Files
heterogeneous-distributed-t…/docs/source/api-guide/pipeline_parallel.rst
tianyutong d6ce507681 Initial Commit of Megatron-LM-0.8.0
Change-Id: Ifb4c061207ee2644a21e161ad52fc6ff40564e39
2025-05-23 09:54:48 +08:00

1.7 KiB

pipeline_parallel package

This package contains implementations for two different pipeline parallelism schedules (one without interleaving and one with interleaving, see Efficient Large-Scale Language Model Training on GPU Clusters Using Megatron-LM for details), and a default no-pipelining schedule. It also contains methods for the point-to-point communication that is needed between pipeline stages.

Submodules

pipeline_parallel.p2p_communication module

Contains implementations for the various point-to-point communication needed (e.g., recv_forward and recv_backward) in the different pipeline parallelism schedules.

core.pipeline_parallel.p2p_communication

pipeline_parallel.schedules module

Contains implementations for two pipeline parallelism schedules (forward_backward_pipelining_with_interleaving`for pipeline parallelism with interleaving, `forward_backward_pipelining_without_interleaving for pipeline parallelism without interleaving) and a default no-pipelining schedule (forward_backward_no_pipelining). get_forward_backward_func returns the right scheduling function to use based on the configuration being trained (e.g., if pipeline-parallel size is 1, use forward_backward_no_pipelining).

core.pipeline_parallel.schedules

Module contents

core.pipeline_parallel