chop.pipelines

Mase pipelines are pass managers that let you do things quicker for common use cases, without the faff of managing individual compiler passes.

AutoPipeline (base class)

class chop.pipelines.AutoPipeline(pass_list=[])

This is the base class for the AutoPipeline.

It takes a list of passes and runs them in order.

The output of each pass is stored in a dictionary and can be accessed by the next pass.

__init__(pass_list=[]) → None

Initializes the AutoPipeline.

Parameters:

pass_list (list, optional) – List of passes to run. Defaults to [].

AutoPipeline for Distributed Inference

class chop.pipelines.AutoPipelineForDistributedInference

This pipeline is used for distributed inference.

It runs the following passes:

• init_metadata_analysis_pass

• report_graph_analysis_pass

• add_common_metadata_analysis_pass

• autosharding_analysis_pass

• resharding_transform_pass

__init__() → None

Initializes the AutoPipeline.