Implementation:Ggml org Llama cpp Model Saver

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

Overview

Implements the `llama_model_saver` class that serializes a loaded model back to GGUF format on disk.

Description

This file provides the model serialization logic for writing llama.cpp models to GGUF files. It initializes an empty GGUF context and provides overloaded `add_kv` methods for writing typed metadata (uint32, int32, float, bool, string, arrays, string vectors). A template `add_kv` method handles per-layer arrays with an optimization to collapse uniform values into scalars. The `add_kv_from_model` method writes all model metadata including architecture, hyperparameters, vocabulary tokens/scores/types, RoPE parameters, and expert configurations, while `add_tensors_from_model` adds all tensor descriptors and `save` writes the complete GGUF file.

Usage

Use this module when exporting or re-serializing models to GGUF format, such as during quantization, conversion, or model manipulation workflows that need to write modified models back to disk.

Code Reference

Source Location

• Repository: Ggml_org_Llama_cpp
• File: src/llama-model-saver.cpp
• Lines: 1-285

Signature

class llama_model_saver {
    llama_model_saver(const struct llama_model & model);
    ~llama_model_saver();

    void add_kv(const enum llm_kv key, const uint32_t value);
    void add_kv(const enum llm_kv key, const int32_t value);
    void add_kv(const enum llm_kv key, const float value);
    void add_kv(const enum llm_kv key, const bool value);
    void add_kv(const enum llm_kv key, const char * value);

    template <typename Container>
    void add_kv(const enum llm_kv key, const Container & value, const bool per_layer);

    void add_kv_from_model();
    void add_tensors_from_model();
    void save(const std::string & path);
};

Import

#include "llama-model-saver.h"

I/O Contract

Inputs

Name	Type	Required	Description
model	const llama_model &	Yes	Reference to the loaded model to serialize
key	enum llm_kv	Yes	Metadata key identifier for KV pairs
value	various (uint32_t, int32_t, float, bool, string, Container)	Yes	Value to write for the given metadata key
per_layer	bool	No	Whether Container values are per-layer (collapses uniform arrays to scalars)
path	const std::string &	Yes	Output file path for the GGUF file (in save method)

Outputs

Name	Type	Description
GGUF file	file on disk	Complete serialized model in GGUF format with metadata and tensor data

Usage Examples

// Create a model saver from a loaded model
llama_model_saver saver(model);

// Write all metadata and tensors
saver.add_kv_from_model();
saver.add_tensors_from_model();

// Save to disk
saver.save("output_model.gguf");

Related Pages

• Principle:Ggml_org_Llama_cpp_Model_Serialization