---
id: traceml
version: "1.3.0"
license: Apache 2.0
license_treatment: permissive
maintenance: active
---

# traceml — Engine for ML/Data tracking, visualization, dashboards, and model UI for Polyaxon.

License: permissive · Maintenance: active · Downloads: 132.8K/mo

## What it is and what it does

TraceML is an experiment tracking engine designed for machine learning workflows. It logs hyperparameters, metrics, data references, artifacts, and model outputs during training runs, with built-in callbacks for popular frameworks like Keras, PyTorch, TensorFlow, Fastai, PyTorch Lightning, and HuggingFace. It supports both online tracking via API and offline mode for local-only tracking.

The package provides a unified logging interface across different ML frameworks, allowing you to record training progress, save visualizations (matplotlib, Plotly, Altair, Bokeh), and organize experiments by project and run name. It has no runtime dependencies and works with Python 3.8 and later, making it lightweight to integrate into existing ML pipelines.

Use it for:

- Track metrics and hyperparameters during Keras model training using the built-in callback
- Log PyTorch training metrics and save model checkpoints with artifact references
- Record TensorFlow experiment metrics and histograms via the integration callback
- Store visualization artifacts (charts, plots, heatmaps) alongside experiment metadata
- Run offline experiment tracking without connecting to an API server

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

TraceML tracks metrics, parameters, artifacts, and data references for machine learning experiments, with integrations for Keras, PyTorch, TensorFlow, Fastai, PyTorch Lightning, and HuggingFace, plus offline mode for tracking without an API.

Yes, if you need lightweight experiment tracking for ML workflows. Low install friction, no dependencies, active maintenance, and integrations with major frameworks make it practical. Best suited for teams already using or planning to use Polyaxon; standalone offline use is possible but the full value emerges with the Polyaxon platform.

## Install

pip install traceml
uv add traceml
poetry add traceml

## Installing traceml

Before you install: Low install friction: pure Python wheel with no runtime dependencies. Active maintenance with a recent release (2026-04-13) and ongoing repository activity.

License in practice: Apache 2.0 is permissive; you can use, modify, and distribute this package freely in commercial and private projects with minimal restrictions.

Quickstart:

pip install traceml

from traceml import tracking

tracking.init(
    is_offline=True,
    project='my-project',
    name='my-run',
    artifacts_path='./artifacts'
)
tracking.log_inputs(learning_rate=0.001, batch_size=64)
tracking.log_metrics(loss=0.5)
tracking.log_outputs(validation_score=0.66)
tracking.stop()

Verify before relying:

- Whether offline mode fully supports all tracking features or if some require API connectivity
- Performance characteristics when logging large numbers of metrics or artifacts
- Whether the package can be used standalone or requires Polyaxon for full functionality

## Package facts

- License: Apache 2.0 (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 132.8K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

ml experiment tracking, metrics logging framework, deep learning experiment management, pytorch keras tensorflow tracking, ml model artifact storage, experiment visualization dashboards, offline ml tracking, ml-experiment-tracking, framework-integrations

[View on SkillFed](https://skillfed.io/packages/traceml) · [View on PyPI](https://pypi.org/project/traceml/)