Implementation:Ggml org Llama cpp GGUF Example

Knowledge Sources	Ggml_org_Llama_cpp
Domains	GGUF, Example
Last Updated	2026-02-15 00:00 GMT

Overview

Demonstrates GGUF file format read and write operations using the low-level gguf API.

Description

This example creates a GGUF context, writes various key-value metadata types (uint8, int8, uint16, float32, string, arrays of int16/float32/string, etc.), creates 10 random tensors with random dimensions and float32 data, and writes everything to a file. It then reads the file back, prints all KV pairs and tensor info, and verifies the data round-trips correctly by comparing tensor values.

Usage

Use this example as a reference and test for the GGUF file format API, demonstrating the full lifecycle of creating, writing, reading, and verifying GGUF files with metadata and tensor data.

Code Reference

Source Location

• Repository: Ggml_org_Llama_cpp
• File: examples/gguf/gguf.cpp
• Lines: 1-270

Signature

static bool gguf_ex_write(const std::string & fname);
static bool gguf_ex_read_0(const std::string & fname);
static bool gguf_ex_read_1(const std::string & fname, bool check_data);
int main(int argc, char ** argv);

Import

#include "ggml.h"
#include "gguf.h"

#include <cstdio>
#include <string>
#include <sstream>
#include <vector>

I/O Contract

Inputs

Name	Type	Required	Description
fname	std::string	Yes	Output/input filename for the GGUF file (default "test.gguf")

Outputs

Name	Type	Description
return	int	Exit code: 0 on success, 1 on failure
test.gguf	file	Written GGUF file containing metadata KV pairs and 10 random tensors

Usage Examples

# Build and run the GGUF example
cmake --build build --target llama-gguf

# Run with default filename "test.gguf"
./build/bin/llama-gguf

Related Pages

• Principle:Ggml_org_Llama_cpp_GGUF_Format