); - } else if (arg == "--perplexity") { - params.perplexity = true; - } else if (arg == "--hellaswag") { - params.hellaswag = true; - } else if (arg == "--hellaswag-tasks") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.hellaswag_tasks = std::stoi(argv[i]); - } else if (arg == "--ignore-eos") { - params.logit_bias[llama_token_eos()] = -INFINITY; - } else if (arg == "--no-penalize-nl") { - params.penalize_nl = false; - } else if (arg == "-l" || arg == "--logit-bias") { - if (++i >= argc) { - invalid_param = true; - break; - } - std::stringstream ss(argv[i]); - llama_token key; - char sign; - std::string value_str; - try { - if (ss >> key && ss >> sign && std::getline(ss, value_str) && (sign == '+' || sign == '-')) { - params.logit_bias[key] = std::stof(value_str) * ((sign == '-') ? -1.0f : 1.0f); - } else { - throw std::exception(); - } - } catch (const std::exception&) { - invalid_param = true; - break; - } - } else if (arg == "-h" || arg == "--help") { - gpt_print_usage(argc, argv, default_params); - exit(0); - } else if (arg == "--random-prompt") { - params.random_prompt = true; - } else if (arg == "--in-prefix-bos") { - params.input_prefix_bos = true; - } else if (arg == "--in-prefix") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.input_prefix = argv[i]; - } else if (arg == "--in-suffix") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.input_suffix = argv[i]; - } else if (arg == "--grammar") { - if (++i >= argc) { - invalid_param = true; - break; - } - params.grammar = argv[i]; - } else if (arg == "--grammar-file") { - if (++i >= argc) { - invalid_param = true; - break; - } - std::ifstream file(argv[i]); - if (!file) { - fprintf(stderr, "error: failed to open file '%s'\n", argv[i]); - invalid_param = true; - break; - } - std::copy( - std::istreambuf_iterator<char>(file), - std::istreambuf_iterator<char>(), - std::back_inserter(params.grammar) - ); - } else { - fprintf(stderr, "error: unknown argument: %s\n", arg.c_str()); - gpt_print_usage(argc, argv, default_params); - exit(1); - } - } - if (invalid_param) { - fprintf(stderr, "error: invalid parameter for argument: %s\n", arg.c_str()); - gpt_print_usage(argc, argv, default_params); - exit(1); - } - if (params.prompt_cache_all && - (params.interactive || params.interactive_first || - params.instruct)) { - fprintf(stderr, "error: --prompt-cache-all not supported in interactive mode yet\n"); - gpt_print_usage(argc, argv, default_params); - exit(1); - } - - if (escape_prompt) { - process_escapes(params.prompt); - process_escapes(params.input_prefix); - process_escapes(params.input_suffix); - } - - return true; -} - -void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) { - fprintf(stdout, "usage: %s [options]\n", argv[0]); - fprintf(stdout, "\n"); - fprintf(stdout, "options:\n"); - fprintf(stdout, " -h, --help show this help message and exit\n"); - fprintf(stdout, " -i, --interactive run in interactive mode\n"); - fprintf(stdout, " --interactive-first run in interactive mode and wait for input right away\n"); - fprintf(stdout, " -ins, --instruct run in instruction mode (use with Alpaca models)\n"); - fprintf(stdout, " --multiline-input allows you to write or paste multiple lines without ending each in '\\'\n"); - fprintf(stdout, " -r PROMPT, --reverse-prompt PROMPT\n"); - fprintf(stdout, " halt generation at PROMPT, return control in interactive mode\n"); - fprintf(stdout, " (can be specified more than once for multiple prompts).\n"); - fprintf(stdout, " --color colorise output to distinguish prompt and user input from generations\n"); - fprintf(stdout, " -s SEED, --seed SEED RNG seed (default: -1, use random seed for < 0)\n"); - fprintf(stdout, " -t N, --threads N number of threads to use during computation (default: %d)\n", params.n_threads); - fprintf(stdout, " -p PROMPT, --prompt PROMPT\n"); - fprintf(stdout, " prompt to start generation with (default: empty)\n"); - fprintf(stdout, " -e process prompt escapes sequences (\\n, \\r, \\t, \\', \\\", \\\\)\n"); - fprintf(stdout, " --prompt-cache FNAME file to cache prompt state for faster startup (default: none)\n"); - fprintf(stdout, " --prompt-cache-all if specified, saves user input and generations to cache as well.\n"); - fprintf(stdout, " not supported with --interactive or other interactive options\n"); - fprintf(stdout, " --prompt-cache-ro if specified, uses the prompt cache but does not update it.\n"); - fprintf(stdout, " --random-prompt start with a randomized prompt.\n"); - fprintf(stdout, " --in-prefix-bos prefix BOS to user inputs, preceding the `--in-prefix` string\n"); - fprintf(stdout, " --in-prefix STRING string to prefix user inputs with (default: empty)\n"); - fprintf(stdout, " --in-suffix STRING string to suffix after user inputs with (default: empty)\n"); - fprintf(stdout, " -f FNAME, --file FNAME\n"); - fprintf(stdout, " prompt file to start generation.\n"); - fprintf(stdout, " -n N, --n-predict N number of tokens to predict (default: %d, -1 = infinity, -2 = until context filled)\n", params.n_predict); - fprintf(stdout, " -c N, --ctx-size N size of the prompt context (default: %d)\n", params.n_ctx); - fprintf(stdout, " -b N, --batch-size N batch size for prompt processing (default: %d)\n", params.n_batch); - fprintf(stdout, " -gqa N, --gqa N grouped-query attention factor (TEMP!!! use 8 for LLaMAv2 70B) (default: %d)\n", params.n_gqa); - fprintf(stdout, " -eps N, --rms-norm-eps N rms norm eps (TEMP!!! use 1e-5 for LLaMAv2) (default: %.1e)\n", params.rms_norm_eps); - fprintf(stdout, " --top-k N top-k sampling (default: %d, 0 = disabled)\n", params.top_k); - fprintf(stdout, " --top-p N top-p sampling (default: %.1f, 1.0 = disabled)\n", (double)params.top_p); - fprintf(stdout, " --tfs N tail free sampling, parameter z (default: %.1f, 1.0 = disabled)\n", (double)params.tfs_z); - fprintf(stdout, " --typical N locally typical sampling, parameter p (default: %.1f, 1.0 = disabled)\n", (double)params.typical_p); - fprintf(stdout, " --repeat-last-n N last n tokens to consider for penalize (default: %d, 0 = disabled, -1 = ctx_size)\n", params.repeat_last_n); - fprintf(stdout, " --repeat-penalty N penalize repeat sequence of tokens (default: %.1f, 1.0 = disabled)\n", (double)params.repeat_penalty); - fprintf(stdout, " --presence-penalty N repeat alpha presence penalty (default: %.1f, 0.0 = disabled)\n", (double)params.presence_penalty); - fprintf(stdout, " --frequency-penalty N repeat alpha frequency penalty (default: %.1f, 0.0 = disabled)\n", (double)params.frequency_penalty); - fprintf(stdout, " --mirostat N use Mirostat sampling.\n"); - fprintf(stdout, " Top K, Nucleus, Tail Free and Locally Typical samplers are ignored if used.\n"); - fprintf(stdout, " (default: %d, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)\n", params.mirostat); - fprintf(stdout, " --mirostat-lr N Mirostat learning rate, parameter eta (default: %.1f)\n", (double)params.mirostat_eta); - fprintf(stdout, " --mirostat-ent N Mirostat target entropy, parameter tau (default: %.1f)\n", (double)params.mirostat_tau); - fprintf(stdout, " -l TOKEN_ID(+/-)BIAS, --logit-bias TOKEN_ID(+/-)BIAS\n"); - fprintf(stdout, " modifies the likelihood of token appearing in the completion,\n"); - fprintf(stdout, " i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',\n"); - fprintf(stdout, " or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'\n"); - fprintf(stdout, " --grammar GRAMMAR BNF-like grammar to constrain generations (see samples in grammars/ dir)\n"); - fprintf(stdout, " --grammar-file FNAME file to read grammar from\n"); - fprintf(stdout, " --cfg-negative-prompt PROMPT\n"); - fprintf(stdout, " negative prompt to use for guidance. (default: empty)\n"); - fprintf(stdout, " --cfg-negative-prompt-file FNAME\n"); - fprintf(stdout, " negative prompt file to use for guidance. (default: empty)\n"); - fprintf(stdout, " --cfg-scale N strength of guidance (default: %f, 1.0 = disable)\n", params.cfg_scale); - fprintf(stdout, " --rope-scale N RoPE context linear scaling factor, inverse of --rope-freq-scale (default: %g)\n", 1.0f/params.rope_freq_scale); - fprintf(stdout, " --rope-freq-base N RoPE base frequency, used by NTK-aware scaling (default: %.1f)\n", params.rope_freq_base); - fprintf(stdout, " --rope-freq-scale N RoPE frequency linear scaling factor, inverse of --rope-scale (default: %g)\n", params.rope_freq_scale); - fprintf(stdout, " --ignore-eos ignore end of stream token and continue generating (implies --logit-bias 2-inf)\n"); - fprintf(stdout, " --no-penalize-nl do not penalize newline token\n"); - fprintf(stdout, " --memory-f32 use f32 instead of f16 for memory key+value (default: disabled)\n"); - fprintf(stdout, " not recommended: doubles context memory required and no measurable increase in quality\n"); - fprintf(stdout, " --temp N temperature (default: %.1f)\n", (double)params.temp); - fprintf(stdout, " --perplexity compute perplexity over each ctx window of the prompt\n"); - fprintf(stdout, " --hellaswag compute HellaSwag score over random tasks from datafile supplied with -f\n"); - fprintf(stdout, " --hellaswag-tasks N number of tasks to use when computing the HellaSwag score (default: %zu)\n", params.hellaswag_tasks); - fprintf(stdout, " --keep N number of tokens to keep from the initial prompt (default: %d, -1 = all)\n", params.n_keep); - fprintf(stdout, " --chunks N max number of chunks to process (default: %d, -1 = all)\n", params.n_chunks); - if (llama_mlock_supported()) { - fprintf(stdout, " --mlock force system to keep model in RAM rather than swapping or compressing\n"); - } - if (llama_mmap_supported()) { - fprintf(stdout, " --no-mmap do not memory-map model (slower load but may reduce pageouts if not using mlock)\n"); - } - fprintf(stdout, " --numa attempt optimizations that help on some NUMA systems\n"); - fprintf(stdout, " if run without this previously, it is recommended to drop the system page cache before using this\n"); - fprintf(stdout, " see https://github.com/ggerganov/llama.cpp/issues/1437\n"); -#ifdef LLAMA_SUPPORTS_GPU_OFFLOAD - fprintf(stdout, " -ngl N, --n-gpu-layers N\n"); - fprintf(stdout, " number of layers to store in VRAM\n"); - fprintf(stdout, " -ts SPLIT --tensor-split SPLIT\n"); - fprintf(stdout, " how to split tensors across multiple GPUs, comma-separated list of proportions, e.g. 3,1\n"); - fprintf(stdout, " -mg i, --main-gpu i the GPU to use for scratch and small tensors\n" ); - fprintf(stdout, " -lv, --low-vram don't allocate VRAM scratch buffer\n" ); - fprintf(stdout, " -mmq, --mul-mat-q use experimental mul_mat_q CUDA kernels instead of cuBLAS. TEMP!!!\n" ); - fprintf(stdout, " Reduces VRAM usage by 700/970/1430 MiB for 7b/13b/33b but prompt processing speed\n" ); - fprintf(stdout, " is still suboptimal, especially q2_K, q3_K, q5_K, and q6_K.\n" ); -#endif - fprintf(stdout, " --mtest compute maximum memory usage\n"); - fprintf(stdout, " --export export the computation graph to 'llama.ggml'\n"); - fprintf(stdout, " --verbose-prompt print prompt before generation\n"); - fprintf(stderr, " --simple-io use basic IO for better compatibility in subprocesses and limited consoles\n"); - fprintf(stdout, " --lora FNAME apply LoRA adapter (implies --no-mmap)\n"); - fprintf(stdout, " --lora-base FNAME optional model to use as a base for the layers modified by the LoRA adapter\n"); - fprintf(stdout, " -m FNAME, --model FNAME\n"); - fprintf(stdout, " model path (default: %s)\n", params.model.c_str()); - fprintf(stdout, "\n"); -} - -std::string gpt_random_prompt(std::mt19937 & rng) { - const int r = rng() % 10; - switch (r) { - case 0: return "So"; - case 1: return "Once upon a time"; - case 2: return "When"; - case 3: return "The"; - case 4: return "After"; - case 5: return "If"; - case 6: return "import"; - case 7: return "He"; - case 8: return "She"; - case 9: return "They"; - default: return "To"; - } - - return "The"; -} - -// TODO: not great allocating this every time -std::vector<llama_token> llama_tokenize(struct llama_context * ctx, const std::string & text, bool add_bos) { - // initialize to prompt numer of chars, since n_tokens <= n_prompt_chars - std::vector<llama_token> res(text.size() + (int) add_bos); - const int n = llama_tokenize(ctx, text.c_str(), res.data(), res.size(), add_bos); - assert(n >= 0); - res.resize(n); - - return res; -} - -struct llama_context_params llama_context_params_from_gpt_params(const gpt_params & params) { - auto lparams = llama_context_default_params(); - - lparams.n_ctx = params.n_ctx; - lparams.n_batch = params.n_batch; - lparams.n_gqa = params.n_gqa; - lparams.rms_norm_eps = params.rms_norm_eps; - lparams.n_gpu_layers = params.n_gpu_layers; - lparams.main_gpu = params.main_gpu; - lparams.tensor_split = params.tensor_split; - lparams.low_vram = params.low_vram; - lparams.mul_mat_q = params.mul_mat_q; - lparams.seed = params.seed; - lparams.f16_kv = params.memory_f16; - lparams.use_mmap = params.use_mmap; - lparams.use_mlock = params.use_mlock; - lparams.logits_all = params.perplexity; - lparams.embedding = params.embedding; - lparams.rope_freq_base = params.rope_freq_base; - lparams.rope_freq_scale = params.rope_freq_scale; - - return lparams; -} - -std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_params(const gpt_params & params) { - auto lparams = llama_context_params_from_gpt_params(params); - - llama_model * model = llama_load_model_from_file(params.model.c_str(), lparams); - if (model == NULL) { - fprintf(stderr, "%s: error: failed to load model '%s'\n", __func__, params.model.c_str()); - return std::make_tuple(nullptr, nullptr); - } - - llama_context * lctx = llama_new_context_with_model(model, lparams); - if (lctx == NULL) { - fprintf(stderr, "%s: error: failed to create context with model '%s'\n", __func__, params.model.c_str()); - llama_free_model(model); - return std::make_tuple(nullptr, nullptr); - } - - if (!params.lora_adapter.empty()) { - int err = llama_model_apply_lora_from_file(model, - params.lora_adapter.c_str(), - params.lora_base.empty() ? NULL : params.lora_base.c_str(), - params.n_threads); - if (err != 0) { - fprintf(stderr, "%s: error: failed to apply lora adapter\n", __func__); - llama_free(lctx); - llama_free_model(model); - return std::make_tuple(nullptr, nullptr); - } - } - - return std::make_tuple(model, lctx); -} diff --git a/examples/common.h b/examples/common.h deleted file mode 100644 index 375bc0a3..00000000 --- a/examples/common.h +++ /dev/null @@ -1,114 +0,0 @@ -// Various helper functions and utilities - -#pragma once - -#include "llama.h" - -#include <string> -#include <vector> -#include <random> -#include <thread> -#include <unordered_map> -#include <tuple> - -// -// CLI argument parsing -// -int32_t get_num_physical_cores(); - -struct gpt_params { - uint32_t seed = -1; // RNG seed - int32_t n_threads = get_num_physical_cores(); - int32_t n_predict = -1; // new tokens to predict - int32_t n_ctx = 512; // context size - int32_t n_batch = 512; // batch size for prompt processing (must be >=32 to use BLAS) - int32_t n_gqa = 1; // grouped-query attention factor (TODO: move to hparams) - int32_t n_keep = 0; // number of tokens to keep from initial prompt - int32_t n_chunks = -1; // max number of chunks to process (-1 = unlimited) - int32_t n_gpu_layers = 0; // number of layers to store in VRAM - int32_t main_gpu = 0; // the GPU that is used for scratch and small tensors - float tensor_split[LLAMA_MAX_DEVICES] = {0}; // how split tensors should be distributed across GPUs - int32_t n_probs = 0; // if greater than 0, output the probabilities of top n_probs tokens. - float rms_norm_eps = LLAMA_DEFAULT_RMS_EPS; // rms norm epsilon - float rope_freq_base = 10000.0f; // RoPE base frequency - float rope_freq_scale = 1.0f; // RoPE frequency scaling factor - - // sampling parameters - std::unordered_map<llama_token, float> logit_bias; // logit bias for specific tokens - int32_t top_k = 40; // <= 0 to use vocab size - float top_p = 0.95f; // 1.0 = disabled - float tfs_z = 1.00f; // 1.0 = disabled - float typical_p = 1.00f; // 1.0 = disabled - float temp = 0.80f; // 1.0 = disabled - float repeat_penalty = 1.10f; // 1.0 = disabled - int32_t repeat_last_n = 64; // last n tokens to penalize (0 = disable penalty, -1 = context size) - float frequency_penalty = 0.00f; // 0.0 = disabled - float presence_penalty = 0.00f; // 0.0 = disabled - int32_t mirostat = 0; // 0 = disabled, 1 = mirostat, 2 = mirostat 2.0 - float mirostat_tau = 5.00f; // target entropy - float mirostat_eta = 0.10f; // learning rate - - // Classifier-Free Guidance - // https://arxiv.org/abs/2306.17806 - std::string cfg_negative_prompt; // string to help guidance - float cfg_scale = 1.f; // How strong is guidance - - std::string model = "models/7B/ggml-model.bin"; // model path - std::string model_alias = "unknown"; // model alias - std::string prompt = ""; - std::string path_prompt_cache = ""; // path to file for saving/loading prompt eval state - std::string input_prefix = ""; // string to prefix user inputs with - std::string input_suffix = ""; // string to suffix user inputs with - std::string grammar = ""; // optional BNF-like grammar to constrain sampling - std::vector<std::string> antiprompt; // string upon seeing which more user input is prompted - - std::string lora_adapter = ""; // lora adapter path - std::string lora_base = ""; // base model path for the lora adapter - - bool hellaswag = false; // compute HellaSwag score over random tasks from datafile supplied in prompt - size_t hellaswag_tasks = 400; // number of tasks to use when computing the HellaSwag score - - bool low_vram = false; // if true, reduce VRAM usage at the cost of performance - bool mul_mat_q = false; // if true, use experimental mul_mat_q kernels - bool memory_f16 = true; // use f16 instead of f32 for memory kv - bool random_prompt = false; // do not randomize prompt if none provided - bool use_color = false; // use color to distinguish generations and inputs - bool interactive = false; // interactive mode - bool prompt_cache_all = false; // save user input and generations to prompt cache - bool prompt_cache_ro = false; // open the prompt cache read-only and do not update it - - bool embedding = false; // get only sentence embedding - bool interactive_first = false; // wait for user input immediately - bool multiline_input = false; // reverse the usage of `\` - bool simple_io = false; // improves compatibility with subprocesses and limited consoles - - bool input_prefix_bos = false; // prefix BOS to user inputs, preceding input_prefix - bool instruct = false; // instruction mode (used for Alpaca models) - bool penalize_nl = true; // consider newlines as a repeatable token - bool perplexity = false; // compute perplexity over the prompt - bool use_mmap = true; // use mmap for faster loads - bool use_mlock = false; // use mlock to keep model in memory - bool mem_test = false; // compute maximum memory usage - bool numa = false; // attempt optimizations that help on some NUMA systems - bool export_cgraph = false; // export the computation graph - bool verbose_prompt = false; // print prompt tokens before generation -}; - -bool gpt_params_parse(int argc, char ** argv, gpt_params & params); - -void gpt_print_usage(int argc, char ** argv, const gpt_params & params); - -std::string gpt_random_prompt(std::mt19937 & rng); - -// -// Vocab utils -// - -std::vector<llama_token> llama_tokenize(struct llama_context * ctx, const std::string & text, bool add_bos); - -// -// Model utils -// - -std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_params(const gpt_params & params); -struct llama_context_params llama_context_params_from_gpt_params(const gpt_params & params); diff --git a/examples/console.cpp b/examples/console.cpp deleted file mode 100644 index 8efa2a67..00000000 --- a/examples/console.cpp +++ /dev/null @@ -1,500 +0,0 @@ -#include "console.h" -#include <vector> -#include <iostream> - -#if defined(_WIN32) -#define WIN32_LEAN_AND_MEAN -#ifndef NOMINMAX -#define NOMINMAX -#endif -#include <windows.h> -#include <fcntl.h> -#include <io.h> -#ifndef ENABLE_VIRTUAL_TERMINAL_PROCESSING -#define ENABLE_VIRTUAL_TERMINAL_PROCESSING 0x0004 -#endif -#else -#include <climits> -#include <sys/ioctl.h> -#include <unistd.h> -#include <wchar.h> -#include <stdio.h> -#include <stdlib.h> -#include <signal.h> -#include <termios.h> -#endif - -#define ANSI_COLOR_RED "\x1b[31m" -#define ANSI_COLOR_GREEN "\x1b[32m" -#define ANSI_COLOR_YELLOW "\x1b[33m" -#define ANSI_COLOR_BLUE "\x1b[34m" -#define ANSI_COLOR_MAGENTA "\x1b[35m" -#define ANSI_COLOR_CYAN "\x1b[36m" -#define ANSI_COLOR_RESET "\x1b[0m" -#define ANSI_BOLD "\x1b[1m" - -namespace console { - - // - // Console state - // - - static bool advanced_display = false; - static bool simple_io = true; - static display_t current_display = reset; - - static FILE* out = stdout; - -#if defined (_WIN32) - static void* hConsole; -#else - static FILE* tty = nullptr; - static termios initial_state; -#endif - - // - // Init and cleanup - // - - void init(bool use_simple_io, bool use_advanced_display) { - advanced_display = use_advanced_display; - simple_io = use_simple_io; -#if defined(_WIN32) - // Windows-specific console initialization - DWORD dwMode = 0; - hConsole = GetStdHandle(STD_OUTPUT_HANDLE); - if (hConsole == INVALID_HANDLE_VALUE || !GetConsoleMode(hConsole, &dwMode)) { - hConsole = GetStdHandle(STD_ERROR_HANDLE); - if (hConsole != INVALID_HANDLE_VALUE && (!GetConsoleMode(hConsole, &dwMode))) { - hConsole = nullptr; - simple_io = true; - } - } - if (hConsole) { - // Check conditions combined to reduce nesting - if (advanced_display && !(dwMode & ENABLE_VIRTUAL_TERMINAL_PROCESSING) && - !SetConsoleMode(hConsole, dwMode | ENABLE_VIRTUAL_TERMINAL_PROCESSING)) { - advanced_display = false; - } - // Set console output codepage to UTF8 - SetConsoleOutputCP(CP_UTF8); - } - HANDLE hConIn = GetStdHandle(STD_INPUT_HANDLE); - if (hConIn != INVALID_HANDLE_VALUE && GetConsoleMode(hConIn, &dwMode)) { - // Set console input codepage to UTF16 - _setmode(_fileno(stdin), _O_WTEXT); - - // Set ICANON (ENABLE_LINE_INPUT) and ECHO (ENABLE_ECHO_INPUT) - if (simple_io) { - dwMode |= ENABLE_LINE_INPUT | ENABLE_ECHO_INPUT; - } else { - dwMode &= ~(ENABLE_LINE_INPUT | ENABLE_ECHO_INPUT); - } - if (!SetConsoleMode(hConIn, dwMode)) { - simple_io = true; - } - } -#else - // POSIX-specific console initialization - if (!simple_io) { - struct termios new_termios; - tcgetattr(STDIN_FILENO, &initial_state); - new_termios = initial_state; - new_termios.c_lflag &= ~(ICANON | ECHO); - new_termios.c_cc[VMIN] = 1; - new_termios.c_cc[VTIME] = 0; - tcsetattr(STDIN_FILENO, TCSANOW, &new_termios); - - tty = fopen("/dev/tty", "w+"); - if (tty != nullptr) { - out = tty; - } - } - - setlocale(LC_ALL, ""); -#endif - } - - void cleanup() { - // Reset console display - set_display(reset); - -#if !defined(_WIN32) - // Restore settings on POSIX systems - if (!simple_io) { - if (tty != nullptr) { - out = stdout; - fclose(tty); - tty = nullptr; - } - tcsetattr(STDIN_FILENO, TCSANOW, &initial_state); - } -#endif - } - - // - // Display and IO - // - - // Keep track of current display and only emit ANSI code if it changes - void set_display(display_t display) { - if (advanced_display && current_display != display) { - fflush(stdout); - switch(display) { - case reset: - fprintf(out, ANSI_COLOR_RESET); - break; - case prompt: - fprintf(out, ANSI_COLOR_YELLOW); - break; - case user_input: - fprintf(out, ANSI_BOLD ANSI_COLOR_GREEN); - break; - case error: - fprintf(out, ANSI_BOLD ANSI_COLOR_RED); - } - current_display = display; - fflush(out); - } - } - - char32_t getchar32() { -#if defined(_WIN32) - HANDLE hConsole = GetStdHandle(STD_INPUT_HANDLE); - wchar_t high_surrogate = 0; - - while (true) { - INPUT_RECORD record; - DWORD count; - if (!ReadConsoleInputW(hConsole, &record, 1, &count) || count == 0) { - return WEOF; - } - - if (record.EventType == KEY_EVENT && record.Event.KeyEvent.bKeyDown) { - wchar_t wc = record.Event.KeyEvent.uChar.UnicodeChar; - if (wc == 0) { - continue; - } - - if ((wc >= 0xD800) && (wc <= 0xDBFF)) { // Check if wc is a high surrogate - high_surrogate = wc; - continue; - } - if ((wc >= 0xDC00) && (wc <= 0xDFFF)) { // Check if wc is a low surrogate - if (high_surrogate != 0) { // Check if we have a high surrogate - return ((high_surrogate - 0xD800) << 10) + (wc - 0xDC00) + 0x10000; - } - } - - high_surrogate = 0; // Reset the high surrogate - return static_cast<char32_t>(wc); - } - } -#else - wchar_t wc = getwchar(); - if (static_cast<wint_t>(wc) == WEOF) { - return WEOF; - } - -#if WCHAR_MAX == 0xFFFF - if ((wc >= 0xD800) && (wc <= 0xDBFF)) { // Check if wc is a high surrogate - wchar_t low_surrogate = getwchar(); - if ((low_surrogate >= 0xDC00) && (low_surrogate <= 0xDFFF)) { // Check if the next wchar is a low surrogate - return (static_cast<char32_t>(wc & 0x03FF) << 10) + (low_surrogate & 0x03FF) + 0x10000; - } - } - if ((wc >= 0xD800) && (wc <= 0xDFFF)) { // Invalid surrogate pair - return 0xFFFD; // Return the replacement character U+FFFD - } -#endif - - return static_cast<char32_t>(wc); -#endif - } - - void pop_cursor() { -#if defined(_WIN32) - if (hConsole != NULL) { - CONSOLE_SCREEN_BUFFER_INFO bufferInfo; - GetConsoleScreenBufferInfo(hConsole, &bufferInfo); - - COORD newCursorPosition = bufferInfo.dwCursorPosition; - if (newCursorPosition.X == 0) { - newCursorPosition.X = bufferInfo.dwSize.X - 1; - newCursorPosition.Y -= 1; - } else { - newCursorPosition.X -= 1; - } - - SetConsoleCursorPosition(hConsole, newCursorPosition); - return; - } -#endif - putc('\b', out); - } - - int estimateWidth(char32_t codepoint) { -#if defined(_WIN32) - return 1; -#else - return wcwidth(codepoint); -#endif - } - - int put_codepoint(const char* utf8_codepoint, size_t length, int expectedWidth) { -#if defined(_WIN32) - CONSOLE_SCREEN_BUFFER_INFO bufferInfo; - if (!GetConsoleScreenBufferInfo(hConsole, &bufferInfo)) { - // go with the default - return expectedWidth; - } - COORD initialPosition = bufferInfo.dwCursorPosition; - DWORD nNumberOfChars = length; - WriteConsole(hConsole, utf8_codepoint, nNumberOfChars, &nNumberOfChars, NULL); - - CONSOLE_SCREEN_BUFFER_INFO newBufferInfo; - GetConsoleScreenBufferInfo(hConsole, &newBufferInfo); - - // Figure out our real position if we're in the last column - if (utf8_codepoint[0] != 0x09 && initialPosition.X == newBufferInfo.dwSize.X - 1) { - DWORD nNumberOfChars; - WriteConsole(hConsole, &" \b", 2, &nNumberOfChars, NULL); - GetConsoleScreenBufferInfo(hConsole, &newBufferInfo); - } - - int width = newBufferInfo.dwCursorPosition.X - initialPosition.X; - if (width < 0) { - width += newBufferInfo.dwSize.X; - } - return width; -#else - // We can trust expectedWidth if we've got one - if (expectedWidth >= 0 || tty == nullptr) { - fwrite(utf8_codepoint, length, 1, out); - return expectedWidth; - } - - fputs("\033[6n", tty); // Query cursor position - int x1; - int y1; - int x2; - int y2; - int results = 0; - results = fscanf(tty, "\033[%d;%dR", &y1, &x1); - - fwrite(utf8_codepoint, length, 1, tty); - - fputs("\033[6n", tty); // Query cursor position - results += fscanf(tty, "\033[%d;%dR", &y2, &x2); - - if (results != 4) { - return expectedWidth; - } - - int width = x2 - x1; - if (width < 0) { - // Calculate the width considering text wrapping - struct winsize w; - ioctl(STDOUT_FILENO, TIOCGWINSZ, &w); - width += w.ws_col; - } - return width; -#endif - } - - void replace_last(char ch) { -#if defined(_WIN32) - pop_cursor(); - put_codepoint(&ch, 1, 1); -#else - fprintf(out, "\b%c", ch); -#endif - } - - void append_utf8(char32_t ch, std::string & out) { - if (ch <= 0x7F) { - out.push_back(static_cast<unsigned char>(ch)); - } else if (ch <= 0x7FF) { - out.push_back(static_cast<unsigned char>(0xC0 | ((ch >> 6) & 0x1F))); - out.push_back(static_cast<unsigned char>(0x80 | (ch & 0x3F))); - } else if (ch <= 0xFFFF) { - out.push_back(static_cast<unsigned char>(0xE0 | ((ch >> 12) & 0x0F))); - out.push_back(static_cast<unsigned char>(0x80 | ((ch >> 6) & 0x3F))); - out.push_back(static_cast<unsigned char>(0x80 | (ch & 0x3F))); - } else if (ch <= 0x10FFFF) { - out.push_back(static_cast<unsigned char>(0xF0 | ((ch >> 18) & 0x07))); - out.push_back(static_cast<unsigned char>(0x80 | ((ch >> 12) & 0x3F))); - out.push_back(static_cast<unsigned char>(0x80 | ((ch >> 6) & 0x3F))); - out.push_back(static_cast<unsigned char>(0x80 | (ch & 0x3F))); - } else { - // Invalid Unicode code point - } - } - - // Helper function to remove the last UTF-8 character from a string - void pop_back_utf8_char(std::string & line) { - if (line.empty()) { - return; - } - - size_t pos = line.length() - 1; - - // Find the start of the last UTF-8 character (checking up to 4 bytes back) - for (size_t i = 0; i < 3 && pos > 0; ++i, --pos) { - if ((line[pos] & 0xC0) != 0x80) { - break; // Found the start of the character - } - } - line.erase(pos); - } - - bool readline_advanced(std::string & line, bool multiline_input) { - if (out != stdout) { - fflush(stdout); - } - - line.clear(); - std::vector<int> widths; - bool is_special_char = false; - bool end_of_stream = false; - - char32_t input_char; - while (true) { - fflush(out); // Ensure all output is displayed before waiting for input - input_char = getchar32(); - - if (input_char == '\r' || input_char == '\n') { - break; - } - - if (input_char == (char32_t) WEOF || input_char == 0x04 /* Ctrl+D*/) { - end_of_stream = true; - break; - } - - if (is_special_char) { - set_display(user_input); - replace_last(line.back()); - is_special_char = false; - } - - if (input_char == '\033') { // Escape sequence - char32_t code = getchar32(); - if (code == '[' || code == 0x1B) { - // Discard the rest of the escape sequence - while ((code = getchar32()) != (char32_t) WEOF) { - if ((code >= 'A' && code <= 'Z') || (code >= 'a' && code <= 'z') || code == '~') { - break; - } - } - } - } else if (input_char == 0x08 || input_char == 0x7F) { // Backspace - if (!widths.empty()) { - int count; - do { - count = widths.back(); - widths.pop_back(); - // Move cursor back, print space, and move cursor back again - for (int i = 0; i < count; i++) { - replace_last(' '); - pop_cursor(); - } - pop_back_utf8_char(line); - } while (count == 0 && !widths.empty()); - } - } else { - int offset = line.length(); - append_utf8(input_char, line); - int width = put_codepoint(line.c_str() + offset, line.length() - offset, estimateWidth(input_char)); - if (width < 0) { - width = 0; - } - widths.push_back(width); - } - - if (!line.empty() && (line.back() == '\\' || line.back() == '/')) { - set_display(prompt); - replace_last(line.back()); - is_special_char = true; - } - } - - bool has_more = multiline_input; - if (is_special_char) { - replace_last(' '); - pop_cursor(); - - char last = line.back(); - line.pop_back(); - if (last == '\\') { - line += '\n'; - fputc('\n', out); - has_more = !has_more; - } else { - // llama will just eat the single space, it won't act as a space - if (line.length() == 1 && line.back() == ' ') { - line.clear(); - pop_cursor(); - } - has_more = false; - } - } else { - if (end_of_stream) { - has_more = false; - } else { - line += '\n'; - fputc('\n', out); - } - } - - fflush(out); - return has_more; - } - - bool readline_simple(std::string & line, bool multiline_input) { -#if defined(_WIN32) - std::wstring wline; - if (!std::getline(std::wcin, wline)) { - // Input stream is bad or EOF received - line.clear(); - GenerateConsoleCtrlEvent(CTRL_C_EVENT, 0); - return false; - } - - int size_needed = WideCharToMultiByte(CP_UTF8, 0, &wline[0], (int)wline.size(), NULL, 0, NULL, NULL); - line.resize(size_needed); - WideCharToMultiByte(CP_UTF8, 0, &wline[0], (int)wline.size(), &line[0], size_needed, NULL, NULL); -#else - if (!std::getline(std::cin, line)) { - // Input stream is bad or EOF received - line.clear(); - return false; - } -#endif - if (!line.empty()) { - char last = line.back(); - if (last == '/') { // Always return control on '/' symbol - line.pop_back(); - return false; - } - if (last == '\\') { // '\\' changes the default action - line.pop_back(); - multiline_input = !multiline_input; - } - } - line += '\n'; - - // By default, continue input if multiline_input is set - return multiline_input; - } - - bool readline(std::string & line, bool multiline_input) { - set_display(user_input); - - if (simple_io) { - return readline_simple(line, multiline_input); - } - return readline_advanced(line, multiline_input); - } - -} diff --git a/examples/console.h b/examples/console.h deleted file mode 100644 index ec175269..00000000 --- a/examples/console.h +++ /dev/null @@ -1,19 +0,0 @@ -// Console functions - -#pragma once - -#include <string> - -namespace console { - enum display_t { - reset = 0, - prompt, - user_input, - error - }; - - void init(bool use_simple_io, bool use_advanced_display); - void cleanup(); - void set_display(display_t display); - bool readline(std::string & line, bool multiline_input); -} diff --git a/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp b/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp index 1a238c4d..469d6e3d 100644 --- a/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp +++ b/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp @@ -1,5 +1,6 @@ #include "ggml.h" #include "llama.h" + #include <unordered_map> #include <vector> #include <cassert> @@ -138,14 +139,16 @@ void print_sample_weights(TransformerWeights *w){ struct llama_vocab { using id = int32_t; using token = std::string; + using ttype = llama_token_type; - struct token_score { - token tok; + struct token_data { + token text; float score; + ttype type; }; std::unordered_map<token, id> token_to_id; - std::vector<token_score> id_to_token; + std::vector<token_data> id_to_token; }; struct my_llama_hparams { @@ -502,7 +505,7 @@ bool is_ggml_file(const char *filename) { return false; } uint32_t magic = file.read_u32(); - return magic == LLAMA_FILE_MAGIC; + return magic == GGUF_MAGIC; } void load_vocab(const char *filename, Config *config, struct llama_vocab *vocab) { @@ -515,36 +518,30 @@ void load_vocab(const char *filename, Config *config, struct llama_vocab *vocab) struct llama_model * lmodel = llama_load_model_from_file(filename, llama_params); struct llama_context * lctx = llama_new_context_with_model(lmodel, llama_params); - std::vector<const char *> strings; - std::vector<float> scores; - int n_vocab = llama_n_vocab(lctx); - strings.resize(n_vocab, NULL); - scores.resize(n_vocab, 0); - n_vocab = llama_get_vocab(lctx, strings.data(), scores.data(), n_vocab); - GGML_ASSERT(n_vocab == llama_n_vocab(lctx)); + const int n_vocab = llama_n_vocab(lctx); vocab->id_to_token.resize(n_vocab); for (int i=0; i<n_vocab; ++i) { - std::string tok = std::string(strings[i]); - float score = scores[i]; - vocab->id_to_token[i].tok = tok; - vocab->id_to_token[i].score = score; - vocab->token_to_id.emplace(tok, i); + vocab->id_to_token[i].text = llama_token_get_text(lctx, i); + vocab->id_to_token[i].score = llama_token_get_score(lctx, i); + vocab->id_to_token[i].type = llama_token_get_type(lctx, i); + vocab->token_to_id.emplace(vocab->id_to_token[i].text, i); } llama_free(lctx); llama_free_model(lmodel); } else { // assume llama2.c vocabulary printf("Assuming llama2.c vocabulary since %s is not a ggml file\n", filename); llama_file file(filename, "rb"); - uint32_t n_vocab = config->vocab_size; + const int n_vocab = config->vocab_size; /* uint32_t max_token_length = */ file.read_u32(); // unused vocab->id_to_token.resize(n_vocab); - for (uint32_t i=0; i<n_vocab; ++i) { + for (int i=0; i<n_vocab; ++i) { float_t score = file.read_f32(); uint32_t len = file.read_u32(); - std::string tok = file.read_string(len); - vocab->id_to_token[i].tok = tok; + std::string text = file.read_string(len); + vocab->id_to_token[i].text = text; vocab->id_to_token[i].score = score; - vocab->token_to_id.emplace(tok, i); + vocab->id_to_token[i].type = LLAMA_TOKEN_TYPE_UNDEFINED; + vocab->token_to_id.emplace(text, i); } } } @@ -590,75 +587,80 @@ void save_as_llama_model(struct llama_vocab * vocab, struct my_llama_model * mod if (file.fp == NULL) { return; } - // write_magic - file.write_u32(LLAMA_FILE_MAGIC); // magic - file.write_u32(LLAMA_FILE_VERSION); // version - // write_hparams - file.write_u32(model->hparams.n_vocab); - file.write_u32(model->hparams.n_embd); - file.write_u32(model->hparams.n_mult); - file.write_u32(model->hparams.n_head); - file.write_u32(model->hparams.n_layer); - file.write_u32(model->hparams.n_rot); - file.write_u32(LLAMA_FTYPE_ALL_F32); - - // write_vocab - for now we are just writing the existing BPE voc. assuming karpathy's vocabulary is the same. idk. - uint32_t n_vocab = model->hparams.n_vocab; - for (uint32_t i = 0; i < n_vocab; i++) { - const auto & token_score = vocab->id_to_token.at(i); - file.write_u32((uint32_t) token_score.tok.size()); - file.write_raw(token_score.tok.data(), token_score.tok.size()); - file.write_raw(&token_score.score, sizeof(token_score.score)); - } - // stuff AK weights into GG weights one by one. - // w->token_embedding_table -> model->tok_embeddings - // float* -> struct ggml_tensor - stuff_karpathy_weights_into_gg(model->tok_embeddings, w->token_embedding_table); - stuff_karpathy_weights_into_gg(model->output, w->token_embedding_table); - - stuff_karpathy_weights_into_gg(model->norm, w->rms_final_weight); - //print_row(model->norm, 0); - - // for rms-att-weight - int row_length = model->hparams.n_embd; - const auto & hparams = model->hparams; - //int n_ff = model->hparams.n_embd; - int n_ff = get_n_ff(&hparams); - - for (uint32_t i = 0; i < model->hparams.n_layer; ++i){ - auto & layer = model->layers[i]; - // 1d - stuff_karpathy_weights_into_gg(layer.attention_norm, &w->rms_att_weight[i*row_length]); - stuff_karpathy_weights_into_gg(layer.ffn_norm , &w->rms_ffn_weight[i*row_length]); - - // from 3d matrix layer x dim x dim to 2d matrix dim x dim - stuff_karpathy_weights_into_gg(layer.wq , &w->wq[i*row_length*row_length]); - stuff_karpathy_weights_into_gg(layer.wk , &w->wk[i*row_length*row_length]); - stuff_karpathy_weights_into_gg(layer.wv , &w->wv[i*row_length*row_length]); - stuff_karpathy_weights_into_gg(layer.wo , &w->wo[i*row_length*row_length]); - - stuff_karpathy_weights_into_gg(layer.w1 , &w->w1[i*row_length*n_ff]); - stuff_karpathy_weights_into_gg(layer.w2 , &w->w2[i*n_ff*row_length]); - stuff_karpathy_weights_into_gg(layer.w3 , &w->w3[i*row_length*n_ff]); - } - // write tensors - write_tensor(&file, model->tok_embeddings); - write_tensor(&file, model->norm); - write_tensor(&file, model->output); // ? - for (uint32_t i = 0; i < model->hparams.n_layer; ++i) { - auto & layer = model->layers[i]; - - write_tensor(&file, layer.attention_norm); - write_tensor(&file, layer.wq); - write_tensor(&file, layer.wk); - write_tensor(&file, layer.wv); - write_tensor(&file, layer.wo); - write_tensor(&file, layer.ffn_norm); - write_tensor(&file, layer.w1); - write_tensor(&file, layer.w2); - write_tensor(&file, layer.w3); - } +#pragma message("TODO: implement file saving using gguf") + (void) vocab; + (void) model; + (void) w; +// // write_magic +// file.write_u32(LLAMA_FILE_MAGIC); // magic +// file.write_u32(LLAMA_FILE_VERSION); // version +// // write_hparams +// file.write_u32(model->hparams.n_vocab); +// file.write_u32(model->hparams.n_embd); +// file.write_u32(model->hparams.n_mult); +// file.write_u32(model->hparams.n_head); +// file.write_u32(model->hparams.n_layer); +// file.write_u32(model->hparams.n_rot); +// file.write_u32(LLAMA_FTYPE_ALL_F32); +// +// // write_vocab - for now we are just writing the existing BPE voc. assuming karpathy's vocabulary is the same. idk. +// uint32_t n_vocab = model->hparams.n_vocab; +// for (uint32_t i = 0; i < n_vocab; i++) { +// const auto & token_data = vocab->id_to_token.at(i); +// file.write_u32((uint32_t) token_data.tok.size()); +// file.write_raw(token_data.tok.data(), token_data.tok.size()); +// file.write_raw(&token_data.score, sizeof(token_data.score)); +// } +// +// // stuff AK weights into GG weights one by one. +// // w->token_embedding_table -> model->tok_embeddings +// // float* -> struct ggml_tensor +// stuff_karpathy_weights_into_gg(model->tok_embeddings, w->token_embedding_table); +// stuff_karpathy_weights_into_gg(model->output, w->token_embedding_table); +// +// stuff_karpathy_weights_into_gg(model->norm, w->rms_final_weight); +// //print_row(model->norm, 0); +// +// // for rms-att-weight +// int row_length = model->hparams.n_embd; +// const auto & hparams = model->hparams; +// //int n_ff = model->hparams.n_embd; +// int n_ff = get_n_ff(&hparams); +// +// for (uint32_t i = 0; i < model->hparams.n_layer; ++i){ +// auto & layer = model->layers[i]; +// // 1d +// stuff_karpathy_weights_into_gg(layer.attention_norm, &w->rms_att_weight[i*row_length]); +// stuff_karpathy_weights_into_gg(layer.ffn_norm , &w->rms_ffn_weight[i*row_length]); +// +// // from 3d matrix layer x dim x dim to 2d matrix dim x dim +// stuff_karpathy_weights_into_gg(layer.wq , &w->wq[i*row_length*row_length]); +// stuff_karpathy_weights_into_gg(layer.wk , &w->wk[i*row_length*row_length]); +// stuff_karpathy_weights_into_gg(layer.wv , &w->wv[i*row_length*row_length]); +// stuff_karpathy_weights_into_gg(layer.wo , &w->wo[i*row_length*row_length]); +// +// stuff_karpathy_weights_into_gg(layer.w1 , &w->w1[i*row_length*n_ff]); +// stuff_karpathy_weights_into_gg(layer.w2 , &w->w2[i*n_ff*row_length]); +// stuff_karpathy_weights_into_gg(layer.w3 , &w->w3[i*row_length*n_ff]); +// } +// // write tensors +// write_tensor(&file, model->tok_embeddings); +// write_tensor(&file, model->norm); +// write_tensor(&file, model->output); // ? +// for (uint32_t i = 0; i < model->hparams.n_layer; ++i) { +// auto & layer = model->layers[i]; +// +// write_tensor(&file, layer.attention_norm); +// write_tensor(&file, layer.wq); +// write_tensor(&file, layer.wk); +// write_tensor(&file, layer.wv); +// write_tensor(&file, layer.wo); +// write_tensor(&file, layer.ffn_norm); +// write_tensor(&file, layer.w1); +// write_tensor(&file, layer.w2); +// write_tensor(&file, layer.w3); +// } } struct train_params get_default_train_params() { diff --git a/examples/embd-input/embd-input-lib.cpp b/examples/embd-input/embd-input-lib.cpp index 2185b9b0..8a6ad882 100644 --- a/examples/embd-input/embd-input-lib.cpp +++ b/examples/embd-input/embd-input-lib.cpp @@ -167,7 +167,7 @@ llama_token sampling_id(struct MyModel* mymodel) { llama_token_data_array candidates_p = { candidates.data(), candidates.size(), false }; // TODO: Apply penalties - // float nl_logit = logits[llama_token_nl()]; + // float nl_logit = logits[llama_token_nl(ctx)]; // auto last_n_repeat = std::min(std::min((int)last_n_tokens.size(), repeat_last_n), n_ctx); // llama_sample_repetition_penalty(ctx, &candidates_p, // last_n_tokens.data() + last_n_tokens.size() - last_n_repeat, @@ -176,7 +176,7 @@ llama_token sampling_id(struct MyModel* mymodel) { // last_n_tokens.data() + last_n_tokens.size() - last_n_repeat, // last_n_repeat, alpha_frequency, alpha_presence); // if (!penalize_nl) { - // logits[llama_token_nl()] = nl_logit; + // logits[llama_token_nl(ctx)] = nl_logit; // } if (temp <= 0) { @@ -211,7 +211,7 @@ const char * sampling(struct MyModel * mymodel) { llama_context * ctx = mymodel->ctx; int id = sampling_id(mymodel); static std::string ret; - if (id == llama_token_eos()) { + if (id == llama_token_eos(ctx)) { ret = "</s>"; } else { ret = llama_token_to_str(ctx, id); diff --git a/examples/embedding/embedding.cpp b/examples/embedding/embedding.cpp index 5192d6df..8788571c 100644 --- a/examples/embedding/embedding.cpp +++ b/examples/embedding/embedding.cpp @@ -67,7 +67,7 @@ int main(int argc, char ** argv) { fprintf(stderr, "%s: prompt: '%s'\n", __func__, params.prompt.c_str()); fprintf(stderr, "%s: number of tokens in prompt = %zu\n", __func__, embd_inp.size()); for (int i = 0; i < (int) embd_inp.size(); i++) { - fprintf(stderr, "%6d -> '%s'\n", embd_inp[i], llama_token_to_str(ctx, embd_inp[i])); + fprintf(stderr, "%6d -> '%s'\n", embd_inp[i], llama_token_to_str(ctx, embd_inp[i]).c_str()); } fprintf(stderr, "\n"); } diff --git a/examples/gguf/gguf.cpp b/examples/gguf/gguf.cpp new file mode 100644 index 00000000..dee00df8 --- /dev/null +++ b/examples/gguf/gguf.cpp @@ -0,0 +1,246 @@ +#include "ggml.h" +#include "llama.h" + +#include <cstdio> +#include <cinttypes> +#include <string> +#include <sstream> +#include <fstream> +#include <vector> + +#undef MIN +#undef MAX +#define MIN(a, b) ((a) < (b) ? (a) : (b)) +#define MAX(a, b) ((a) > (b) ? (a) : (b)) + +template<typename T> +static std::string to_string(const T & val) { + std::stringstream ss; + ss << val; + return ss.str(); +} + +bool gguf_ex_write(const std::string & fname) { + struct gguf_context * ctx = gguf_init_empty(); + + gguf_set_val_u8 (ctx, "some.parameter.uint8", 0x12); + gguf_set_val_i8 (ctx, "some.parameter.int8", -0x13); + gguf_set_val_u16 (ctx, "some.parameter.uint16", 0x1234); + gguf_set_val_i16 (ctx, "some.parameter.int16", -0x1235); + gguf_set_val_u32 (ctx, "some.parameter.uint32", 0x12345678); + gguf_set_val_i32 (ctx, "some.parameter.int32", -0x12345679); + gguf_set_val_f32 (ctx, "some.parameter.float32", 0.123456789f); + gguf_set_val_bool(ctx, "some.parameter.bool", true); + gguf_set_val_str (ctx, "some.parameter.string", "hello world"); + + gguf_set_arr_data(ctx, "some.parameter.arr.i16", GGUF_TYPE_INT16, std::vector<int16_t>{ 1, 2, 3, 4, }.data(), 4); + gguf_set_arr_data(ctx, "some.parameter.arr.f32", GGUF_TYPE_FLOAT32, std::vector<float>{ 3.145f, 2.718f, 1.414f, }.data(), 3); + gguf_set_arr_str (ctx, "some.parameter.arr.str", std::vector<const char *>{ "hello", "world", "!" }.data(), 3); + + struct ggml_init_params params = { + /*.mem_size =*/ 128ull*1024ull*1024ull, + /*.mem_buffer =*/ NULL, + /*.no_alloc =*/ false, + }; + + struct ggml_context * ctx_data = ggml_init(params); + + const int n_tensors = 10; + + // tensor infos + for (int i = 0; i < n_tensors; ++i) { + const std::string name = "tensor_" + to_string(i); + + int64_t ne[GGML_MAX_DIMS] = { 1 }; + int32_t n_dims = rand() % GGML_MAX_DIMS + 1; + + for (int j = 0; j < n_dims; ++j) { + ne[j] = rand() % 10 + 1; + } + + struct ggml_tensor * cur = ggml_new_tensor(ctx_data, GGML_TYPE_F32, n_dims, ne); + ggml_set_name(cur, name.c_str()); + + { + float * data = (float *) cur->data; + for (int j = 0; j < ggml_nelements(cur); ++j) { + data[j] = 100 + i; + } + } + + gguf_add_tensor(ctx, cur); + } + + gguf_write_to_file(ctx, fname.c_str(), false); + + fprintf(stdout, "%s: wrote file '%s;\n", __func__, fname.c_str()); + + ggml_free(ctx_data); + gguf_free(ctx); + + return true; +} + +// just read tensor info +bool gguf_ex_read_0(const std::string & fname) { + struct gguf_init_params params = { + /*.no_alloc = */ false, + /*.ctx = */ NULL, + }; + + struct gguf_context * ctx = gguf_init_from_file(fname.c_str(), params); + + fprintf(stdout, "%s: version: %d\n", __func__, gguf_get_version(ctx)); + fprintf(stdout, "%s: alignment: %zu\n", __func__, gguf_get_alignment(ctx)); + fprintf(stdout, "%s: data offset: %zu\n", __func__, gguf_get_data_offset(ctx)); + + // kv + { + const int n_kv = gguf_get_n_kv(ctx); + + fprintf(stdout, "%s: n_kv: %d\n", __func__, n_kv); + + for (int i = 0; i < n_kv; ++i) { + const char * key = gguf_get_key(ctx, i); + + fprintf(stdout, "%s: kv[%d]: key = %s\n", __func__, i, key); + } + } + + // find kv string + { + const char * findkey = "some.parameter.string"; + + const int keyidx = gguf_find_key(ctx, findkey); + if (keyidx == -1) { + fprintf(stdout, "%s: find key: %s not found.\n", __func__, findkey); + } else { + const char * key_value = gguf_get_val_str(ctx, keyidx); + fprintf(stdout, "%s: find key: %s found, kv[%d] value = %s\n", __func__, findkey, keyidx, key_value); + } + } + + // tensor info + { + const int n_tensors = gguf_get_n_tensors(ctx); + + fprintf(stdout, "%s: n_tensors: %d\n", __func__, n_tensors); + + for (int i = 0; i < n_tensors; ++i) { + const char * name = gguf_get_tensor_name (ctx, i); + const size_t offset = gguf_get_tensor_offset(ctx, i); + + fprintf(stdout, "%s: tensor[%d]: name = %s, offset = %zu\n", __func__, i, name, offset); + } + } + + gguf_free(ctx); + + return true; +} + +// read and create ggml_context containing the tensors and their data +bool gguf_ex_read_1(const std::string & fname) { + struct ggml_context * ctx_data = NULL; + + struct gguf_init_params params = { + /*.no_alloc = */ false, + /*.ctx = */ &ctx_data, + }; + + struct gguf_context * ctx = gguf_init_from_file(fname.c_str(), params); + + fprintf(stdout, "%s: version: %d\n", __func__, gguf_get_version(ctx)); + fprintf(stdout, "%s: alignment: %zu\n", __func__, gguf_get_alignment(ctx)); + fprintf(stdout, "%s: data offset: %zu\n", __func__, gguf_get_data_offset(ctx)); + + // kv + { + const int n_kv = gguf_get_n_kv(ctx); + + fprintf(stdout, "%s: n_kv: %d\n", __func__, n_kv); + + for (int i = 0; i < n_kv; ++i) { + const char * key = gguf_get_key(ctx, i); + + fprintf(stdout, "%s: kv[%d]: key = %s\n", __func__, i, key); + } + } + + // tensor info + { + const int n_tensors = gguf_get_n_tensors(ctx); + + fprintf(stdout, "%s: n_tensors: %d\n", __func__, n_tensors); + + for (int i = 0; i < n_tensors; ++i) { + const char * name = gguf_get_tensor_name (ctx, i); + const size_t offset = gguf_get_tensor_offset(ctx, i); + + fprintf(stdout, "%s: tensor[%d]: name = %s, offset = %zu\n", __func__, i, name, offset); + } + } + + // data + { + const int n_tensors = gguf_get_n_tensors(ctx); + + for (int i = 0; i < n_tensors; ++i) { + fprintf(stdout, "%s: reading tensor %d data\n", __func__, i); + + const char * name = gguf_get_tensor_name(ctx, i); + + struct ggml_tensor * cur = ggml_get_tensor(ctx_data, name); + + fprintf(stdout, "%s: tensor[%d]: n_dims = %d, name = %s, data = %p\n", __func__, i, cur->n_dims, cur->name, cur->data); + + // print first 10 elements + const float * data = (const float *) cur->data; + + printf("%s data[:10] : ", name); + for (int j = 0; j < MIN(10, ggml_nelements(cur)); ++j) { + printf("%f ", data[j]); + } + printf("\n\n"); + + // check data + { + const float * data = (const float *) cur->data; + for (int j = 0; j < ggml_nelements(cur); ++j) { + if (data[j] != 100 + i) { + fprintf(stderr, "%s: tensor[%d]: data[%d] = %f\n", __func__, i, j, data[j]); + return false; + } + } + } + } + } + + fprintf(stdout, "%s: ctx_data size: %zu\n", __func__, ggml_get_mem_size(ctx_data)); + + ggml_free(ctx_data); + gguf_free(ctx); + + return true; +} + +int main(int argc, char ** argv) { + if (argc < 3) { + fprintf(stdout, "usage: %s data.gguf r|w\n", argv[0]); + return -1; + } + + const std::string fname(argv[1]); + const std::string mode (argv[2]); + + GGML_ASSERT((mode == "r" || mode == "w") && "mode must be r or w"); + + if (mode == "w") { + GGML_ASSERT(gguf_ex_write(fname) && "failed to write gguf file"); + } else if (mode == "r") { + GGML_ASSERT(gguf_ex_read_0(fname) && "failed to read gguf file"); + GGML_ASSERT(gguf_ex_read_1(fname) && "failed to read gguf file"); + } + + return 0; +} diff --git a/examples/gptneox-wip/cmpnct_gpt2bpe.hpp b/examples/gptneox-wip/cmpnct_gpt2bpe.hpp new file mode 100644 index 00000000..9d433f4b --- /dev/null +++ b/examples/gptneox-wip/cmpnct_gpt2bpe.hpp @@ -0,0 +1,1133 @@ +#ifndef CMPNCT_GPT2BPE +#define CMPNCT_GPT2BPE + +#include <vector> +#include <string> +#include <algorithm> +#include <utility> +#include <iostream> +#include <map> +#include <unordered_map> +#include <queue> +#include <cstring> + + +// Unicode GPT2 Byte Pair Encoding Tokenizer +// Adapted from https://github.com/cmp-nct/ggllm.cpp [MIT License] +// Removed loading of merges from HF json and parts made for a specific vocab + + +//----------------- +// Unicode library (from cmpnct_unicode.cpp) +//----------------- + +// Minimal library for high performance handling and categorization of UTF8 strings and characters +// Using std::string + +enum CNCTCharType { + DIGIT, // a numerical char in any language + LETTER, // a letter in any language + WHITESPACE, // any form of whitespace + ACCENT_MARK, // letter modifiers like ´ in é + PUNCTUATION, // punctuation including brackets + SYMBOL, // math, currency, other symbols + CONTROL, // control characters + MIXED, // a mix of the above + UNIDENTIFIED // something more exotic like emoji or separators +}; + +struct CNCTUnicode; + +struct CNCTString { + std::string str; + size_t utf8_chars; + + CNCTCharType char_type=UNIDENTIFIED; + bool is_sequential=false; + + size_t seq_offset_bytes=0; + size_t seq_offset_utf8_chars=0; + + bool operator==(const std::string &other) const; + bool operator==(const char other) const; + bool operator==(const CNCTString &other) const; + CNCTString &operator+=(const std::string &other); + CNCTString &operator+=(const char other); + friend CNCTString operator+(CNCTString lhs, const std::string &rhs); + friend CNCTString operator+(CNCTString lhs, const char rhs); + CNCTString& operator+=(const CNCTString& other); + friend CNCTString operator+(CNCTString lhs, const CNCTString& rhs); +}; + +struct CNCTUnicode { + static bool check_code_range(int c, const std::vector<std::pair<int, int>>& ranges); + static CNCTCharType get_code_type(int c); + static CNCTCharType get_code_type(const std::string &utf8_char); + static int utf8_len(const char c); + static int strlen_utf8(std::string src); + static std::vector<std::string> split_utf8(const std::string &src); + static std::vector<CNCTString> split_utf8_enhanced(const std::string &src); + static CNCTCharType string_identify(const std::string& str); + static bool string_test(const std::string& str, CNCTCharType chartype); +}; + +static const std::vector<std::pair<int, int>> digit_ranges = { +{0x30, 0x39}, {0xB2, 0xB3}, {0xB9, 0xB9}, {0x660, 0x669}, {0x6F0, 0x6F9}, {0x7C0, 0x7C9}, {0x966, 0x96F}, {0x9E6, 0x9EF}, {0xA66, 0xA6F}, {0xAE6, 0xAEF}, {0xB66, 0xB6F}, {0xBE6, 0xBEF}, {0xC66, 0xC6F}, +{0xCE6, 0xCEF}, {0xD66, 0xD6F}, {0xDE6, 0xDEF}, {0xE50, 0xE59}, {0xED0, 0xED9}, {0xF20, 0xF29}, {0x1040, 0x1049}, {0x1090, 0x1099}, {0x1369, 0x1371}, {0x17E0, 0x17E9}, {0x1810, 0x1819}, {0x1946, 0x194F}, +{0x19D0, 0x19DA}, {0x1A80, 0x1A89}, {0x1A90, 0x1A99}, {0x1B50, 0x1B59}, {0x1BB0, 0x1BB9}, {0x1C40, 0x1C49}, {0x1C50, 0x1C59}, {0x2070, 0x2070}, {0x2074, 0x2079}, {0x2080, 0x2089}, {0x2460, 0x2468}, +{0x2474, 0x247C}, {0x2488, 0x2490}, {0x24EA, 0x24EA}, {0x24F5, 0x24FD}, {0x24FF, 0x24FF}, {0x2776, 0x277E}, {0x2780, 0x2788}, {0x278A, 0x2792}, {0xA620, 0xA629}, {0xA8D0, 0xA8D9}, {0xA900, 0xA909}, +{0xA9D0, 0xA9D9}, {0xA9F0, 0xA9F9}, {0xAA50, 0xAA59}, {0xABF0, 0xABF9}, {0xFF10, 0xFF19}, {0x104A0, 0x104A9}, {0x10A40, 0x10A43}, {0x10D30, 0x10D39}, {0x10E60, 0x10E68}, {0x11052, 0x1105A}, +{0x11066, 0x1106F}, {0x110F0, 0x110F9}, {0x11136, 0x1113F}, {0x111D0, 0x111D9}, {0x112F0, 0x112F9}, {0x11450, 0x11459}, {0x114D0, 0x114D9}, {0x11650, 0x11659}, {0x116C0, 0x116C9}, {0x11730, 0x11739}, +{0x118E0, 0x118E9}, {0x11950, 0x11959}, {0x11C50, 0x11C59}, {0x11D50, 0x11D59}, {0x11DA0, 0x11DA9}, {0x16A60, 0x16A69}, {0x16B50, 0x16B59}, {0x1D7CE, 0x1D7FF}, {0x1E140, 0x1E149}, {0x1E2F0, 0x1E2F9}, +{0x1E950, 0x1E959}, {0x1F100, 0x1F10A}, {0x1FBF0, 0x1FBF9}, +}; + +static const std::vector<std::pair<int, int>> letter_ranges = { +{0x41, 0x5A}, {0x61, 0x7A}, {0xAA, 0xAA}, {0xB5, 0xB5}, {0xBA, 0xBA}, {0xC0, 0xD6}, {0xD8, 0xF6}, {0xF8, 0x2C1}, {0x2C6, 0x2D1}, {0x2E0, 0x2E4}, {0x2EC, 0x2EC}, {0x2EE, 0x2EE}, {0x370, 0x374}, +{0x376, 0x377}, {0x37A, 0x37D}, {0x37F, 0x37F}, {0x386, 0x386}, {0x388, 0x38A}, {0x38C, 0x38C}, {0x38E, 0x3A1}, {0x3A3, 0x3F5}, {0x3F7, 0x481}, {0x48A, 0x52F}, {0x531, 0x556}, {0x559, 0x559}, +{0x560, 0x588}, {0x5D0, 0x5EA}, {0x5EF, 0x5F2}, {0x620, 0x64A}, {0x66E, 0x66F}, {0x671, 0x6D3}, {0x6D5, 0x6D5}, {0x6E5, 0x6E6}, {0x6EE, 0x6EF}, {0x6FA, 0x6FC}, {0x6FF, 0x6FF}, {0x710, 0x710}, +{0x712, 0x72F}, {0x74D, 0x7A5}, {0x7B1, 0x7B1}, {0x7CA, 0x7EA}, {0x7F4, 0x7F5}, {0x7FA, 0x7FA}, {0x800, 0x815}, {0x81A, 0x81A}, {0x824, 0x824}, {0x828, 0x828}, {0x840, 0x858}, {0x860, 0x86A}, +{0x8A0, 0x8B4}, {0x8B6, 0x8C7}, {0x904, 0x939}, {0x93D, 0x93D}, {0x950, 0x950}, {0x958, 0x961}, {0x971, 0x980}, {0x985, 0x98C}, {0x98F, 0x990}, {0x993, 0x9A8}, {0x9AA, 0x9B0}, {0x9B2, 0x9B2}, +{0x9B6, 0x9B9}, {0x9BD, 0x9BD}, {0x9CE, 0x9CE}, {0x9DC, 0x9DD}, {0x9DF, 0x9E1}, {0x9F0, 0x9F1}, {0x9FC, 0x9FC}, {0xA05, 0xA0A}, {0xA0F, 0xA10}, {0xA13, 0xA28}, {0xA2A, 0xA30}, {0xA32, 0xA33}, +{0xA35, 0xA36}, {0xA38, 0xA39}, {0xA59, 0xA5C}, {0xA5E, 0xA5E}, {0xA72, 0xA74}, {0xA85, 0xA8D}, {0xA8F, 0xA91}, {0xA93, 0xAA8}, {0xAAA, 0xAB0}, {0xAB2, 0xAB3}, {0xAB5, 0xAB9}, {0xABD, 0xABD}, +{0xAD0, 0xAD0}, {0xAE0, 0xAE1}, {0xAF9, 0xAF9}, {0xB05, 0xB0C}, {0xB0F, 0xB10}, {0xB13, 0xB28}, {0xB2A, 0xB30}, {0xB32, 0xB33}, {0xB35, 0xB39}, {0xB3D, 0xB3D}, {0xB5C, 0xB5D}, {0xB5F, 0xB61}, +{0xB71, 0xB71}, {0xB83, 0xB83}, {0xB85, 0xB8A}, {0xB8E, 0xB90}, {0xB92, 0xB95}, {0xB99, 0xB9A}, {0xB9C, 0xB9C}, {0xB9E, 0xB9F}, {0xBA3, 0xBA4}, {0xBA8, 0xBAA}, {0xBAE, 0xBB9}, {0xBD0, 0xBD0}, +{0xC05, 0xC0C}, {0xC0E, 0xC10}, {0xC12, 0xC28}, {0xC2A, 0xC39}, {0xC3D, 0xC3D}, {0xC58, 0xC5A}, {0xC60, 0xC61}, {0xC80, 0xC80}, {0xC85, 0xC8C}, {0xC8E, 0xC90}, {0xC92, 0xCA8}, {0xCAA, 0xCB3}, +{0xCB5, 0xCB9}, {0xCBD, 0xCBD}, {0xCDE, 0xCDE}, {0xCE0, 0xCE1}, {0xCF1, 0xCF2}, {0xD04, 0xD0C}, {0xD0E, 0xD10}, {0xD12, 0xD3A}, {0xD3D, 0xD3D}, {0xD4E, 0xD4E}, {0xD54, 0xD56}, {0xD5F, 0xD61}, +{0xD7A, 0xD7F}, {0xD85, 0xD96}, {0xD9A, 0xDB1}, {0xDB3, 0xDBB}, {0xDBD, 0xDBD}, {0xDC0, 0xDC6}, {0xE01, 0xE30}, {0xE32, 0xE33}, {0xE40, 0xE46}, {0xE81, 0xE82}, {0xE84, 0xE84}, {0xE86, 0xE8A}, +{0xE8C, 0xEA3}, {0xEA5, 0xEA5}, {0xEA7, 0xEB0}, {0xEB2, 0xEB3}, {0xEBD, 0xEBD}, {0xEC0, 0xEC4}, {0xEC6, 0xEC6}, {0xEDC, 0xEDF}, {0xF00, 0xF00}, {0xF40, 0xF47}, {0xF49, 0xF6C}, {0xF88, 0xF8C}, +{0x1000, 0x102A}, {0x103F, 0x103F}, {0x1050, 0x1055}, {0x105A, 0x105D}, {0x1061, 0x1061}, {0x1065, 0x1066}, {0x106E, 0x1070}, {0x1075, 0x1081}, {0x108E, 0x108E}, {0x10A0, 0x10C5}, {0x10C7, 0x10C7}, +{0x10CD, 0x10CD}, {0x10D0, 0x10FA}, {0x10FC, 0x1248}, {0x124A, 0x124D}, {0x1250, 0x1256}, {0x1258, 0x1258}, {0x125A, 0x125D}, {0x1260, 0x1288}, {0x128A, 0x128D}, {0x1290, 0x12B0}, {0x12B2, 0x12B5}, +{0x12B8, 0x12BE}, {0x12C0, 0x12C0}, {0x12C2, 0x12C5}, {0x12C8, 0x12D6}, {0x12D8, 0x1310}, {0x1312, 0x1315}, {0x1318, 0x135A}, {0x1380, 0x138F}, {0x13A0, 0x13F5}, {0x13F8, 0x13FD}, {0x1401, 0x166C}, +{0x166F, 0x167F}, {0x1681, 0x169A}, {0x16A0, 0x16EA}, {0x16F1, 0x16F8}, {0x1700, 0x170C}, {0x170E, 0x1711}, {0x1720, 0x1731}, {0x1740, 0x1751}, {0x1760, 0x176C}, {0x176E, 0x1770}, {0x1780, 0x17B3}, +{0x17D7, 0x17D7}, {0x17DC, 0x17DC}, {0x1820, 0x1878}, {0x1880, 0x1884}, {0x1887, 0x18A8}, {0x18AA, 0x18AA}, {0x18B0, 0x18F5}, {0x1900, 0x191E}, {0x1950, 0x196D}, {0x1970, 0x1974}, {0x1980, 0x19AB}, +{0x19B0, 0x19C9}, {0x1A00, 0x1A16}, {0x1A20, 0x1A54}, {0x1AA7, 0x1AA7}, {0x1B05, 0x1B33}, {0x1B45, 0x1B4B}, {0x1B83, 0x1BA0}, {0x1BAE, 0x1BAF}, {0x1BBA, 0x1BE5}, {0x1C00, 0x1C23}, {0x1C4D, 0x1C4F}, +{0x1C5A, 0x1C7D}, {0x1C80, 0x1C88}, {0x1C90, 0x1CBA}, {0x1CBD, 0x1CBF}, {0x1CE9, 0x1CEC}, {0x1CEE, 0x1CF3}, {0x1CF5, 0x1CF6}, {0x1CFA, 0x1CFA}, {0x1D00, 0x1DBF}, {0x1E00, 0x1F15}, {0x1F18, 0x1F1D}, +{0x1F20, 0x1F45}, {0x1F48, 0x1F4D}, {0x1F50, 0x1F57}, {0x1F59, 0x1F59}, {0x1F5B, 0x1F5B}, {0x1F5D, 0x1F5D}, {0x1F5F, 0x1F7D}, {0x1F80, 0x1FB4}, {0x1FB6, 0x1FBC}, {0x1FBE, 0x1FBE}, {0x1FC2, 0x1FC4}, +{0x1FC6, 0x1FCC}, {0x1FD0, 0x1FD3}, {0x1FD6, 0x1FDB}, {0x1FE0, 0x1FEC}, {0x1FF2, 0x1FF4}, {0x1FF6, 0x1FFC}, {0x2071, 0x2071}, {0x207F, 0x207F}, {0x2090, 0x209C}, {0x2102, 0x2102}, {0x2107, 0x2107}, +{0x210A, 0x2113}, {0x2115, 0x2115}, {0x2119, 0x211D}, {0x2124, 0x2124}, {0x2126, 0x2126}, {0x2128, 0x2128}, {0x212A, 0x212D}, {0x212F, 0x2139}, {0x213C, 0x213F}, {0x2145, 0x2149}, {0x214E, 0x214E}, +{0x2183, 0x2184}, {0x2C00, 0x2C2E}, {0x2C30, 0x2C5E}, {0x2C60, 0x2CE4}, {0x2CEB, 0x2CEE}, {0x2CF2, 0x2CF3}, {0x2D00, 0x2D25}, {0x2D27, 0x2D27}, {0x2D2D, 0x2D2D}, {0x2D30, 0x2D67}, {0x2D6F, 0x2D6F}, +{0x2D80, 0x2D96}, {0x2DA0, 0x2DA6}, {0x2DA8, 0x2DAE}, {0x2DB0, 0x2DB6}, {0x2DB8, 0x2DBE}, {0x2DC0, 0x2DC6}, {0x2DC8, 0x2DCE}, {0x2DD0, 0x2DD6}, {0x2DD8, 0x2DDE}, {0x2E2F, 0x2E2F}, {0x3005, 0x3006}, +{0x3031, 0x3035}, {0x303B, 0x303C}, {0x3041, 0x3096}, {0x309D, 0x309F}, {0x30A1, 0x30FA}, {0x30FC, 0x30FF}, {0x3105, 0x312F}, {0x3131, 0x318E}, {0x31A0, 0x31BF}, {0x31F0, 0x31FF}, {0x3400, 0x4DBF}, +{0x4E00, 0x9FFC}, {0xA000, 0xA48C}, {0xA4D0, 0xA4FD}, {0xA500, 0xA60C}, {0xA610, 0xA61F}, {0xA62A, 0xA62B}, {0xA640, 0xA66E}, {0xA67F, 0xA69D}, {0xA6A0, 0xA6E5}, {0xA717, 0xA71F}, {0xA722, 0xA788}, +{0xA78B, 0xA7BF}, {0xA7C2, 0xA7CA}, {0xA7F5, 0xA801}, {0xA803, 0xA805}, {0xA807, 0xA80A}, {0xA80C, 0xA822}, {0xA840, 0xA873}, {0xA882, 0xA8B3}, {0xA8F2, 0xA8F7}, {0xA8FB, 0xA8FB}, {0xA8FD, 0xA8FE}, +{0xA90A, 0xA925}, {0xA930, 0xA946}, {0xA960, 0xA97C}, {0xA984, 0xA9B2}, {0xA9CF, 0xA9CF}, {0xA9E0, 0xA9E4}, {0xA9E6, 0xA9EF}, {0xA9FA, 0xA9FE}, {0xAA00, 0xAA28}, {0xAA40, 0xAA42}, {0xAA44, 0xAA4B}, +{0xAA60, 0xAA76}, {0xAA7A, 0xAA7A}, {0xAA7E, 0xAAAF}, {0xAAB1, 0xAAB1}, {0xAAB5, 0xAAB6}, {0xAAB9, 0xAABD}, {0xAAC0, 0xAAC0}, {0xAAC2, 0xAAC2}, {0xAADB, 0xAADD}, {0xAAE0, 0xAAEA}, {0xAAF2, 0xAAF4}, +{0xAB01, 0xAB06}, {0xAB09, 0xAB0E}, {0xAB11, 0xAB16}, {0xAB20, 0xAB26}, {0xAB28, 0xAB2E}, {0xAB30, 0xAB5A}, {0xAB5C, 0xAB69}, {0xAB70, 0xABE2}, {0xAC00, 0xD7A3}, {0xD7B0, 0xD7C6}, {0xD7CB, 0xD7FB}, +{0xF900, 0xFA6D}, {0xFA70, 0xFAD9}, {0xFB00, 