---
id: dvclive
version: "3.49.1"
license: Apache-2.0
license_treatment: permissive
maintenance: active
---

# dvclive — Experiments logger for ML projects.

License: permissive · Maintenance: active · Downloads: 569.8K/mo

## What it is and what it does

DVCLive is a Python library that logs machine learning experiment metadata—metrics, parameters, and plots—to plain text files compatible with DVC. Unlike heavier ML logging platforms, it stores everything locally as versioned files that can be tracked by Git or DVC, eliminating the need for external services or servers. It integrates with the DVC ecosystem to enable experiment comparison and visualization through the DVC CLI, VS Code extension, or DVC Studio.

The library is designed for teams already using DVC for data versioning and experiment management. It provides a simple context-manager API for logging during training loops, supports framework integrations, and can save experiment runs as hidden Git commits for reproducibility. The main trade-off is that it assumes you're working within a DVC-managed repository rather than offering a standalone logging solution.

Use it for:

- Log training metrics and parameters during model development, then compare across multiple experiment runs using dvc exp show.
- Track validation metrics in a Git-versioned format without setting up external services.
- Integrate with PyTorch Lightning or Scikit-learn workflows to automatically capture metrics alongside your training code.
- Visualize experiment results in VS Code using the DVC extension or push to DVC Studio for team collaboration.
- Store plots and metrics as plain text files that can be reviewed in version control history without external dependencies.

## Worth the install?

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

DVCLive logs machine learning metrics, parameters, and plots to simple file formats that integrate with DVC for experiment tracking and comparison.

Yes, if you are already using DVC or planning to adopt it for experiment management. DVCLive is a natural fit for teams that want lightweight, Git-friendly metric logging without external services. If you are not using DVC and prefer a standalone logger, the nine runtime dependencies and DVC integration requirement make it less attractive. No known security vulnerabilities and active maintenance support the recommendation.

## Install

pip install dvclive
uv add dvclive
poetry add dvclive

## Installing dvclive

Before you install: Low friction install with a pure-Python wheel. Active maintenance (last commit 2026-08-10) and recent release (2026-06-05). Depends on nine runtime packages including dvc, dvc-render, and dvc-studio-client, which may add setup complexity if not already present.

License in practice: Apache-2.0 permissive license allows commercial and private use with minimal restrictions.

Quickstart:

pip install dvclive

from dvclive import Live

with Live() as live:
    live.log_param("learning_rate", 0.002)
    live.log_metric("accuracy", 0.5)
    live.next_step()

Requires Python 3.9 or later. A DVC repository (git init && dvc init) is expected for full functionality.

Verify before relying:

- Whether nvidia-ml-py is required for all use cases or only when GPU metrics are needed
- Performance characteristics when logging high-frequency metrics in long-running experiments

## Package facts

- License: Apache-2.0 (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 569.8K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

ml metrics logging, experiment tracking, dvc integration, machine learning logger, training metrics storage, experiment-tracking, dvc-ecosystem, ml-logging

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