Implementation:Ggml org Llama cpp Download Header

Knowledge Sources	Ggml_org_Llama_cpp
Domains	Networking, Model_Management
Last Updated	2026-02-15 00:00 GMT

Overview

Declares the API for downloading models from HuggingFace Hub, Docker registries, and direct URLs with caching support.

Description

This header defines `common_remote_params` for HTTP configuration (headers, timeout, max_size) and `common_remote_get_content` for raw URL fetching. It provides `common_download_split_repo_tag` to parse "user/model:tag" strings, `common_get_hf_file` to resolve HF repo tags to specific GGUF files (with optional mmproj files), and `common_download_model` for the full download workflow. Additionally, `common_list_cached_models` returns locally cached models, `common_download_file_single` downloads individual files, and `common_docker_resolve_model` resolves models from Docker/OCI registries.

Usage

Use this header when implementing model acquisition features such as `--hf-repo` and `--model-url` CLI options. Include it to download models from HuggingFace, resolve Docker registry references, or manage the local model cache.

Code Reference

Source Location

• Repository: Ggml_org_Llama_cpp
• File: common/download.h
• Lines: 1-84

Signature

struct common_remote_params {
    common_header_list headers;
    long timeout  = 0;
    long max_size = 0;
};

std::pair<long, std::vector<char>> common_remote_get_content(const std::string & url, const common_remote_params & params);

std::pair<std::string, std::string> common_download_split_repo_tag(const std::string & hf_repo_with_tag);

common_hf_file_res common_get_hf_file(
    const std::string & hf_repo_with_tag,
    const std::string & bearer_token,
    bool offline,
    const common_header_list & headers = {});

bool common_download_model(
    const common_params_model & model,
    const std::string & bearer_token,
    bool offline,
    const common_header_list & headers = {});

std::vector<common_cached_model_info> common_list_cached_models();

int common_download_file_single(const std::string & url,
                                const std::string & path,
                                const std::string & bearer_token,
                                bool offline,
                                const common_header_list & headers = {});

std::string common_docker_resolve_model(const std::string & docker);

Import

#include "download.h"

I/O Contract

Inputs

Name	Type	Required	Description
url	const std::string &	Yes	URL to fetch content from or download a file from
params	const common_remote_params &	Yes	HTTP configuration including headers, timeout, and max size
hf_repo_with_tag	const std::string &	Yes	HuggingFace repo string in "user/model:tag" format
bearer_token	const std::string &	Yes	Authentication token for HuggingFace API
offline	bool	Yes	If true, only use cached data without network requests
headers	const common_header_list &	No	Additional HTTP headers for requests
model	const common_params_model &	Yes	Model parameters for download_model
path	const std::string &	Yes	Local file path for single file download
docker	const std::string &	Yes	Docker registry reference string

Outputs

Name	Type	Description
remote_get_content return	std::pair<long, std::vector<char>>	HTTP status code and raw response body
split_repo_tag return	std::pair<std::string, std::string>	Separated repo name and tag
get_hf_file return	common_hf_file_res	Resolved repo, GGUF file path, and optional mmproj file path
download_model return	bool	True if download succeeded
list_cached_models return	std::vector<common_cached_model_info>	List of locally cached model information
download_file_single return	int	HTTP status code or -1 on error
docker_resolve_model return	std::string	Local path to the downloaded model file

Usage Examples

#include "download.h"

// Parse a HuggingFace repo with tag
auto [repo, tag] = common_download_split_repo_tag("bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M");
// repo = "bartowski/Llama-3.2-3B-Instruct-GGUF", tag = "Q4_K_M"

// Resolve HF repo to specific GGUF file
auto hf_file = common_get_hf_file("bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M", token, false);

// Download a model
common_params_model model_params;
bool success = common_download_model(model_params, token, false);

// List cached models
auto cached = common_list_cached_models();

Related Pages

• Principle:Ggml_org_Llama_cpp_Model_Download