dmlcloud.Pipeline

class dmlcloud.Pipeline(config=None, name=None)

A training pipeline that consists of multiple stages.

This is the main entry point for training with dmlcloud. The pipeline manages the training process and orchestrates the execution of multiple stages. It also provides a way to add callbacks that are executed at different points during the training process.

Use the append() method to add stages to the pipeline and add_callback() to add callbacks.

Checkpointing can be enabled with enable_checkpointing() and Weights & Biases integration with enable_wandb().

Once the pipeline is set up, call run() to start the training process.

__init__(config=None, name=None)

Methods

__init__([config, name])
add_callback(callback[, priority])	Adds a callback to this pipeline.
append(stage)
barrier([timeout])
enable_checkpointing(root[, resume])
enable_wandb([project, entity, group, tags, ...])
post_run()
pre_run()
resume_run()
run()	Starts the training and runs all registered stages.

Attributes

device
has_checkpointing
has_tensorboard
has_wandb