0xFB06}, {0xFB13, 0xFB17}, {0xFB1D, 0xFB1D}, {0xFB1F, 0xFB28}, {0xFB2A, 0xFB36}, {0xFB38, 0xFB3C}, {0xFB3E, 0xFB3E}, {0xFB40, 0xFB41}, {0xFB43, 0xFB44}, +{0xFB46, 0xFBB1}, {0xFBD3, 0xFD3D}, {0xFD50, 0xFD8F}, {0xFD92, 0xFDC7}, {0xFDF0, 0xFDFB}, {0xFE70, 0xFE74}, {0xFE76, 0xFEFC}, {0xFF21, 0xFF3A}, {0xFF41, 0xFF5A}, {0xFF66, 0xFFBE}, {0xFFC2, 0xFFC7}, +{0xFFCA, 0xFFCF}, {0xFFD2, 0xFFD7}, {0xFFDA, 0xFFDC}, {0x10000, 0x1000B}, {0x1000D, 0x10026}, {0x10028, 0x1003A}, {0x1003C, 0x1003D}, {0x1003F, 0x1004D}, {0x10050, 0x1005D}, {0x10080, 0x100FA}, +{0x10280, 0x1029C}, {0x102A0, 0x102D0}, {0x10300, 0x1031F}, {0x1032D, 0x10340}, {0x10342, 0x10349}, {0x10350, 0x10375}, {0x10380, 0x1039D}, {0x103A0, 0x103C3}, {0x103C8, 0x103CF}, {0x10400, 0x1049D}, +{0x104B0, 0x104D3}, {0x104D8, 0x104FB}, {0x10500, 0x10527}, {0x10530, 0x10563}, {0x10600, 0x10736}, {0x10740, 0x10755}, {0x10760, 0x10767}, {0x10800, 0x10805}, {0x10808, 0x10808}, {0x1080A, 0x10835}, +{0x10837, 0x10838}, {0x1083C, 0x1083C}, {0x1083F, 0x10855}, {0x10860, 0x10876}, {0x10880, 0x1089E}, {0x108E0, 0x108F2}, {0x108F4, 0x108F5}, {0x10900, 0x10915}, {0x10920, 0x10939}, {0x10980, 0x109B7}, +{0x109BE, 0x109BF}, {0x10A00, 0x10A00}, {0x10A10, 0x10A13}, {0x10A15, 0x10A17}, {0x10A19, 0x10A35}, {0x10A60, 0x10A7C}, {0x10A80, 0x10A9C}, {0x10AC0, 0x10AC7}, {0x10AC9, 0x10AE4}, {0x10B00, 0x10B35}, +{0x10B40, 0x10B55}, {0x10B60, 0x10B72}, {0x10B80, 0x10B91}, {0x10C00, 0x10C48}, {0x10C80, 0x10CB2}, {0x10CC0, 0x10CF2}, {0x10D00, 0x10D23}, {0x10E80, 0x10EA9}, {0x10EB0, 0x10EB1}, {0x10F00, 0x10F1C}, +{0x10F27, 0x10F27}, {0x10F30, 0x10F45}, {0x10FB0, 0x10FC4}, {0x10FE0, 0x10FF6}, {0x11003, 0x11037}, {0x11083, 0x110AF}, {0x110D0, 0x110E8}, {0x11103, 0x11126}, {0x11144, 0x11144}, {0x11147, 0x11147}, +{0x11150, 0x11172}, {0x11176, 0x11176}, {0x11183, 0x111B2}, {0x111C1, 0x111C4}, {0x111DA, 0x111DA}, {0x111DC, 0x111DC}, {0x11200, 0x11211}, {0x11213, 0x1122B}, {0x11280, 0x11286}, {0x11288, 0x11288}, +{0x1128A, 0x1128D}, {0x1128F, 0x1129D}, {0x1129F, 0x112A8}, {0x112B0, 0x112DE}, {0x11305, 0x1130C}, {0x1130F, 0x11310}, {0x11313, 0x11328}, {0x1132A, 0x11330}, {0x11332, 0x11333}, {0x11335, 0x11339}, +{0x1133D, 0x1133D}, {0x11350, 0x11350}, {0x1135D, 0x11361}, {0x11400, 0x11434}, {0x11447, 0x1144A}, {0x1145F, 0x11461}, {0x11480, 0x114AF}, {0x114C4, 0x114C5}, {0x114C7, 0x114C7}, {0x11580, 0x115AE}, +{0x115D8, 0x115DB}, {0x11600, 0x1162F}, {0x11644, 0x11644}, {0x11680, 0x116AA}, {0x116B8, 0x116B8}, {0x11700, 0x1171A}, {0x11800, 0x1182B}, {0x118A0, 0x118DF}, {0x118FF, 0x11906}, {0x11909, 0x11909}, +{0x1190C, 0x11913}, {0x11915, 0x11916}, {0x11918, 0x1192F}, {0x1193F, 0x1193F}, {0x11941, 0x11941}, {0x119A0, 0x119A7}, {0x119AA, 0x119D0}, {0x119E1, 0x119E1}, {0x119E3, 0x119E3}, {0x11A00, 0x11A00}, +{0x11A0B, 0x11A32}, {0x11A3A, 0x11A3A}, {0x11A50, 0x11A50}, {0x11A5C, 0x11A89}, {0x11A9D, 0x11A9D}, {0x11AC0, 0x11AF8}, {0x11C00, 0x11C08}, {0x11C0A, 0x11C2E}, {0x11C40, 0x11C40}, {0x11C72, 0x11C8F}, +{0x11D00, 0x11D06}, {0x11D08, 0x11D09}, {0x11D0B, 0x11D30}, {0x11D46, 0x11D46}, {0x11D60, 0x11D65}, {0x11D67, 0x11D68}, {0x11D6A, 0x11D89}, {0x11D98, 0x11D98}, {0x11EE0, 0x11EF2}, {0x11FB0, 0x11FB0}, +{0x12000, 0x12399}, {0x12480, 0x12543}, {0x13000, 0x1342E}, {0x14400, 0x14646}, {0x16800, 0x16A38}, {0x16A40, 0x16A5E}, {0x16AD0, 0x16AED}, {0x16B00, 0x16B2F}, {0x16B40, 0x16B43}, {0x16B63, 0x16B77}, +{0x16B7D, 0x16B8F}, {0x16E40, 0x16E7F}, {0x16F00, 0x16F4A}, {0x16F50, 0x16F50}, {0x16F93, 0x16F9F}, {0x16FE0, 0x16FE1}, {0x16FE3, 0x16FE3}, {0x17000, 0x187F7}, {0x18800, 0x18CD5}, {0x18D00, 0x18D08}, +{0x1B000, 0x1B11E}, {0x1B150, 0x1B152}, {0x1B164, 0x1B167}, {0x1B170, 0x1B2FB}, {0x1BC00, 0x1BC6A}, {0x1BC70, 0x1BC7C}, {0x1BC80, 0x1BC88}, {0x1BC90, 0x1BC99}, {0x1D400, 0x1D454}, {0x1D456, 0x1D49C}, +{0x1D49E, 0x1D49F}, {0x1D4A2, 0x1D4A2}, {0x1D4A5, 0x1D4A6}, {0x1D4A9, 0x1D4AC}, {0x1D4AE, 0x1D4B9}, {0x1D4BB, 0x1D4BB}, {0x1D4BD, 0x1D4C3}, {0x1D4C5, 0x1D505}, {0x1D507, 0x1D50A}, {0x1D50D, 0x1D514}, +{0x1D516, 0x1D51C}, {0x1D51E, 0x1D539}, {0x1D53B, 0x1D53E}, {0x1D540, 0x1D544}, {0x1D546, 0x1D546}, {0x1D54A, 0x1D550}, {0x1D552, 0x1D6A5}, {0x1D6A8, 0x1D6C0}, {0x1D6C2, 0x1D6DA}, {0x1D6DC, 0x1D6FA}, +{0x1D6FC, 0x1D714}, {0x1D716, 0x1D734}, {0x1D736, 0x1D74E}, {0x1D750, 0x1D76E}, {0x1D770, 0x1D788}, {0x1D78A, 0x1D7A8}, {0x1D7AA, 0x1D7C2}, {0x1D7C4, 0x1D7CB}, {0x1E100, 0x1E12C}, {0x1E137, 0x1E13D}, +{0x1E14E, 0x1E14E}, {0x1E2C0, 0x1E2EB}, {0x1E800, 0x1E8C4}, {0x1E900, 0x1E943}, {0x1E94B, 0x1E94B}, {0x1EE00, 0x1EE03}, {0x1EE05, 0x1EE1F}, {0x1EE21, 0x1EE22}, {0x1EE24, 0x1EE24}, {0x1EE27, 0x1EE27}, +{0x1EE29, 0x1EE32}, {0x1EE34, 0x1EE37}, {0x1EE39, 0x1EE39}, {0x1EE3B, 0x1EE3B}, {0x1EE42, 0x1EE42}, {0x1EE47, 0x1EE47}, {0x1EE49, 0x1EE49}, {0x1EE4B, 0x1EE4B}, {0x1EE4D, 0x1EE4F}, {0x1EE51, 0x1EE52}, +{0x1EE54, 0x1EE54}, {0x1EE57, 0x1EE57}, {0x1EE59, 0x1EE59}, {0x1EE5B, 0x1EE5B}, {0x1EE5D, 0x1EE5D}, {0x1EE5F, 0x1EE5F}, {0x1EE61, 0x1EE62}, {0x1EE64, 0x1EE64}, {0x1EE67, 0x1EE6A}, {0x1EE6C, 0x1EE72}, +{0x1EE74, 0x1EE77}, {0x1EE79, 0x1EE7C}, {0x1EE7E, 0x1EE7E}, {0x1EE80, 0x1EE89}, {0x1EE8B, 0x1EE9B}, {0x1EEA1, 0x1EEA3}, {0x1EEA5, 0x1EEA9}, {0x1EEAB, 0x1EEBB}, {0x20000, 0x2A6DD}, {0x2A700, 0x2B734}, +{0x2B740, 0x2B81D}, {0x2B820, 0x2CEA1}, {0x2CEB0, 0x2EBE0}, {0x2F800, 0x2FA1D}, {0x30000, 0x3134A}, +}; + +static const std::vector<std::pair<int, int>> whitespace_ranges = { +{0x9, 0xD}, {0x1C, 0x20}, {0x85, 0x85}, {0xA0, 0xA0}, {0x1680, 0x1680}, {0x2000, 0x200A}, {0x2028, 0x2029}, {0x202F, 0x202F}, {0x205F, 0x205F}, {0x3000, 0x3000}, +}; + +static const std::vector<std::pair<int, int>> accent_mark_ranges = { +{0x300, 0x36F}, {0x483, 0x489}, {0x591, 0x5BD}, {0x5BF, 0x5BF}, {0x5C1, 0x5C2}, {0x5C4, 0x5C5}, {0x5C7, 0x5C7}, {0x610, 0x61A}, {0x64B, 0x65F}, {0x670, 0x670}, {0x6D6, 0x6DC}, {0x6DF, 0x6E4}, +{0x6E7, 0x6E8}, {0x6EA, 0x6ED}, {0x711, 0x711}, {0x730, 0x74A}, {0x7A6, 0x7B0}, {0x7EB, 0x7F3}, {0x7FD, 0x7FD}, {0x816, 0x819}, {0x81B, 0x823}, {0x825, 0x827}, {0x829, 0x82D}, {0x859, 0x85B}, +{0x8D3, 0x8E1}, {0x8E3, 0x903}, {0x93A, 0x93C}, {0x93E, 0x94F}, {0x951, 0x957}, {0x962, 0x963}, {0x981, 0x983}, {0x9BC, 0x9BC}, {0x9BE, 0x9C4}, {0x9C7, 0x9C8}, {0x9CB, 0x9CD}, {0x9D7, 0x9D7}, +{0x9E2, 0x9E3}, {0x9FE, 0x9FE}, {0xA01, 0xA03}, {0xA3C, 0xA3C}, {0xA3E, 0xA42}, {0xA47, 0xA48}, {0xA4B, 0xA4D}, {0xA51, 0xA51}, {0xA70, 0xA71}, {0xA75, 0xA75}, {0xA81, 0xA83}, {0xABC, 0xABC}, +{0xABE, 0xAC5}, {0xAC7, 0xAC9}, {0xACB, 0xACD}, {0xAE2, 0xAE3}, {0xAFA, 0xAFF}, {0xB01, 0xB03}, {0xB3C, 0xB3C}, {0xB3E, 0xB44}, {0xB47, 0xB48}, {0xB4B, 0xB4D}, {0xB55, 0xB57}, {0xB62, 0xB63}, +{0xB82, 0xB82}, {0xBBE, 0xBC2}, {0xBC6, 0xBC8}, {0xBCA, 0xBCD}, {0xBD7, 0xBD7}, {0xC00, 0xC04}, {0xC3E, 0xC44}, {0xC46, 0xC48}, {0xC4A, 0xC4D}, {0xC55, 0xC56}, {0xC62, 0xC63}, {0xC81, 0xC83}, +{0xCBC, 0xCBC}, {0xCBE, 0xCC4}, {0xCC6, 0xCC8}, {0xCCA, 0xCCD}, {0xCD5, 0xCD6}, {0xCE2, 0xCE3}, {0xD00, 0xD03}, {0xD3B, 0xD3C}, {0xD3E, 0xD44}, {0xD46, 0xD48}, {0xD4A, 0xD4D}, {0xD57, 0xD57}, +{0xD62, 0xD63}, {0xD81, 0xD83}, {0xDCA, 0xDCA}, {0xDCF, 0xDD4}, {0xDD6, 0xDD6}, {0xDD8, 0xDDF}, {0xDF2, 0xDF3}, {0xE31, 0xE31}, {0xE34, 0xE3A}, {0xE47, 0xE4E}, {0xEB1, 0xEB1}, {0xEB4, 0xEBC}, +{0xEC8, 0xECD}, {0xF18, 0xF19}, {0xF35, 0xF35}, {0xF37, 0xF37}, {0xF39, 0xF39}, {0xF3E, 0xF3F}, {0xF71, 0xF84}, {0xF86, 0xF87}, {0xF8D, 0xF97}, {0xF99, 0xFBC}, {0xFC6, 0xFC6}, {0x102B, 0x103E}, +{0x1056, 0x1059}, {0x105E, 0x1060}, {0x1062, 0x1064}, {0x1067, 0x106D}, {0x1071, 0x1074}, {0x1082, 0x108D}, {0x108F, 0x108F}, {0x109A, 0x109D}, {0x135D, 0x135F}, {0x1712, 0x1714}, {0x1732, 0x1734}, +{0x1752, 0x1753}, {0x1772, 0x1773}, {0x17B4, 0x17D3}, {0x17DD, 0x17DD}, {0x180B, 0x180D}, {0x1885, 0x1886}, {0x18A9, 0x18A9}, {0x1920, 0x192B}, {0x1930, 0x193B}, {0x1A17, 0x1A1B}, {0x1A55, 0x1A5E}, +{0x1A60, 0x1A7C}, {0x1A7F, 0x1A7F}, {0x1AB0, 0x1AC0}, {0x1B00, 0x1B04}, {0x1B34, 0x1B44}, {0x1B6B, 0x1B73}, {0x1B80, 0x1B82}, {0x1BA1, 0x1BAD}, {0x1BE6, 0x1BF3}, {0x1C24, 0x1C37}, {0x1CD0, 0x1CD2}, +{0x1CD4, 0x1CE8}, {0x1CED, 0x1CED}, {0x1CF4, 0x1CF4}, {0x1CF7, 0x1CF9}, {0x1DC0, 0x1DF9}, {0x1DFB, 0x1DFF}, {0x20D0, 0x20F0}, {0x2CEF, 0x2CF1}, {0x2D7F, 0x2D7F}, {0x2DE0, 0x2DFF}, {0x302A, 0x302F}, +{0x3099, 0x309A}, {0xA66F, 0xA672}, {0xA674, 0xA67D}, {0xA69E, 0xA69F}, {0xA6F0, 0xA6F1}, {0xA802, 0xA802}, {0xA806, 0xA806}, {0xA80B, 0xA80B}, {0xA823, 0xA827}, {0xA82C, 0xA82C}, {0xA880, 0xA881}, +{0xA8B4, 0xA8C5}, {0xA8E0, 0xA8F1}, {0xA8FF, 0xA8FF}, {0xA926, 0xA92D}, {0xA947, 0xA953}, {0xA980, 0xA983}, {0xA9B3, 0xA9C0}, {0xA9E5, 0xA9E5}, {0xAA29, 0xAA36}, {0xAA43, 0xAA43}, {0xAA4C, 0xAA4D}, +{0xAA7B, 0xAA7D}, {0xAAB0, 0xAAB0}, {0xAAB2, 0xAAB4}, {0xAAB7, 0xAAB8}, {0xAABE, 0xAABF}, {0xAAC1, 0xAAC1}, {0xAAEB, 0xAAEF}, {0xAAF5, 0xAAF6}, {0xABE3, 0xABEA}, {0xABEC, 0xABED}, {0xFB1E, 0xFB1E}, +{0xFE00, 0xFE0F}, {0xFE20, 0xFE2F}, {0x101FD, 0x101FD}, {0x102E0, 0x102E0}, {0x10376, 0x1037A}, {0x10A01, 0x10A03}, {0x10A05, 0x10A06}, {0x10A0C, 0x10A0F}, {0x10A38, 0x10A3A}, {0x10A3F, 0x10A3F}, +{0x10AE5, 0x10AE6}, {0x10D24, 0x10D27}, {0x10EAB, 0x10EAC}, {0x10F46, 0x10F50}, {0x11000, 0x11002}, {0x11038, 0x11046}, {0x1107F, 0x11082}, {0x110B0, 0x110BA}, {0x11100, 0x11102}, {0x11127, 0x11134}, +{0x11145, 0x11146}, {0x11173, 0x11173}, {0x11180, 0x11182}, {0x111B3, 0x111C0}, {0x111C9, 0x111CC}, {0x111CE, 0x111CF}, {0x1122C, 0x11237}, {0x1123E, 0x1123E}, {0x112DF, 0x112EA}, {0x11300, 0x11303}, +{0x1133B, 0x1133C}, {0x1133E, 0x11344}, {0x11347, 0x11348}, {0x1134B, 0x1134D}, {0x11357, 0x11357}, {0x11362, 0x11363}, {0x11366, 0x1136C}, {0x11370, 0x11374}, {0x11435, 0x11446}, {0x1145E, 0x1145E}, +{0x114B0, 0x114C3}, {0x115AF, 0x115B5}, {0x115B8, 0x115C0}, {0x115DC, 0x115DD}, {0x11630, 0x11640}, {0x116AB, 0x116B7}, {0x1171D, 0x1172B}, {0x1182C, 0x1183A}, {0x11930, 0x11935}, {0x11937, 0x11938}, +{0x1193B, 0x1193E}, {0x11940, 0x11940}, {0x11942, 0x11943}, {0x119D1, 0x119D7}, {0x119DA, 0x119E0}, {0x119E4, 0x119E4}, {0x11A01, 0x11A0A}, {0x11A33, 0x11A39}, {0x11A3B, 0x11A3E}, {0x11A47, 0x11A47}, +{0x11A51, 0x11A5B}, {0x11A8A, 0x11A99}, {0x11C2F, 0x11C36}, {0x11C38, 0x11C3F}, {0x11C92, 0x11CA7}, {0x11CA9, 0x11CB6}, {0x11D31, 0x11D36}, {0x11D3A, 0x11D3A}, {0x11D3C, 0x11D3D}, {0x11D3F, 0x11D45}, +{0x11D47, 0x11D47}, {0x11D8A, 0x11D8E}, {0x11D90, 0x11D91}, {0x11D93, 0x11D97}, {0x11EF3, 0x11EF6}, {0x16AF0, 0x16AF4}, {0x16B30, 0x16B36}, {0x16F4F, 0x16F4F}, {0x16F51, 0x16F87}, {0x16F8F, 0x16F92}, +{0x16FE4, 0x16FE4}, {0x16FF0, 0x16FF1}, {0x1BC9D, 0x1BC9E}, {0x1D165, 0x1D169}, {0x1D16D, 0x1D172}, {0x1D17B, 0x1D182}, {0x1D185, 0x1D18B}, {0x1D1AA, 0x1D1AD}, {0x1D242, 0x1D244}, {0x1DA00, 0x1DA36}, +{0x1DA3B, 0x1DA6C}, {0x1DA75, 0x1DA75}, {0x1DA84, 0x1DA84}, {0x1DA9B, 0x1DA9F}, {0x1DAA1, 0x1DAAF}, {0x1E000, 0x1E006}, {0x1E008, 0x1E018}, {0x1E01B, 0x1E021}, {0x1E023, 0x1E024}, {0x1E026, 0x1E02A}, +{0x1E130, 0x1E136}, {0x1E2EC, 0x1E2EF}, {0x1E8D0, 0x1E8D6}, {0x1E944, 0x1E94A}, {0xE0100, 0xE01EF}, +}; + +static const std::vector<std::pair<int, int>> punctuation_ranges = { +{0x21, 0x23}, {0x25, 0x2A}, {0x2C, 0x2F}, {0x3A, 0x3B}, {0x3F, 0x40}, {0x5B, 0x5D}, {0x5F, 0x5F}, {0x7B, 0x7B}, {0x7D, 0x7D}, {0xA1, 0xA1}, {0xA7, 0xA7}, {0xAB, 0xAB}, {0xB6, 0xB7}, {0xBB, 0xBB}, +{0xBF, 0xBF}, {0x37E, 0x37E}, {0x387, 0x387}, {0x55A, 0x55F}, {0x589, 0x58A}, {0x5BE, 0x5BE}, {0x5C0, 0x5C0}, {0x5C3, 0x5C3}, {0x5C6, 0x5C6}, {0x5F3, 0x5F4}, {0x609, 0x60A}, {0x60C, 0x60D}, +{0x61B, 0x61B}, {0x61E, 0x61F}, {0x66A, 0x66D}, {0x6D4, 0x6D4}, {0x700, 0x70D}, {0x7F7, 0x7F9}, {0x830, 0x83E}, {0x85E, 0x85E}, {0x964, 0x965}, {0x970, 0x970}, {0x9FD, 0x9FD}, {0xA76, 0xA76}, +{0xAF0, 0xAF0}, {0xC77, 0xC77}, {0xC84, 0xC84}, {0xDF4, 0xDF4}, {0xE4F, 0xE4F}, {0xE5A, 0xE5B}, {0xF04, 0xF12}, {0xF14, 0xF14}, {0xF3A, 0xF3D}, {0xF85, 0xF85}, {0xFD0, 0xFD4}, {0xFD9, 0xFDA}, +{0x104A, 0x104F}, {0x10FB, 0x10FB}, {0x1360, 0x1368}, {0x1400, 0x1400}, {0x166E, 0x166E}, {0x169B, 0x169C}, {0x16EB, 0x16ED}, {0x1735, 0x1736}, {0x17D4, 0x17D6}, {0x17D8, 0x17DA}, {0x1800, 0x180A}, +{0x1944, 0x1945}, {0x1A1E, 0x1A1F}, {0x1AA0, 0x1AA6}, {0x1AA8, 0x1AAD}, {0x1B5A, 0x1B60}, {0x1BFC, 0x1BFF}, {0x1C3B, 0x1C3F}, {0x1C7E, 0x1C7F}, {0x1CC0, 0x1CC7}, {0x1CD3, 0x1CD3}, {0x2010, 0x2027}, +{0x2030, 0x2043}, {0x2045, 0x2051}, {0x2053, 0x205E}, {0x207D, 0x207E}, {0x208D, 0x208E}, {0x2308, 0x230B}, {0x2329, 0x232A}, {0x2768, 0x2775}, {0x27C5, 0x27C6}, {0x27E6, 0x27EF}, {0x2983, 0x2998}, +{0x29D8, 0x29DB}, {0x29FC, 0x29FD}, {0x2CF9, 0x2CFC}, {0x2CFE, 0x2CFF}, {0x2D70, 0x2D70}, {0x2E00, 0x2E2E}, {0x2E30, 0x2E4F}, {0x2E52, 0x2E52}, {0x3001, 0x3003}, {0x3008, 0x3011}, {0x3014, 0x301F}, +{0x3030, 0x3030}, {0x303D, 0x303D}, {0x30A0, 0x30A0}, {0x30FB, 0x30FB}, {0xA4FE, 0xA4FF}, {0xA60D, 0xA60F}, {0xA673, 0xA673}, {0xA67E, 0xA67E}, {0xA6F2, 0xA6F7}, {0xA874, 0xA877}, {0xA8CE, 0xA8CF}, +{0xA8F8, 0xA8FA}, {0xA8FC, 0xA8FC}, {0xA92E, 0xA92F}, {0xA95F, 0xA95F}, {0xA9C1, 0xA9CD}, {0xA9DE, 0xA9DF}, {0xAA5C, 0xAA5F}, {0xAADE, 0xAADF}, {0xAAF0, 0xAAF1}, {0xABEB, 0xABEB}, {0xFD3E, 0xFD3F}, +{0xFE10, 0xFE19}, {0xFE30, 0xFE52}, {0xFE54, 0xFE61}, {0xFE63, 0xFE63}, {0xFE68, 0xFE68}, {0xFE6A, 0xFE6B}, {0xFF01, 0xFF03}, {0xFF05, 0xFF0A}, {0xFF0C, 0xFF0F}, {0xFF1A, 0xFF1B}, {0xFF1F, 0xFF20}, +{0xFF3B, 0xFF3D}, {0xFF3F, 0xFF3F}, {0xFF5B, 0xFF5B}, {0xFF5D, 0xFF5D}, {0xFF5F, 0xFF65}, {0x10100, 0x10102}, {0x1039F, 0x1039F}, {0x103D0, 0x103D0}, {0x1056F, 0x1056F}, {0x10857, 0x10857}, +{0x1091F, 0x1091F}, {0x1093F, 0x1093F}, {0x10A50, 0x10A58}, {0x10A7F, 0x10A7F}, {0x10AF0, 0x10AF6}, {0x10B39, 0x10B3F}, {0x10B99, 0x10B9C}, {0x10EAD, 0x10EAD}, {0x10F55, 0x10F59}, {0x11047, 0x1104D}, +{0x110BB, 0x110BC}, {0x110BE, 0x110C1}, {0x11140, 0x11143}, {0x11174, 0x11175}, {0x111C5, 0x111C8}, {0x111CD, 0x111CD}, {0x111DB, 0x111DB}, {0x111DD, 0x111DF}, {0x11238, 0x1123D}, {0x112A9, 0x112A9}, +{0x1144B, 0x1144F}, {0x1145A, 0x1145B}, {0x1145D, 0x1145D}, {0x114C6, 0x114C6}, {0x115C1, 0x115D7}, {0x11641, 0x11643}, {0x11660, 0x1166C}, {0x1173C, 0x1173E}, {0x1183B, 0x1183B}, {0x11944, 0x11946}, +{0x119E2, 0x119E2}, {0x11A3F, 0x11A46}, {0x11A9A, 0x11A9C}, {0x11A9E, 0x11AA2}, {0x11C41, 0x11C45}, {0x11C70, 0x11C71}, {0x11EF7, 0x11EF8}, {0x11FFF, 0x11FFF}, {0x12470, 0x12474}, {0x16A6E, 0x16A6F}, +{0x16AF5, 0x16AF5}, {0x16B37, 0x16B3B}, {0x16B44, 0x16B44}, {0x16E97, 0x16E9A}, {0x16FE2, 0x16FE2}, {0x1BC9F, 0x1BC9F}, {0x1DA87, 0x1DA8B}, {0x1E95E, 0x1E95F}, +}; + +static const std::vector<std::pair<int, int>> symbol_ranges = { +{0x24, 0x24}, {0x2B, 0x2B}, {0x3C, 0x3E}, {0x5E, 0x5E}, {0x60, 0x60}, {0x7C, 0x7C}, {0x7E, 0x7E}, {0xA2, 0xA6}, {0xA8, 0xA9}, {0xAC, 0xAC}, {0xAE, 0xB1}, {0xB4, 0xB4}, {0xB8, 0xB8}, {0xD7, 0xD7}, +{0xF7, 0xF7}, {0x2C2, 0x2C5}, {0x2D2, 0x2DF}, {0x2E5, 0x2EB}, {0x2ED, 0x2ED}, {0x2EF, 0x2FF}, {0x375, 0x375}, {0x384, 0x385}, {0x3F6, 0x3F6}, {0x482, 0x482}, {0x58D, 0x58F}, {0x606, 0x608}, +{0x60B, 0x60B}, {0x60E, 0x60F}, {0x6DE, 0x6DE}, {0x6E9, 0x6E9}, {0x6FD, 0x6FE}, {0x7F6, 0x7F6}, {0x7FE, 0x7FF}, {0x9F2, 0x9F3}, {0x9FA, 0x9FB}, {0xAF1, 0xAF1}, {0xB70, 0xB70}, {0xBF3, 0xBFA}, +{0xC7F, 0xC7F}, {0xD4F, 0xD4F}, {0xD79, 0xD79}, {0xE3F, 0xE3F}, {0xF01, 0xF03}, {0xF13, 0xF13}, {0xF15, 0xF17}, {0xF1A, 0xF1F}, {0xF34, 0xF34}, {0xF36, 0xF36}, {0xF38, 0xF38}, {0xFBE, 0xFC5}, +{0xFC7, 0xFCC}, {0xFCE, 0xFCF}, {0xFD5, 0xFD8}, {0x109E, 0x109F}, {0x1390, 0x1399}, {0x166D, 0x166D}, {0x17DB, 0x17DB}, {0x1940, 0x1940}, {0x19DE, 0x19FF}, {0x1B61, 0x1B6A}, {0x1B74, 0x1B7C}, +{0x1FBD, 0x1FBD}, {0x1FBF, 0x1FC1}, {0x1FCD, 0x1FCF}, {0x1FDD, 0x1FDF}, {0x1FED, 0x1FEF}, {0x1FFD, 0x1FFE}, {0x2044, 0x2044}, {0x2052, 0x2052}, {0x207A, 0x207C}, {0x208A, 0x208C}, {0x20A0, 0x20BF}, +{0x2100, 0x2101}, {0x2103, 0x2106}, {0x2108, 0x2109}, {0x2114, 0x2114}, {0x2116, 0x2118}, {0x211E, 0x2123}, {0x2125, 0x2125}, {0x2127, 0x2127}, {0x2129, 0x2129}, {0x212E, 0x212E}, {0x213A, 0x213B}, +{0x2140, 0x2144}, {0x214A, 0x214D}, {0x214F, 0x214F}, {0x218A, 0x218B}, {0x2190, 0x2307}, {0x230C, 0x2328}, {0x232B, 0x2426}, {0x2440, 0x244A}, {0x249C, 0x24E9}, {0x2500, 0x2767}, {0x2794, 0x27C4}, +{0x27C7, 0x27E5}, {0x27F0, 0x2982}, {0x2999, 0x29D7}, {0x29DC, 0x29FB}, {0x29FE, 0x2B73}, {0x2B76, 0x2B95}, {0x2B97, 0x2BFF}, {0x2CE5, 0x2CEA}, {0x2E50, 0x2E51}, {0x2E80, 0x2E99}, {0x2E9B, 0x2EF3}, +{0x2F00, 0x2FD5}, {0x2FF0, 0x2FFB}, {0x3004, 0x3004}, {0x3012, 0x3013}, {0x3020, 0x3020}, {0x3036, 0x3037}, {0x303E, 0x303F}, {0x309B, 0x309C}, {0x3190, 0x3191}, {0x3196, 0x319F}, {0x31C0, 0x31E3}, +{0x3200, 0x321E}, {0x322A, 0x3247}, {0x3250, 0x3250}, {0x3260, 0x327F}, {0x328A, 0x32B0}, {0x32C0, 0x33FF}, {0x4DC0, 0x4DFF}, {0xA490, 0xA4C6}, {0xA700, 0xA716}, {0xA720, 0xA721}, {0xA789, 0xA78A}, +{0xA828, 0xA82B}, {0xA836, 0xA839}, {0xAA77, 0xAA79}, {0xAB5B, 0xAB5B}, {0xAB6A, 0xAB6B}, {0xFB29, 0xFB29}, {0xFBB2, 0xFBC1}, {0xFDFC, 0xFDFD}, {0xFE62, 0xFE62}, {0xFE64, 0xFE66}, {0xFE69, 0xFE69}, +{0xFF04, 0xFF04}, {0xFF0B, 0xFF0B}, {0xFF1C, 0xFF1E}, {0xFF3E, 0xFF3E}, {0xFF40, 0xFF40}, {0xFF5C, 0xFF5C}, {0xFF5E, 0xFF5E}, {0xFFE0, 0xFFE6}, {0xFFE8, 0xFFEE}, {0xFFFC, 0xFFFD}, {0x10137, 0x1013F}, +{0x10179, 0x10189}, {0x1018C, 0x1018E}, {0x10190, 0x1019C}, {0x101A0, 0x101A0}, {0x101D0, 0x101FC}, {0x10877, 0x10878}, {0x10AC8, 0x10AC8}, {0x1173F, 0x1173F}, {0x11FD5, 0x11FF1}, {0x16B3C, 0x16B3F}, +{0x16B45, 0x16B45}, {0x1BC9C, 0x1BC9C}, {0x1D000, 0x1D0F5}, {0x1D100, 0x1D126}, {0x1D129, 0x1D164}, {0x1D16A, 0x1D16C}, {0x1D183, 0x1D184}, {0x1D18C, 0x1D1A9}, {0x1D1AE, 0x1D1E8}, {0x1D200, 0x1D241}, +{0x1D245, 0x1D245}, {0x1D300, 0x1D356}, {0x1D6C1, 0x1D6C1}, {0x1D6DB, 0x1D6DB}, {0x1D6FB, 0x1D6FB}, {0x1D715, 0x1D715}, {0x1D735, 0x1D735}, {0x1D74F, 0x1D74F}, {0x1D76F, 0x1D76F}, {0x1D789, 0x1D789}, +{0x1D7A9, 0x1D7A9}, {0x1D7C3, 0x1D7C3}, {0x1D800, 0x1D9FF}, {0x1DA37, 0x1DA3A}, {0x1DA6D, 0x1DA74}, {0x1DA76, 0x1DA83}, {0x1DA85, 0x1DA86}, {0x1E14F, 0x1E14F}, {0x1E2FF, 0x1E2FF}, {0x1ECAC, 0x1ECAC}, +{0x1ECB0, 0x1ECB0}, {0x1ED2E, 0x1ED2E}, {0x1EEF0, 0x1EEF1}, {0x1F000, 0x1F02B}, {0x1F030, 0x1F093}, {0x1F0A0, 0x1F0AE}, {0x1F0B1, 0x1F0BF}, {0x1F0C1, 0x1F0CF}, {0x1F0D1, 0x1F0F5}, {0x1F10D, 0x1F1AD}, +{0x1F1E6, 0x1F202}, {0x1F210, 0x1F23B}, {0x1F240, 0x1F248}, {0x1F250, 0x1F251}, {0x1F260, 0x1F265}, {0x1F300, 0x1F6D7}, {0x1F6E0, 0x1F6EC}, {0x1F6F0, 0x1F6FC}, {0x1F700, 0x1F773}, {0x1F780, 0x1F7D8}, +{0x1F7E0, 0x1F7EB}, {0x1F800, 0x1F80B}, {0x1F810, 0x1F847}, {0x1F850, 0x1F859}, {0x1F860, 0x1F887}, {0x1F890, 0x1F8AD}, {0x1F8B0, 0x1F8B1}, {0x1F900, 0x1F978}, {0x1F97A, 0x1F9CB}, {0x1F9CD, 0x1FA53}, +{0x1FA60, 0x1FA6D}, {0x1FA70, 0x1FA74}, {0x1FA78, 0x1FA7A}, {0x1FA80, 0x1FA86}, {0x1FA90, 0x1FAA8}, {0x1FAB0, 0x1FAB6}, {0x1FAC0, 0x1FAC2}, {0x1FAD0, 0x1FAD6}, {0x1FB00, 0x1FB92}, {0x1FB94, 0x1FBCA}, +}; + +static const std::vector<std::pair<int, int>> control_ranges = { +{0x0, 0x8}, {0xE, 0x1B}, {0x7F, 0x84}, {0x86, 0x9F}, {0xAD, 0xAD}, {0x378, 0x379}, {0x380, 0x383}, {0x38B, 0x38B}, {0x38D, 0x38D}, {0x3A2, 0x3A2}, {0x530, 0x530}, {0x557, 0x558}, {0x58B, 0x58C}, +{0x590, 0x590}, {0x5C8, 0x5CF}, {0x5EB, 0x5EE}, {0x5F5, 0x605}, {0x61C, 0x61D}, {0x6DD, 0x6DD}, {0x70E, 0x70F}, {0x74B, 0x74C}, {0x7B2, 0x7BF}, {0x7FB, 0x7FC}, {0x82E, 0x82F}, {0x83F, 0x83F}, +{0x85C, 0x85D}, {0x85F, 0x85F}, {0x86B, 0x89F}, {0x8B5, 0x8B5}, {0x8C8, 0x8D2}, {0x8E2, 0x8E2}, {0x984, 0x984}, {0x98D, 0x98E}, {0x991, 0x992}, {0x9A9, 0x9A9}, {0x9B1, 0x9B1}, {0x9B3, 0x9B5}, +{0x9BA, 0x9BB}, {0x9C5, 0x9C6}, {0x9C9, 0x9CA}, {0x9CF, 0x9D6}, {0x9D8, 0x9DB}, {0x9DE, 0x9DE}, {0x9E4, 0x9E5}, {0x9FF, 0xA00}, {0xA04, 0xA04}, {0xA0B, 0xA0E}, {0xA11, 0xA12}, {0xA29, 0xA29}, +{0xA31, 0xA31}, {0xA34, 0xA34}, {0xA37, 0xA37}, {0xA3A, 0xA3B}, {0xA3D, 0xA3D}, {0xA43, 0xA46}, {0xA49, 0xA4A}, {0xA4E, 0xA50}, {0xA52, 0xA58}, {0xA5D, 0xA5D}, {0xA5F, 0xA65}, {0xA77, 0xA80}, +{0xA84, 0xA84}, {0xA8E, 0xA8E}, {0xA92, 0xA92}, {0xAA9, 0xAA9}, {0xAB1, 0xAB1}, {0xAB4, 0xAB4}, {0xABA, 0xABB}, {0xAC6, 0xAC6}, {0xACA, 0xACA}, {0xACE, 0xACF}, {0xAD1, 0xADF}, {0xAE4, 0xAE5}, +{0xAF2, 0xAF8}, {0xB00, 0xB00}, {0xB04, 0xB04}, {0xB0D, 0xB0E}, {0xB11, 0xB12}, {0xB29, 0xB29}, {0xB31, 0xB31}, {0xB34, 0xB34}, {0xB3A, 0xB3B}, {0xB45, 0xB46}, {0xB49, 0xB4A}, {0xB4E, 0xB54}, +{0xB58, 0xB5B}, {0xB5E, 0xB5E}, {0xB64, 0xB65}, {0xB78, 0xB81}, {0xB84, 0xB84}, {0xB8B, 0xB8D}, {0xB91, 0xB91}, {0xB96, 0xB98}, {0xB9B, 0xB9B}, {0xB9D, 0xB9D}, {0xBA0, 0xBA2}, {0xBA5, 0xBA7}, +{0xBAB, 0xBAD}, {0xBBA, 0xBBD}, {0xBC3, 0xBC5}, {0xBC9, 0xBC9}, {0xBCE, 0xBCF}, {0xBD1, 0xBD6}, {0xBD8, 0xBE5}, {0xBFB, 0xBFF}, {0xC0D, 0xC0D}, {0xC11, 0xC11}, {0xC29, 0xC29}, {0xC3A, 0xC3C}, +{0xC45, 0xC45}, {0xC49, 0xC49}, {0xC4E, 0xC54}, {0xC57, 0xC57}, {0xC5B, 0xC5F}, {0xC64, 0xC65}, {0xC70, 0xC76}, {0xC8D, 0xC8D}, {0xC91, 0xC91}, {0xCA9, 0xCA9}, {0xCB4, 0xCB4}, {0xCBA, 0xCBB}, +{0xCC5, 0xCC5}, {0xCC9, 0xCC9}, {0xCCE, 0xCD4}, {0xCD7, 0xCDD}, {0xCDF, 0xCDF}, {0xCE4, 0xCE5}, {0xCF0, 0xCF0}, {0xCF3, 0xCFF}, {0xD0D, 0xD0D}, {0xD11, 0xD11}, {0xD45, 0xD45}, {0xD49, 0xD49}, +{0xD50, 0xD53}, {0xD64, 0xD65}, {0xD80, 0xD80}, {0xD84, 0xD84}, {0xD97, 0xD99}, {0xDB2, 0xDB2}, {0xDBC, 0xDBC}, {0xDBE, 0xDBF}, {0xDC7, 0xDC9}, {0xDCB, 0xDCE}, {0xDD5, 0xDD5}, {0xDD7, 0xDD7}, +{0xDE0, 0xDE5}, {0xDF0, 0xDF1}, {0xDF5, 0xE00}, {0xE3B, 0xE3E}, {0xE5C, 0xE80}, {0xE83, 0xE83}, {0xE85, 0xE85}, {0xE8B, 0xE8B}, {0xEA4, 0xEA4}, {0xEA6, 0xEA6}, {0xEBE, 0xEBF}, {0xEC5, 0xEC5}, +{0xEC7, 0xEC7}, {0xECE, 0xECF}, {0xEDA, 0xEDB}, {0xEE0, 0xEFF}, {0xF48, 0xF48}, {0xF6D, 0xF70}, {0xF98, 0xF98}, {0xFBD, 0xFBD}, {0xFCD, 0xFCD}, {0xFDB, 0xFFF}, {0x10C6, 0x10C6}, {0x10C8, 0x10CC}, +{0x10CE, 0x10CF}, {0x1249, 0x1249}, {0x124E, 0x124F}, {0x1257, 0x1257}, {0x1259, 0x1259}, {0x125E, 0x125F}, {0x1289, 0x1289}, {0x128E, 0x128F}, {0x12B1, 0x12B1}, {0x12B6, 0x12B7}, {0x12BF, 0x12BF}, +{0x12C1, 0x12C1}, {0x12C6, 0x12C7}, {0x12D7, 0x12D7}, {0x1311, 0x1311}, {0x1316, 0x1317}, {0x135B, 0x135C}, {0x137D, 0x137F}, {0x139A, 0x139F}, {0x13F6, 0x13F7}, {0x13FE, 0x13FF}, {0x169D, 0x169F}, +{0x16F9, 0x16FF}, {0x170D, 0x170D}, {0x1715, 0x171F}, {0x1737, 0x173F}, {0x1754, 0x175F}, {0x176D, 0x176D}, {0x1771, 0x1771}, {0x1774, 0x177F}, {0x17DE, 0x17DF}, {0x17EA, 0x17EF}, {0x17FA, 0x17FF}, +{0x180E, 0x180F}, {0x181A, 0x181F}, {0x1879, 0x187F}, {0x18AB, 0x18AF}, {0x18F6, 0x18FF}, {0x191F, 0x191F}, {0x192C, 0x192F}, {0x193C, 0x193F}, {0x1941, 0x1943}, {0x196E, 0x196F}, {0x1975, 0x197F}, +{0x19AC, 0x19AF}, {0x19CA, 0x19CF}, {0x19DB, 0x19DD}, {0x1A1C, 0x1A1D}, {0x1A5F, 0x1A5F}, {0x1A7D, 0x1A7E}, {0x1A8A, 0x1A8F}, {0x1A9A, 0x1A9F}, {0x1AAE, 0x1AAF}, {0x1AC1, 0x1AFF}, {0x1B4C, 0x1B4F}, +{0x1B7D, 0x1B7F}, {0x1BF4, 0x1BFB}, {0x1C38, 0x1C3A}, {0x1C4A, 0x1C4C}, {0x1C89, 0x1C8F}, {0x1CBB, 0x1CBC}, {0x1CC8, 0x1CCF}, {0x1CFB, 0x1CFF}, {0x1DFA, 0x1DFA}, {0x1F16, 0x1F17}, {0x1F1E, 0x1F1F}, +{0x1F46, 0x1F47}, {0x1F4E, 0x1F4F}, {0x1F58, 0x1F58}, {0x1F5A, 0x1F5A}, {0x1F5C, 0x1F5C}, {0x1F5E, 0x1F5E}, {0x1F7E, 0x1F7F}, {0x1FB5, 0x1FB5}, {0x1FC5, 0x1FC5}, {0x1FD4, 0x1FD5}, {0x1FDC, 0x1FDC}, +{0x1FF0, 0x1FF1}, {0x1FF5, 0x1FF5}, {0x1FFF, 0x1FFF}, {0x200B, 0x200F}, {0x202A, 0x202E}, {0x2060, 0x206F}, {0x2072, 0x2073}, {0x208F, 0x208F}, {0x209D, 0x209F}, {0x20C0, 0x20CF}, {0x20F1, 0x20FF}, +{0x218C, 0x218F}, {0x2427, 0x243F}, {0x244B, 0x245F}, {0x2B74, 0x2B75}, {0x2B96, 0x2B96}, {0x2C2F, 0x2C2F}, {0x2C5F, 0x2C5F}, {0x2CF4, 0x2CF8}, {0x2D26, 0x2D26}, {0x2D28, 0x2D2C}, {0x2D2E, 0x2D2F}, +{0x2D68, 0x2D6E}, {0x2D71, 0x2D7E}, {0x2D97, 0x2D9F}, {0x2DA7, 0x2DA7}, {0x2DAF, 0x2DAF}, {0x2DB7, 0x2DB7}, {0x2DBF, 0x2DBF}, {0x2DC7, 0x2DC7}, {0x2DCF, 0x2DCF}, {0x2DD7, 0x2DD7}, {0x2DDF, 0x2DDF}, +{0x2E53, 0x2E7F}, {0x2E9A, 0x2E9A}, {0x2EF4, 0x2EFF}, {0x2FD6, 0x2FEF}, {0x2FFC, 0x2FFF}, {0x3040, 0x3040}, {0x3097, 0x3098}, {0x3100, 0x3104}, {0x3130, 0x3130}, {0x318F, 0x318F}, {0x31E4, 0x31EF}, +{0x321F, 0x321F}, {0x9FFD, 0x9FFF}, {0xA48D, 0xA48F}, {0xA4C7, 0xA4CF}, {0xA62C, 0xA63F}, {0xA6F8, 0xA6FF}, {0xA7C0, 0xA7C1}, {0xA7CB, 0xA7F4}, {0xA82D, 0xA82F}, {0xA83A, 0xA83F}, {0xA878, 0xA87F}, +{0xA8C6, 0xA8CD}, {0xA8DA, 0xA8DF}, {0xA954, 0xA95E}, {0xA97D, 0xA97F}, {0xA9CE, 0xA9CE}, {0xA9DA, 0xA9DD}, {0xA9FF, 0xA9FF}, {0xAA37, 0xAA3F}, {0xAA4E, 0xAA4F}, {0xAA5A, 0xAA5B}, {0xAAC3, 0xAADA}, +{0xAAF7, 0xAB00}, {0xAB07, 0xAB08}, {0xAB0F, 0xAB10}, {0xAB17, 0xAB1F}, {0xAB27, 0xAB27}, {0xAB2F, 0xAB2F}, {0xAB6C, 0xAB6F}, {0xABEE, 0xABEF}, {0xABFA, 0xABFF}, {0xD7A4, 0xD7AF}, {0xD7C7, 0xD7CA}, +{0xD7FC, 0xF8FF}, {0xFA6E, 0xFA6F}, {0xFADA, 0xFAFF}, {0xFB07, 0xFB12}, {0xFB18, 0xFB1C}, {0xFB37, 0xFB37}, {0xFB3D, 0xFB3D}, {0xFB3F, 0xFB3F}, {0xFB42, 0xFB42}, {0xFB45, 0xFB45}, {0xFBC2, 0xFBD2}, +{0xFD40, 0xFD4F}, {0xFD90, 0xFD91}, {0xFDC8, 0xFDEF}, {0xFDFE, 0xFDFF}, {0xFE1A, 0xFE1F}, {0xFE53, 0xFE53}, {0xFE67, 0xFE67}, {0xFE6C, 0xFE6F}, {0xFE75, 0xFE75}, {0xFEFD, 0xFF00}, {0xFFBF, 0xFFC1}, +{0xFFC8, 0xFFC9}, {0xFFD0, 0xFFD1}, {0xFFD8, 0xFFD9}, {0xFFDD, 0xFFDF}, {0xFFE7, 0xFFE7}, {0xFFEF, 0xFFFB}, {0xFFFE, 0xFFFF}, {0x1000C, 0x1000C}, {0x10027, 0x10027}, {0x1003B, 0x1003B}, +{0x1003E, 0x1003E}, {0x1004E, 0x1004F}, {0x1005E, 0x1007F}, {0x100FB, 0x100FF}, {0x10103, 0x10106}, {0x10134, 0x10136}, {0x1018F, 0x1018F}, {0x1019D, 0x1019F}, {0x101A1, 0x101CF}, {0x101FE, 0x1027F}, +{0x1029D, 0x1029F}, {0x102D1, 0x102DF}, {0x102FC, 0x102FF}, {0x10324, 0x1032C}, {0x1034B, 0x1034F}, {0x1037B, 0x1037F}, {0x1039E, 0x1039E}, {0x103C4, 0x103C7}, {0x103D6, 0x103FF}, {0x1049E, 0x1049F}, +{0x104AA, 0x104AF}, {0x104D4, 0x104D7}, {0x104FC, 0x104FF}, {0x10528, 0x1052F}, {0x10564, 0x1056E}, {0x10570, 0x105FF}, {0x10737, 0x1073F}, {0x10756, 0x1075F}, {0x10768, 0x107FF}, {0x10806, 0x10807}, +{0x10809, 0x10809}, {0x10836, 0x10836}, {0x10839, 0x1083B}, {0x1083D, 0x1083E}, {0x10856, 0x10856}, {0x1089F, 0x108A6}, {0x108B0, 0x108DF}, {0x108F3, 0x108F3}, {0x108F6, 0x108FA}, {0x1091C, 0x1091E}, +{0x1093A, 0x1093E}, {0x10940, 0x1097F}, {0x109B8, 0x109BB}, {0x109D0, 0x109D1}, {0x10A04, 0x10A04}, {0x10A07, 0x10A0B}, {0x10A14, 0x10A14}, {0x10A18, 0x10A18}, {0x10A36, 0x10A37}, {0x10A3B, 0x10A3E}, +{0x10A49, 0x10A4F}, {0x10A59, 0x10A5F}, {0x10AA0, 0x10ABF}, {0x10AE7, 0x10AEA}, {0x10AF7, 0x10AFF}, {0x10B36, 0x10B38}, {0x10B56, 0x10B57}, {0x10B73, 0x10B77}, {0x10B92, 0x10B98}, {0x10B9D, 0x10BA8}, +{0x10BB0, 0x10BFF}, {0x10C49, 0x10C7F}, {0x10CB3, 0x10CBF}, {0x10CF3, 0x10CF9}, {0x10D28, 0x10D2F}, {0x10D3A, 0x10E5F}, {0x10E7F, 0x10E7F}, {0x10EAA, 0x10EAA}, {0x10EAE, 0x10EAF}, {0x10EB2, 0x10EFF}, +{0x10F28, 0x10F2F}, {0x10F5A, 0x10FAF}, {0x10FCC, 0x10FDF}, {0x10FF7, 0x10FFF}, {0x1104E, 0x11051}, {0x11070, 0x1107E}, {0x110BD, 0x110BD}, {0x110C2, 0x110CF}, {0x110E9, 0x110EF}, {0x110FA, 0x110FF}, +{0x11135, 0x11135}, {0x11148, 0x1114F}, {0x11177, 0x1117F}, {0x111E0, 0x111E0}, {0x111F5, 0x111FF}, {0x11212, 0x11212}, {0x1123F, 0x1127F}, {0x11287, 0x11287}, {0x11289, 0x11289}, {0x1128E, 0x1128E}, +{0x1129E, 0x1129E}, {0x112AA, 0x112AF}, {0x112EB, 0x112EF}, {0x112FA, 0x112FF}, {0x11304, 0x11304}, {0x1130D, 0x1130E}, {0x11311, 0x11312}, {0x11329, 0x11329}, {0x11331, 0x11331}, {0x11334, 0x11334}, +{0x1133A, 0x1133A}, {0x11345, 0x11346}, {0x11349, 0x1134A}, {0x1134E, 0x1134F}, {0x11351, 0x11356}, {0x11358, 0x1135C}, {0x11364, 0x11365}, {0x1136D, 0x1136F}, {0x11375, 0x113FF}, {0x1145C, 0x1145C}, +{0x11462, 0x1147F}, {0x114C8, 0x114CF}, {0x114DA, 0x1157F}, {0x115B6, 0x115B7}, {0x115DE, 0x115FF}, {0x11645, 0x1164F}, {0x1165A, 0x1165F}, {0x1166D, 0x1167F}, {0x116B9, 0x116BF}, {0x116CA, 0x116FF}, +{0x1171B, 0x1171C}, {0x1172C, 0x1172F}, {0x11740, 0x117FF}, {0x1183C, 0x1189F}, {0x118F3, 0x118FE}, {0x11907, 0x11908}, {0x1190A, 0x1190B}, {0x11914, 0x11914}, {0x11917, 0x11917}, {0x11936, 0x11936}, +{0x11939, 0x1193A}, {0x11947, 0x1194F}, {0x1195A, 0x1199F}, {0x119A8, 0x119A9}, {0x119D8, 0x119D9}, {0x119E5, 0x119FF}, {0x11A48, 0x11A4F}, {0x11AA3, 0x11ABF}, {0x11AF9, 0x11BFF}, {0x11C09, 0x11C09}, +{0x11C37, 0x11C37}, {0x11C46, 0x11C4F}, {0x11C6D, 0x11C6F}, {0x11C90, 0x11C91}, {0x11CA8, 0x11CA8}, {0x11CB7, 0x11CFF}, {0x11D07, 0x11D07}, {0x11D0A, 0x11D0A}, {0x11D37, 0x11D39}, {0x11D3B, 0x11D3B}, +{0x11D3E, 0x11D3E}, {0x11D48, 0x11D4F}, {0x11D5A, 0x11D5F}, {0x11D66, 0x11D66}, {0x11D69, 0x11D69}, {0x11D8F, 0x11D8F}, {0x11D92, 0x11D92}, {0x11D99, 0x11D9F}, {0x11DAA, 0x11EDF}, {0x11EF9, 0x11FAF}, +{0x11FB1, 0x11FBF}, {0x11FF2, 0x11FFE}, {0x1239A, 0x123FF}, {0x1246F, 0x1246F}, {0x12475, 0x1247F}, {0x12544, 0x12FFF}, {0x1342F, 0x143FF}, {0x14647, 0x167FF}, {0x16A39, 0x16A3F}, {0x16A5F, 0x16A5F}, +{0x16A6A, 0x16A6D}, {0x16A70, 0x16ACF}, {0x16AEE, 0x16AEF}, {0x16AF6, 0x16AFF}, {0x16B46, 0x16B4F}, {0x16B5A, 0x16B5A}, {0x16B62, 0x16B62}, {0x16B78, 0x16B7C}, {0x16B90, 0x16E3F}, {0x16E9B, 0x16EFF}, +{0x16F4B, 0x16F4E}, {0x16F88, 0x16F8E}, {0x16FA0, 0x16FDF}, {0x16FE5, 0x16FEF}, {0x16FF2, 0x16FFF}, {0x187F8, 0x187FF}, {0x18CD6, 0x18CFF}, {0x18D09, 0x1AFFF}, {0x1B11F, 0x1B14F}, {0x1B153, 0x1B163}, +{0x1B168, 0x1B16F}, {0x1B2FC, 0x1BBFF}, {0x1BC6B, 0x1BC6F}, {0x1BC7D, 0x1BC7F}, {0x1BC89, 0x1BC8F}, {0x1BC9A, 0x1BC9B}, {0x1BCA0, 0x1CFFF}, {0x1D0F6, 0x1D0FF}, {0x1D127, 0x1D128}, {0x1D173, 0x1D17A}, +{0x1D1E9, 0x1D1FF}, {0x1D246, 0x1D2DF}, {0x1D2F4, 0x1D2FF}, {0x1D357, 0x1D35F}, {0x1D379, 0x1D3FF}, {0x1D455, 0x1D455}, {0x1D49D, 0x1D49D}, {0x1D4A0, 0x1D4A1}, {0x1D4A3, 0x1D4A4}, {0x1D4A7, 0x1D4A8}, +{0x1D4AD, 0x1D4AD}, {0x1D4BA, 0x1D4BA}, {0x1D4BC, 0x1D4BC}, {0x1D4C4, 0x1D4C4}, {0x1D506, 0x1D506}, {0x1D50B, 0x1D50C}, {0x1D515, 0x1D515}, {0x1D51D, 0x1D51D}, {0x1D53A, 0x1D53A}, {0x1D53F, 0x1D53F}, +{0x1D545, 0x1D545}, {0x1D547, 0x1D549}, {0x1D551, 0x1D551}, {0x1D6A6, 0x1D6A7}, {0x1D7CC, 0x1D7CD}, {0x1DA8C, 0x1DA9A}, {0x1DAA0, 0x1DAA0}, {0x1DAB0, 0x1DFFF}, {0x1E007, 0x1E007}, {0x1E019, 0x1E01A}, +{0x1E022, 0x1E022}, {0x1E025, 0x1E025}, {0x1E02B, 0x1E0FF}, {0x1E12D, 0x1E12F}, {0x1E13E, 0x1E13F}, {0x1E14A, 0x1E14D}, {0x1E150, 0x1E2BF}, {0x1E2FA, 0x1E2FE}, {0x1E300, 0x1E7FF}, {0x1E8C5, 0x1E8C6}, +{0x1E8D7, 0x1E8FF}, {0x1E94C, 0x1E94F}, {0x1E95A, 0x1E95D}, {0x1E960, 0x1EC70}, {0x1ECB5, 0x1ED00}, {0x1ED3E, 0x1EDFF}, {0x1EE04, 0x1EE04}, {0x1EE20, 0x1EE20}, {0x1EE23, 0x1EE23}, {0x1EE25, 0x1EE26}, +{0x1EE28, 0x1EE28}, {0x1EE33, 0x1EE33}, {0x1EE38, 0x1EE38}, {0x1EE3A, 0x1EE3A}, {0x1EE3C, 0x1EE41}, {0x1EE43, 0x1EE46}, {0x1EE48, 0x1EE48}, {0x1EE4A, 0x1EE4A}, {0x1EE4C, 0x1EE4C}, {0x1EE50, 0x1EE50}, +{0x1EE53, 0x1EE53}, {0x1EE55, 0x1EE56}, {0x1EE58, 0x1EE58}, {0x1EE5A, 0x1EE5A}, {0x1EE5C, 0x1EE5C}, {0x1EE5E, 0x1EE5E}, {0x1EE60, 0x1EE60}, {0x1EE63, 0x1EE63}, {0x1EE65, 0x1EE66}, {0x1EE6B, 0x1EE6B}, +{0x1EE73, 0x1EE73}, {0x1EE78, 0x1EE78}, {0x1EE7D, 0x1EE7D}, {0x1EE7F, 0x1EE7F}, {0x1EE8A, 0x1EE8A}, {0x1EE9C, 0x1EEA0}, {0x1EEA4, 0x1EEA4}, {0x1EEAA, 0x1EEAA}, {0x1EEBC, 0x1EEEF}, {0x1EEF2, 0x1EFFF}, +{0x1F02C, 0x1F02F}, {0x1F094, 0x1F09F}, {0x1F0AF, 0x1F0B0}, {0x1F0C0, 0x1F0C0}, {0x1F0D0, 0x1F0D0}, {0x1F0F6, 0x1F0FF}, {0x1F1AE, 0x1F1E5}, {0x1F203, 0x1F20F}, {0x1F23C, 0x1F23F}, {0x1F249, 0x1F24F}, +{0x1F252, 0x1F25F}, {0x1F266, 0x1F2FF}, {0x1F6D8, 0x1F6DF}, {0x1F6ED, 0x1F6EF}, {0x1F6FD, 0x1F6FF}, {0x1F774, 0x1F77F}, {0x1F7D9, 0x1F7DF}, {0x1F7EC, 0x1F7FF}, {0x1F80C, 0x1F80F}, {0x1F848, 0x1F84F}, +{0x1F85A, 0x1F85F}, {0x1F888, 0x1F88F}, {0x1F8AE, 