tracelet 0.0.9

Tracelet is an automagic pytorch metric exporter

Tracelet

Tracelet is a Python library for tracking machine learning experiments. It works with popular backends like MLflow, Weights & Biases, ClearML, and AIM, and can automatically detect hyperparameters from your code.

What it does

• Tracks metrics, parameters, and artifacts across ML backends
• Automatically captures hyperparameters without manual logging
• Integrates with PyTorch, PyTorch Lightning, and TensorBoard
• Handles models, checkpoints, images, and other ML artifacts
• Monitors system resources during training

Installation

pip install tracelet

Install backend support as needed:

pip install tracelet[mlflow]     # For MLflow
pip install tracelet[wandb]      # For Weights & Biases
pip install tracelet[clearml]    # For ClearML
pip install tracelet[aim]        # For AIM
pip install tracelet[all]        # Everything

Usage

Basic Example

from tracelet import Experiment

# Start tracking
exp = Experiment(name="my_experiment", backend=["mlflow"])
exp.start()

# Train your model
for epoch in range(10):
    loss = train_model()
    exp.log_metric("loss", loss, epoch)

exp.stop()

Automatic Detection

from tracelet import Experiment

# Enable automagic to detect hyperparameters automatically
exp = Experiment(name="auto_experiment", backend=["wandb"], automagic=True)
exp.start()

# Define hyperparameters normally - they're captured automatically
learning_rate = 0.001
batch_size = 32

# Your training code here
for epoch in range(epochs):
    loss = train_model()
    # Metrics logged automatically if using TensorBoard

exp.stop()

PyTorch Lightning

from tracelet import Experiment
import pytorch_lightning as pl

# Add to existing Lightning code
exp = Experiment(name="lightning_model", backend=["clearml"], automagic=True)
exp.start()

# Your existing code - all self.log() calls tracked automatically
trainer = pl.Trainer(max_epochs=10)
trainer.fit(model, datamodule)

exp.stop()

Multiple Backends

# Log to multiple backends simultaneously
exp = Experiment(
    name="comparison",
    backend=["mlflow", "wandb", "clearml"]
)

Configuration

Set defaults with environment variables:

export TRACELET_PROJECT="my_project"
export TRACELET_BACKEND="mlflow"
export TRACELET_ENABLE_AUTOMAGIC="true"

Or programmatically:

from tracelet.settings import TraceletSettings

settings = TraceletSettings(
    project="my_project",
    backend=["mlflow"],
    enable_automagic=True,
    track_system=True
)

Examples

The examples/ directory contains runnable examples:

• 01_manual_tracking/ - Basic usage examples
• 02_automagic_tracking/ - Automatic hyperparameter detection
• 03_backend_integrations/ - Backend-specific features
• 05_lightning_automagic/ - PyTorch Lightning integration

cd examples
python 01_manual_tracking/01_basic_manual.py

Documentation

• Documentation
• API Reference
• Configuration Guide

Development

git clone https://github.com/prassanna-ravishankar/tracelet.git
cd tracelet
uv sync
uv run pytest

License

Apache License 2.0 - see LICENSE file.

Support

• Issues for bug reports
• Discussions for questions