Skip to main content

Real time TensorFlow™ visualization

Monitor and compare training runs in real time with
Guild AI's easy to use command line toolset

Get Started

Currently 0.1.1

Monitor TensorFlow in real time

Guild AI provides a high-level dashboard for your models that lets you view important details at-a-glance. As you train your models, Guild automatically tracks run progress and updates your dashboard in real time. In addition to tracking TensorFlow metrics, Guild captures system details such as GPU utilization and memory allocation to give you a more complete picture of your training operations.

Compare and analyze

Guild AI maintains detailed records of each TensorFlow experiment and lets you quickly compare results over time. Guild provides high level summaries, TensorFlow and system metric comparisons per run, and configuration level diffing.

TensorBoard integration

Guild AI seamlessly integrates TensorBoard, letting you drill into TensorFlow event logs when you need lower level details. TensorBoard includes visualizations for event scalars, distributions, histograms, images, audio, and graphs. There's no need to run TensorBoard separately as it's built directly into Guild AI.

More about Guild AI features

Guild AI works with your TensorFlow code

Train, evaluate, and serve your models using simple command line operations.

Command line interface

Guild reduces complex model operations to single commands. This makes your life easier and it also makes your models easier to use by others.

prepare
prepares a model for training (e.g. download and pre-process a dataset)
train
trains a model defined in your Guild project file
view
start Guild View - a browser interface for viewing training results for your project
evaluate
evaluate a trained model's performance
serve
serve a trained model locally over HTTP + JSON

View all Guild commands

Guild project file

To enable Guild features, add a Guild file to your project.

[project]

name            = MNIST

[model]

runtime         = tensorflow
train           = intro
prepare         = intro --prepare
evaluate        = intro --evaluate

[flags]

batch-size      = 100
epochs          = 5

Measure system resources used by your model

Guild runs efficiently in the background logging CPU, GPU, and other system metrics to give you a more complete picture of model performance.