0x1F8AF}, {0x1F8B2, 0x1F8FF}, {0x1F979, 0x1F979}, {0x1F9CC, 0x1F9CC}, {0x1FA54, 0x1FA5F}, {0x1FA6E, 0x1FA6F}, {0x1FA75, 0x1FA77}, {0x1FA7B, 0x1FA7F}, +{0x1FA87, 0x1FA8F}, {0x1FAA9, 0x1FAAF}, {0x1FAB7, 0x1FABF}, {0x1FAC3, 0x1FACF}, {0x1FAD7, 0x1FAFF}, {0x1FB93, 0x1FB93}, {0x1FBCB, 0x1FBEF}, {0x1FBFA, 0x1FFFF}, {0x2A6DE, 0x2A6FF}, {0x2B735, 0x2B73F}, +{0x2B81E, 0x2B81F}, {0x2CEA2, 0x2CEAF}, {0x2EBE1, 0x2F7FF}, {0x2FA1E, 0x2FFFF}, {0x3134B, 0xE00FF}, {0xE01F0, 0x10FFFF}, +}; + +//String +bool CNCTString::operator==(const std::string& other) const { + return str.compare(other) == 0; +} +bool CNCTString::operator==(const char other) const { + return str.compare(std::string(1, other)) == 0; +} +bool CNCTString::operator==(const CNCTString& other) const { + return str.compare(other.str) == 0; +} +// + operators +CNCTString& CNCTString::operator+=(const std::string& other) { + str += other; + int new_len = CNCTUnicode::strlen_utf8(other); + utf8_chars += new_len; + char_type = CNCTUnicode::string_identify(str); + seq_offset_bytes += other.size(); + seq_offset_utf8_chars += new_len; + return *this; +} + +CNCTString& CNCTString::operator+=(const char other) { + std::string str = std::string(1, other); + *this += str; + return *this; +} + +CNCTString& CNCTString::operator+=(const CNCTString& other) { + str += other.str; + utf8_chars += other.utf8_chars; + char_type = CNCTUnicode::string_identify(str); + seq_offset_bytes += other.str.size(); + seq_offset_utf8_chars += other.utf8_chars; + return *this; +} + +struct CRCompare { + bool operator()(const std::pair<int, int>& p, int i) { + return p.second < i; + } + bool operator()(int i, const std::pair<int, int>& p) { + return i < p.first; + } +}; + +// binary search for code range +bool CNCTUnicode::check_code_range(int c, const std::vector<std::pair<int, int>> &ranges) { + auto it = std::upper_bound(ranges.begin(), ranges.end(), c, CRCompare()); + if (it != ranges.begin()) { + --it; + } + return c >= it->first && c <= it->second; +} + +// these are binary searches, it takes only a few operations +CNCTCharType CNCTUnicode::get_code_type(int c) { + if (check_code_range(c, letter_ranges)) { + return LETTER; + } + if (check_code_range(c, digit_ranges)) { + return DIGIT; + } + if (check_code_range(c, whitespace_ranges)) { + return WHITESPACE; + } + if (check_code_range(c, punctuation_ranges)) { + return PUNCTUATION; + } + if (check_code_range(c, symbol_ranges)) { + return SYMBOL; + } + if (check_code_range(c, accent_mark_ranges)) { + return ACCENT_MARK; + } + if (check_code_range(c, control_ranges)) { + return CONTROL; + } + return UNIDENTIFIED; +} + +static int utf8_to_unicode(const std::string& utf8_char) { + int c = 0; + int len = (int)utf8_char.size(); + if (len == 1) { + c = utf8_char[0]; + } else if (len == 2) { + c = ((utf8_char[0] & 0x1F) << 6) | (utf8_char[1] & 0x3F); + } else if (len == 3) { + c = ((utf8_char[0] & 0x0F) << 12) | ((utf8_char[1] & 0x3F) << 6) | (utf8_char[2] & 0x3F); + } else if (len == 4) { + c = ((utf8_char[0] & 0x07) << 18) | ((utf8_char[1] & 0x3F) << 12) | ((utf8_char[2] & 0x3F) << 6) | (utf8_char[3] & 0x3F); + } + return c; +} + +CNCTCharType CNCTUnicode::get_code_type(const std::string &utf8_char) { + return get_code_type(utf8_to_unicode(utf8_char)); +} + +int CNCTUnicode::utf8_len(const char c) +{ + if ((c & 0x80) == 0) { + return 1; // ASCII character + } + if ((c & 0xE0) == 0xC0) { + return 2; // 2-byte character + } + if ((c & 0xF0) == 0xE0) { + return 3; // 3-byte character + } + if ((c & 0xF0) == 0xF0) { + return 4; // 4-byte character + } + return 1; // not valid utf8 + // static const uint8_t lookup[] = { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 3, 4 }; + // return lookup[static_cast<uint8_t>(c) >> 4]; +} + +int CNCTUnicode::strlen_utf8(const std::string src) { + int len = 0; + for (std::string::const_iterator it = src.begin(); it != src.end(); ++it) { + int char_len = utf8_len(*it); + if (char_len > 1) { + it += char_len - 1; + } + len += 1; + } + return len; +} + +// split a string into unicode strings +std::vector<std::string> CNCTUnicode::split_utf8(const std::string &src) { + std::vector<std::string> result; + for (std::string::const_iterator it = src.begin(); it != src.end(); ++it) { + int char_len = utf8_len(*it); + std::string str(it, it + char_len); + result.push_back(str); + if (char_len > 1) { + it += char_len - 1; + } + } + return result; +} + +// split a string into unicode strings (CNCTString) with sequence information +std::vector<CNCTString> CNCTUnicode::split_utf8_enhanced(const std::string &src) { + std::vector<CNCTString> result; + int seq_offset_bytes=0; + int seq_offset_utf8_chars=0; + for (std::string::const_iterator it = src.begin(); it != src.end(); ++it) { + int char_len = utf8_len(*it); + std::string str(it, it + char_len); + CNCTString cnct_str; + cnct_str.seq_offset_bytes = seq_offset_bytes; + cnct_str.seq_offset_utf8_chars = seq_offset_utf8_chars; + cnct_str.str = str; + cnct_str.utf8_chars = 1; + cnct_str.char_type = get_code_type(str); + #if 0 + switch (cnct_str.char_type) + { + case DIGIT: + printf("%s = DIGIT\n", str.c_str()); + break; + case LETTER: + printf("%s = LETTER\n", str.c_str()); + break; + case WHITESPACE: + printf("%s = WHITESPACE\n", str.c_str()); + break; + case PUNCTUATION: + printf("%s = PUNCTUATION\n", str.c_str()); + break; + case UNIDENTIFIED: + printf("%s = UNIDENTIFIED\n", str.c_str()); + break; + case SYMBOL: + printf("%s = SYMBOL\n", str.c_str()); + break; + case CONTROL: + printf("%s = CONTROL\n", str.c_str()); + break; + } + #endif + + result.push_back(cnct_str); + seq_offset_bytes += char_len; + seq_offset_utf8_chars += 1; + if (char_len > 1) { + it += char_len - 1; + } + + } + return result; +} + +// return the type of the string +CNCTCharType CNCTUnicode::string_identify(const std::string &str) { + CNCTCharType result = UNIDENTIFIED; + std::string::const_iterator it = str.begin(); + while (it != str.end()) { + int len = utf8_len(*it); + int c = 0; + for (int i = 0; i < len && it != str.end(); ++i, ++it) { + c = (c << 8) | static_cast<unsigned char>(*it); + } + switch (get_code_type(c)) { + case DIGIT: + if (result == UNIDENTIFIED) { + result = DIGIT; + } else if (result != DIGIT) { + return MIXED; + } + break; + case LETTER: + if (result == UNIDENTIFIED) { + result = LETTER; + } else if (result != LETTER) { + return MIXED; + } + break; + case WHITESPACE: + if (result == UNIDENTIFIED) { + result = WHITESPACE; + } else if (result != WHITESPACE) { + return MIXED; + } + break; + case PUNCTUATION: + if (result == UNIDENTIFIED) { + result = PUNCTUATION; + } else if (result != PUNCTUATION) { + return MIXED; + } + break; + default: + return MIXED; + break; + } + } + return result; +} + +// verify the content of a string +bool CNCTUnicode::string_test(const std::string &str, CNCTCharType chartype) +{ + std::string::const_iterator it = str.begin(); + while (it != str.end()) { + int len = utf8_len(*it); + int c = 0; + for (int i = 0; i < len && it != str.end(); ++i, ++it) { + c = (c << 8) | static_cast<unsigned char>(*it); + } + if (get_code_type(c) != chartype) { + return false; + } + } + return true; +} + +//----------------- +// llama.cpp GPT2 vocab (from libfalcon.cpp) +//----------------- + +std::string replaceAll(std::string str, const std::string& from, const std::string& to) { + size_t start_pos = 0; + while((start_pos = str.find(from, start_pos)) != std::string::npos) { + str.replace(start_pos, from.length(), to); + start_pos += to.length(); // Handles case where 'to' is a substring of 'from' + } + return str; +} + +struct TrieNode { + std::map<char, TrieNode*> map; + int32_t Id = -1; +}; + +struct Trie { + TrieNode *root; + + Trie() : root(new TrieNode()) {} + + ~Trie() { + if(root) + deleteTrie(root); + } + + // Move constructor + Trie(Trie&& other) noexcept : root(other.root) { + other.root = nullptr; + } + + // Move assignment operator + Trie& operator=(Trie&& other) noexcept { + if (this != &other) { + if(root) + deleteTrie(root); + root = other.root; + other.root = nullptr; + } + return *this; + } + + void insert(const std::string &token, int32_t Id) { + TrieNode* current = root; + for(auto ch : token) { + if(current->map.find(ch) == current->map.end()) { + current->map[ch] = new TrieNode(); + } + current = current->map[ch]; + } + current->Id = Id; + } + + void reset() { + deleteTrie(root); + root = new TrieNode(); + } + +private: + void deleteTrie(TrieNode* node) { + for(auto &it: node->map) { + deleteTrie(it.second); + } + delete node; + } + +}; + +struct gpt2bpe_vocab { + using id = int32_t; + using token = std::string; + + std::map<std::string, uint32_t> max_token_length; // max length, for each 2byte prefix + std::map<std::pair<std::string,std::string>, int> bpe_ranks; + std::vector<std::pair<std::string, std::string>> bpe_merges; + + id special_bos_id = -1; + id special_eos_id = -1; + id special_unk_id = -1; + id special_sep_id = -1; + id special_pad_id = -1; + + id linefeed_id = -1; + + std::unordered_map<token, id> token_to_id; + std::unordered_map<id, token> id_to_token; + + Trie trie; // highspeed access to tokens by prefix tree + + // populate trie from map + void populate_trie_from_map() { + trie.reset(); + for (const auto& pair : token_to_id) { + trie.insert(pair.first, pair.second); + if (pair.first.size() >= 2) { + std::string prefix = pair.first.substr(0, 2); + max_token_length[prefix] = std::max(max_token_length[prefix], (uint32_t)pair.first.size()); + } + } + } + // populate token ranks map + int populate_bpe_ranks(std::vector<std::pair<std::string, std::string>> bpe_merges_) { + for (int i = 0; i < (int)bpe_merges_.size(); i++) { + bpe_ranks.emplace(bpe_merges_[i], i); + } + bpe_merges = bpe_merges_; + return bpe_merges_.size(); + } + + // Trim whitespace characters from the beginning and end of the string + void trim(std::string& str) { + // Remove whitespace characters from the beginning of the string + str.erase(str.begin(), std::find_if(str.begin(), str.end(), [](int ch) { + return !std::isspace(ch); + })); + + // Remove whitespace characters from the end of the string + str.erase(std::find_if(str.rbegin(), str.rend(), [](int ch) { + return !std::isspace(ch); + }).base(), str.end()); + } + + // get max token length available for a prefix of 2 bytes (string at least 2 bytes long) + int get_max_token_length(const std::string& string) const { + if (string.size() < 2) { + return -1; + } + std::string prefix = string.substr(0, 2); + if (max_token_length.find(prefix) == max_token_length.end()) { + return 0; + } + return max_token_length.at(prefix); + } + + // function to find if two tokens match in bpe_rank, return rank or -1 + int find_bpe_rank(const std::string& token1, const std::string& token2) const { + std::string left_token = token1; + std::string right_token = token2; + left_token = replaceAll(left_token, " ", "Ġ"); + left_token = replaceAll(left_token, "\n", "Ċ"); + right_token = replaceAll(right_token, " ", "Ġ"); + right_token = replaceAll(right_token, "\n", "Ċ"); + + auto it = bpe_ranks.find(std::make_pair(left_token, right_token)); + if (it == bpe_ranks.end()) { + return -1; + } + return it->second; + } + + std::pair<gpt2bpe_vocab::id, std::string> find_longest_match(const std::string& snippet) const { + TrieNode* current = trie.root; + gpt2bpe_vocab::id last_matched_id = -1; + std::string last_matched_token = ""; + std::string current_token = ""; + for (auto ch : snippet) { + if (current->map.find(ch) == current->map.end()) { + break; + } + current = current->map[ch]; + current_token += ch; + if (current->Id != -1) { + last_matched_id = current->Id; + last_matched_token = current_token; + } + } + return {last_matched_id, last_matched_token}; + } + +}; + + +// +// tokenizer - bpe type, gpt2 tokenization compatible +// + +struct ggllm_bpe_symbol { + using index = int; + index prev; + index next; + const char * text; + size_t n; +}; + +static_assert(std::is_trivially_copyable<ggllm_bpe_symbol>::value, "ggllm_bpe_symbol is not trivially copyable"); + +struct ggllm_bpe_bigram { + struct comparator { + bool operator()(ggllm_bpe_bigram & l, ggllm_bpe_bigram & r) { + return l.rank > r.rank || (l.rank == r.rank && l.left > r.left); + } + }; + + using queue_storage = std::vector<ggllm_bpe_bigram>; + using queue = std::priority_queue<ggllm_bpe_bigram, queue_storage, comparator>; + ggllm_bpe_symbol::index left; + ggllm_bpe_symbol::index right; + std::string text; + int rank; + size_t size; +}; + +struct gpt2bpe_tokenizer { + gpt2bpe_tokenizer(const gpt2bpe_vocab & vocab, bool g2ws_): vocab_(vocab) { flag_g2ws = g2ws_; } + + void tokenize(const std::string & text, std::vector<gpt2bpe_vocab::id> & output) { + int final_prev_index = -1; + // auto start = ggml_time_us(); + auto word_collection = bpe_gpt2_preprocess(text); + // auto end = ggml_time_us(); + // fprintf(stderr, "%s: preprocessing took %0.3f ms\n", __func__, (end - start) / 1000.0); + + symbols_final.clear(); + + for (auto & word : word_collection) { + work_queue_ = ggllm_bpe_bigram::queue(); + symbols_.clear(); + + int index = 0; + size_t offset = 0; + + while (offset < word.size()) { + ggllm_bpe_symbol sym; + size_t char_len = std::min(word.size() - offset, (size_t) CNCTUnicode::utf8_len(word[offset])); + sym.text = word.c_str() + offset; + sym.n = 1; + sym.n = char_len; + offset += sym.n; + sym.prev = index - 1; + sym.next = offset == word.size() ? -1 : index + 1; + index++; + symbols_.emplace_back(sym); + } + for (size_t i = 1; i < symbols_.size(); ++i) { + add_new_bigram(i - 1, i); + } + + // build token(s) + while (!work_queue_.empty()) { + auto bigram = work_queue_.top(); + work_queue_.pop(); + + auto & left_symbol = symbols_[bigram.left]; + auto & right_symbol = symbols_[bigram.right]; + + if (left_symbol.n == 0 || right_symbol.n == 0) { + continue; + } + std::string left_token = std::string(left_symbol.text, left_symbol.n); + std::string right_token = std::string(right_symbol.text, right_symbol.n); + if (left_token + right_token != bigram.text) { + continue; // Skip this bigram if it's outdated + } + + // merge the right sym into the left one + left_symbol.n += right_symbol.n; + right_symbol.n = 0; + + // remove the right sym from the chain + left_symbol.next = right_symbol.next; + if (right_symbol.next >= 0) { + symbols_[right_symbol.next].prev = bigram.left; + } + + add_new_bigram(left_symbol.prev, bigram.left); // left side of current symbol + add_new_bigram(bigram.left, left_symbol.next); // right side of current symbol + } + + // add the fnished tokens to the final list keeping correct order for next and prev + for (auto & sym : symbols_) { + if (sym.n > 0) { + sym.prev = final_prev_index; + sym.next = -1; + if (final_prev_index != -1) { + symbols_final[final_prev_index].next = symbols_final.size(); + } + symbols_final.emplace_back(sym); + final_prev_index = symbols_final.size() - 1; + } + } + } + + symbols_ = symbols_final; + if (symbols_.size()) + for (int i = 0; i != -1; i = symbols_[i].next) { + auto & symbol = symbols_[i]; + if (symbol.n == 0) { + continue; + } + std::string str = std::string(symbol.text, symbol.n); + std::string str_decoded = decode_token(str); + auto token = vocab_.token_to_id.find(str_decoded); + + if (token == vocab_.token_to_id.end()) { + for (auto j = str_decoded.begin(); j != str_decoded.end(); ++j) { + std::string byte_str(1, *j); + auto token_multibyte = vocab_.token_to_id.find(byte_str); + if (token_multibyte == vocab_.token_to_id.end()) { + fprintf(stderr,"ERROR: byte not found in vocab: '%s'\n", byte_str.c_str()); + } + output.push_back((*token_multibyte).second); + } + } else { + output.push_back((*token).second); + } + } + } + +private: + void add_new_bigram(int left, int right) { + if (left == -1 || right == -1) return; + + std::string left_token = std::string(symbols_[left].text, symbols_[left].n); + std::string right_token = std::string(symbols_[right].text, symbols_[right].n); + + int rank_found = -1; + rank_found = vocab_.find_bpe_rank(left_token, right_token); + + if (rank_found < 0) { + return; + } + + ggllm_bpe_bigram bigram; + bigram.left = left; + bigram.right = right; + bigram.rank = rank_found; + bigram.size = left_token.size() + right_token.size(); + bigram.text = left_token + right_token; + work_queue_.push(bigram); + } + + std::unordered_map<unsigned char, std::string> bytes_to_unicode() { + static std::unordered_map<unsigned char, std::string> hex_map = { + { 0x21, "\x21" }, { 0x22, "\x22" }, { 0x23, "\x23" }, { 0x24, "\x24" }, { 0x25, "\x25" }, { 0x26, "\x26" }, { 0x27, "\x27" }, { 0x28, "\x28" }, { 0x29, "\x29" }, { 0x2A, "\x2A" }, + { 0x2B, "\x2B" }, { 0x2C, "\x2C" }, { 0x2D, "\x2D" }, { 0x2E, "\x2E" }, { 0x2F, "\x2F" }, { 0x30, "\x30" }, { 0x31, "\x31" }, { 0x32, "\x32" }, { 0x33, "\x33" }, { 0x34, "\x34" }, + { 0x35, "\x35" }, { 0x36, "\x36" }, { 0x37, "\x37" }, { 0x38, "\x38" }, { 0x39, "\x39" }, { 0x3A, "\x3A" }, { 0x3B, "\x3B" }, { 0x3C, "\x3C" }, { 0x3D, "\x3D" }, { 0x3E, "\x3E" }, + { 0x3F, "\x3F" }, { 0x40, "\x40" }, { 0x41, "\x41" }, { 0x42, "\x42" }, { 0x43, "\x43" }, { 0x44, "\x44" }, { 0x45, "\x45" }, { 0x46, "\x46" }, { 0x47, "\x47" }, { 0x48, "\x48" }, + { 0x49, "\x49" }, { 0x4A, "\x4A" }, { 0x4B, "\x4B" }, { 0x4C, "\x4C" }, { 0x4D, "\x4D" }, { 0x4E, "\x4E" }, { 0x4F, "\x4F" }, { 0x50, "\x50" }, { 0x51, "\x51" }, { 0x52, "\x52" }, + { 0x53, "\x53" }, { 0x54, "\x54" }, { 0x55, "\x55" }, { 0x56, "\x56" }, { 0x57, "\x57" }, { 0x58, "\x58" }, { 0x59, "\x59" }, { 0x5A, "\x5A" }, { 0x5B, "\x5B" }, { 0x5C, "\x5C" }, + { 0x5D, "\x5D" }, { 0x5E, "\x5E" }, { 0x5F, "\x5F" }, { 0x60, "\x60" }, { 0x61, "\x61" }, { 0x62, "\x62" }, { 0x63, "\x63" }, { 0x64, "\x64" }, { 0x65, "\x65" }, { 0x66, "\x66" }, + { 0x67, "\x67" }, { 0x68, "\x68" }, { 0x69, "\x69" }, { 0x6A, "\x6A" }, { 0x6B, "\x6B" }, { 0x6C, "\x6C" }, { 0x6D, "\x6D" }, { 0x6E, "\x6E" }, { 0x6F, "\x6F" }, { 0x70, "\x70" }, + { 0x71, "\x71" }, { 0x72, "\x72" }, { 0x73, "\x73" }, { 0x74, "\x74" }, { 0x75, "\x75" }, { 0x76, "\x76" }, { 0x77, "\x77" }, { 0x78, "\x78" }, { 0x79, "\x79" }, { 0x7A, "\x7A" }, + { 0x7B, "\x7B" }, { 0x7C, "\x7C" }, { 0x7D, "\x7D" }, { 0x7E, "\x7E" }, { 0xA1, "\xC2\xA1" }, { 0xA2, "\xC2\xA2" }, { 0xA3, "\xC2\xA3" }, { 0xA4, "\xC2\xA4" }, { 0xA5, "\xC2\xA5" }, + { 0xA6, "\xC2\xA6" }, { 0xA7, "\xC2\xA7" }, { 0xA8, "\xC2\xA8" }, { 0xA9, "\xC2\xA9" }, { 0xAA, "\xC2\xAA" }, { 0xAB, "\xC2\xAB" }, { 0xAC, "\xC2\xAC" }, { 0xAE, "\xC2\xAE" }, + { 0xAF, "\xC2\xAF" }, { 0xB0, "\xC2\xB0" }, { 0xB1, "\xC2\xB1" }, { 0xB2, "\xC2\xB2" }, { 0xB3, "\xC2\xB3" }, { 0xB4, "\xC2\xB4" }, { 0xB5, "\xC2\xB5" }, { 0xB6, "\xC2\xB6" }, + { 0xB7, "\xC2\xB7" }, { 0xB8, "\xC2\xB8" }, { 0xB9, "\xC2\xB9" }, { 0xBA, "\xC2\xBA" }, { 0xBB, "\xC2\xBB" }, { 0xBC, "\xC2\xBC" }, { 0xBD, "\xC2\xBD" }, { 0xBE, "\xC2\xBE" }, + { 0xBF, "\xC2\xBF" }, { 0xC0, "\xC3\x80" }, { 0xC1, "\xC3\x81" }, { 0xC2, "\xC3\x82" }, { 0xC3, "\xC3\x83" }, { 0xC4, "\xC3\x84" }, { 0xC5, "\xC3\x85" }, { 0xC6, "\xC3\x86" }, + { 0xC7, "\xC3\x87" }, { 0xC8, "\xC3\x88" }, { 0xC9, "\xC3\x89" }, { 0xCA, "\xC3\x8A" }, { 0xCB, "\xC3\x8B" }, { 0xCC, "\xC3\x8C" }, { 0xCD, "\xC3\x8D" }, { 0xCE, "\xC3\x8E" }, + { 0xCF, "\xC3\x8F" }, { 0xD0, "\xC3\x90" }, { 0xD1, "\xC3\x91" }, { 0xD2, "\xC3\x92" }, { 0xD3, "\xC3\x93" }, { 0xD4, "\xC3\x94" }, { 0xD5, "\xC3\x95" }, { 0xD6, "\xC3\x96" }, + { 0xD7, "\xC3\x97" }, { 0xD8, "\xC3\x98" }, { 0xD9, "\xC3\x99" }, { 0xDA, "\xC3\x9A" }, { 0xDB, "\xC3\x9B" }, { 0xDC, "\xC3\x9C" }, { 0xDD, "\xC3\x9D" }, { 0xDE, "\xC3\x9E" }, + { 0xDF, "\xC3\x9F" }, { 0xE0, "\xC3\xA0" }, { 0xE1, "\xC3\xA1" }, { 0xE2, "\xC3\xA2" }, { 0xE3, "\xC3\xA3" }, { 0xE4, "\xC3\xA4" }, { 0xE5, "\xC3\xA5" }, { 0xE6, "\xC3\xA6" }, + { 0xE7, "\xC3\xA7" }, { 0xE8, "\xC3\xA8" }, { 0xE9, "\xC3\xA9" }, { 0xEA, "\xC3\xAA" }, { 0xEB, "\xC3\xAB" }, { 0xEC, "\xC3\xAC" }, { 0xED, "\xC3\xAD" }, { 0xEE, "\xC3\xAE" }, + { 0xEF, "\xC3\xAF" }, { 0xF0, "\xC3\xB0" }, { 0xF1, "\xC3\xB1" }, { 0xF2, "\xC3\xB2" }, { 0xF3, "\xC3\xB3" }, { 0xF4, "\xC3\xB4" }, { 0xF5, "\xC3\xB5" }, { 0xF6, "\xC3\xB6" }, + { 0xF7, "\xC3\xB7" }, { 0xF8, "\xC3\xB8" }, { 0xF9, "\xC3\xB9" }, { 0xFA, "\xC3\xBA" }, { 0xFB, "\xC3\xBB" }, { 0xFC, "\xC3\xBC" }, { 0xFD, "\xC3\xBD" }, { 0xFE, "\xC3\xBE" }, + { 0xFF, "\xC3\xBF" }, { 0x00, "\xC4\x80" }, { 0x01, "\xC4\x81" }, { 0x02, "\xC4\x82" }, { 0x03, "\xC4\x83" }, { 0x04, "\xC4\x84" }, { 0x05, "\xC4\x85" }, { 0x06, "\xC4\x86" }, + { 0x07, "\xC4\x87" }, { 0x08, "\xC4\x88" }, { 0x09, "\xC4\x89" }, { 0x0A, "\xC4\x8A" }, { 0x0B, "\xC4\x8B" }, { 0x0C, "\xC4\x8C" }, { 0x0D, "\xC4\x8D" }, { 0x0E, "\xC4\x8E" }, + { 0x0F, "\xC4\x8F" }, { 0x10, "\xC4\x90" }, { 0x11, "\xC4\x91" }, { 0x12, "\xC4\x92" }, { 0x13, "\xC4\x93" }, { 0x14, "\xC4\x94" }, { 0x15, "\xC4\x95" }, { 0x16, "\xC4\x96" }, + { 0x17, "\xC4\x97" }, { 0x18, "\xC4\x98" }, { 0x19, "\xC4\x99" }, { 0x1A, "\xC4\x9A" }, { 0x1B, "\xC4\x9B" }, { 0x1C, "\xC4\x9C" }, { 0x1D, "\xC4\x9D" }, { 0x1E, "\xC4\x9E" }, + { 0x1F, "\xC4\x9F" }, { 0x20, "\xC4\xA0" }, { 0x7F, "\xC4\xA1" }, { 0x80, "\xC4\xA2" }, { 0x81, "\xC4\xA3" }, { 0x82, "\xC4\xA4" }, { 0x83, "\xC4\xA5" }, { 0x84, "\xC4\xA6" }, + { 0x85, "\xC4\xA7" }, { 0x86, "\xC4\xA8" }, { 0x87, "\xC4\xA9" }, { 0x88, "\xC4\xAA" }, { 0x89, "\xC4\xAB" }, { 0x8A, "\xC4\xAC" }, { 0x8B, "\xC4\xAD" }, { 0x8C, "\xC4\xAE" }, + { 0x8D, "\xC4\xAF" }, { 0x8E, "\xC4\xB0" }, { 0x8F, "\xC4\xB1" }, { 0x90, "\xC4\xB2" }, { 0x91, "\xC4\xB3" }, { 0x92, "\xC4\xB4" }, { 0x93, "\xC4\xB5" }, { 0x94, "\xC4\xB6" }, + { 0x95, "\xC4\xB7" }, { 0x96, "\xC4\xB8" }, { 0x97, "\xC4\xB9" }, { 0x98, "\xC4\xBA" }, { 0x99, "\xC4\xBB" }, { 0x9A, "\xC4\xBC" }, { 0x9B, "\xC4\xBD" }, { 0x9C, "\xC4\xBE" }, + { 0x9D, "\xC4\xBF" }, { 0x9E, "\xC5\x80" }, { 0x9F, "\xC5\x81" }, { 0xA0, "\xC5\x82" }, { 0xAD, "\xC5\x83" } + }; + return hex_map; + } + + std::unordered_map<std::string, unsigned char> unicode_to_bytes() { + static std::unordered_map<std::string, unsigned char> hex_map = { + { "\x21", 0x21 }, { "\x22", 0x22 }, { "\x23", 0x23 }, { "\x24", 0x24 }, { "\x25", 0x25 }, { "\x26", 0x26 }, { "\x27", 0x27 }, { "\x28", 0x28 }, { "\x29", 0x29 }, { "\x2A", 0x2A }, + { "\x2B", 0x2B }, { "\x2C", 0x2C }, { "\x2D", 0x2D }, { "\x2E", 0x2E }, { "\x2F", 0x2F }, { "\x30", 0x30 }, { "\x31", 0x31 }, { "\x32", 0x32 }, { "\x33", 0x33 }, { "\x34", 0x34 }, + { "\x35", 0x35 }, { "\x36", 0x36 }, { "\x37", 0x37 }, { "\x38", 0x38 }, { "\x39", 0x39 }, { "\x3A", 0x3A }, { "\x3B", 0x3B }, { "\x3C", 0x3C }, { "\x3D", 0x3D }, { "\x3E", 0x3E }, + { "\x3F", 0x3F }, { "\x40", 0x40 }, { "\x41", 0x41 }, { "\x42", 0x42 }, { "\x43", 0x43 }, { "\x44", 0x44 }, { "\x45", 0x45 }, { "\x46", 0x46 }, { "\x47", 0x47 }, { "\x48", 0x48 }, + { "\x49", 0x49 }, { "\x4A", 0x4A }, { "\x4B", 0x4B }, { "\x4C", 0x4C }, { "\x4D", 0x4D }, { "\x4E", 0x4E }, { "\x4F", 0x4F }, { "\x50", 0x50 }, { "\x51", 0x51 }, { "\x52", 0x52 }, + { "\x53", 0x53 }, { "\x54", 0x54 }, { "\x55", 0x55 }, { "\x56", 0x56 }, { "\x57", 0x57 }, { "\x58", 0x58 }, { "\x59", 0x59 }, { "\x5A", 0x5A }, { "\x5B", 0x5B }, { "\x5C", 0x5C }, + { "\x5D", 0x5D }, { "\x5E", 0x5E }, { "\x5F", 0x5F }, { "\x60", 0x60 }, { "\x61", 0x61 }, { "\x62", 0x62 }, { "\x63", 0x63 }, { "\x64", 0x64 }, { "\x65", 0x65 }, { "\x66", 0x66 }, + { "\x67", 0x67 }, { "\x68", 0x68 }, { "\x69", 0x69 }, { "\x6A", 0x6A }, { "\x6B", 0x6B }, { "\x6C", 0x6C }, { "\x6D", 0x6D }, { "\x6E", 0x6E }, { "\x6F", 0x6F }, { "\x70", 0x70 }, + { "\x71", 0x71 }, { "\x72", 0x72 }, { "\x73", 0x73 }, { "\x74", 0x74 }, { "\x75", 0x75 }, { "\x76", 0x76 }, { "\x77", 0x77 }, { "\x78", 0x78 }, { "\x79", 0x79 }, { "\x7A", 0x7A }, + { "\x7B", 0x7B }, { "\x7C", 0x7C }, { "\x7D", 0x7D }, { "\x7E", 0x7E }, { "\xC2\xA1", 0xA1 }, { "\xC2\xA2", 0xA2 }, { "\xC2\xA3", 0xA3 }, { "\xC2\xA4", 0xA4 }, { "\xC2\xA5", 0xA5 }, + { "\xC2\xA6", 0xA6 }, { "\xC2\xA7", 0xA7 }, { "\xC2\xA8", 0xA8 }, { "\xC2\xA9", 0xA9 }, { "\xC2\xAA", 0xAA }, { "\xC2\xAB", 0xAB }, { "\xC2\xAC", 0xAC }, { "\xC2\xAE", 0xAE }, + { "\xC2\xAF", 0xAF }, { "\xC2\xB0", 0xB0 }, { "\xC2\xB1", 0xB1 }, { "\xC2\xB2", 0xB2 }, { "\xC2\xB3", 0xB3 }, { "\xC2\xB4", 0xB4 }, { "\xC2\xB5", 0xB5 }, { "\xC2\xB6", 0xB6 }, + { "\xC2\xB7", 0xB7 }, { "\xC2\xB8", 0xB8 }, { "\xC2\xB9", 0xB9 }, { "\xC2\xBA", 0xBA }, { "\xC2\xBB", 0xBB }, { "\xC2\xBC", 0xBC }, { "\xC2\xBD", 0xBD }, { "\xC2\xBE", 0xBE }, + { "\xC2\xBF", 0xBF }, { "\xC3\x80", 0xC0 }, { "\xC3\x81", 0xC1 }, { "\xC3\x82", 0xC2 }, { "\xC3\x83", 0xC3 }, { "\xC3\x84", 0xC4 }, { "\xC3\x85", 0xC5 }, { "\xC3\x86", 0xC6 }, + { "\xC3\x87", 0xC7 }, { "\xC3\x88", 0xC8 }, { "\xC3\x89", 0xC9 }, { "\xC3\x8A", 0xCA }, { "\xC3\x8B", 0xCB }, { "\xC3\x8C", 0xCC }, { "\xC3\x8D", 0xCD }, { "\xC3\x8E", 0xCE }, + { "\xC3\x8F", 0xCF }, { "\xC3\x90", 0xD0 }, { "\xC3\x91", 0xD1 }, { "\xC3\x92", 0xD2 }, { "\xC3\x93", 0xD3 }, { "\xC3\x94", 0xD4 }, { "\xC3\x95", 0xD5 }, { "\xC3\x96", 0xD6 }, + { "\xC3\x97", 0xD7 }, { "\xC3\x98", 0xD8 }, { "\xC3\x99", 0xD9 }, { "\xC3\x9A", 0xDA }, { "\xC3\x9B", 0xDB }, { "\xC3\x9C", 0xDC }, { "\xC3\x9D", 0xDD }, { "\xC3\x9E", 0xDE }, + { "\xC3\x9F", 0xDF }, { "\xC3\xA0", 0xE0 }, { "\xC3\xA1", 0xE1 }, { "\xC3\xA2", 0xE2 }, { "\xC3\xA3", 0xE3 }, { "\xC3\xA4", 0xE4 }, { "\xC3\xA5", 0xE5 }, { "\xC3\xA6", 0xE6 }, + { "\xC3\xA7", 0xE7 }, { "\xC3\xA8", 0xE8 }, { "\xC3\xA9", 0xE9 }, { "\xC3\xAA", 0xEA }, { "\xC3\xAB", 0xEB }, { "\xC3\xAC", 0xEC }, { "\xC3\xAD", 0xED }, { "\xC3\xAE", 0xEE }, + { "\xC3\xAF", 0xEF }, { "\xC3\xB0", 0xF0 }, { "\xC3\xB1", 0xF1 }, { "\xC3\xB2", 0xF2 }, { "\xC3\xB3", 0xF3 }, { "\xC3\xB4", 0xF4 }, { "\xC3\xB5", 0xF5 }, { "\xC3\xB6", 0xF6 }, + { "\xC3\xB7", 0xF7 }, { "\xC3\xB8", 0xF8 }, { "\xC3\xB9", 0xF9 }, { "\xC3\xBA", 0xFA }, { "\xC3\xBB", 0xFB }, { "\xC3\xBC", 0xFC }, { "\xC3\xBD", 0xFD }, { "\xC3\xBE", 0xFE }, + { "\xC3\xBF", 0xFF }, { "\xC4\x80", 0x00 }, { "\xC4\x81", 0x01 }, { "\xC4\x82", 0x02 }, { "\xC4\x83", 0x03 }, { "\xC4\x84", 0x04 }, { "\xC4\x85", 0x05 }, { "\xC4\x86", 0x06 }, + { "\xC4\x87", 0x07 }, { "\xC4\x88", 0x08 }, { "\xC4\x89", 0x09 }, { "\xC4\x8A", 0x0A }, { "\xC4\x8B", 0x0B }, { "\xC4\x8C", 0x0C }, { "\xC4\x8D", 0x0D }, { "\xC4\x8E", 0x0E }, + { "\xC4\x8F", 0x0F }, { "\xC4\x90", 0x10 }, { "\xC4\x91", 0x11 }, { "\xC4\x92", 0x12 }, { "\xC4\x93", 0x13 }, { "\xC4\x94", 0x14 }, { "\xC4\x95", 0x15 }, { "\xC4\x96", 0x16 }, + { "\xC4\x97", 0x17 }, { "\xC4\x98", 0x18 }, { "\xC4\x99", 0x19 }, { "\xC4\x9A", 0x1A }, { "\xC4\x9B", 0x1B }, { "\xC4\x9C", 0x1C }, { "\xC4\x9D", 0x1D }, { "\xC4\x9E", 0x1E }, + { "\xC4\x9F", 0x1F }, { "\xC4\xA0", 0x20 }, { "\xC4\xA1", 0x7F }, { "\xC4\xA2", 0x80 }, { "\xC4\xA3", 0x81 }, { "\xC4\xA4", 0x82 }, { "\xC4\xA5", 0x83 }, { "\xC4\xA6", 0x84 }, + { "\xC4\xA7", 0x85 }, { "\xC4\xA8", 0x86 }, { "\xC4\xA9", 0x87 }, { "\xC4\xAA", 0x88 }, { "\xC4\xAB", 0x89 }, { "\xC4\xAC", 0x8A }, { "\xC4\xAD", 0x8B }, { "\xC4\xAE", 0x8C }, + { "\xC4\xAF", 0x8D }, { "\xC4\xB0", 0x8E }, { "\xC4\xB1", 0x8F }, { "\xC4\xB2", 0x90 }, { "\xC4\xB3", 0x91 }, { "\xC4\xB4", 0x92 }, { "\xC4\xB5", 0x93 }, { "\xC4\xB6", 0x94 }, + { "\xC4\xB7", 0x95 }, { "\xC4\xB8", 0x96 }, { "\xC4\xB9", 0x97 }, { "\xC4\xBA", 0x98 }, { "\xC4\xBB", 0x99 }, { "\xC4\xBC", 0x9A }, { "\xC4\xBD", 0x9B }, { "\xC4\xBE", 0x9C }, + { "\xC4\xBF", 0x9D }, { "\xC5\x80", 0x9E }, { "\xC5\x81", 0x9F }, { "\xC5\x82", 0xA0 }, { "\xC5\x83", 0xAD } + }; + return hex_map; + } + + // len must be available + bool inline str_is_equal(const char* str1, const char* str2, size_t len) { + for (size_t i = 0; i < len; ++i) { + if (str1[i] != str2[i]) { + return false; + } + } + return true; + } + + std::vector<std::string> bpe_gpt2_preprocess(const std::string& text) { + static std::unordered_map< unsigned char, std::string> byte_encoder = bytes_to_unicode(); + std::vector<std::string> bpe_words; + std::vector<std::string> bpe_encoded_words; + + std::string token=""; + const char *raw_text_p = text.c_str(); + // GPT2 system regex: 's|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+ + bool collecting_numeric = false; + bool collecting_letter = false; + bool collecting_special = false; + bool collecting_whitespace_lookahead = false; + bool collecting=false; + + std::vector<CNCTString> text_utf; + text_utf.reserve(text.size()); + bpe_words.reserve(text.size()); + bpe_encoded_words.reserve(text.size()); + + text_utf = CNCTUnicode::split_utf8_enhanced(text); + + for (int i = 0; i < (int)text_utf.size(); i++) { + const CNCTString &utf_char = text_utf[i]; + bool split_condition = false; + const char *text_pos = raw_text_p + utf_char.seq_offset_bytes; + int bytes_remain = strlen(text_pos); + // forward backward lookups + const CNCTString &utf_char_next = (i+1 < (int)text_utf.size()) ? text_utf[i+1] : CNCTString(); + const CNCTString &utf_char_next_next = (i+2 < (int)text_utf.size()) ? text_utf[i+2] : CNCTString(); + // const CNCTString &utf_char_prev = (i > 0) ? text_utf[i-1] : CNCTString(); + + // handling contractions + if (!split_condition && bytes_remain >= 2) { + // 's|'t|'m|'d + if (utf_char == '\'' && (utf_char_next == 's' || utf_char_next == 't' || utf_char_next == 'm' || utf_char_next == 'd')) { + split_condition = true; + } + if (split_condition) { + if (token.size()) { + bpe_words.emplace_back(token); // push previous content as token + } + token = utf_char.str + utf_char_next.str; + bpe_words.emplace_back(token); + token=""; + i++; + continue; + } + } + if (!split_condition && bytes_remain >= 3) { + // 're|'ve|'ll + if (utf_char == '\'' && ( + (utf_char_next == 'r' || utf_char_next_next == 'e') || + (utf_char_next == 'v' || utf_char_next_next == 'e') || + (utf_char_next == 'l' || utf_char_next_next == 'l')) + ) { + split_condition = true; + } + if (split_condition) { + // current token + next token can be defined + if (token.size()) { + bpe_words.emplace_back(token); // push previous content as token + } + token = utf_char.str + utf_char_next.str + utf_char_next_next.str; + bpe_words.emplace_back(token); // the contraction + token=""; + i+=2; + continue; + } + } + + if (!split_condition && !collecting) { + if (utf_char.char_type == CNCTCharType::LETTER || (!token.size() && utf_char==" " && utf_char_next.char_type == CNCTCharType::LETTER)) { + collecting_letter = true; + collecting = true; + } else if (utf_char.char_type == CNCTCharType::DIGIT || (!token.size() && utf_char==" " && utf_char_next.char_type == CNCTCharType::DIGIT)) { + collecting_numeric = true; + collecting = true; + } else if ( + ((utf_char.char_type != CNCTCharType::LETTER && utf_char.char_type != CNCTCharType::DIGIT) && (utf_char.char_type != CNCTCharType::WHITESPACE)) || + (!token.size() && utf_char==" " && utf_char_next.char_type != CNCTCharType::LETTER && utf_char_next.char_type != CNCTCharType::DIGIT && utf_char_next.char_type != CNCTCharType::WHITESPACE) + ) { + collecting_special = true; + collecting = true; + } else if (utf_char.char_type == CNCTCharType::WHITESPACE && utf_char_next.char_type == CNCTCharType::WHITESPACE) { + collecting_whitespace_lookahead = true; + collecting = true; + } else if (utf_char.char_type == CNCTCharType::WHITESPACE) { + split_condition = true; + } + } else if (!split_condition && collecting) { + if (collecting_letter && utf_char.char_type != CNCTCharType::LETTER) { + split_condition = true; + } else if (collecting_numeric && utf_char.char_type != CNCTCharType::DIGIT) { + split_condition = true; + } else if (collecting_special && (utf_char.char_type == CNCTCharType::LETTER || utf_char.char_type == CNCTCharType::DIGIT || utf_char.char_type == CNCTCharType::WHITESPACE)) { + split_condition = true; + } else if (collecting_whitespace_lookahead && utf_char_next.char_type != CNCTCharType::WHITESPACE) { + split_condition = true; + } + } + + if(utf_char_next.str.size() == 0) { + split_condition = true; // final + token += utf_char.str; + } + + if (split_condition) { + if (token.size()) { + bpe_words.emplace_back(token); + } + token = utf_char.str; + collecting = false; + collecting_letter = false; + collecting_numeric = false; + collecting_special = false; + collecting_whitespace_lookahead = false; + } else { + token += utf_char.str; + } + } + + for (std::string& word : bpe_words) { + std::string encoded_token=""; + for (char& c : word) { + encoded_token += byte_encoder[c]; + } + bpe_encoded_words.emplace_back(encoded_token); + } + + return bpe_encoded_words; + } + + // decoder (for one token) + std::string decode_token(const std::string& token) { + static std::unordered_map< std::string, unsigned char> byte_decoder = unicode_to_bytes(); + std::string decoded_token=""; + auto unicode_seqeunces = CNCTUnicode::split_utf8(token); + for (auto& unicode_sequence : unicode_seqeunces) { + decoded_token += byte_decoder[unicode_sequence]; + } + + return decoded_token; + } + + const gpt2bpe_vocab & vocab_; + std::vector<ggllm_bpe_symbol> symbols_; + std::vector<ggllm_bpe_symbol> symbols_final; + ggllm_bpe_bigram::queue work_queue_; + bool flag_g2ws=false; +}; + +static std::vector<gpt2bpe_vocab::id> gpt2bpe_tokenize(const gpt2bpe_vocab & vocab, const std::string & text, bool bos, bool g2ws ) { + gpt2bpe_tokenizer tokenizer(vocab, g2ws); + std::vector<gpt2bpe_vocab::id> output; + + if (text.empty()) { + return output; + } + + if (bos && vocab.special_bos_id != -1) { + output.push_back(vocab.special_bos_id); + } + + tokenizer.tokenize(text, output); + return output; +} + +#endif // CMPNCT_GPT2BPE diff --git a/examples/gptneox-wip/falcon-main.cpp b/examples/gptneox-wip/falcon-main.cpp new file mode 100644 index 00000000..43b6a29f --- /dev/null +++ b/examples/gptneox-wip/falcon-main.cpp @@ -0,0 +1,1111 @@ +#include "ggml.h" +#include "cmpnct_gpt2bpe.hpp" + +#include <cassert> +#include <cmath> +#include <cstdio> +#include <cstring> +#include <cinttypes> +#include <fstream> +#include <map> +#include <string> +#include <vector> +#include <thread> +#include <random> + +#if defined(_MSC_VER) +#pragma warning(disable: 4244 4267) // possible loss of data +#endif + +// default hparams +struct falcon_hparams { + size_t n_merges = 0; + size_t n_vocab = 0; + uint32_t n_ctx = 0; + uint32_t n_embd = 0; + uint32_t n_head = 0; + uint32_t n_head_kv = 1; // Needs to be 1 for 7B model + uint32_t n_ff = 0; + uint32_t n_block = 0; + float norm_eps = 1e-5; +}; +struct falcon_block { + // normalization + struct ggml_tensor* input_layernorm; + struct ggml_tensor* input_layernorm_b; + struct ggml_tensor* attention_norm; // Falcon-40B only + struct ggml_tensor* attention_norm_b; // Falcon-40B only + + // attention + struct ggml_tensor* query_key_value; + struct ggml_tensor* wo; + + // ff + struct ggml_tensor* ffn_up; + struct ggml_tensor* ffn_down; +}; + +struct falcon_model { + falcon_hparams hparams; + + struct ggml_tensor* tok_embeddings; + struct ggml_tensor* output_norm; + struct ggml_tensor* output_norm_b; + struct ggml_tensor* lm_head; + + std::vector<falcon_block> blocks; + + // key + value memory + struct ggml_tensor* memory_k; + struct ggml_tensor* memory_v; + + struct gguf_context * ggufctx; + struct ggml_context * ctx; + struct ggml_context * kvctx; + + std::map<std::string, struct ggml_tensor*> tensors; +}; + +struct gpt_params { + int32_t seed = -1; // RNG seed + int32_t n_threads = std::min(4, (int32_t) std::thread::hardware_concurrency()); + uint32_t n_predict = 200; // new tokens to predict + uint32_t n_batch = 512; // batch size for prompt processing + + // sampling parameters + int32_t top_k = 40; + float top_p = 1.0f; + float temp = 0.8f; + int32_t repeat_last_n = 64; + float repeat_penalty = 1.02f; + + std::string model = ""; // model path + std::string prompt = ""; + + std::string token_test = ""; + bool interactive = false; + int32_t interactive_port = -1; + int32_t n_gpu_layers = 0; +}; + +void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) { + fprintf(stderr, "usage: %s [options]\n", argv[0]); + fprintf(stderr, "\n"); + fprintf(stderr, "options:\n"); + fprintf(stderr, " -h, --help show this help message and exit\n"); + fprintf(stderr, " -s SEED, --seed SEED RNG seed (default: -1)\n"); + fprintf(stderr, " -t N, --threads N number of threads to use during computation (default: %d)\n", params.n_threads); + fprintf(stderr, " -ngl N, --gpu-layers N number of layers to offload to GPU on supported models (default: %d)\n", params.n_gpu_layers); + fprintf(stderr, " -p PROMPT, --prompt PROMPT\n"); + fprintf(stderr, " prompt to start generation with (default: random)\n"); + fprintf(stderr, " -f FNAME, --file FNAME\n"); + fprintf(stderr, " load prompt from a file\n"); + fprintf(stderr, " -tt TOKEN_TEST, --token_test TOKEN_TEST\n"); + fprintf(stderr, " test tokenization\n"); + fprintf(stderr, " -n N, --n_predict N number of tokens to predict (default: %d)\n", params.n_predict); + fprintf(stderr, " --top_k N top-k sampling, 0 = n_vocab (default: %d)\n", params.top_k); + fprintf(stderr, " --top_p N top-p sampling (default: %.1f)\n", params.top_p); + fprintf(stderr, " --temp N temperature (default: %.1f)\n", params.temp); + fprintf(stderr, " --repeat-last-n N last n tokens to consider for penalize (default: %d, 0 = disabled)\n", params.repeat_last_n); + fprintf(stderr, " --repeat-penalty N penalize repeat sequence of tokens (default: %.2f, 1.0 = disabled)\n", (double)params.repeat_penalty); + fprintf(stderr, " -b N, --batch_size N batch size for prompt processing (default: %d)\n", params.n_batch); + fprintf(stderr, " -m FNAME, --model FNAME\n"); + fprintf(stderr, " model path (default: %s)\n", params.model.c_str()); + fprintf(stderr, "\n"); +} + +// Function to check if the next argument exists +std::string get_next_arg(int& i, int argc, char** argv, const std::string& flag, gpt_params& params) { + if (i + 1 < argc && argv[i + 1][0] != '-') { + return argv[++i]; + } else { + fprintf(stderr, "error: %s requires one argument.\n", flag.c_str()); + gpt_print_usage(argc, argv, params); + exit(0); + } +} + +bool gpt_params_parse(int argc, char ** argv, gpt_params & params) { + for (int i = 1; i < argc; i++) { + std::string arg = argv[i]; + + if (arg == "-s" || arg == "--seed") { + params.seed = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "-t" || arg == "--threads") { + params.n_threads = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "-ngl" || arg == "--gpu-layers" || arg == "--n-gpu-layers") { + params.n_gpu_layers = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "-p" || arg == "--prompt") { + params.prompt = get_next_arg(i, argc, argv, arg, params); + } else if (arg == "-n" || arg == "--n_predict") { + params.n_predict = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "--top_k") { + params.top_k = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "--top_p") { + params.top_p = std::stof(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "--temp") { + params.temp = std::stof(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "--repeat-last-n") { + params.repeat_last_n = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "--repeat-penalty") { + params.repeat_penalty = std::stof(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "-b" || arg == "--batch_size") { + params.n_batch= std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "-m" || arg == "--model") { + params.model = get_next_arg(i, argc, argv, arg, params); + } else if (arg == "-i" || arg == "--interactive") { + params.interactive = true; + } else if (arg == "-ip" || arg == "--interactive-port") { + params.interactive = true; + params.interactive_port = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "-h" || arg == "--help") { + gpt_print_usage(argc, argv, params); + exit(0); + } else if (arg == "-f" || arg == "--file") { + get_next_arg(i, argc, argv, arg, params); + std::ifstream file(argv[i]); + if (!file) { + fprintf(stderr, "error: failed to open file '%s'\n", argv[i]); + break; + } + std::copy(std::istreambuf_iterator<char>(file), std::istreambuf_iterator<char>(), back_inserter(params.prompt)); + if (params.prompt.back() == '\n') { + params.prompt.pop_back(); + } + } else if (arg == "-tt" || arg == "--token_test") { + params.token_test = get_next_arg(i, argc, argv, arg, params); + } + else { + fprintf(stderr, "error: unknown argument: %s\n", arg.c_str()); + gpt_print_usage(argc, argv, params); + exit(0); + } + } + + return true; +} + +gpt2bpe_vocab::id sample_top_k_top_p_repeat( + const gpt2bpe_vocab & vocab, + const float * logits, + const int32_t * last_n_tokens_data, + size_t last_n_tokens_data_size, + int top_k, + double top_p, + double temp, + int repeat_last_n, + float repeat_penalty, + std::mt19937 & rng) { + + int n_logits = vocab.id_to_token.size(); + + const auto * plogits = logits; + + const auto last_n_tokens = std::vector<int32_t>(last_n_tokens_data, last_n_tokens_data + last_n_tokens_data_size); + + if (temp <= 0) { + // select the token with the highest logit directly + float max_logit = plogits[0]; + gpt2bpe_vocab::id max_id = 0; + + for (int i = 1; i < n_logits; ++i) { + if (plogits[i] > max_logit) { + max_logit = plogits[i]; + max_id = i; + } + } + return max_id; + } + + + std::vector<std::pair<double, gpt2bpe_vocab::id>> logits_id; + logits_id.reserve(n_logits); + + { + const float scale = 1.0f/temp; + for (int i = 0; i < n_logits; ++i) { + // repetition penalty from ctrl paper (https://arxiv.org/abs/1909.05858) + // credit https://github.com/facebookresearch/llama/compare/main...shawwn:llama:main + if (repeat_last_n > 0 && std::find(last_n_tokens.end()-repeat_last_n, last_n_tokens.end(), i) != last_n_tokens.end()) { + // if score < 0 then repetition penalty has to multiplied to reduce the previous token probability + if (plogits[i] < 0.0f) { + logits_id.push_back(std::make_pair(plogits[i]*scale*repeat_penalty, i)); + } else { + logits_id.push_back(std::make_pair(plogits[i]*scale/repeat_penalty, i)); + } + } else { + logits_id.push_back(std::make_pair(plogits[i]*scale, i)); + } + } + } + + // find the top K tokens + std::partial_sort( + logits_id.begin(), + logits_id.begin() + top_k, logits_id.end(), + [](const std::pair<double, gpt2bpe_vocab::id> & a, const std::pair<double, gpt2bpe_vocab::id> & b) { + return a.first > b.first; + }); + + logits_id.resize(top_k); + + double maxl = -INFINITY; + for (const auto & kv : logits_id) { + maxl = std::max(maxl, kv.first); + } + + // compute probs for the top K tokens + std::vector<double> probs; + probs.reserve(logits_id.size()); + + double sum = 0.0; + for (const auto & kv : logits_id) { + double p = exp(kv.first - maxl); + probs.push_back(p); + sum += p; + } + + // normalize the probs + for (auto & p : probs) { + p /= sum; + } + + if (top_p < 1.0f) { + double cumsum = 0.0f; + for (int i = 0; i < top_k; i++) { + cumsum += probs[i]; + if (cumsum >= top_p) { + top_k = i + 1; + probs.resize(top_k); + logits_id.resize(top_k); + break; + } + } + + cumsum = 1.0/cumsum; + for (int i = 0; i < (int) probs.size(); i++) { + probs[i] *= cumsum; + } + } + +// printf("\n"); +// for (int i = 0; i < (int) probs.size(); i++) { +// for (int i = 0; i < 10; i++) { +// printf("%d: '%s' %f\n", i, vocab.id_to_token.at(logits_id[i].second).c_str(), probs[i]); +// } + + std::discrete_distribution<> dist(probs.begin(), probs.end()); + int idx = dist(rng); + + return logits_id[idx].second; + +} + +struct ggml_tensor * get_tensor_ex( struct ggml_context * ctx, std::string name){ + + struct ggml_tensor * cur = ggml_get_tensor(ctx, name.c_str()); + if( cur == NULL ) { + fprintf(stdout, "%s: tensor '%s' not found!\n", __func__, name.c_str()); + } else { +// fprintf(stdout, "%s: n_dims = %d, name = '%s'\n", __func__, cur->n_dims, cur->name); + } + + return cur; +} + +// load the model's weights from a file +bool falcon_model_load(const std::string & fname, falcon_model & model, gpt2bpe_vocab & vocab) { + printf("%s: loading model from '%s'..\n", __func__, fname.c_str()); + + model.ctx = NULL; + + struct gguf_init_params ggufparams = { + /*.no_alloc = */ false, + /*.ctx = */ &model.ctx, + }; + + auto & ggufctx = model.ggufctx; + + ggufctx = gguf_init_from_file(fname.c_str(), ggufparams); + + if (!ggufctx) { + fprintf(stderr, "%s: gguf_init_from_file() failed\n", __func__); + return false; + } + + fprintf(stdout, "%s: gguf version = %d\n", __func__, gguf_get_version(ggufctx)); + fprintf(stdout, "%s: gguf alignment = %zu\n", __func__, gguf_get_alignment(ggufctx)); + fprintf(stdout, "%s: gguf data offset = %zu\n", __func__, gguf_get_data_offset(ggufctx)); + + // print all kv + #if 0 + { + const int n_kv = gguf_get_n_kv(ggufctx); + + fprintf(stdout, "%s: n_kv: %d\n", __func__, n_kv); + + for (int i = 0; i < n_kv; ++i) { + const char * key = gguf_get_key(ggufctx, i); + + fprintf(stdout, "%s: kv[%d]: key = %s\n", __func__, i, key); + } + } + #endif + + // print some standard metadata + { + int keyidx; + + keyidx = gguf_find_key(ggufctx, "general.name"); + if (keyidx != -1) { fprintf(stdout, "%s: model name = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "general.description"); + if (keyidx != -1) { fprintf(stdout, "%s: model description = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "general.author"); + if (keyidx != -1) { fprintf(stdout, "%s: model author = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "general.license"); + if (keyidx != -1) { fprintf(stdout, "%s: model license = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "general.architecture"); + if (keyidx != -1) { fprintf(stdout, "%s: model architecture = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "general.file_type"); + if (keyidx != -1) { fprintf(stdout, "%s: model file type = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "gptneox.tensor_data_layout"); + if (keyidx != -1) { fprintf(stdout, "%s: model data layout = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "general.source.hugginface.repository"); + if (keyidx != -1) { fprintf(stdout, "%s: model source HF repo = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + } + + // check required metadata + { + int keyidx; + + // check model architecture kv + keyidx = gguf_find_key(ggufctx, "general.architecture"); + if (keyidx != -1) { + if ( strcmp(gguf_get_val_str(ggufctx, keyidx), "falcon") != 0) { + fprintf(stdout, "%s: model architecture not supported!\n", __func__); + return false; + } + } else { + fprintf(stdout, "%s: gguf model architecture not found!\n", __func__); + return false; + } + + // check model tensor data layout kv + keyidx = gguf_find_key(ggufctx, "falcon.tensor_data_layout"); + if (keyidx != -1) { + if ( strcmp(gguf_get_val_str(ggufctx, keyidx), "jploski") != 0) { + fprintf(stdout, "%s: model tensor data layout not supported!\n", __func__); + return false; + } + } else { + fprintf(stdout, "%s: gguf model tensor data layout not found!\n", __func__); + return false; + } + + } + + // load hparams + { + auto & hparams = model.hparams; + + bool ok = true; + int keyidx; + + if (ok) { keyidx = gguf_find_key(ggufctx, "falcon.context_length"); + if (keyidx != -1) { hparams.n_ctx = gguf_get_val_u32(ggufctx, keyidx); } else { ok = false; } } + + if (ok) { keyidx = gguf_find_key(ggufctx, "falcon.embedding_length"); + if (keyidx != -1) { hparams.n_embd = gguf_get_val_u32(ggufctx, keyidx); } else { ok = false; } } + + if (ok) { keyidx = gguf_find_key(ggufctx, "falcon.attention.head_count"); + if (keyidx != -1) { hparams.n_head = gguf_get_val_u32(ggufctx, keyidx); } else { ok = false; } } + + if (ok) { keyidx = gguf_find_key(ggufctx, "falcon.feed_forward_length"); + if (keyidx != -1) { hparams.n_ff = gguf_get_val_u32(ggufctx, keyidx); } else { ok = false; } } + + if (ok) { keyidx = gguf_find_key(ggufctx, "falcon.block_count"); + if (keyidx != -1) { hparams.n_block = gguf_get_val_u32(ggufctx, keyidx); } else { ok = false; } } + + if (ok) { keyidx = gguf_find_key(ggufctx, "falcon.attention.layer_norm_epsilon"); + if (keyidx != -1) { hparams.norm_eps= gguf_get_val_f32(ggufctx, keyidx); } else { ok = false; } } + + if (!ok) { + fprintf(stderr, "%s: required hparam missing!\n", __func__); + return false; + } + + keyidx = gguf_find_key(ggufctx, "falcon.attention.head_count_kv"); + if (keyidx != -1) { hparams.n_head_kv = gguf_get_val_u32(ggufctx, keyidx); } + + + printf("%s: n_ctx = %d\n", __func__, hparams.n_ctx); + printf("%s: n_embd = %d\n", __func__, hparams.n_embd); + printf("%s: n_head = %d\n", __func__, hparams.n_head); + printf("%s: n_head_kv = %d\n", __func__, hparams.n_head_kv); + printf("%s: n_block = %d\n", __func__, hparams.n_block); + printf("%s: norm_eps = %g\n", __func__, hparams.norm_eps); + + } + + // load vocab + { + auto & hparams = model.hparams; + + int keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.model"); + + if (keyidx != -1) { + if ( strcmp(gguf_get_val_str(ggufctx, keyidx), "gpt2") != 0) { + fprintf(stdout, "%s: tokenizer model not supported!\n", __func__); + return false; + } + } else { + fprintf(stdout, "%s: tokenizer model not found!\n", __func__); + return false; + } + + + int tokens_keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.tokens"); + + if (tokens_keyidx == -1) { + fprintf(stdout, "%s: gpt2 tokenizer vocab not found!\n", __func__); + return false; + } + + int merges_keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.merges"); + + if (merges_keyidx == -1) { + fprintf(stdout, "%s: gpt2 tokenizer merges not found!\n", __func__); + return false; + } + + hparams.n_vocab = gguf_get_arr_n(ggufctx,tokens_keyidx); + hparams.n_merges = gguf_get_arr_n(ggufctx,merges_keyidx); + + fprintf(stdout, "%s: gpt2 tokenizer vocab = %zu\n", __func__, hparams.n_vocab); + fprintf(stdout, "%s: gpt2 tokenizer merges = %zu\n", __func__, hparams.n_merges); + + for (size_t i = 0; i < hparams.n_vocab; i++) { + std::string word = gguf_get_arr_str(ggufctx, tokens_keyidx, i); + +// printf("token %d = '%s'\n",i,word.c_str() ); + + vocab.token_to_id[word] = i; + vocab.id_to_token[i] = word; + + if( vocab.id_to_token[i] == "\n" ) { + vocab.linefeed_id = i; + } + } + + std::vector<std::pair<std::string, std::string>> bpe_merges; + + for (size_t i = 0; i < hparams.n_merges; i++) { + + std::string word = gguf_get_arr_str(ggufctx, merges_keyidx, i); + + // Split the merges + std::string first, second; + size_t pos = word.find(' ', 1); // Start the search from the second character + if (pos != std::string::npos) { + first = word.substr(0, pos); + second = word.substr(pos + 1); + } + + bpe_merges.push_back(std::make_pair(first, second)); + } + + vocab.populate_bpe_ranks(bpe_merges); + + + keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.bos_token_id"); if( keyidx != -1 ) { vocab.special_bos_id = (int32_t)gguf_get_val_u32(ggufctx, keyidx); } + keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.eos_token_id"); if( keyidx != -1 ) { vocab.special_eos_id = (int32_t)gguf_get_val_u32(ggufctx, keyidx); } + keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.unknown_token_id"); if( keyidx != -1 ) { vocab.special_unk_id = (int32_t)gguf_get_val_u32(ggufctx, keyidx); } + keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.separator_token_id"); if( keyidx != -1 ) { vocab.special_sep_id = (int32_t)gguf_get_val_u32(ggufctx, keyidx); } + keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.padding_token_id"); if( keyidx != -1 ) { vocab.special_pad_id = (int32_t)gguf_get_val_u32(ggufctx, keyidx); } + + if( vocab.special_bos_id != -1 ) { fprintf(stdout, "%s: BOS token = %d '%s'\n", __func__, vocab.special_bos_id, vocab.id_to_token[vocab.special_bos_id].c_str() ); } + if( vocab.special_eos_id != -1 ) { fprintf(stdout, "%s: EOS token = %d '%s'\n", __func__, vocab.special_eos_id, vocab.id_to_token[vocab.special_eos_id].c_str() ); } + if( vocab.special_unk_id != -1 ) { fprintf(stdout, "%s: UNK token = %d '%s'\n", __func__, vocab.special_unk_id, vocab.id_to_token[vocab.special_unk_id].c_str() ); } + if( vocab.special_sep_id != -1 ) { fprintf(stdout, "%s: SEP token = %d '%s'\n", __func__, vocab.special_sep_id, vocab.id_to_token[vocab.special_sep_id].c_str() ); } + if( vocab.special_pad_id != -1 ) { fprintf(stdout, "%s: PAD token = %d '%s'\n", __func__, vocab.special_pad_id, vocab.id_to_token[vocab.special_pad_id].c_str() ); } + if( vocab.linefeed_id != -1 ) { fprintf(stdout, "%s: LF token = %d\n", __func__, vocab.linefeed_id ); } + + } + + + auto & ctx = model.ctx; + size_t ctx_size = ggml_get_mem_size(ctx); + + printf("%s: ggml ctx size = %6.2f MB\n", __func__, ctx_size/(1024.0*1024.0)); + + // print tensor info + #if 0 + { + const int n_tensors = gguf_get_n_tensors(ggufctx); + + fprintf(stdout, "%s: n_tensors: %d\n", __func__, n_tensors); + + for (int i = 0; i < n_tensors; ++i) { + const char * name = gguf_get_tensor_name (ggufctx, i); + const size_t offset = gguf_get_tensor_offset(ggufctx, i); + + fprintf(stdout, "%s: tensor[%d]: name = %s, offset = %zu\n", __func__, i, name, offset); + } + } + #endif + + // prepare memory for the weights + { + + auto & hparams = model.hparams; + + const int n_block = hparams.n_block; + + model.blocks.resize(n_block); + + model.tok_embeddings = ggml_get_tensor(ctx, "token_embd.weight"); + + model.output_norm = ggml_get_tensor(ctx, "output_norm.weight"); + model.output_norm_b = ggml_get_tensor(ctx, "output_norm.bias"); + model.lm_head = ggml_get_tensor(ctx, "output.weight"); + + // map by name + model.tensors["token_embd.weight"] = model.tok_embeddings; + model.tensors["output_norm.weight"] = model.output_norm; + model.tensors["output_norm.bias"] = model.output_norm_b; + model.tensors["output.weight"] = model.lm_head; + + for (int i = 0; i < n_block; ++i) { + + auto& block = model.blocks[i]; + std::string blocknamestart = "blk." + std::to_string(i) + "."; + + block.input_layernorm = get_tensor_ex(ctx, blocknamestart + "attn_norm.weight" ); + block.input_layernorm_b = get_tensor_ex(ctx, blocknamestart + "attn_norm.bias" ); + + if ( hparams.n_head_kv == 8 ) { // Falcon-40B + block.attention_norm = get_tensor_ex(ctx, blocknamestart + "attn_norm_2.weight" ); + block.attention_norm_b = get_tensor_ex(ctx, blocknamestart + "attn_norm_2.bias" ); + } + + // query_key_value shape for config.multi_query == True: + block.query_key_value = get_tensor_ex(ctx, blocknamestart + "attn_qkv.weight" ); + block.wo = get_tensor_ex(ctx, blocknamestart + "attn_output.weight" ); + + block.ffn_up = get_tensor_ex(ctx, blocknamestart + "ffn_up.weight" ); + block.ffn_down = get_tensor_ex(ctx, blocknamestart + "ffn_down.weight" ); + + // map by name + if ( hparams.n_head_kv == 8 ) { // Falcon-40B + // Falcon-40B: + model.tensors[blocknamestart + "attn_norm.weight"] = block.input_layernorm; + model.tensors[blocknamestart + "attn_norm.bias"] = block.input_layernorm_b; + model.tensors[blocknamestart + "attn_norm_2.weight"] = block.attention_norm; + model.tensors[blocknamestart + "attn_norm_2.bias"] = block.attention_norm_b; + } else { + // Falcon-7B: + model.tensors[blocknamestart + "attn_norm.weight"] = block.input_layernorm; + model.tensors[blocknamestart + "attn_norm.bias"] = block.input_layernorm_b; + } + + model.tensors[blocknamestart + "attn_qkv.weight"] = block.query_key_value; + model.tensors[blocknamestart + "attn_output.weight"] = block.wo; + + model.tensors[blocknamestart + "ffn_up.weight"] = block.ffn_up; + model.tensors[blocknamestart + "ffn_down.weight"] = block.ffn_down; + } + } + + // key + value memory + { + const auto & kvctx = model.kvctx; + const auto & hparams = model.hparams; + + const int n_block = hparams.n_block; + const int n_ctx = hparams.n_ctx; + const int n_embd = hparams.n_embd; + + const int64_t n_mem = n_block*n_ctx; + const int64_t n_elements = n_embd*n_mem; + + // create the ggml context + { + struct ggml_init_params params = { + /*.mem_size =*/ size_t(n_elements*4+ggml_tensor_overhead()*2), + /*.mem_buffer =*/ NULL, + /*.no_alloc =*/ false, + }; + + model.kvctx = ggml_init(params); + if (!model.kvctx) { + fprintf(stderr, "%s: kv ggml_init() failed\n", __func__); + return false; + } + + } + + + model.memory_k = ggml_new_tensor_1d(kvctx, GGML_TYPE_F16, n_elements); + model.memory_v = ggml_new_tensor_1d(kvctx, GGML_TYPE_F16, n_elements); + + const size_t memory_size = ggml_nbytes(model.memory_k) + ggml_nbytes(model.memory_v); + + printf("%s: memory_size = %8.2f MB, n_mem = %" PRId64 "\n", __func__, memory_size/1024.0/1024.0, n_mem); + } + + return true; +} + + +// evaluate the transformer +// +// - model: the model +// - n_threads: number of threads to use +// - n_past: the context size so far +// - embd_inp: the embeddings of the tokens in the context +// - embd_w: the predicted logits for the next token +// +bool falcon_eval( + const falcon_model & model, + const int n_threads, + const int n_past, + const std::vector<gpt2bpe_vocab::id> & embd_inp, + std::vector<float> & embd_w, + size_t & mem_per_token) { + + + const int N = embd_inp.size(); + + const auto & hparams = model.hparams; + + const int n_embd = hparams.n_embd; + const int n_block = hparams.n_block; + const int n_ctx = hparams.n_ctx; + const int n_head = hparams.n_head; + const int n_head_kv = hparams.n_head_kv; + const int n_vocab = hparams.n_vocab; + const size_t head_dim = n_embd / n_head; + + static size_t buf_size = 256u*1024*1024; + static void * buf = malloc(buf_size); + + // use 2 scratch buffers + // TODO: very hacky solution - reimplement in a more elegant way + static size_t scr0_size = 256u*1024*1024; + static void * scr0 = malloc(scr0_size); + + static size_t scr1_size = 256u*1024*1024; + static void * scr1 = malloc(scr1_size); + + if (mem_per_token > 0 && mem_per_token*N > buf_size) { + const size_t buf_size_new = 1.1*(mem_per_token*N); // add 10% to account for ggml object overhead + //printf("\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_size, buf_size_new); + + // reallocate + buf_size = buf_size_new; + buf = realloc(buf, buf_size); + if (buf == nullptr) { + fprintf(stderr, "%s: failed to allocate %zu bytes\n", __func__, buf_size); + return false; + } + } + + struct ggml_init_params params = { + /*.mem_size =*/ buf_size, + /*.mem_buffer =*/ buf, + /*.no_alloc =*/ false, + }; + + struct ggml_context * ctx0 = ggml_init(params); + struct ggml_cgraph gf = {}; +// gf.n_threads = n_threads; + + struct ggml_tensor * embd = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, N); + memcpy(embd->data, embd_inp.data(), N*ggml_element_size(embd)); + + // wte + struct ggml_tensor * inpL = ggml_get_rows(ctx0, model.tok_embeddings, embd); +// struct ggml_tensor* repeat_dummy = ggml_new_tensor_3d(ctx0, inpL->type, head_dim, N + n_past, n_head); + + ggml_type wtype = GGML_TYPE_F32; + const int sizeof_wtype = ggml_type_sizef(wtype); + + for (int il = 0; il < n_block; ++il) { + struct ggml_tensor * cur; + struct ggml_tensor * layernorm_output; + + ggml_set_scratch(ctx0, { 0, scr0_size, scr0, }); + + // self-attention + { + layernorm_output = ggml_norm(ctx0, inpL); + + layernorm_output = ggml_add(ctx0, + ggml_mul(ctx0, + ggml_repeat(ctx0, model.blocks[il].input_layernorm, layernorm_output), + layernorm_output), + ggml_repeat(ctx0, model.blocks[il].input_layernorm_b, layernorm_output)); + + if ( hparams.n_head_kv == 8 ) { // Falcon-40B + cur = ggml_norm(ctx0, inpL); + + cur = ggml_add(ctx0, + ggml_mul(ctx0, + ggml_repeat(ctx0, model.blocks[il].attention_norm, cur), + cur), + ggml_repeat(ctx0, model.blocks[il].attention_norm_b, cur)); + } + else { // Falcon 7B + cur = layernorm_output; + } + + // compute QKV + + cur = ggml_mul_mat(ctx0, model.blocks[il].query_key_value, cur); + + // Note that the strides for Kcur, Vcur are set up so that the + // resulting views are misaligned with the tensor's storage + // (by applying the K/V offset we shift the tensor's original + // view to stick out behind the viewed QKV tensor's allocated + // memory, so to say). This is ok because no actual accesses + // happen to that out-of-range memory, but it can require some + // trickery when trying to accurately dump these views for + // debugging. + + struct ggml_tensor * Qcur = ggml_view_3d( + ctx0, cur, head_dim, n_head, N, + head_dim * sizeof_wtype, + head_dim * (n_head + 2 * n_head_kv) * sizeof_wtype, + 0); + + struct ggml_tensor * Kcur = ggml_view_3d( + ctx0, cur, head_dim, n_head_kv, N, + head_dim * sizeof_wtype, + head_dim * (n_head + 2 * n_head_kv) * sizeof_wtype, + head_dim * n_head * sizeof_wtype); + + struct ggml_tensor * Vcur = ggml_view_3d( + ctx0, cur, head_dim, n_head_kv, N, + head_dim * sizeof_wtype, + head_dim * (n_head + 2 * n_head_kv) * sizeof_wtype, + head_dim * (n_head + n_head_kv) * sizeof_wtype); + + // using mode = 2 for neox mode + Qcur = ggml_rope_inplace(ctx0, Qcur, n_past, head_dim, 2, 0); + Kcur = ggml_rope_inplace(ctx0, Kcur, n_past, head_dim, 2, 0); + + // store key and value to memory + { + struct ggml_tensor* k = ggml_view_1d( + ctx0, model.memory_k, N * n_head_kv * head_dim, + (ggml_element_size(model.memory_k) * n_head_kv * head_dim) * + (il * n_ctx + n_past)); + struct ggml_tensor* v = ggml_view_1d( + ctx0, model.memory_v, N * n_head_kv * head_dim, + (ggml_element_size(model.memory_v) * n_head_kv * head_dim) * + (il * n_ctx + n_past)); + + ggml_build_forward_expand(&gf, ggml_cpy(ctx0, Kcur, k)); + ggml_build_forward_expand(&gf, ggml_cpy(ctx0, Vcur, v)); + } + + struct ggml_tensor * K = ggml_permute( + ctx0, + ggml_reshape_3d( + ctx0, + ggml_view_1d(ctx0, model.memory_k, (n_past + N) * n_head_kv * head_dim, + il * n_ctx * + ggml_element_size(model.memory_k) * + n_head_kv * + head_dim), + head_dim, n_head_kv, n_past + N), + 0, 2, 1, 3); + + // K * Q + +// K = ggml_cont(ctx0, ggml_repeat2(ctx0, K, repeat_dummy)); + + struct ggml_tensor * Q = ggml_permute(ctx0, Qcur, 0, 2, 1, 