Implementation:Ggml org Llama cpp Log

Knowledge Sources	Ggml_org_Llama_cpp
Domains	Logging, Infrastructure
Last Updated	2026-02-15 00:00 GMT

Overview

Implements the asynchronous logging system with a worker thread that buffers and dispatches log messages to stderr and optional file output.

Description

The common_log struct uses a worker thread that reads from a message queue (vector of common_log_entry). Log entries carry a level (debug/info/warn/error/output), timestamp, prefix flag, and message text. common_log_add formats messages via vsnprintf and pushes them to the queue, with the worker thread printing to stderr (filtered by verbosity threshold) and optionally to a log file (unfiltered). The system supports colored output via ANSI codes (auto-detected or forced), timestamp prefixes, and log level indicators (D/I/W/E/O).

Usage

Use this module through the LOG_DBG, LOG_INF, LOG_WRN, LOG_ERR macros throughout the codebase. The singleton common_log_main() provides a global instance with automatic cleanup at exit. The async design prevents logging from blocking inference performance.

Code Reference

Source Location

• Repository: Ggml_org_Llama_cpp
• File: common/log.cpp
• Lines: 1-446

Signature

int common_log_verbosity_thold = LOG_DEFAULT_LLAMA;

void common_log_set_verbosity_thold(int verbosity);

struct common_log_entry {
    enum ggml_log_level level;
    bool prefix;
    int64_t timestamp;
    std::vector<char> msg;
    bool is_end;
    void print(FILE * file = nullptr) const;
};

struct common_log {
    // Worker thread, message queue, mutex, condition variable
    // Methods: add, resume, pause, set_file, set_colors, set_prefix, set_timestamps
};

common_log * common_log_main();

Import

#include "common.h"
#include "log.h"
#include <chrono>
#include <condition_variable>
#include <mutex>
#include <thread>
#include <vector>

I/O Contract

Inputs

Name	Type	Required	Description
level	enum ggml_log_level	Yes	Log level: GGML_LOG_LEVEL_DEBUG, _INFO, _WARN, _ERROR
prefix	bool	Yes	Whether to prepend timestamp and level indicator
msg	const char * (format string)	Yes	Printf-style format string for the log message
verbosity	int	No	Threshold for filtering messages on stderr output

Outputs

Name	Type	Description
stderr output	text	Formatted log messages printed to stderr (filtered by verbosity)
file output	text	All log messages written to optional log file (unfiltered)

Usage Examples

#include "log.h"

// Use logging macros (most common usage)
LOG_DBG("Loading model from %s\n", path.c_str());
LOG_INF("Model loaded: %d layers, %d parameters\n", n_layers, n_params);
LOG_WRN("Context size %d exceeds maximum, clamping to %d\n", ctx_size, max_ctx);
LOG_ERR("Failed to allocate %zu bytes for KV cache\n", required_bytes);

// Configure logging
common_log_set_verbosity_thold(LOG_DEFAULT_DEBUG);  // Show debug messages
common_log * log = common_log_main();
log->set_file("llama.log");   // Enable file logging
log->set_colors(true);        // Enable ANSI color output
log->set_timestamps(true);    // Enable timestamp prefixes

Related Pages

• Principle:Ggml_org_Llama_cpp_LoggingInfrastructure