labml 0.5.3

Organize Machine Learning Experiments

Monitor deep learning model training and hardware usage from mobile.

🔥 Features

• Monitor running experiments from mobile phone or laptop
• Monitor hardware usage on any computer with a single command
• Integrate with just 2 lines of code (see examples below)
• Keeps track of experiments including infomation like git commit, configurations and hyper-parameters
• API for custom visualizations
• Pretty logs of training progress
• Open source!

Hosting the experiments server

Prerequisites

To install MongoDB, refer to the official documentation here.

Installation

Install the package using pip:

pip install labml-app

Starting the server

# Start the server on the default port (5005)
labml app-server

# To start the server on a different port, use the following command
labml app-server --port PORT

Optional: to setup and configure Nginx in your server, please refer to this.

You can access the user interface either by visiting http://localhost:{port} or, if configured on a separate machine, by navigating to http://{server-ip}:{port}.

Monitor Experiments

Installation

1. Install the package using pip.

pip install labml

2. Create a file named .labml.yaml at the top level of your project folder, and add the following line to the file:

app_url: http://localhost:{port}/api/v1/default

# If you are setting up the project on a different machine, include the following line instead,
app_url: http://{server-ip}:{port}/api/v1/default

PyTorch example

from labml import tracker, experiment

with experiment.record(name='sample', exp_conf=conf):
    for i in range(50):
        loss, accuracy = train()
        tracker.save(i, {'loss': loss, 'accuracy': accuracy})

Distributed training example

from labml import tracker, experiment

uuid = experiment.generate_uuid() # make sure to sync this in every machine
experiment.create(uuid=uuid,
                  name='distributed training sample',
                  distributed_rank=0,
                  distributed_world_size=8,
                  )
with experiment.start():
    for i in range(50):
        loss, accuracy = train()
        tracker.save(i, {'loss': loss, 'accuracy': accuracy})

📚 Documentation

• Python API Reference
• Samples

Guides

• API to create experiments
• Track training metrics
• Monitored training loop and other iterators
• API for custom visualizations
• Configurations management API
• Logger for stylized logging

🖥 Screenshots

Formatted training loop output

Custom visualizations based on Tensorboard logs

Monitoring hardware usage

# Install packages and dependencies
pip install labml psutil py3nvml

# Start monitoring
labml monitor

Citing

If you use LabML for academic research, please cite the library using the following BibTeX entry.

@misc{labml,
 author = {Varuna Jayasiri, Nipun Wijerathne, Adithya Narasinghe, Lakshith Nishshanke},
 title = {labml.ai: A library to organize machine learning experiments},
 year = {2020},
 url = {https://labml.ai/},
}