3); + struct ggml_tensor * KQ = ggml_mul_mat(ctx0, K, Q); + + // KQ_scaled = KQ / sqrt(n_embd/n_head) + struct ggml_tensor * KQ_scaled = + ggml_scale_inplace(ctx0, + KQ, + ggml_new_f32(ctx0, 1.0f/sqrt(float(head_dim))) + ); + + // KQ_masked = mask_past(KQ_scaled) + struct ggml_tensor * KQ_masked = ggml_diag_mask_inf_inplace(ctx0, KQ_scaled, n_past); + + // KQ = soft_max(KQ_masked) + struct ggml_tensor * KQ_soft_max = ggml_soft_max_inplace(ctx0, KQ_masked); + + // V_trans = Vmem.view(n_embd/n_head, n_head, n_past + N).permute(1, 2, 0, 3).contiguous() + struct ggml_tensor* V = ggml_permute( + ctx0, + ggml_reshape_3d( + ctx0, + ggml_view_1d(ctx0, model.memory_v, (n_past + N) * n_head_kv * head_dim, + il * n_ctx * + ggml_element_size(model.memory_v) * + n_head_kv * + head_dim), + head_dim, n_head_kv, n_past + N), + 0, 2, 1, 3); + +// V = ggml_cont(ctx0, ggml_transpose(ctx0, ggml_repeat2(ctx0, V, repeat_dummy))); + V = ggml_cont(ctx0, ggml_transpose(ctx0, V)); + + // KQV = transpose(V) * KQ_soft_max + struct ggml_tensor * KQV = ggml_mul_mat(ctx0, V, KQ_soft_max); + + // KQV_merged = KQV.permute(0, 2, 1, 3) + struct ggml_tensor * KQV_merged = ggml_permute(ctx0, KQV, 0, 2, 1, 3); + + // cur = KQV_merged.contiguous().view(n_embd, N) + cur = ggml_cpy(ctx0, + KQV_merged, + ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_embd, N)); + + // projection + { + cur = ggml_mul_mat(ctx0, + model.blocks[il].wo, + cur); + } + } + + ggml_set_scratch(ctx0, { 0, scr1_size, scr1, }); + + struct ggml_tensor* inpFF = layernorm_output; + struct ggml_tensor* attn_out = ggml_cpy( + ctx0, cur, ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_embd, N)); + + { + cur = ggml_mul_mat(ctx0, model.blocks[il].ffn_up, inpFF); + cur = ggml_gelu(ctx0, cur); + cur = ggml_mul_mat(ctx0, model.blocks[il].ffn_down, cur); + } + + cur = ggml_add(ctx0, cur, attn_out); + cur = ggml_add(ctx0, cur, inpL); + // input for next layer + inpL = cur; + } + + ggml_set_scratch(ctx0, { 0, scr0_size, scr0, }); + + // norm + { + inpL = ggml_norm(ctx0, inpL); + + // inpL = ln_f_g*inpL + ln_f_b + inpL = ggml_add(ctx0, + ggml_mul(ctx0, + ggml_repeat(ctx0, model.output_norm, inpL), + inpL), + ggml_repeat(ctx0, model.output_norm_b, inpL)); + } + + ggml_set_scratch(ctx0, { 0, 0, nullptr, }); + + // lm_head + { + inpL = ggml_mul_mat(ctx0, model.lm_head, inpL); + + //inpL = ggml_add(ctx0, + // ggml_repeat(ctx0, model.lmh_b, inpL), + // inpL); + } + + // logits -> probs + //inpL = ggml_soft_max_inplace(ctx0, inpL); + + // run the computation + ggml_build_forward_expand(&gf, inpL); +// ggml_graph_compute (ctx0, &gf); + ggml_graph_compute_with_ctx(ctx0, &gf, n_threads); + + //if (n_past%100 == 0) { + // ggml_graph_print (&gf); + // ggml_graph_dump_dot(&gf, NULL, "gpt-2.dot"); + //} + + // return result for just the last token + embd_w.resize(n_vocab); + memcpy(embd_w.data(), (float *)ggml_get_data(inpL) + (n_vocab * (N - 1)), sizeof(float) * n_vocab); + + if (mem_per_token == 0) { + mem_per_token = ggml_used_mem(ctx0)/N; + } + //printf("used_mem = %zu\n", ggml_used_mem(ctx0)); + + ggml_free(ctx0); + + return true; +} + +int main(int argc, char ** argv) { + ggml_time_init(); + + const int64_t t_main_start_us = ggml_time_us(); + + gpt_params params; + + if (gpt_params_parse(argc, argv, params) == false) { + return 1; + } + + int64_t t_load_us = 0; + + gpt2bpe_vocab vocab; + falcon_model model; + + // load the model + { + const int64_t t_start_us = ggml_time_us(); + + if (!falcon_model_load(params.model, model, vocab)) { + fprintf(stderr, "%s: failed to load model from '%s'\n", __func__, params.model.c_str()); + return 1; + } + + t_load_us = ggml_time_us() - t_start_us; + + } + + if (params.seed < 0) { + params.seed = time(NULL); + } + + if (params.top_k == 0) { + params.top_k = model.hparams.n_vocab; + } + + printf("%s: seed = %d\n", __func__, params.seed); + printf("%s: temp = %.3f\n", __func__, params.temp); + printf("%s: top_k = %d\n", __func__, params.top_k); + printf("%s: top_p = %.3f\n", __func__, params.top_p); + printf("%s: repeat_last_n = %d\n", __func__, params.repeat_last_n); + printf("%s: repeat_penalty = %.3f\n", __func__, params.repeat_penalty); + + std::mt19937 rng(params.seed); + + if (params.prompt.empty()) { + params.prompt = "Once upon"; + } + + std::vector<int32_t> last_n_tokens(model.hparams.n_ctx); + std::fill(last_n_tokens.begin(), last_n_tokens.end(), 0); + + int n_past = 0; + + int64_t t_sample_us = 0; + int64_t t_predict_us = 0; + + std::vector<float> logits; + + // tokenize the prompt + std::vector<gpt2bpe_vocab::id> embd_inp = gpt2bpe_tokenize(vocab, params.prompt,false, false); + + params.n_predict = std::min(params.n_predict, model.hparams.n_ctx - (int) embd_inp.size()); + + printf("%s: number of tokens in prompt = %zu\n", __func__, embd_inp.size()); +// for (size_t i = 0; i < embd_inp.size(); i++) { +// printf("%s: token[%zu] = %6d, %s\n", __func__, i, embd_inp[i], vocab.id_to_token[embd_inp[i]].c_str()); +// } + + if( model.hparams.n_ctx < params.n_predict+embd_inp.size() ) { + params.n_predict = model.hparams.n_ctx-embd_inp.size(); + } + + printf("%s: n_predict = %d\n", __func__, params.n_predict); + printf("\n"); + + std::vector<gpt2bpe_vocab::id> embd; + + // determine the required inference memory per token: + size_t mem_per_token = 0; + falcon_eval(model, params.n_threads, 0, { 0, 1, 2, 3 }, logits, mem_per_token); + + for (size_t i = embd.size(); i < embd_inp.size() + params.n_predict; i++) { + // predict + if (embd.size() > 0) { + const int64_t t_start_us = ggml_time_us(); + + if (!falcon_eval(model, params.n_threads, n_past, embd, logits, mem_per_token)) { + printf("Failed to predict\n"); + return 1; + } + + t_predict_us += ggml_time_us() - t_start_us; + } + + n_past += embd.size(); + embd.clear(); + + if (i >= embd_inp.size()) { + // sample next token + const int top_k = params.top_k; + const float top_p = params.top_p; + const float temp = params.temp; + const int repeat_last_n = params.repeat_last_n; + const float repeat_penalty = params.repeat_penalty; + + const int n_vocab = model.hparams.n_vocab; + + gpt2bpe_vocab::id id = 0; + + { + const int64_t t_start_sample_us = ggml_time_us(); + + id = sample_top_k_top_p_repeat(vocab, logits.data() + (logits.size() - n_vocab), last_n_tokens.data(), last_n_tokens.size(), top_k, top_p, temp, repeat_last_n, repeat_penalty, rng); + + last_n_tokens.erase(last_n_tokens.begin()); + last_n_tokens.push_back(id); + + t_sample_us += ggml_time_us() - t_start_sample_us; + } + + // add it to the context + embd.push_back(id); + } else { + // if here, it means we are still processing the input prompt + for (size_t k = i; k < embd_inp.size(); k++) { + embd.push_back(embd_inp[k]); + if (embd.size() > params.n_batch) { + break; + } + } + i += embd.size() - 1; + } + + // display text + for (auto id : embd) { + printf("%s", vocab.id_to_token[id].c_str() ); + } + fflush(stdout); + + // end of text token + if (vocab.special_eos_id != -1 && embd.back() == vocab.special_eos_id) { + break; + } + } + + // report timing + { + const int64_t t_main_end_us = ggml_time_us(); + + printf("\n\n"); + printf("%s: mem per token = %8zu bytes\n", __func__, mem_per_token); + printf("%s: load time = %8.2f ms\n", __func__, t_load_us/1000.0f); + printf("%s: sample time = %8.2f ms\n", __func__, t_sample_us/1000.0f); + printf("%s: predict time = %8.2f ms / %.2f ms per token\n", __func__, t_predict_us/1000.0f, t_predict_us/1000.0f/n_past); + printf("%s: total time = %8.2f ms\n", __func__, (t_main_end_us - t_main_start_us)/1000.0f); + } + + ggml_free(model.ctx); + + return 0; +} diff --git a/examples/gptneox-wip/gptneox-main.cpp b/examples/gptneox-wip/gptneox-main.cpp new file mode 100644 index 00000000..04af5024 --- /dev/null +++ b/examples/gptneox-wip/gptneox-main.cpp @@ -0,0 +1,1082 @@ +#include "ggml.h" +#include "cmpnct_gpt2bpe.hpp" + +#include <cassert> +#include <cmath> +#include <cstdio> +#include <cstring> +#include <cinttypes> +#include <fstream> +#include <map> +#include <string> +#include <vector> +#include <thread> +#include <random> + +#if defined(_MSC_VER) +#pragma warning(disable: 4244 4267) // possible loss of data +#endif + +// default hparams +struct gpt_neox_hparams { + size_t n_merges = 0; + size_t n_vocab = 0; + uint32_t n_ctx = 0; + uint32_t n_embd = 0; + uint32_t n_head = 0; + uint32_t n_block = 0; + uint32_t n_rot = 0; // rotary_pct * (n_embd / n_head) + bool par_res = true; + float norm_eps = 1e-5; +}; + +struct gpt_neox_block { + // pre normalization + struct ggml_tensor * ln_1_g; + struct ggml_tensor * ln_1_b; + + // attention + struct ggml_tensor * c_attn_attn_w; + struct ggml_tensor * c_attn_attn_b; + + struct ggml_tensor * c_attn_proj_w; + struct ggml_tensor * c_attn_proj_b; + + // post normalization + struct ggml_tensor * ln_2_g; + struct ggml_tensor * ln_2_b; + + // ff + struct ggml_tensor * c_mlp_fc_w; + struct ggml_tensor * c_mlp_fc_b; + + struct ggml_tensor * c_mlp_proj_w; + struct ggml_tensor * c_mlp_proj_b; +}; + +struct gpt_neox_model { + gpt_neox_hparams hparams; + + // normalization + struct ggml_tensor * ln_f_g; + struct ggml_tensor * ln_f_b; + + struct ggml_tensor * wte; // position embedding + + struct ggml_tensor * lmh_g; // language model head + + std::vector<gpt_neox_block> blocks; + + // key + value memory + struct ggml_tensor * memory_k; + struct ggml_tensor * memory_v; + + // + struct gguf_context * ggufctx; + struct ggml_context * ctx; + struct ggml_context * kvctx; + + std::map<std::string, struct ggml_tensor *> tensors; +}; + +struct gpt_params { + int32_t seed = -1; // RNG seed + int32_t n_threads = std::min(4, (int32_t) std::thread::hardware_concurrency()); + uint32_t n_predict = 200; // new tokens to predict + uint32_t n_batch = 512; // batch size for prompt processing + + // sampling parameters + int32_t top_k = 40; + float top_p = 1.0f; + float temp = 0.8f; + int32_t repeat_last_n = 64; + float repeat_penalty = 1.02f; + + std::string model = ""; // model path + std::string prompt = ""; + + std::string token_test = ""; + bool interactive = false; + int32_t interactive_port = -1; + int32_t n_gpu_layers = 0; +}; + +void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) { + fprintf(stderr, "usage: %s [options]\n", argv[0]); + fprintf(stderr, "\n"); + fprintf(stderr, "options:\n"); + fprintf(stderr, " -h, --help show this help message and exit\n"); + fprintf(stderr, " -s SEED, --seed SEED RNG seed (default: -1)\n"); + fprintf(stderr, " -t N, --threads N number of threads to use during computation (default: %d)\n", params.n_threads); + fprintf(stderr, " -ngl N, --gpu-layers N number of layers to offload to GPU on supported models (default: %d)\n", params.n_gpu_layers); + fprintf(stderr, " -p PROMPT, --prompt PROMPT\n"); + fprintf(stderr, " prompt to start generation with (default: random)\n"); + fprintf(stderr, " -f FNAME, --file FNAME\n"); + fprintf(stderr, " load prompt from a file\n"); + fprintf(stderr, " -tt TOKEN_TEST, --token_test TOKEN_TEST\n"); + fprintf(stderr, " test tokenization\n"); + fprintf(stderr, " -n N, --n_predict N number of tokens to predict (default: %d)\n", params.n_predict); + fprintf(stderr, " --top_k N top-k sampling, 0 = n_vocab (default: %d)\n", params.top_k); + fprintf(stderr, " --top_p N top-p sampling (default: %.1f)\n", params.top_p); + fprintf(stderr, " --temp N temperature (default: %.1f)\n", params.temp); + fprintf(stderr, " --repeat-last-n N last n tokens to consider for penalize (default: %d, 0 = disabled)\n", params.repeat_last_n); + fprintf(stderr, " --repeat-penalty N penalize repeat sequence of tokens (default: %.2f, 1.0 = disabled)\n", (double)params.repeat_penalty); + fprintf(stderr, " -b N, --batch_size N batch size for prompt processing (default: %d)\n", params.n_batch); + fprintf(stderr, " -m FNAME, --model FNAME\n"); + fprintf(stderr, " model path (default: %s)\n", params.model.c_str()); + fprintf(stderr, "\n"); +} + +// Function to check if the next argument exists +std::string get_next_arg(int& i, int argc, char** argv, const std::string& flag, gpt_params& params) { + if (i + 1 < argc && argv[i + 1][0] != '-') { + return argv[++i]; + } else { + fprintf(stderr, "error: %s requires one argument.\n", flag.c_str()); + gpt_print_usage(argc, argv, params); + exit(0); + } +} + +bool gpt_params_parse(int argc, char ** argv, gpt_params & params) { + for (int i = 1; i < argc; i++) { + std::string arg = argv[i]; + + if (arg == "-s" || arg == "--seed") { + params.seed = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "-t" || arg == "--threads") { + params.n_threads = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "-ngl" || arg == "--gpu-layers" || arg == "--n-gpu-layers") { + params.n_gpu_layers = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "-p" || arg == "--prompt") { + params.prompt = get_next_arg(i, argc, argv, arg, params); + } else if (arg == "-n" || arg == "--n_predict") { + params.n_predict = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "--top_k") { + params.top_k = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "--top_p") { + params.top_p = std::stof(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "--temp") { + params.temp = std::stof(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "--repeat-last-n") { + params.repeat_last_n = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "--repeat-penalty") { + params.repeat_penalty = std::stof(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "-b" || arg == "--batch_size") { + params.n_batch= std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "-m" || arg == "--model") { + params.model = get_next_arg(i, argc, argv, arg, params); + } else if (arg == "-i" || arg == "--interactive") { + params.interactive = true; + } else if (arg == "-ip" || arg == "--interactive-port") { + params.interactive = true; + params.interactive_port = std::stoi(get_next_arg(i, argc, argv, arg, params)); + } else if (arg == "-h" || arg == "--help") { + gpt_print_usage(argc, argv, params); + exit(0); + } else if (arg == "-f" || arg == "--file") { + get_next_arg(i, argc, argv, arg, params); + std::ifstream file(argv[i]); + if (!file) { + fprintf(stderr, "error: failed to open file '%s'\n", argv[i]); + break; + } + std::copy(std::istreambuf_iterator<char>(file), std::istreambuf_iterator<char>(), back_inserter(params.prompt)); + if (params.prompt.back() == '\n') { + params.prompt.pop_back(); + } + } else if (arg == "-tt" || arg == "--token_test") { + params.token_test = get_next_arg(i, argc, argv, arg, params); + } + else { + fprintf(stderr, "error: unknown argument: %s\n", arg.c_str()); + gpt_print_usage(argc, argv, params); + exit(0); + } + } + + return true; +} + +gpt2bpe_vocab::id sample_top_k_top_p_repeat( + const gpt2bpe_vocab & vocab, + const float * logits, + const int32_t * last_n_tokens_data, + size_t last_n_tokens_data_size, + int top_k, + double top_p, + double temp, + int repeat_last_n, + float repeat_penalty, + std::mt19937 & rng) { + + int n_logits = vocab.id_to_token.size(); + + const auto * plogits = logits; + + const auto last_n_tokens = std::vector<int32_t>(last_n_tokens_data, last_n_tokens_data + last_n_tokens_data_size); + + if (temp <= 0) { + // select the token with the highest logit directly + float max_logit = plogits[0]; + gpt2bpe_vocab::id max_id = 0; + + for (int i = 1; i < n_logits; ++i) { + if (plogits[i] > max_logit) { + max_logit = plogits[i]; + max_id = i; + } + } + return max_id; + } + + + std::vector<std::pair<double, gpt2bpe_vocab::id>> logits_id; + logits_id.reserve(n_logits); + + { + const float scale = 1.0f/temp; + for (int i = 0; i < n_logits; ++i) { + // repetition penalty from ctrl paper (https://arxiv.org/abs/1909.05858) + // credit https://github.com/facebookresearch/llama/compare/main...shawwn:llama:main + if (repeat_last_n > 0 && std::find(last_n_tokens.end()-repeat_last_n, last_n_tokens.end(), i) != last_n_tokens.end()) { + // if score < 0 then repetition penalty has to multiplied to reduce the previous token probability + if (plogits[i] < 0.0f) { + logits_id.push_back(std::make_pair(plogits[i]*scale*repeat_penalty, i)); + } else { + logits_id.push_back(std::make_pair(plogits[i]*scale/repeat_penalty, i)); + } + } else { + logits_id.push_back(std::make_pair(plogits[i]*scale, i)); + } + } + } + + // find the top K tokens + std::partial_sort( + logits_id.begin(), + logits_id.begin() + top_k, logits_id.end(), + [](const std::pair<double, gpt2bpe_vocab::id> & a, const std::pair<double, gpt2bpe_vocab::id> & b) { + return a.first > b.first; + }); + + logits_id.resize(top_k); + + double maxl = -INFINITY; + for (const auto & kv : logits_id) { + maxl = std::max(maxl, kv.first); + } + + // compute probs for the top K tokens + std::vector<double> probs; + probs.reserve(logits_id.size()); + + double sum = 0.0; + for (const auto & kv : logits_id) { + double p = exp(kv.first - maxl); + probs.push_back(p); + sum += p; + } + + // normalize the probs + for (auto & p : probs) { + p /= sum; + } + + if (top_p < 1.0f) { + double cumsum = 0.0f; + for (int i = 0; i < top_k; i++) { + cumsum += probs[i]; + if (cumsum >= top_p) { + top_k = i + 1; + probs.resize(top_k); + logits_id.resize(top_k); + break; + } + } + + cumsum = 1.0/cumsum; + for (int i = 0; i < (int) probs.size(); i++) { + probs[i] *= cumsum; + } + } + +// printf("\n"); +// for (int i = 0; i < (int) probs.size(); i++) { +// for (int i = 0; i < 10; i++) { +// printf("%d: '%s' %f\n", i, vocab.id_to_token.at(logits_id[i].second).c_str(), probs[i]); +// } + + std::discrete_distribution<> dist(probs.begin(), probs.end()); + int idx = dist(rng); + + return logits_id[idx].second; + +} + +struct ggml_tensor * get_tensor_ex( struct ggml_context * ctx, std::string name){ + + struct ggml_tensor * cur = ggml_get_tensor(ctx, name.c_str()); + if( cur == NULL ) { + fprintf(stdout, "%s: tensor '%s' not found!\n", __func__, name.c_str()); + } else { +// fprintf(stdout, "%s: n_dims = %d, name = '%s'\n", __func__, cur->n_dims, cur->name); + } + + return cur; +} + +// load the model's weights from a file +bool gpt_neox_model_load(const std::string & fname, gpt_neox_model & model, gpt2bpe_vocab & vocab) { + printf("%s: loading model from '%s'..\n", __func__, fname.c_str()); + + model.ctx = NULL; + + struct gguf_init_params ggufparams = { + /*.no_alloc = */ false, + /*.ctx = */ &model.ctx, + }; + + auto & ggufctx = model.ggufctx; + + ggufctx = gguf_init_from_file(fname.c_str(), ggufparams); + + if (!ggufctx) { + fprintf(stderr, "%s: gguf_init_from_file() failed\n", __func__); + return false; + } + + fprintf(stdout, "%s: gguf version = %d\n", __func__, gguf_get_version(ggufctx)); + fprintf(stdout, "%s: gguf alignment = %zu\n", __func__, gguf_get_alignment(ggufctx)); + fprintf(stdout, "%s: gguf data offset = %zu\n", __func__, gguf_get_data_offset(ggufctx)); + + // print all kv + #if 0 + { + const int n_kv = gguf_get_n_kv(ggufctx); + + fprintf(stdout, "%s: n_kv: %d\n", __func__, n_kv); + + for (int i = 0; i < n_kv; ++i) { + const char * key = gguf_get_key(ggufctx, i); + + fprintf(stdout, "%s: kv[%d]: key = %s\n", __func__, i, key); + } + } + #endif + + // print some standard metadata + { + int keyidx; + + keyidx = gguf_find_key(ggufctx, "general.name"); + if (keyidx != -1) { fprintf(stdout, "%s: model name = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "general.description"); + if (keyidx != -1) { fprintf(stdout, "%s: model description = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "general.author"); + if (keyidx != -1) { fprintf(stdout, "%s: model author = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "general.license"); + if (keyidx != -1) { fprintf(stdout, "%s: model license = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "general.architecture"); + if (keyidx != -1) { fprintf(stdout, "%s: model architecture = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "general.file_type"); + if (keyidx != -1) { fprintf(stdout, "%s: model file type = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "gptneox.tensor_data_layout"); + if (keyidx != -1) { fprintf(stdout, "%s: model data layout = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + keyidx = gguf_find_key(ggufctx, "general.source.hugginface.repository"); + if (keyidx != -1) { fprintf(stdout, "%s: model source HF repo = %s\n", __func__, gguf_get_val_str(ggufctx, keyidx)); } + } + + // check required metadata + { + int keyidx; + + // check model architecture kv + keyidx = gguf_find_key(ggufctx, "general.architecture"); + if (keyidx != -1) { + if ( strcmp(gguf_get_val_str(ggufctx, keyidx), "gptneox") != 0) { + fprintf(stdout, "%s: model architecture not supported!\n", __func__); + return false; + } + } else { + fprintf(stdout, "%s: gguf model architecture not found!\n", __func__); + return false; + } + + } + + // load hparams + { + auto & hparams = model.hparams; + + bool ok = true; + int keyidx; + + if (ok) { keyidx = gguf_find_key(ggufctx, "gptneox.context_length"); + if (keyidx != -1) { hparams.n_ctx = gguf_get_val_u32(ggufctx, keyidx); } else { ok = false; } } + + if (ok) { keyidx = gguf_find_key(ggufctx, "gptneox.embedding_length"); + if (keyidx != -1) { hparams.n_embd = gguf_get_val_u32(ggufctx, keyidx); } else { ok = false; } } + + if (ok) { keyidx = gguf_find_key(ggufctx, "gptneox.attention.head_count"); + if (keyidx != -1) { hparams.n_head = gguf_get_val_u32(ggufctx, keyidx); } else { ok = false; } } + + if (ok) { keyidx = gguf_find_key(ggufctx, "gptneox.block_count"); + if (keyidx != -1) { hparams.n_block = gguf_get_val_u32(ggufctx, keyidx); } else { ok = false; } } + + if (ok) { keyidx = gguf_find_key(ggufctx, "gptneox.rope.dimension_count"); + if (keyidx != -1) { hparams.n_rot = gguf_get_val_u32(ggufctx, keyidx); } else { ok = false; } } + + if (ok) { keyidx = gguf_find_key(ggufctx, "gptneox.use_parallel_residual"); + if (keyidx != -1) { hparams.par_res = gguf_get_val_bool(ggufctx, keyidx); } else { ok = false; } } + + if (ok) { keyidx = gguf_find_key(ggufctx, "gptneox.attention.layer_norm_epsilon"); + if (keyidx != -1) { hparams.norm_eps= gguf_get_val_f32(ggufctx, keyidx); } else { ok = false; } } + + if (!ok) { + fprintf(stderr, "%s: required hparam missing!\n", __func__); + return false; + } + + printf("%s: n_ctx = %d\n", __func__, hparams.n_ctx); + printf("%s: n_embd = %d\n", __func__, hparams.n_embd); + printf("%s: n_head = %d\n", __func__, hparams.n_head); + printf("%s: n_block = %d\n", __func__, hparams.n_block); + printf("%s: n_rot = %d\n", __func__, hparams.n_rot); + printf("%s: par_res = %d\n", __func__, hparams.par_res); + printf("%s: norm_eps = %g\n", __func__, hparams.norm_eps); + + } + + // load vocab + { + auto & hparams = model.hparams; + + int keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.model"); + + if (keyidx != -1) { + if ( strcmp(gguf_get_val_str(ggufctx, keyidx), "gpt2") != 0) { + fprintf(stdout, "%s: tokenizer model not supported!\n", __func__); + return false; + } + } else { + fprintf(stdout, "%s: tokenizer model not found!\n", __func__); + return false; + } + + + int tokens_keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.tokens"); + + if (tokens_keyidx == -1) { + fprintf(stdout, "%s: gpt2 tokenizer vocab not found!\n", __func__); + return false; + } + + int merges_keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.merges"); + + if (merges_keyidx == -1) { + fprintf(stdout, "%s: gpt2 tokenizer merges not found!\n", __func__); + return false; + } + + hparams.n_vocab = gguf_get_arr_n(ggufctx,tokens_keyidx); + hparams.n_merges = gguf_get_arr_n(ggufctx,merges_keyidx); + + fprintf(stdout, "%s: gpt2 tokenizer vocab = %zu\n", __func__, hparams.n_vocab); + fprintf(stdout, "%s: gpt2 tokenizer merges = %zu\n", __func__, hparams.n_merges); + + for (size_t i = 0; i < hparams.n_vocab; i++) { + std::string word = gguf_get_arr_str(ggufctx, tokens_keyidx, i); + +// printf("token %d = '%s'\n",i,word.c_str() ); + + vocab.token_to_id[word] = i; + vocab.id_to_token[i] = word; + + if( vocab.id_to_token[i] == "\n" ) { + vocab.linefeed_id = i; + } + } + + std::vector<std::pair<std::string, std::string>> bpe_merges; + + for (size_t i = 0; i < hparams.n_merges; i++) { + + std::string word = gguf_get_arr_str(ggufctx, merges_keyidx, i); + + // Split the merges + std::string first, second; + size_t pos = word.find(' ', 1); // Start the search from the second character + if (pos != std::string::npos) { + first = word.substr(0, pos); + second = word.substr(pos + 1); + } + + bpe_merges.push_back(std::make_pair(first, second)); + } + + vocab.populate_bpe_ranks(bpe_merges); + + + keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.bos_token_id"); if( keyidx != -1 ) { vocab.special_bos_id = (int32_t)gguf_get_val_u32(ggufctx, keyidx); } + keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.eos_token_id"); if( keyidx != -1 ) { vocab.special_eos_id = (int32_t)gguf_get_val_u32(ggufctx, keyidx); } + keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.unknown_token_id"); if( keyidx != -1 ) { vocab.special_unk_id = (int32_t)gguf_get_val_u32(ggufctx, keyidx); } + keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.separator_token_id"); if( keyidx != -1 ) { vocab.special_sep_id = (int32_t)gguf_get_val_u32(ggufctx, keyidx); } + keyidx = gguf_find_key(ggufctx, "tokenizer.ggml.padding_token_id"); if( keyidx != -1 ) { vocab.special_pad_id = (int32_t)gguf_get_val_u32(ggufctx, keyidx); } + + if( vocab.special_bos_id != -1 ) { fprintf(stdout, "%s: BOS token = %d '%s'\n", __func__, vocab.special_bos_id, vocab.id_to_token[vocab.special_bos_id].c_str() ); } + if( vocab.special_eos_id != -1 ) { fprintf(stdout, "%s: EOS token = %d '%s'\n", __func__, vocab.special_eos_id, vocab.id_to_token[vocab.special_eos_id].c_str() ); } + if( vocab.special_unk_id != -1 ) { fprintf(stdout, "%s: UNK token = %d '%s'\n", __func__, vocab.special_unk_id, vocab.id_to_token[vocab.special_unk_id].c_str() ); } + if( vocab.special_sep_id != -1 ) { fprintf(stdout, "%s: SEP token = %d '%s'\n", __func__, vocab.special_sep_id, vocab.id_to_token[vocab.special_sep_id].c_str() ); } + if( vocab.special_pad_id != -1 ) { fprintf(stdout, "%s: PAD token = %d '%s'\n", __func__, vocab.special_pad_id, vocab.id_to_token[vocab.special_pad_id].c_str() ); } + if( vocab.linefeed_id != -1 ) { fprintf(stdout, "%s: LF token = %d\n", __func__, vocab.linefeed_id ); } + } + + + auto & ctx = model.ctx; + size_t ctx_size = ggml_get_mem_size(ctx); + + printf("%s: ggml ctx size = %6.2f MB\n", __func__, ctx_size/(1024.0*1024.0)); + + // print tensor info + #if 0 + { + const int n_tensors = gguf_get_n_tensors(ggufctx); + + fprintf(stdout, "%s: n_tensors: %d\n", __func__, n_tensors); + + for (int i = 0; i < n_tensors; ++i) { + const char * name = gguf_get_tensor_name (ggufctx, i); + const size_t offset = gguf_get_tensor_offset(ggufctx, i); + + fprintf(stdout, "%s: tensor[%d]: name = %s, offset = %zu\n", __func__, i, name, offset); + } + } + #endif + + // prepare memory for the weights + { + const int n_block = model.hparams.n_block; + + model.blocks.resize(n_block); + + model.wte = ggml_get_tensor(ctx, "token_embd.weight"); + model.ln_f_g = ggml_get_tensor(ctx, "output_norm.weight"); + model.ln_f_b = ggml_get_tensor(ctx, "output_norm.bias"); + model.lmh_g = ggml_get_tensor(ctx, "output.weight"); + + // map by name + model.tensors["token_embd.weight"] = model.wte; + model.tensors["output_norm.weight"] = model.ln_f_g; + model.tensors["output_norm.bias"] = model.ln_f_b; + model.tensors["output.weight"] = model.lmh_g; + + for (int i = 0; i < n_block; ++i) { + auto & block = model.blocks[i]; + + std::string blocknamestart = "blk." + std::to_string(i) + "."; + + block.ln_1_g = get_tensor_ex(ctx, blocknamestart + "attn_norm.weight" ); + block.ln_1_b = get_tensor_ex(ctx, blocknamestart + "attn_norm.bias" ); + + block.c_attn_attn_w = get_tensor_ex(ctx, blocknamestart + "attn_qkv.weight" ); + block.c_attn_attn_b = get_tensor_ex(ctx ,blocknamestart + "attn_qkv.bias" ); + + block.c_attn_proj_w = get_tensor_ex(ctx, blocknamestart + "attn_output.weight" ); + block.c_attn_proj_b = get_tensor_ex(ctx, blocknamestart + "attn_output.bias" ); + + block.ln_2_g = get_tensor_ex(ctx, blocknamestart + "ffn_norm.weight" ); + block.ln_2_b = get_tensor_ex(ctx, blocknamestart + "ffn_norm.bias"); + + block.c_mlp_fc_w = get_tensor_ex(ctx, blocknamestart + "ffn_up.weight" ); + block.c_mlp_fc_b = get_tensor_ex(ctx, blocknamestart + "ffn_up.bias" ); + + block.c_mlp_proj_w = get_tensor_ex(ctx, blocknamestart + "ffn_down.weight" ); + block.c_mlp_proj_b = get_tensor_ex(ctx, blocknamestart + "ffn_down.bias" ); + + // map by name + model.tensors[blocknamestart + "attn_norm.weight"] = block.ln_1_g; + model.tensors[blocknamestart + "attn_norm.bias"] = block.ln_1_b; + + model.tensors[blocknamestart + "attn_qkv.weight"] = block.c_attn_attn_w; + model.tensors[blocknamestart + "attn_qkv.bias"] = block.c_attn_attn_b; + + model.tensors[blocknamestart + "attn_output.weight"] = block.c_attn_proj_w; + model.tensors[blocknamestart + "attn_output.bias"] = block.c_attn_proj_b; + + model.tensors[blocknamestart + "ffn_norm.weight"] = block.ln_2_g; + model.tensors[blocknamestart + "ffn_norm.bias"] = block.ln_2_b; + + model.tensors[blocknamestart + "ffn_up.weight"] = block.c_mlp_fc_w; + model.tensors[blocknamestart + "ffn_up.bias"] = block.c_mlp_fc_b; + + model.tensors[blocknamestart + "ffn_down.weight"] = block.c_mlp_proj_w; + model.tensors[blocknamestart + "ffn_down.bias"] = block.c_mlp_proj_b; + } + } + + // key + value memory + { + const auto & kvctx = model.kvctx; + const auto & hparams = model.hparams; + + const int n_embd = hparams.n_embd; + const int n_block = hparams.n_block; + const int n_ctx = hparams.n_ctx; + + const int64_t n_mem = n_block*n_ctx; + const int64_t n_elements = n_embd*n_mem; + + // create the ggml context + { + struct ggml_init_params params = { + /*.mem_size =*/ size_t(n_elements*4+ggml_tensor_overhead()*2), + /*.mem_buffer =*/ NULL, + /*.no_alloc =*/ false, + }; + + model.kvctx = ggml_init(params); + if (!model.kvctx) { + fprintf(stderr, "%s: kv ggml_init() failed\n", __func__); + return false; + } + + } + + + model.memory_k = ggml_new_tensor_1d(kvctx, GGML_TYPE_F16, n_elements); + model.memory_v = ggml_new_tensor_1d(kvctx, GGML_TYPE_F16, n_elements); + + const size_t memory_size = ggml_nbytes(model.memory_k) + ggml_nbytes(model.memory_v); + + printf("%s: memory_size = %8.2f MB, n_mem = %" PRId64 "\n", __func__, memory_size/1024.0/1024.0, n_mem); + } + + return true; +} + + +// feed-forward network +ggml_tensor * gpt_neox_ff( + const gpt_neox_block &block, + ggml_context * ctx0, + ggml_tensor * inp) { + + ggml_tensor * cur = ggml_norm(ctx0, inp); + + cur = ggml_add(ctx0, ggml_mul(ctx0, ggml_repeat(ctx0, block.ln_2_g, cur), cur), ggml_repeat(ctx0, block.ln_2_b, cur)); + cur = ggml_mul_mat(ctx0, block.c_mlp_fc_w, cur); + cur = ggml_add(ctx0, ggml_repeat(ctx0, block.c_mlp_fc_b, cur), cur); + + // GELU activation + cur = ggml_gelu(ctx0, cur); + + // projection + // cur = proj_w*cur + proj_b + cur = ggml_mul_mat(ctx0, block.c_mlp_proj_w, cur); + + cur = ggml_add(ctx0, ggml_repeat(ctx0, block.c_mlp_proj_b, cur), cur); + return cur; +} + +// evaluate the transformer +// +// - model: the model +// - n_threads: number of threads to use +// - n_past: the context size so far +// - embd_inp: the embeddings of the tokens in the context +// - embd_w: the predicted logits for the next token +// +bool gpt_neox_eval( + const gpt_neox_model & model, + const int n_threads, + const int n_past, + const std::vector<gpt2bpe_vocab::id> & embd_inp, + std::vector<float> & embd_w, + size_t & mem_per_token) { + const int N = embd_inp.size(); + + const auto & hparams = model.hparams; + + const int n_embd = hparams.n_embd; + const int n_block = hparams.n_block; + const int n_ctx = hparams.n_ctx; + const int n_head = hparams.n_head; + const int n_vocab = hparams.n_vocab; + const int n_rot = hparams.n_rot; + + static size_t buf_size = 256u*1024*1024; + static void * buf = malloc(buf_size); + + // use 2 scratch buffers + // TODO: very hacky solution - reimplement in a more elegant way + static size_t scr0_size = 256u*1024*1024; + static void * scr0 = malloc(scr0_size); + + static size_t scr1_size = 256u*1024*1024; + static void * scr1 = malloc(scr1_size); + + if (mem_per_token > 0 && mem_per_token*N > buf_size) { + const size_t buf_size_new = 1.1*(mem_per_token*N); // add 10% to account for ggml object overhead + //printf("\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_size, buf_size_new); + + // reallocate + buf_size = buf_size_new; + buf = realloc(buf, buf_size); + if (buf == nullptr) { + fprintf(stderr, "%s: failed to allocate %zu bytes\n", __func__, buf_size); + return false; + } + } + + struct ggml_init_params params = { + /*.mem_size =*/ buf_size, + /*.mem_buffer =*/ buf, + /*.no_alloc =*/ false, + }; + + struct ggml_context * ctx0 = ggml_init(params); + struct ggml_cgraph gf = {}; + + struct ggml_tensor * embd = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, N); + memcpy(embd->data, embd_inp.data(), N*ggml_element_size(embd)); + + + // wte + struct ggml_tensor * inpL = ggml_get_rows(ctx0, model.wte, embd); + + for (int il = 0; il < n_block; ++il) { + struct ggml_tensor * cur; + + ggml_set_scratch(ctx0, { 0, scr0_size, scr0, }); + + // self-attention + { + { + cur = ggml_norm(ctx0, inpL); + + cur = ggml_add(ctx0, + ggml_mul(ctx0, ggml_repeat(ctx0, model.blocks[il].ln_1_g, cur), cur), + ggml_repeat(ctx0, model.blocks[il].ln_1_b, cur)); + } + + // compute QKV + { + + cur = ggml_mul_mat(ctx0, model.blocks[il].c_attn_attn_w, cur); + cur = ggml_add(ctx0, ggml_repeat(ctx0, model.blocks[il].c_attn_attn_b, cur), cur); + } + + struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_3d(ctx0, cur, n_embd/n_head, n_head, N, cur->nb[1]/n_head, cur->nb[1], 0*sizeof(float)*n_embd/n_head)); + struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_3d(ctx0, cur, n_embd/n_head, n_head, N, cur->nb[1]/n_head, cur->nb[1], 1*sizeof(float)*n_embd/n_head)); + struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_3d(ctx0, cur, n_embd/n_head, n_head, N, cur->nb[1]/n_head, cur->nb[1], 2*sizeof(float)*n_embd/n_head)); + + // using mode = 2 for GPT-NeoX mode + Qcur = ggml_rope_inplace(ctx0, Qcur, n_past, n_rot, 2, 0); + Kcur = ggml_rope_inplace(ctx0, Kcur, n_past, n_rot, 2, 0); + + // store key and value to memory + { + Vcur = ggml_transpose(ctx0, ggml_reshape_2d(ctx0, Vcur, n_embd, N)); + + struct ggml_tensor * k = ggml_view_1d(ctx0, model.memory_k, N*n_embd, (ggml_element_size(model.memory_k)*n_embd)*(il*n_ctx + n_past)); + struct ggml_tensor * v = ggml_view_2d(ctx0, model.memory_v, N, n_embd, + ( n_ctx)*ggml_element_size(model.memory_v), + (il*n_ctx)*ggml_element_size(model.memory_v)*n_embd + n_past*ggml_element_size(model.memory_v)); + + ggml_build_forward_expand(&gf, ggml_cpy(ctx0, Kcur, k)); + ggml_build_forward_expand(&gf, ggml_cpy(ctx0, Vcur, v)); + } + + // Q = Qcur.contiguous().view(n_embd/n_head, n_head, N).permute(0, 2, 1, 3) + struct ggml_tensor * Q = ggml_permute(ctx0, Qcur, 0, 2, 1, 3); + + // K = Kmem.view(n_embd/n_head, n_head, n_past + N).permute(0, 2, 1, 3) + struct ggml_tensor * K = + ggml_permute(ctx0, + ggml_reshape_3d(ctx0, + ggml_view_1d(ctx0, model.memory_k, (n_past + N)*n_embd, il*n_ctx*ggml_element_size(model.memory_k)*n_embd), + n_embd/n_head, n_head, n_past + N), + 0, 2, 1, 3); + + // K * Q + struct ggml_tensor * KQ = ggml_mul_mat(ctx0, K, Q); + + // KQ_scaled = KQ / sqrt(n_embd/n_head) + struct ggml_tensor * KQ_scaled = + ggml_scale_inplace(ctx0, + KQ, + ggml_new_f32(ctx0, 1.0f/sqrt(float(n_embd)/n_head)) + ); + + // KQ_masked = mask_past(KQ_scaled) + struct ggml_tensor * KQ_masked = ggml_diag_mask_inf_inplace(ctx0, KQ_scaled, n_past); + + // KQ = soft_max(KQ_masked) + struct ggml_tensor * KQ_soft_max = ggml_soft_max_inplace(ctx0, KQ_masked); + + // V_trans = Vmem.view(n_embd/n_head, n_head, n_past + N).permute(1, 2, 0, 3).contiguous() + struct ggml_tensor * V = + ggml_view_3d(ctx0, model.memory_v, + n_past + N, n_embd/n_head, n_head, + n_ctx*ggml_element_size(model.memory_v), + n_ctx*ggml_element_size(model.memory_v)*n_embd/n_head, + il*n_ctx*ggml_element_size(model.memory_v)*n_embd); + + // KQV = transpose(V) * KQ_soft_max + struct ggml_tensor * KQV = ggml_mul_mat(ctx0, V, KQ_soft_max); + + // KQV_merged = KQV.permute(0, 2, 1, 3) + struct ggml_tensor * KQV_merged = ggml_permute(ctx0, KQV, 0, 2, 1, 3); + + // cur = KQV_merged.contiguous().view(n_embd, N) + cur = ggml_cpy(ctx0, KQV_merged, ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_embd, N)); + + // projection + { + cur = ggml_mul_mat(ctx0, model.blocks[il].c_attn_proj_w, cur); + cur = ggml_add(ctx0, ggml_repeat(ctx0, model.blocks[il].c_attn_proj_b, cur), cur); + } + } + + ggml_set_scratch(ctx0, { 0, scr1_size, scr1, }); + + if (hparams.par_res == 0) { + struct ggml_tensor * inpFF = ggml_add(ctx0, cur, inpL); + + cur = gpt_neox_ff(model.blocks[il], ctx0, inpFF); + + // input for next layer + inpL = ggml_add(ctx0, cur, inpFF); + } else { + struct ggml_tensor * inpFF = cur; + + // this is independent of the self-attention result, so it could be done in parallel to the self-attention + // note here we pass inpL instead of cur + cur = gpt_neox_ff(model.blocks[il], ctx0, inpL); + + // layer input + FF + cur = ggml_add(ctx0, cur, inpFF); + + // input for next layer + inpL = ggml_add(ctx0, cur, inpL); + } + } + + ggml_set_scratch(ctx0, { 0, scr0_size, scr0, }); + + // norm + { + inpL = ggml_norm(ctx0, inpL); + + // inpL = ln_f_g*inpL + ln_f_b + inpL = ggml_add(ctx0, + ggml_mul(ctx0, + ggml_repeat(ctx0, model.ln_f_g, inpL), + inpL), + ggml_repeat(ctx0, model.ln_f_b, inpL)); + } + + ggml_set_scratch(ctx0, { 0, 0, nullptr, }); + + // lm_head + { + inpL = ggml_mul_mat(ctx0, model.lmh_g, inpL); + + //inpL = ggml_add(ctx0, + // ggml_repeat(ctx0, model.lmh_b, inpL), + // inpL); + } + + // logits -> probs + //inpL = ggml_soft_max_inplace(ctx0, inpL); + + // run the computation + ggml_build_forward_expand(&gf, inpL); + ggml_graph_compute_with_ctx(ctx0, &gf, n_threads); + + //if (n_past%100 == 0) { + // ggml_graph_print (&gf); + // ggml_graph_dump_dot(&gf, NULL, "gpt-2.dot"); + //} + + //embd_w.resize(n_vocab*N); + //memcpy(embd_w.data(), ggml_get_data(inpL), sizeof(float)*n_vocab*N); + + // return result for just the last token + embd_w.resize(n_vocab); + memcpy(embd_w.data(), (float *) ggml_get_data(inpL) + (n_vocab*(N-1)), sizeof(float)*n_vocab); + + if (mem_per_token == 0) { + mem_per_token = ggml_used_mem(ctx0)/N; + } + //printf("used_mem = %zu\n", ggml_used_mem(ctx0)); + + ggml_free(ctx0); + + return true; +} + +int main(int argc, char ** argv) { + ggml_time_init(); + + const int64_t t_main_start_us = ggml_time_us(); + + gpt_params params; + + if (gpt_params_parse(argc, argv, params) == false) { + return 1; + } + + int64_t t_load_us = 0; + + gpt2bpe_vocab vocab; + gpt_neox_model model; + + // load the model + { + const int64_t t_start_us = ggml_time_us(); + + if (!gpt_neox_model_load(params.model, model, vocab)) { + fprintf(stderr, "%s: failed to load model from '%s'\n", __func__, params.model.c_str()); + return 1; + } + + t_load_us = ggml_time_us() - t_start_us; + + } + + if (params.seed < 0) { + params.seed = time(NULL); + } + + if (params.top_k == 0) { + params.top_k = model.hparams.n_vocab; + } + + printf("%s: seed = %d\n", __func__, params.seed); + printf("%s: temp = %.3f\n", __func__, params.temp); + printf("%s: top_k = %d\n", __func__, params.top_k); + printf("%s: top_p = %.3f\n", __func__, params.top_p); + printf("%s: repeat_last_n = %d\n", __func__, params.repeat_last_n); + printf("%s: repeat_penalty = %.3f\n", __func__, params.repeat_penalty); + + std::mt19937 rng(params.seed); + + if (params.prompt.empty()) { + params.prompt = "Once upon"; + } + + std::vector<int32_t> last_n_tokens(model.hparams.n_ctx); + std::fill(last_n_tokens.begin(), last_n_tokens.end(), 0); + + int n_past = 0; + + int64_t t_sample_us = 0; + int64_t t_predict_us = 0; + + std::vector<float> logits; + + // tokenize the prompt + std::vector<gpt2bpe_vocab::id> embd_inp = gpt2bpe_tokenize(vocab, params.prompt,false, false); + + params.n_predict = std::min(params.n_predict, model.hparams.n_ctx - (int) embd_inp.size()); + + printf("%s: number of tokens in prompt = %zu\n", __func__, embd_inp.size()); +// for (size_t i = 0; i < embd_inp.size(); i++) { +// printf("%s: token[%zu] = %6d, %s\n", __func__, i, embd_inp[i], vocab.id_to_token[embd_inp[i]].c_str()); +// } + + if( model.hparams.n_ctx < params.n_predict+embd_inp.size() ) { + params.n_predict = model.hparams.n_ctx-embd_inp.size(); + } + + printf("%s: n_predict = %d\n", __func__, params.n_predict); + printf("\n"); + + std::vector<gpt2bpe_vocab::id> embd; + + // determine the required inference memory per token: + size_t mem_per_token = 0; + gpt_neox_eval(model, params.n_threads, 0, { 0, 1, 2, 3 }, logits, mem_per_token); + + for (size_t i = embd.size(); i < embd_inp.size() + params.n_predict; i++) { + // predict + if (embd.size() > 0) { + const int64_t t_start_us = ggml_time_us(); + + if (!gpt_neox_eval(model, params.n_threads, n_past, embd, logits, mem_per_token)) { + printf("Failed to predict\n"); + return 1; + } + + t_predict_us += ggml_time_us() - t_start_us; + } + + n_past += embd.size(); + embd.clear(); + + if (i >= embd_inp.size()) { + // sample next token + const int top_k = params.top_k; + const float top_p = params.top_p; + const float temp = params.temp; + const int repeat_last_n = params.repeat_last_n; + const float repeat_penalty = params.repeat_penalty; + + const int n_vocab = model.hparams.n_vocab; + + gpt2bpe_vocab::id id = 0; + + { + const int64_t t_start_sample_us = ggml_time_us(); + + id = sample_top_k_top_p_repeat(vocab, logits.data() + (logits.size() - n_vocab), last_n_tokens.data(), last_n_tokens.size(), top_k, top_p, temp, repeat_last_n, repeat_penalty, rng); + + last_n_tokens.erase(last_n_tokens.begin()); + last_n_tokens.push_back(id); + + t_sample_us += ggml_time_us() - t_start_sample_us; + } + + // add it to the context + embd.push_back(id); + } else { + // if here, it means we are still processing the input prompt + for (size_t k = i; k < embd_inp.size(); k++) { + embd.push_back(embd_inp[k]); + if (embd.size() > params.n_batch) { + break; + } + } + i += embd.size() - 1; + } + + // display text + for (auto id : embd) { + printf("%s", vocab.id_to_token[id].c_str() ); + } + fflush(stdout); + + // end of text token + if (vocab.special_eos_id != -1 && embd.back() == vocab.special_eos_id) { + break; + } + } + + // report timing + { + const int64_t t_main_end_us = ggml_time_us(); + + printf("\n\n"); + printf("%s: mem per token = %8zu bytes\n", __func__, mem_per_token); + printf("%s: load time = %8.2f ms\n", __func__, t_load_us/1000.0f); + printf("%s: sample time = %8.2f ms\n", __func__, t_sample_us/1000.0f); + printf("%s: predict time = %8.2f ms / %.2f ms per token\n", __func__, t_predict_us/1000.0f, t_predict_us/1000.0f/n_past); + printf("%s: total time = %8.2f ms\n", __func__, (t_main_end_us - t_main_start_us)/1000.0f); + } + + ggml_free(model.ctx); + + return 0; +} diff --git a/examples/grammar-parser.cpp b/examples/grammar-parser.cpp deleted file mode 100644 index e76bd11c..00000000 --- a/examples/grammar-parser.cpp +++ /dev/null @@ -1,423 +0,0 @@ -#include "grammar-parser.h" -#include <cstdint> -#include <cwchar> -#include <string> -#include <utility> -#include <stdexcept> -#include <exception> - -namespace grammar_parser { - // NOTE: assumes valid utf8 (but checks for overrun) - // copied from llama.cpp - std::pair<uint32_t, const char *> decode_utf8(const char * src) { - static const int lookup[] = { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 3, 4 }; - uint8_t first_byte = static_cast<uint8_t>(*src); - uint8_t highbits = first_byte >> 4; - int len = lookup[highbits]; - uint8_t mask = (1 << (8 - len)) - 1; - uint32_t value = first_byte & mask; - const char * end = src + len; // may overrun! - const char * pos = src + 1; - for ( ; pos < end && *pos; pos++) { - value = (value << 6) + (static_cast<uint8_t>(*pos) & 0x3F); - } - return std::make_pair(value, pos); - } - - uint32_t get_symbol_id(parse_state & state, const char * src, size_t len) { - uint32_t next_id = static_cast<uint32_t>(state.symbol_ids.size()); - auto result = state.symbol_ids.insert(std::make_pair(std::string(src, len), next_id)); - return result.first->second; - } - - uint32_t generate_symbol_id(parse_state & state, const std::string & base_name) { - uint32_t next_id = static_cast<uint32_t>(state.symbol_ids.size()); - state.symbol_ids[base_name + '_' + std::to_string(next_id)] = next_id; - return next_id; - } - - void add_rule( - parse_state & state, - uint32_t rule_id, - const std::vector<llama_grammar_element> & rule) { - if (state.rules.size() <= rule_id) { - state.rules.resize(rule_id + 1); - } - state.rules[rule_id] = rule; - } - - bool is_word_char(char c) { - return ('a' <= c && c <= 'z') || ('A' <= c && c <= 'Z') || c == '-' || ('0' <= c && c <= '9'); - } - - std::pair<uint32_t, const char *> parse_hex(const char * src, int size) { - const char * pos = src; - const char * end = src + size; - uint32_t value = 0; - for ( ; pos < end && *pos; pos++) { - value <<= 4; - char c = *pos; - if ('a' <= c && c <= 'f') { - value += c - 'a' + 10; - } else if ('A' <= c && c <= 'F') { - value += c - 'A' + 10; - } else if ('0' <= c && c <= '9') { - value += c - '0'; - } else { - break; - } - } - if (pos != end) { - throw std::runtime_error("expecting " + std::to_string(size) + " hex chars at " + src); - } - return std::make_pair(value, pos); - } - - const char * parse_space(const char * src, bool newline_ok) { - const char * pos = src; - while (*pos == ' ' || *pos == '\t' || *pos == '#' || - (newline_ok && (*pos == '\r' || *pos == '\n'))) { - if (*pos == '#') { - while (*pos && *pos != '\r' && *pos != '\n') { - pos++; - } - } else { - pos++; - } - } - return pos; - } - - const char * parse_name(const char * src) { - const char * pos = src; - while (is_word_char(*pos)) { - pos++; - } - if (pos == src) { - throw std::runtime_error(std::string("expecting name at ") + src); - } - return pos; - } - - std::pair<uint32_t, const char *> parse_char(const char * src) { - if (*src == '\\') { - switch (src[1]) { - case 'x': return parse_hex(src + 2, 2); - case 'u': return parse_hex(src + 2, 4); - case 'U': return parse_hex(src + 2, 8); - case 't': return std::make_pair('\t', src + 2); - case 'r': return std::make_pair('\r', src + 2); - case 'n': return std::make_pair('\n', src + 2); - case '\\': - case '"': - case '[': - case ']': - return std::make_pair(src[1], src + 2); - default: - throw std::runtime_error(std::string("unknown escape at ") + src); - } - } else if (*src) { - return decode_utf8(src); - } - throw std::runtime_error("unexpected end of input"); - } - - const char * parse_alternates( - parse_state & state, - const char * src, - const std::string & rule_name, - uint32_t rule_id, - bool is_nested); - - const char * parse_sequence( - parse_state & state, - const char * src, - const std::string & rule_name, - std::vector<llama_grammar_element> & out_elements, - bool is_nested) { - size_t last_sym_start = out_elements.size(); - const char * pos = src; - while (*pos) { - if (*pos == '"') { // literal string - pos++; - last_sym_start = out_elements.size(); - while (*pos != '"') { - auto char_pair = parse_char(pos); - pos = char_pair.second; - out_elements.push_back({LLAMA_GRETYPE_CHAR, char_pair.first}); - } - pos = parse_space(pos + 1, is_nested); - } else if (*pos == '[') { // char range(s) - pos++; - enum llama_gretype start_type = LLAMA_GRETYPE_CHAR; - if (*pos == '^') { - pos++; - start_type = LLAMA_GRETYPE_CHAR_NOT; - } - last_sym_start = out_elements.size(); - while (*pos != ']') { - auto char_pair = parse_char(pos); - pos = char_pair.second; - enum llama_gretype type = last_sym_start < out_elements.size() - ? LLAMA_GRETYPE_CHAR_ALT - : start_type; - - out_elements.push_back({type, char_pair.first}); - if (pos[0] == '-' && pos[1] != ']') { - auto endchar_pair = parse_char(pos + 1); - pos = endchar_pair.second; - out_elements.push_back({LLAMA_GRETYPE_CHAR_RNG_UPPER, endchar_pair.first}); - } - } - pos = parse_space(pos + 1, is_nested); - } else if (is_word_char(*pos)) { // rule reference - const char * name_end = parse_name(pos); - uint32_t ref_rule_id = get_symbol_id(state, pos, name_end - pos); - pos = parse_space(name_end, is_nested); - last_sym_start = out_elements.size(); - out_elements.push_back({LLAMA_GRETYPE_RULE_REF, ref_rule_id}); - } else if (*pos == '(') { // grouping - // parse nested alternates into synthesized rule - pos = parse_space(pos + 1, true); - uint32_t sub_rule_id = generate_symbol_id(state, rule_name); - pos = parse_alternates(state, pos, rule_name, sub_rule_id, true); - last_sym_start = out_elements.size(); - // output reference to synthesized rule - out_elements.push_back({LLAMA_GRETYPE_RULE_REF, sub_rule_id}); - if (*pos != ')') { - throw std::runtime_error(std::string("expecting ')' at ") + pos); - } - pos = parse_space(pos + 1, is_nested); - } else if (*pos == '*' || *pos == '+' || *pos == '?') { // repetition operator - if (last_sym_start == out_elements.size()) { - throw std::runtime_error(std::string("expecting preceeding item to */+/? at ") + pos); - } - - // apply transformation to previous symbol (last_sym_start to end) according to - // rewrite rules: - // S* --> S' ::= S S' | - // S+ --> S' ::= S S' | S - // S? --> S' ::= S | - uint32_t sub_rule_id = generate_symbol_id(state, rule_name); - std::vector<llama_grammar_element> sub_rule; - // add preceding symbol to generated rule - sub_rule.insert( - sub_rule.end(), out_elements.begin() + last_sym_start, out_elements.end()); - if (*pos == '*' || *pos == '+') { - // cause generated rule to recurse - sub_rule.push_back({LLAMA_GRETYPE_RULE_REF, sub_rule_id}); - } - // mark start of alternate def - sub_rule.push_back({LLAMA_GRETYPE_ALT, 0}); - if (*pos == '+') { - // add preceding symbol as alternate only for '+' (otherwise empty) - sub_rule.insert( - sub_rule.end(), out_elements.begin() + last_sym_start, out_elements.end()); - } - sub_rule.push_back({LLAMA_GRETYPE_END, 0}); - add_rule(state, sub_rule_id, sub_rule); - - // in original rule, replace previous symbol with reference to generated rule - out_elements.resize(last_sym_start); - out_elements.push_back({LLAMA_GRETYPE_RULE_REF, sub_rule_id}); - - pos = parse_space(pos + 1, is_nested); - } else { - break; - } - } - return pos; - } - - const char * parse_alternates( - parse_state & state, - const char * src, - const std::string & rule_name, - uint32_t rule_id, - bool is_nested) { - std::vector<llama_grammar_element> rule; - const char * pos = parse_sequence(state, src, rule_name, rule, is_nested); - while (*pos == '|') { - rule.push_back({LLAMA_GRETYPE_ALT, 0}); - pos = parse_space(pos + 1, true); - pos = parse_sequence(state, pos, rule_name, rule, is_nested); - } - rule.push_back({LLAMA_GRETYPE_END, 0}); - add_rule(state, rule_id, rule); - return pos; - } - - const char * parse_rule(parse_state & state, const char * src) { - const char * name_end = parse_name(src); - const char * pos = parse_space(name_end, false); - size_t name_len = name_end - src; - uint32_t rule_id = get_symbol_id(state, src, name_len); - const std::string name(src, name_len); - - if (!(pos[0] == ':' && pos[1] == ':' && pos[2] == '=')) { - throw std::runtime_error(std::string("expecting ::= at ") + pos); - } - pos = parse_space(pos + 3, true); - - pos = parse_alternates(state, pos, name, rule_id, false); - - if (*pos == '\r') { - pos += pos[1] == '\n' ? 2 : 1; - } else if (*pos == '\n') { - pos++; - } else if (*pos) { - throw std::runtime_error(std::string("expecting newline or end at ") + pos); - } - return parse_space(pos, true); - } - - parse_state parse(const char * src) { - try { - parse_state state; - const char * pos = parse_space(src, true); - while (*pos) { - pos = parse_rule(state, pos); - } - return state; - } catch (const std::exception & err) { - fprintf(stderr, "%s: error parsing grammar: %s\n", __func__, err.what()); - return parse_state(); - } - } - - void print_grammar_char(FILE * file, uint32_t c) { - if (0x20 <= c && c <= 0x7f) { - fprintf(file, "%c", static_cast<char>(c)); - } else { - // cop out of encoding UTF-8 - fprintf(file, "<U+%04X>", c); - } - } - - bool is_char_element(llama_grammar_element elem) { - switch (elem.type) { - case LLAMA_GRETYPE_CHAR: return true; - case LLAMA_GRETYPE_CHAR_NOT: return true; - case LLAMA_GRETYPE_CHAR_ALT: return true; - case LLAMA_GRETYPE_CHAR_RNG_UPPER: return true; - default: return false; - } - } - - void print_rule_binary(FILE * file, const std::vector<llama_grammar_element> & rule) { - for (auto elem : rule) { - switch (elem.type) { - case LLAMA_GRETYPE_END: fprintf(file, "END"); break; - case LLAMA_GRETYPE_ALT: fprintf(file, "ALT"); break; - case LLAMA_GRETYPE_RULE_REF: fprintf(file, "RULE_REF"); break; - case LLAMA_GRETYPE_CHAR: fprintf(file, "CHAR"); break; - case LLAMA_GRETYPE_CHAR_NOT: fprintf(file, "CHAR_NOT"); break; - case LLAMA_GRETYPE_CHAR_RNG_UPPER: fprintf(file, "CHAR_RNG_UPPER"); break; - case LLAMA_GRETYPE_CHAR_ALT: fprintf(file, "CHAR_ALT"); break; - } - switch (elem.type) { - case LLAMA_GRETYPE_END: - case LLAMA_GRETYPE_ALT: - case LLAMA_GRETYPE_RULE_REF: - fprintf(file, "(%u) ", elem.value); - break; - case LLAMA_GRETYPE_CHAR: - case LLAMA_GRETYPE_CHAR_NOT: - case LLAMA_GRETYPE_CHAR_RNG_UPPER: - case LLAMA_GRETYPE_CHAR_ALT: - fprintf(file, "(\""); - print_grammar_char(file, elem.value); - fprintf(file, "\") "); - break; - } - } - fprintf(file, "\n"); - } - - void print_rule( - FILE * file, - uint32_t rule_id, - const std::vector<llama_grammar_element> & rule, - const std::map<uint32_t, std::string> & symbol_id_names) { - if (rule.empty() || rule.back().type != LLAMA_GRETYPE_END) { - throw std::runtime_error( - "malformed rule, does not end with LLAMA_GRETYPE_END: " + std::to_string(rule_id)); - } - fprintf(file, "%s ::= ", symbol_id_names.at(rule_id).c_str()); - for (size_t i = 0, end = rule.size() - 1; i < end; i++) { - llama_grammar_element elem = rule[i]; - switch (elem.type) { - case LLAMA_GRETYPE_END: - throw std::runtime_error( - "unexpected end of rule: " + std::to_string(rule_id) + "," + - std::to_string(i)); - case LLAMA_GRETYPE_ALT: - fprintf(file, "| "); - break; - case LLAMA_GRETYPE_RULE_REF: - fprintf(file, "%s ", symbol_id_names.at(elem.value).c_str()); - break; - case LLAMA_GRETYPE_CHAR: - fprintf(file, "["); - print_grammar_char(file, elem.value); - break; - case LLAMA_GRETYPE_CHAR_NOT: - fprintf(file, "[^"); - print_grammar_char(file, elem.value); - break; - case LLAMA_GRETYPE_CHAR_RNG_UPPER: - if (i == 0 || !is_char_element(rule[i - 1])) { - throw std::runtime_error( - "LLAMA_GRETYPE_CHAR_RNG_UPPER without preceding char: " + - std::to_string(rule_id) + "," + std::to_string(i)); - } - fprintf(file, "-"); - print_grammar_char(file, elem.value); - break; - case LLAMA_GRETYPE_CHAR_ALT: - if (i == 0 || !is_char_element(rule[i - 1])) { - throw std::runtime_error( - "LLAMA_GRETYPE_CHAR_ALT without preceding char: " + - std::to_string(rule_id) + "," + std::to_string(i)); - } - print_grammar_char(file, elem.value); - break; - } - if (is_char_element(elem)) { - switch (rule[i + 1].type) { - case LLAMA_GRETYPE_CHAR_ALT: - case LLAMA_GRETYPE_CHAR_RNG_UPPER: - break; - default: - fprintf(file, "] "); - } - } - } - fprintf(file, "\n"); - } - - void print_grammar(FILE * file, const parse_state & state) { - try { - std::map<uint32_t, std::string> symbol_id_names; - for (auto kv : state.symbol_ids) { - symbol_id_names[kv.second] = kv.first; - } - for (size_t i = 0, end = state.rules.size(); i < end; i++) { - // fprintf(file, "%zu: ", i); - // print_rule_binary(file, state.rules[i]); - print_rule(file, uint32_t(i), state.rules[i], symbol_id_names); - // fprintf(file, "\n"); - } - } catch (const std::exception & err) { - fprintf(stderr, "\n%s: error printing grammar: %s\n", __func__, err.what()); - } - } - - std::vector<const llama_grammar_element *> parse_state::c_rules() { - std::vector<const llama_grammar_element *> ret; - for (const auto & rule : rules) { - ret.push_back(rule.data()); - } - return ret; - } -} diff --git a/examples/grammar-parser.h b/examples/grammar-parser.h deleted file mode 100644 index 9037d727..00000000 --- a/examples/grammar-parser.h +++ /dev/null @@ -1,29 +0,0 @@ -// Implements a parser for an extended Backus-Naur form (BNF), producing the -// binary context-free grammar format specified by llama.h. Supports character -// ranges, grouping, and repetition operators. As an example, a grammar for -// arithmetic might look like: -// -// root ::= expr -// expr ::= term ([-+*/] term)* -// term ::= num | "(" space expr ")" space -// num ::= [0-9]+ space -// space ::= [ \t\n]* - -#pragma once -#include "llama.h" -#include <vector> -#include <map> -#include <cstdint> -#include <string> - -namespace grammar_parser { - struct parse_state { - std::map<std::string, uint32_t> symbol_ids; - std::vector<std::vector<llama_grammar_element>> rules; - - std::vector<const llama_grammar_element *> c_rules(); - }; - - parse_state parse(const char * src); - void print_grammar(FILE * file, const parse_state & state); -} diff --git a/examples/llama-bench/llama-bench.cpp b/examples/llama-bench/llama-bench.cpp index 266c8eab..d11fff28 100755 --- a/examples/llama-bench/llama-bench.cpp +++ b/examples/llama-bench/llama-bench.cpp @@ -606,6 +606,8 @@ const std::string test::cpu_info = get_cpu_info(); const std::string test::gpu_info = get_gpu_info(); struct printer { + virtual ~printer() {} + FILE * fout; virtual void print_header(const cmd_params & params) { (void) params; }; virtual void print_test(const test & t) = 0; @@ -849,7 +851,7 @@ struct sql_printer : public printer { }; static void test_prompt(llama_context * ctx, int n_prompt, int n_past, int n_batch, int n_threads) { - std::vector<llama_token> tokens(n_batch, llama_token_bos()); + std::vector<llama_token> tokens(n_batch, llama_token_bos(ctx)); int n_processed = 0; while (n_processed < n_prompt) { int n_tokens = std::min(n_prompt - n_processed, n_batch); @@ -859,7 +861,7 @@ static void test_prompt(llama_context * ctx, int n_prompt, int n_past, int n_bat } static void test_gen(llama_context * ctx, int n_gen, int n_past, int n_threads) { - llama_token token = llama_token_bos(); + llama_token token = llama_token_bos(ctx); for (int i = 0; i < n_gen; i++) { llama_eval(ctx, &token, 1, n_past + i, n_threads); } diff --git a/examples/main/main.cpp b/examples/main/main.cpp index a632bea1..388e1f7d 100644 --- a/examples/main/main.cpp +++ b/examples/main/main.cpp @@ -143,7 +143,7 @@ int main(int argc, char ** argv) { { fprintf(stderr, "%s: testing memory usage for n_batch = %d, n_ctx = %d\n", __func__, params.n_batch, params.n_ctx); - const std::vector<llama_token> tmp(params.n_batch, llama_token_bos()); + const std::vector<llama_token> tmp(params.n_batch, llama_token_bos(ctx)); llama_eval(ctx, tmp.data(), tmp.size(), params.n_ctx, params.n_threads); } @@ -191,10 +191,6 @@ int main(int argc, char ** argv) { // tokenize the prompt std::vector<llama_token> embd_inp; - - // Add a space in front of the first character to match OG llama tokenizer behavior - params.prompt.insert(0, 1, ' '); - if (params.interactive_first || params.instruct || !params.prompt.empty() || session_tokens.empty()) { embd_inp = ::llama_tokenize(ctx, params.prompt, true); } else { @@ -270,15 +266,12 @@ int main(int argc, char ** argv) { params.interactive = true; } - // determine newline token - auto llama_token_newline = ::llama_tokenize(ctx, "\n", false); - if (params.verbose_prompt) { fprintf(stderr, "\n"); fprintf(stderr, "%s: prompt: '%s'\n", __func__, params.prompt.c_str()); fprintf(stderr, "%s: number of tokens in prompt = %zu\n", __func__, embd_inp.size()); for (int i = 0; i < (int) embd_inp.size(); i++) { - fprintf(stderr, "%6d -> '%s'\n", embd_inp[i], llama_token_to_str(ctx, embd_inp[i])); + fprintf(stderr, "%6d -> '%s'\n", embd_inp[i], llama_token_to_str(ctx, embd_inp[i]).c_str()); } if (ctx_guidance) { @@ -286,14 +279,14 @@ int main(int argc, char ** argv) { fprintf(stderr, "%s: negative prompt: '%s'\n", __func__, params.cfg_negative_prompt.c_str()); fprintf(stderr, "%s: number of tokens in negative prompt = %zu\n", __func__, guidance_inp.size()); for (int i = 0; i < (int) guidance_inp.size(); i++) { - fprintf(stderr, "%6d -> '%s'\n", guidance_inp[i], llama_token_to_str(ctx, guidance_inp[i])); + fprintf(stderr, "%6d -> '%s'\n", guidance_inp[i], llama_token_to_str(ctx, guidance_inp[i]).c_str()); } } if (params.n_keep > 0) { fprintf(stderr, "%s: static prompt based on n_keep: '", __func__); for (int i = 0; i < params.n_keep; i++) { - fprintf(stderr, "%s", llama_token_to_str(ctx, embd_inp[i])); + fprintf(stderr, "%s", llama_token_to_str(ctx, embd_inp[i]).c_str()); } fprintf(stderr, "'\n"); } @@ -311,7 +304,7 @@ int main(int argc, char ** argv) { auto console_ctrl_handler = +[](DWORD ctrl_type) -> BOOL { return (ctrl_type == CTRL_C_EVENT) ? (sigint_handler(SIGINT), true) : false; }; - SetConsoleCtrlHandler(static_cast<PHANDLER_ROUTINE>(console_ctrl_handler), true); + SetConsoleCtrlHandler(reinterpret_cast<PHANDLER_ROUTINE>(console_ctrl_handler), true); #endif fprintf(stderr, "%s: interactive mode on.