trackio-rs

trackio-rs is the high-performance Rust client for Trackio, the open-source experiment tracker built by Hugging Face.

This SDK enables systems engineers and ML researchers to log metrics from high-concurrency, production-grade environments with the safety and speed of Rust.

Why trackio-rs?

While the Python client is ideal for research, trackio-rs is built for:

• Production Systems: Low-latency metric logging for live model monitoring.
• Thread Safety: Built on parking_lot and reqwest for safe concurrent logging.
• Memory Efficiency: Minimal footprint for edge devices and high-throughput servers.

Quickstart

1. Set Up Your Dashboard

Deploy your Trackio Space on Hugging Face: Deploy Trackio Template

2. Installation

Add this to your Cargo.toml:

[dependencies]
trackio-rs = "0.1.0"

Quickstart

1. Create a Trackio Dashboard Space

Create your dashboard Space: https://huggingface.co/new-space?sdk=gradio&template=gradio-templates%2Ftrackio-dashboard

Once deployed, the iframed Space URL will be something like: https://username-trackio-dashboard.hf.space (you can find the iframed URL by clicking the triple dot menu next to Settings and then clicking "Embed this Space")

2. Log metrics

Set environment variables and run the example:

export TRACKIO_SERVER_URL="https://your-space-url.hf.space"
export HF_TOKEN="hf_…"
export TRACKIO_PROJECT="rs-quickstart"
export TRACKIO_RUN="rs-run-1"

cargo run --example quickstart

3. View in the Dashboard

Open your Space URL and select:

• Project: rs-quickstart
• Run: rs-run-1

Open the "Metrics" tab to view your logged metrics.

Usage

use trackio::Client;
use serde_json::json;

let client = Client::new()
    .with_base_url("https://your-space-url.hf.space")
    .with_project("my-project")
    .with_run("my-run");

client.log(json!({"loss": 0.5, "acc": 0.8}), Some(0), None);
client.log(json!({"loss": 0.4, "acc": 0.82}), Some(1), None);
client.flush()?;

Environment Variables