gguf_parser.h

Go to the documentation of this file.

#pragma once
 
#include <cstddef>
#include <fstream>
#include <map>
#include <string>
#include <vector>
 
#include "gguf_structs.h"
 
constexpr uint32_t GGUF_MAGIC = 0x46554747;
 
constexpr uint64_t GGUF_DEFAULT_ALIGNMENT = 32;          
constexpr uint32_t GGUF_MAX_TENSOR_DIMS = 4;            
constexpr uint64_t GGUF_STRING_MAX_LENGTH = 1ull << 30; 
constexpr float GGUF_EPSILON = 1e-10f;   
constexpr float GGUF_SMALL_VAL = 1e-6f;  
constexpr size_t GGML_QK_K = 256;          
constexpr size_t GGML_QK8_0 = 32;          
constexpr size_t GGML_Q4K_BLOCK_SIZE = 32; 
constexpr size_t GGML_Q6K_BLOCK_SIZE = 256;
constexpr float TENSOR_SCALE_MAX = 1000.0f;  
constexpr float TENSOR_SCALE_MIN = -1000.0f; 
constexpr float Q4K_SCALE_FACTOR = 15.0f;    
constexpr float Q6K_SCALE_FACTOR = 31.0f;    
constexpr float Q8K_SCALE_FACTOR = 127.0f;   
constexpr int8_t Q4K_OFFSET = 8;    
constexpr int8_t Q6K_OFFSET = 32;   
template <typename T>
void read_raw(std::ifstream& file, T& dest);
 
std::string read_gguf_string(std::ifstream& file);
 
GGUFData load_gguf_meta(const std::string& filename, bool use_mmap);