\n", __func__); @@ -352,10 +345,9 @@ int main(int argc, char ** argv) { fprintf(stderr, "\n"); { - auto it = params.logit_bias.find(llama_token_eos()); + auto it = params.logit_bias.find(llama_token_eos(ctx)); if (it != params.logit_bias.end() && it->second == -INFINITY) { - fprintf(stderr, - "%s: warning: EOS token is disabled, which will cause most grammars to fail\n", __func__); + fprintf(stderr, "%s: warning: EOS token is disabled, which will cause most grammars to fail\n", __func__); } } @@ -405,7 +397,7 @@ int main(int argc, char ** argv) { // do one empty run to warm up the model { - const std::vector<llama_token> tmp = { llama_token_bos(), }; + const std::vector<llama_token> tmp = { llama_token_bos(ctx), }; llama_eval(ctx, tmp.data(), tmp.size(), 0, params.n_threads); llama_reset_timings(ctx); } @@ -589,7 +581,7 @@ int main(int argc, char ** argv) { } // Apply penalties - float nl_logit = logits[llama_token_nl()]; + float nl_logit = logits[llama_token_nl(ctx)]; auto last_n_repeat = std::min(std::min((int)last_n_tokens.size(), repeat_last_n), n_ctx); llama_sample_repetition_penalty(ctx, &candidates_p, last_n_tokens.data() + last_n_tokens.size() - last_n_repeat, @@ -598,7 +590,7 @@ int main(int argc, char ** argv) { last_n_tokens.data() + last_n_tokens.size() - last_n_repeat, last_n_repeat, alpha_frequency, alpha_presence); if (!penalize_nl) { - logits[llama_token_nl()] = nl_logit; + logits[llama_token_nl(ctx)] = nl_logit; } if (grammar != NULL) { @@ -662,7 +654,7 @@ int main(int argc, char ** argv) { // display text if (input_echo) { for (auto id : embd) { - printf("%s", llama_token_to_str(ctx, id)); + printf("%s", llama_token_to_str(ctx, id).c_str()); } fflush(stdout); } @@ -704,7 +696,7 @@ int main(int argc, char ** argv) { } // deal with end of text token in interactive mode - if (last_n_tokens.back() == llama_token_eos()) { + if (last_n_tokens.back() == llama_token_eos(ctx)) { if (params.interactive) { if (params.antiprompt.size() != 0) { // tokenize and inject first reverse prompt @@ -728,7 +720,7 @@ int main(int argc, char ** argv) { } if (params.input_prefix_bos) { - embd_inp.push_back(llama_token_bos()); + embd_inp.push_back(llama_token_bos(ctx)); } std::string buffer; @@ -782,8 +774,7 @@ int main(int argc, char ** argv) { if (grammar != NULL) { llama_grammar_free(grammar); - std::vector<const llama_grammar_element *> grammar_rules( - parsed_grammar.c_rules()); + std::vector<const llama_grammar_element *> grammar_rules( parsed_grammar.c_rules()); grammar = llama_grammar_init( grammar_rules.data(), grammar_rules.size(), parsed_grammar.symbol_ids.at("root")); @@ -794,7 +785,7 @@ int main(int argc, char ** argv) { } // end of text token - if (!embd.empty() && embd.back() == llama_token_eos() && !(params.instruct || params.interactive)) { + if (!embd.empty() && embd.back() == llama_token_eos(ctx) && !(params.instruct || params.interactive)) { fprintf(stderr, " [end of text]\n"); break; } diff --git a/examples/metal/metal.cpp b/examples/metal/metal.cpp index 7438defd..c05a4fa9 100644 --- a/examples/metal/metal.cpp +++ b/examples/metal/metal.cpp @@ -2,7 +2,7 @@ // // - First, export a LLaMA graph: // -// $ ./bin/main -m ../models/7B/ggml-model-q4_0.bin --export +// $ ./bin/main -m ../models/7B/ggml-model-q4_0.gguf --export // // - Run this tool to evaluate the exported graph: // diff --git a/examples/perplexity/perplexity.cpp b/examples/perplexity/perplexity.cpp index 2409db69..f3c045ae 100644 --- a/examples/perplexity/perplexity.cpp +++ b/examples/perplexity/perplexity.cpp @@ -64,7 +64,7 @@ void perplexity(llama_context * ctx, const gpt_params & params) { // add BOS token for the first batch of each chunk if (j == 0) { - tokens[batch_start] = llama_token_bos(); + tokens[batch_start] = llama_token_bos(ctx); } if (llama_eval(ctx, tokens.data() + batch_start, batch_size, j * n_batch, params.n_threads)) { diff --git a/examples/quantize-stats/quantize-stats.cpp b/examples/quantize-stats/quantize-stats.cpp index 6aa06ec8..06ce18f0 100644 --- a/examples/quantize-stats/quantize-stats.cpp +++ b/examples/quantize-stats/quantize-stats.cpp @@ -24,7 +24,7 @@ #endif struct quantize_stats_params { - std::string model = "models/7B/ggml-model-f16.bin"; + std::string model = "models/7B/ggml-model-f16.gguf"; bool verbose = false; bool per_layer_stats = false; bool print_histogram = false; diff --git a/examples/quantize/quantize.cpp b/examples/quantize/quantize.cpp index 744f549c..f628d064 100644 --- a/examples/quantize/quantize.cpp +++ b/examples/quantize/quantize.cpp @@ -68,10 +68,10 @@ bool try_parse_ftype(const std::string & ftype_str_in, llama_ftype & ftype, std: } // usage: -// ./quantize [--allow-requantize] [--leave-output-tensor] models/llama/ggml-model.bin [models/llama/ggml-model-quant.bin] type [nthreads] +// ./quantize [--allow-requantize] [--leave-output-tensor] models/llama/ggml-model.gguf [models/llama/ggml-model-quant.gguf] type [nthreads] // void usage(const char * executable) { - fprintf(stderr, "usage: %s [--help] [--allow-requantize] [--leave-output-tensor] model-f32.bin [model-quant.bin] type [nthreads]\n\n", executable); + fprintf(stderr, "usage: %s [--help] [--allow-requantize] [--leave-output-tensor] model-f32.gguf [model-quant.gguf] type [nthreads]\n\n", executable); fprintf(stderr, " --allow-requantize: Allows requantizing tensors that have already been quantized. Warning: This can severely reduce quality compared to quantizing from 16bit or 32bit\n"); fprintf(stderr, " --leave-output-tensor: Will leave output.weight un(re)quantized. Increases model size but may also increase quality, especially when requantizing\n"); fprintf(stderr, "\nAllowed quantization types:\n"); @@ -118,8 +118,8 @@ int main(int argc, char ** argv) { if (pos != std::string::npos) { fpath = fname_inp.substr(0, pos + 1); } - // export as [inp path]/ggml-model-[ftype].bin - fname_out = fpath + "ggml-model-" + ftype_str + ".bin"; + // export as [inp path]/ggml-model-[ftype].gguf + fname_out = fpath + "ggml-model-" + ftype_str + ".gguf"; arg_idx++; } else { diff --git a/examples/save-load-state/save-load-state.cpp b/examples/save-load-state/save-load-state.cpp index 61c71c35..3db61b75 100644 --- a/examples/save-load-state/save-load-state.cpp +++ b/examples/save-load-state/save-load-state.cpp @@ -26,7 +26,6 @@ int main(int argc, char ** argv) { auto lparams = llama_context_default_params(); lparams.n_ctx = params.n_ctx; - lparams.n_gqa = params.n_gqa; lparams.seed = params.seed; lparams.f16_kv = params.memory_f16; lparams.use_mmap = params.use_mmap; @@ -45,9 +44,8 @@ int main(int argc, char ** argv) { llama_free_model(model); return 1; } - auto tokens = std::vector<llama_token>(params.n_ctx); - auto n_prompt_tokens = llama_tokenize(ctx, params.prompt.c_str(), tokens.data(), int(tokens.size()), true); - + auto tokens = llama_tokenize(ctx, params.prompt.c_str(), true); + auto n_prompt_tokens = tokens.size(); if (n_prompt_tokens < 1) { fprintf(stderr, "%s : failed to tokenize prompt\n", __func__); llama_free(ctx); @@ -92,7 +90,7 @@ int main(int argc, char ** argv) { auto next_token_str = llama_token_to_str(ctx, next_token); last_n_tokens_data.push_back(next_token); - printf("%s", next_token_str); + printf("%s", next_token_str.c_str()); if (llama_eval(ctx, &next_token, 1, n_past, params.n_threads)) { fprintf(stderr, "\n%s : failed to evaluate\n", __func__); llama_free(ctx); @@ -152,7 +150,7 @@ int main(int argc, char ** argv) { auto next_token_str = llama_token_to_str(ctx2, next_token); last_n_tokens_data.push_back(next_token); - printf("%s", next_token_str); + printf("%s", next_token_str.c_str()); if (llama_eval(ctx2, &next_token, 1, n_past, params.n_threads)) { fprintf(stderr, "\n%s : failed to evaluate\n", __func__); llama_free(ctx2); diff --git a/examples/server/README.md b/examples/server/README.md index 1559dd3f..4d97db2e 100644 --- a/examples/server/README.md +++ b/examples/server/README.md @@ -5,7 +5,7 @@ This example demonstrates a simple HTTP API server and a simple web front end to Command line options: - `--threads N`, `-t N`: Set the number of threads to use during computation. -- `-m FNAME`, `--model FNAME`: Specify the path to the LLaMA model file (e.g., `models/7B/ggml-model.bin`). +- `-m FNAME`, `--model FNAME`: Specify the path to the LLaMA model file (e.g., `models/7B/ggml-model.gguf`). - `-m ALIAS`, `--alias ALIAS`: Set an alias for the model. The alias will be returned in API responses. - `-c N`, `--ctx-size N`: Set the size of the prompt context. The default is 512, but LLaMA models were built with a context of 2048, which will provide better results for longer input/inference. The size may differ in other models, for example, baichuan models were build with a context of 4096. - `-ngl N`, `--n-gpu-layers N`: When compiled with appropriate support (currently CLBlast or cuBLAS), this option allows offloading some layers to the GPU for computation. Generally results in increased performance. @@ -48,15 +48,14 @@ To get started right away, run the following command, making sure to use the cor ### Unix-based systems (Linux, macOS, etc.): ```bash -./server -m models/7B/ggml-model.bin -c 2048 +./server -m models/7B/ggml-model.gguf -c 2048 ``` ### Windows: ```powershell -server.exe -m models\7B\ggml-model.bin -c 2048 +server.exe -m models\7B\ggml-model.gguf -c 2048 ``` - The above command will start a server that by default listens on `127.0.0.1:8080`. You can consume the endpoints with Postman or NodeJS with axios library. You can visit the web front end at the same url. diff --git a/examples/server/server.cpp b/examples/server/server.cpp index 99660455..a04f1910 100644 --- a/examples/server/server.cpp +++ b/examples/server/server.cpp @@ -279,7 +279,7 @@ struct llama_server_context grammar_parser::print_grammar(stderr, parsed_grammar); { - auto it = params.logit_bias.find(llama_token_eos()); + auto it = params.logit_bias.find(llama_token_eos(ctx)); if (it != params.logit_bias.end() && it->second == -INFINITY) { LOG_WARNING("EOS token is disabled, which will cause most grammars to fail", {}); } @@ -402,7 +402,7 @@ struct llama_server_context if (params.n_predict == 0) { has_next_token = false; - result.tok = llama_token_eos(); + result.tok = llama_token_eos(ctx); return result; } @@ -442,7 +442,7 @@ struct llama_server_context llama_token_data_array candidates_p = {candidates.data(), candidates.size(), false}; // Apply penalties - float nl_logit = logits[llama_token_nl()]; + float nl_logit = logits[llama_token_nl(ctx)]; auto last_n_repeat = std::min(std::min((int)last_n_tokens.size(), repeat_last_n), params.n_ctx); llama_sample_repetition_penalty(ctx, &candidates_p, last_n_tokens.data() + last_n_tokens.size() - last_n_repeat, @@ -452,7 +452,7 @@ struct llama_server_context last_n_repeat, alpha_frequency, alpha_presence); if (!penalize_nl) { - logits[llama_token_nl()] = nl_logit; + logits[llama_token_nl(ctx)] = nl_logit; } if (grammar != nullptr) { @@ -515,7 +515,7 @@ struct llama_server_context // decrement remaining sampling budget --n_remain; - if (!embd.empty() && embd.back() == llama_token_eos()) + if (!embd.empty() && embd.back() == llama_token_eos(ctx)) { // stopping_word = llama_token_to_str(ctx, embd.back()); has_next_token = false; @@ -652,8 +652,6 @@ static void server_print_usage(const char *argv0, const gpt_params ¶ms, fprintf(stdout, " -v, --verbose verbose output (default: %s)\n", server_verbose ? "enabled" : "disabled"); fprintf(stdout, " -t N, --threads N number of threads to use during computation (default: %d)\n", params.n_threads); fprintf(stdout, " -c N, --ctx-size N size of the prompt context (default: %d)\n", params.n_ctx); - fprintf(stdout, " -gqa N, --gqa N grouped-query attention factor (TEMP!!! use 8 for LLaMAv2 70B) (default: %d)\n", params.n_gqa); - fprintf(stdout, " -eps N, --rms-norm-eps N rms norm eps (TEMP!!! use 1e-5 for LLaMAv2) (default: %.1e)\n", params.rms_norm_eps); fprintf(stdout, " --rope-freq-base N RoPE base frequency (default: %.1f)\n", params.rope_freq_base); fprintf(stdout, " --rope-freq-scale N RoPE frequency scaling factor (default: %g)\n", params.rope_freq_scale); fprintf(stdout, " -b N, --batch-size N batch size for prompt processing (default: %d)\n", params.n_batch); @@ -774,23 +772,6 @@ static void server_params_parse(int argc, char **argv, server_params &sparams, } params.n_ctx = std::stoi(argv[i]); } - else if (arg == "-gqa" || arg == "--gqa") - { - if (++i >= argc) - { - invalid_param = true; - break; - } - params.n_gqa = std::stoi(argv[i]); - } - else if (arg == "-eps" || arg == "--rms-norm-eps") { - if (++i >= argc) - { - invalid_param = true; - break; - } - params.rms_norm_eps = std::stof(argv[i]); - } else if (arg == "--rope-freq-base") { if (++i >= argc) @@ -968,7 +949,7 @@ static void server_params_parse(int argc, char **argv, server_params &sparams, static json format_generation_settings(llama_server_context &llama) { - const auto eos_bias = llama.params.logit_bias.find(llama_token_eos()); + const auto eos_bias = llama.params.logit_bias.find(llama_token_eos(llama.ctx)); const bool ignore_eos = eos_bias != llama.params.logit_bias.end() && eos_bias->second < 0.0f && std::isinf(eos_bias->second); @@ -1103,7 +1084,7 @@ static void parse_options_completion(const json &body, llama_server_context &lla llama.params.logit_bias.clear(); if (body.value("ignore_eos", false)) { - llama.params.logit_bias[llama_token_eos()] = -INFINITY; + llama.params.logit_bias[llama_token_eos(llama.ctx)] = -INFINITY; } const auto &logit_bias = body.find("logit_bias"); diff --git a/examples/simple/simple.cpp b/examples/simple/simple.cpp index 97137a65..132f7fbf 100644 --- a/examples/simple/simple.cpp +++ b/examples/simple/simple.cpp @@ -2,180 +2,129 @@ #define _GNU_SOURCE #endif +#include "build-info.h" + #include "common.h" #include "llama.h" -#include "build-info.h" -#include <cassert> -#include <cinttypes> #include <cmath> #include <cstdio> -#include <cstring> -#include <ctime> -#include <fstream> -#include <iostream> #include <string> #include <vector> -#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) -#include <signal.h> -#include <unistd.h> -#elif defined (_WIN32) -#define WIN32_LEAN_AND_MEAN -#define NOMINMAX -#include <windows.h> -#include <signal.h> -#endif - - - -int main(int argc, char ** argv) -{ +int main(int argc, char ** argv) { gpt_params params; - //--------------------------------- - // Print help : - //--------------------------------- - - if ( argc == 1 || argv[1][0] == '-' ) - { - printf( "usage: %s MODEL_PATH [PROMPT]\n" , argv[0] ); + if (argc == 1 || argv[1][0] == '-') { + printf("usage: %s MODEL_PATH [PROMPT]\n" , argv[0]); return 1 ; } - //--------------------------------- - // Load parameters : - //--------------------------------- - - if ( argc >= 2 ) - { + if (argc >= 2) { params.model = argv[1]; } - if ( argc >= 3 ) - { + if (argc >= 3) { params.prompt = argv[2]; } - if ( params.prompt.empty() ) - { + if (params.prompt.empty()) { params.prompt = "Hello my name is"; } - //--------------------------------- - // Init LLM : - //--------------------------------- + // init LLM llama_backend_init(params.numa); - llama_model * model; - llama_context * ctx; + llama_context_params ctx_params = llama_context_default_params(); - std::tie(model, ctx) = llama_init_from_gpt_params( params ); + llama_model * model = llama_load_model_from_file(params.model.c_str(), ctx_params); - if ( model == NULL ) - { - fprintf( stderr , "%s: error: unable to load model\n" , __func__ ); + if (model == NULL) { + fprintf(stderr , "%s: error: unable to load model\n" , __func__); return 1; } - //--------------------------------- - // Tokenize the prompt : - //--------------------------------- + llama_context * ctx = llama_new_context_with_model(model, ctx_params); + + // tokenize the prompt std::vector<llama_token> tokens_list; - tokens_list = ::llama_tokenize( ctx , params.prompt , true ); + tokens_list = ::llama_tokenize(ctx, params.prompt, true); - const int max_context_size = llama_n_ctx( ctx ); - const int max_tokens_list_size = max_context_size - 4 ; + const int max_context_size = llama_n_ctx(ctx); + const int max_tokens_list_size = max_context_size - 4; - if ( (int)tokens_list.size() > max_tokens_list_size ) - { - fprintf( stderr , "%s: error: prompt too long (%d tokens, max %d)\n" , - __func__ , (int)tokens_list.size() , max_tokens_list_size ); + if ((int) tokens_list.size() > max_tokens_list_size) { + fprintf(stderr, "%s: error: prompt too long (%d tokens, max %d)\n", __func__, (int) tokens_list.size(), max_tokens_list_size); return 1; } - fprintf( stderr, "\n\n" ); - - // Print the tokens from the prompt : + fprintf(stderr, "\n\n"); - for( auto id : tokens_list ) - { - printf( "%s" , llama_token_to_str( ctx , id ) ); + for (auto id : tokens_list) { + fprintf(stderr, "%s", llama_token_to_str(ctx, id).c_str()); } - fflush(stdout); - + fflush(stderr); - //--------------------------------- - // Main prediction loop : - //--------------------------------- + // main loop // The LLM keeps a contextual cache memory of previous token evaluation. // Usually, once this cache is full, it is required to recompute a compressed context based on previous // tokens (see "infinite text generation via context swapping" in the main example), but in this minimalist // example, we will just stop the loop once this cache is full or once an end of stream is detected. - while ( llama_get_kv_cache_token_count( ctx ) < max_context_size ) - { - //--------------------------------- - // Evaluate the tokens : - //--------------------------------- + const int n_gen = std::min(32, max_context_size); - if ( llama_eval( ctx , tokens_list.data() , int(tokens_list.size()) , llama_get_kv_cache_token_count( ctx ) , params.n_threads ) ) - { - fprintf( stderr, "%s : failed to eval\n" , __func__ ); + while (llama_get_kv_cache_token_count(ctx) < n_gen) { + // evaluate the transformer + + if (llama_eval(ctx, tokens_list.data(), int(tokens_list.size()), llama_get_kv_cache_token_count(ctx), params.n_threads)) { + fprintf(stderr, "%s : failed to eval\n", __func__); return 1; } tokens_list.clear(); - //--------------------------------- - // Select the best prediction : - //--------------------------------- + // sample the next token llama_token new_token_id = 0; - auto logits = llama_get_logits( ctx ); - auto n_vocab = llama_n_vocab( ctx ); // the size of the LLM vocabulary (in tokens) + auto logits = llama_get_logits(ctx); + auto n_vocab = llama_n_vocab(ctx); std::vector<llama_token_data> candidates; - candidates.reserve( n_vocab ); + candidates.reserve(n_vocab); - for( llama_token token_id = 0 ; token_id < n_vocab ; token_id++ ) - { - candidates.emplace_back( llama_token_data{ token_id , logits[ token_id ] , 0.0f } ); + for (llama_token token_id = 0; token_id < n_vocab; token_id++) { + candidates.emplace_back(llama_token_data{ token_id, logits[token_id], 0.0f }); } llama_token_data_array candidates_p = { candidates.data(), candidates.size(), false }; - // Select it using the "Greedy sampling" method : - new_token_id = llama_sample_token_greedy( ctx , &candidates_p ); - + new_token_id = llama_sample_token_greedy(ctx , &candidates_p); // is it an end of stream ? - if ( new_token_id == llama_token_eos() ) - { + if (new_token_id == llama_token_eos(ctx)) { fprintf(stderr, " [end of text]\n"); break; } - // Print the new token : - printf( "%s" , llama_token_to_str( ctx , new_token_id ) ); - fflush( stdout ); + // print the new token : + printf("%s", llama_token_to_str(ctx, new_token_id).c_str()); + fflush(stdout); - // Push this new token for next evaluation : - tokens_list.push_back( new_token_id ); - - } // wend of main loop + // push this new token for next evaluation + tokens_list.push_back(new_token_id); + } - llama_free( ctx ); - llama_free_model( model ); + llama_free(ctx); + llama_free_model(model); llama_backend_free(); + fprintf(stderr, "\n\n"); + return 0; } - -// EOF diff --git a/examples/train-text-from-scratch/train-text-from-scratch.cpp b/examples/train-text-from-scratch/train-text-from-scratch.cpp index 54dc2bee..31d6620a 100644 --- a/examples/train-text-from-scratch/train-text-from-scratch.cpp +++ b/examples/train-text-from-scratch/train-text-from-scratch.cpp @@ -1,4 +1,5 @@ #include "ggml.h" +#include "common.h" #include "llama.h" #include <unordered_map> #include <vector> @@ -16,7 +17,7 @@ #pragma warning(disable: 4244 4267) // possible loss of data #endif -static const float rms_norm_eps = LLAMA_DEFAULT_RMS_EPS; +static const float rms_norm_eps = 1e-5f; struct random_normal_distribution { std::mt19937 gen; @@ -169,14 +170,16 @@ struct ggml_tensor * randomize_tensor_uniform(struct ggml_tensor * tensor, struc struct llama_vocab { using id = int32_t; using token = std::string; + using ttype = llama_token_type; - struct token_score { - token tok; + struct token_data { + token text; float score; + ttype type; }; std::unordered_map<token, id> token_to_id; - std::vector<token_score> id_to_token; + std::vector<token_data> id_to_token; }; struct my_llama_hparams { @@ -1961,7 +1964,7 @@ void print_matrix(struct ggml_tensor * probs) { void print_token(struct llama_context * ctx, llama_token token) { - printf("%s", llama_token_to_str(ctx, token)); + printf("%s", llama_token_to_str(ctx, token).c_str()); } void print_tokens(struct llama_context* ctx, struct ggml_tensor * tokens) { @@ -1995,7 +1998,7 @@ void print_tokens_batch(struct llama_context* ctx, struct ggml_tensor * tokens) } } -void get_example_targets(const int * train_samples, size_t n_train_samples, const llama_token * train_data, size_t n_train_data, int example_id, struct ggml_tensor * tokens_input, struct ggml_tensor * target_logits, struct ggml_tensor * target_probs) { +void get_example_targets(struct llama_context * lctx, const int * train_samples, size_t n_train_samples, const llama_token * train_data, size_t n_train_data, int example_id, struct ggml_tensor * tokens_input, struct ggml_tensor * target_logits, struct ggml_tensor * target_probs) { int n_tokens = tokens_input->ne[0]; int n_vocab = target_logits->ne[0]; @@ -2004,7 +2007,7 @@ void get_example_targets(const int * train_samples, size_t n_train_samples, cons ggml_set_f32(target_logits, -1.0f/n_vocab); ggml_set_f32(target_probs, 0.0f); - ggml_set_i32_1d(tokens_input, 0, llama_token_bos()); + ggml_set_i32_1d(tokens_input, 0, llama_token_bos(lctx)); for (int i=1; i<n_tokens+1; ++i) { int token = clamp(train_data[sample+i-1], 0, n_vocab-1); set_f32_2d(target_logits, token, i-1, +1.0f); @@ -2015,7 +2018,7 @@ void get_example_targets(const int * train_samples, size_t n_train_samples, cons } } -void get_example_targets_batch(struct llama_context * /*lctx*/, const int * train_samples, size_t n_train_samples, const llama_token * train_data, size_t n_train_data, int example_id, struct ggml_tensor * tokens_input, struct ggml_tensor * target_logits, struct ggml_tensor * target_probs) { +void get_example_targets_batch(struct llama_context * lctx, const int * train_samples, size_t n_train_samples, const llama_token * train_data, size_t n_train_data, int example_id, struct ggml_tensor * tokens_input, struct ggml_tensor * target_logits, struct ggml_tensor * target_probs) { GGML_ASSERT(tokens_input->n_dims == 2); GGML_ASSERT(target_logits->n_dims == 3); GGML_ASSERT(target_probs->n_dims == 3); @@ -2035,7 +2038,7 @@ void get_example_targets_batch(struct llama_context * /*lctx*/, const int * trai size_t sample = train_samples[(example_id*n_batch + k) % n_train_samples]; GGML_ASSERT(sample+n_tokens-1 < n_train_data); - set_i32_2d(tokens_input, 0, k, llama_token_bos()); + set_i32_2d(tokens_input, 0, k, llama_token_bos(lctx)); for (int i=1; i<n_tokens+1; ++i) { int token = clamp(train_data[sample+i-1], 0, n_vocab-1); // print_token(lctx, token); @@ -2188,11 +2191,10 @@ int tokenize_file(struct llama_context * lctx, const char * filename, std::vecto f.read_raw(buf.data(), f.size); buf[f.size] = '\0'; - out.resize(buf.size()); - - int n_tokens = llama_tokenize(lctx, buf.data(), out.data(), buf.size(), false); - if (n_tokens >= 0) { - out.resize(n_tokens); + int n_tokens = llama_tokenize(lctx, buf.data(), out.data(), out.size(), false); + if (n_tokens < 0) { + out.resize(-n_tokens); + llama_tokenize(lctx, buf.data(), out.data(), out.size(), false); } bool verify = false; @@ -2200,17 +2202,17 @@ int tokenize_file(struct llama_context * lctx, const char * filename, std::vecto const char * in = buf.data(); const char * end = buf.data() + buf.size(); for (int i = 0; i < (int) out.size(); ++i) { - const char * s = llama_token_to_str(lctx, out[i]); - int len = strlen(s); + std::string s = llama_token_to_str(lctx, out[i]); + int len = s.length(); if (in >= end) { printf("%s: unexpected end of original text.\n", __func__); break; } - const bool matches = (strncmp(in, s, len) == 0); + const bool matches = (strncmp(in, s.c_str(), len) == 0); if (matches) { in += len; } else { - printf("%s: mismatch: expected '%s', but got '%s'\n", __func__, std::string(in, len).c_str(), s); + printf("%s: mismatch: expected '%s', but got '%s'\n", __func__, std::string(in, len).c_str(), s.c_str()); } } } @@ -2294,7 +2296,7 @@ llama_token sample(struct my_llama_sampler * sampler, float * logits, const llam const auto params = sampler->params; // Apply penalties - const float nl_logit = logits[llama_token_nl()]; + const float nl_logit = logits[llama_token_nl(ctx)]; const int n_last = std::min(std::min(n_last_tokens, params.repeat_last_n), sampler->n_ctx); @@ -2313,7 +2315,7 @@ llama_token sample(struct my_llama_sampler * sampler, float * logits, const llam params.alpha_presence); if (!params.penalize_nl) { - logits[llama_token_nl()] = nl_logit; + logits[llama_token_nl(ctx)] = nl_logit; } llama_token token = 0; @@ -2612,42 +2614,45 @@ void save_as_llama_model(struct llama_vocab * vocab, struct my_llama_model * mod return; } - // write_magic - file.write_u32(LLAMA_FILE_MAGIC); // magic - file.write_u32(LLAMA_FILE_VERSION); // version - // write_hparams - file.write_u32(model->hparams.n_vocab); - file.write_u32(model->hparams.n_embd); - file.write_u32(model->hparams.n_mult); - file.write_u32(model->hparams.n_head); - file.write_u32(model->hparams.n_layer); - file.write_u32(model->hparams.n_rot); - file.write_u32(LLAMA_FTYPE_ALL_F32); - // write_vocab - uint32_t n_vocab = model->hparams.n_vocab; - for (uint32_t i = 0; i < n_vocab; i++) { - const auto & token_score = vocab->id_to_token.at(i); - file.write_u32((uint32_t) token_score.tok.size()); - file.write_raw(token_score.tok.data(), token_score.tok.size()); - file.write_raw(&token_score.score, sizeof(token_score.score)); - } - // write tensors - write_tensor(&file, model->tok_embeddings); - write_tensor(&file, model->norm); - write_tensor(&file, model->output); - for (uint32_t i = 0; i < model->hparams.n_layer; ++i) { - auto & layer = model->layers[i]; - - write_tensor(&file, layer.attention_norm); - write_tensor(&file, layer.wq); - write_tensor(&file, layer.wk); - write_tensor(&file, layer.wv); - write_tensor(&file, layer.wo); - write_tensor(&file, layer.ffn_norm); - write_tensor(&file, layer.w1); - write_tensor(&file, layer.w2); - write_tensor(&file, layer.w3); - } +#pragma message("TODO: implement file saving using gguf") + (void) vocab; + (void) model; +// // write_magic +// file.write_u32(LLAMA_FILE_MAGIC); // magic +// file.write_u32(LLAMA_FILE_VERSION); // version +// // write_hparams +// file.write_u32(model->hparams.n_vocab); +// file.write_u32(model->hparams.n_embd); +// file.write_u32(model->hparams.n_mult); +// file.write_u32(model->hparams.n_head); +// file.write_u32(model->hparams.n_layer); +// file.write_u32(model->hparams.n_rot); +// file.write_u32(LLAMA_FTYPE_ALL_F32); +// // write_vocab +// uint32_t n_vocab = model->hparams.n_vocab; +// for (uint32_t i = 0; i < n_vocab; i++) { +// const auto & token_data = vocab->id_to_token.at(i); +// file.write_u32((uint32_t) token_data.tok.size()); +// file.write_raw(token_data.tok.data(), token_data.tok.size()); +// file.write_raw(&token_data.score, sizeof(token_data.score)); +// } +// // write tensors +// write_tensor(&file, model->tok_embeddings); +// write_tensor(&file, model->norm); +// write_tensor(&file, model->output); +// for (uint32_t i = 0; i < model->hparams.n_layer; ++i) { +// auto & layer = model->layers[i]; +// +// write_tensor(&file, layer.attention_norm); +// write_tensor(&file, layer.wq); +// write_tensor(&file, layer.wk); +// write_tensor(&file, layer.wv); +// write_tensor(&file, layer.wo); +// write_tensor(&file, layer.ffn_norm); +// write_tensor(&file, layer.w1); +// write_tensor(&file, layer.w2); +// write_tensor(&file, layer.w3); +// } } float cosine_decay(const int decay_steps, const float alpha, int step) { @@ -3052,20 +3057,13 @@ int main(int argc, char ** argv) { struct llama_vocab vocab; { - std::vector<const char *> strings; - std::vector<float> scores; - int n_vocab = llama_n_vocab(lctx); - strings.resize(n_vocab, NULL); - scores.resize(n_vocab, 0); - n_vocab = llama_get_vocab(lctx, strings.data(), scores.data(), n_vocab); - GGML_ASSERT(n_vocab == llama_n_vocab(lctx)); + const int n_vocab = llama_n_vocab(lctx); vocab.id_to_token.resize(n_vocab); for (int i=0; i<n_vocab; ++i) { - std::string tok = std::string(strings[i]); - float score = scores[i]; - vocab.id_to_token[i].tok = tok; - vocab.id_to_token[i].score = score; - vocab.token_to_id.emplace(tok, i); + vocab.id_to_token[i].text = llama_token_get_text(lctx, i); + vocab.id_to_token[i].score = llama_token_get_score(lctx, i); + vocab.id_to_token[i].type = llama_token_get_type(lctx, i); + vocab.token_to_id.emplace(vocab.id_to_token[i].text, i); } } @@ -3178,7 +3176,7 @@ int main(int argc, char ** argv) { std::vector<int> train_samples; train_samples.push_back(0); for (int i = 1; i < (int) train_tokens.size() - n_tokens; ++i) { - if (!params.samples_start_after_nl || (train_tokens[i-1] == llama_token_nl())) { + if (!params.samples_start_after_nl || (train_tokens[i-1] == llama_token_nl(lctx))) { train_samples.push_back(i); } } @@ -3338,7 +3336,7 @@ int main(int argc, char ** argv) { struct ggml_tensor * target_logits = ggml_new_tensor_2d(model.ctx, GGML_TYPE_F32, n_vocab, n_tokens); struct ggml_tensor * target_probs = ggml_new_tensor_2d(model.ctx, GGML_TYPE_F32, n_vocab, n_tokens); - get_example_targets(train_samples.data(), train_samples.size(), train_tokens.data(), train_tokens.size(), rand()%train_samples.size(), tokens_input, target_logits, target_probs); + get_example_targets(lctx, train_samples.data(), train_samples.size(), train_tokens.data(), train_tokens.size(), rand()%train_samples.size(), tokens_input, target_logits, target_probs); for (int i=sample_ctx; i<n_tokens; ++i) { ggml_set_i32_1d(tokens_input, i, n_vocab/2); } |