TIREx Tracker¶

Automatic resource and metadata tracking for information retrieval experiments.

The reproducibility and transparency of retrieval experiments depend on properly recording the experimental setup and conditions. Manually curating such metadata is tedious, error-prone, and inconsistent. TIREx Tracker automates this: wrap your experiment in a single call, and get a structured record of hardware, software, timing, energy, and Git state — without changing your code.

What it tracks¶

TIREx Tracker collects over 50 measurements across six categories:

Category	Examples
OS	OS name and version, kernel version
Time	Wall-clock, user-mode, and system-mode elapsed time
CPU	Model, cores, frequency, utilization (process & system), energy
RAM	Usage (process & system), available, DRAM energy
GPU	Model, VRAM, utilization (process & system), GPU energy (NVIDIA)
Git	Commit hash, branch, tags, uncommitted/unpushed changes, code archive

See the full list of tracked measures for details.

How to use it¶

TIREx Tracker is available as a CLI tool and as a library for C/C++, Python, and Java/Kotlin.

CLIPythonJavaKotlinC

tirex-tracker "python train.py --epochs 10"

from tirex_tracker import tracking

with tracking() as results:
    train_model()

print(results)

var result = Tracker.track(() -> trainModel());
System.out.println(result);

val result = track { trainModel() }
println(result)

#include <tirex_tracker.h>

int main() {
    tirexMeasureConf conf[] = {
        {TIREX_TIME_ELAPSED_WALL_CLOCK_MS, TIREX_AGG_NO},
        {TIREX_CPU_USED_PROCESS_PERCENT,   TIREX_AGG_MAX},
        {TIREX_RAM_USED_PROCESS_KB,        TIREX_AGG_MAX},
        tirexNullConf
    };
    tirexTrackingConf trackingConf = {.measures = conf, .pollIntervalMs = 100};
    tirexMeasureHandle* handle;
    tirexStartTracking(trackingConf, &handle);

    run_experiment();

    tirexResult* result;
    tirexStopTracking(handle, &result);
    tirexResultFree(result);
}

Getting started¶

Choose the guide for your language or use case:

CLI Tool

Download a prebuilt binary and wrap any shell command.
Python

pip install tirex-tracker and use the context manager or decorator.
Java / Kotlin

Add a single Gradle or Maven dependency.
C / C++

Integrate via CMake FetchContent and link against the C API.

Citation¶

If you use TIREx Tracker in your research, please cite:

@inproceedings{tirextracker2025,
    author    = {Hagen, Tim and Fr{\"o}be, Maik and Merker, Jan Heinrich and
                 Scells, Harrisen and Hagen, Matthias and Potthast, Martin},
    booktitle = {48th International ACM SIGIR Conference on Research and
                 Development in Information Retrieval (SIGIR 2025)},
    month     = jul,
    publisher = {ACM},
    title     = {{TIREx Tracker: The Information Retrieval Experiment Tracker}},
    year      = {2025}
}