Implementation:Ollama Ollama Llama Log

Knowledge Sources	Ollama
Domains	Logging, Infrastructure
Last Updated	2025-02-15 00:00 GMT

Overview

Implements the asynchronous logging system for llama.cpp with a background worker thread, color support, and configurable verbosity filtering.

Description

Uses a common_log struct with an internal worker thread that processes log entries from a ring buffer queue. Log entries (common_log_entry) carry a timestamp, log level, and message. The worker thread dequeues entries and writes them to stdout/stderr (or an optional file) based on level. Supports ANSI color codes (auto-detected via isatty), optional timestamps and level prefixes, and verbosity filtering via common_log_verbosity_thold. Provides pause/resume control for the worker thread, and a singleton accessor (common_log_main) with automatic cleanup on exit.

Usage

Use this for all logging within llama.cpp. The asynchronous design prevents logging from blocking inference operations.

Code Reference

Source Location

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

Signature

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 {
    common_log();
    common_log(size_t capacity);
    ~common_log();
    void add(enum ggml_log_level level, const char * fmt, va_list args);
    void resume();
    void pause();
    void set_file(const char * file);
    void set_colors(log_colors colors);
    void set_prefix(bool prefix);
    void set_timestamps(bool timestamps);
};

Import

#include "log.h"

I/O Contract

Inputs

Name	Type	Required	Description
level	ggml_log_level	Yes	Log level (DEBUG, INFO, WARN, ERROR, NONE)
fmt	const char *	Yes	Printf-style format string
args	va_list	Yes	Variadic arguments for format string
verbosity	int	No	Verbosity threshold (lower = fewer messages)

Outputs

Name	Type	Description
output	FILE *	Messages written to stdout/stderr or log file

Usage Examples

#include "log.h"

// Use convenience macros (most common)
LOG_INF("Model loaded: %s\n", model_path);
LOG_WRN("Context size reduced to %d\n", n_ctx);
LOG_ERR("Failed to load model\n");
LOG_DBG("Token %d decoded in %lld us\n", token_id, elapsed);

// Configure logging
common_log * log = common_log_main();
common_log_set_file(log, "output.log");
common_log_set_colors(log, LOG_COLORS_ENABLED);
common_log_set_prefix(log, true);
common_log_set_timestamps(log, true);

Related Pages

• Principle:Ollama_Ollama_Llama_Cpp_Common_Library