Skip to main content

PyTorch Abseil

The pytorch_abseil.py example demonstrates the integration of ClearML into code that uses PyTorch and absl.flags.

The example script does the following:

  • Trains a simple deep neural network on the PyTorch built-in MNIST dataset
  • Creates an experiment named pytorch mnist train with abseil in the examples project
  • ClearML automatically logs the absl.flags, and the models (and their snapshots) created by PyTorch
  • Additional metrics are logged by calling Logger.report_scalar()

Scalars​

In the example script's train function, the following code explicitly reports scalars to ClearML:

Logger.current_logger().report_scalar(
    "train", 
    "loss", 
    iteration=(epoch * len(train_loader) + batch_idx), 
    value=loss.item()
)

In the test method, the code explicitly reports loss and accuracy scalars.

Logger.current_logger().report_scalar(
    "test", "loss", iteration=epoch, value=test_loss
)
Logger.current_logger().report_scalar(
    "test", 
    "accuracy", 
    iteration=epoch, 
    value=(correct / len(test_loader.dataset))
)

These scalars can be visualized in plots, which appear in the ClearML web UI, in the experiment's SCALARS tab.

image

Hyperparameters​

ClearML automatically logs command line options defined with abseil flags. They appear in CONFIGURATION > HYPERPARAMETERS > TF_DEFINE.

image

Console​

Text printed to the console for training progress, as well as all other console output, appear in CONSOLE.

image

Artifacts​

Models created by the experiment appear in the experiment's ARTIFACTS tab. ClearML automatically logs and tracks models and any snapshots created using PyTorch.

image

Clicking on the model name takes you to the model's page, where you can view the model's details and access the model.

image