deepspeedai / deepspeedai/DeepSpeed

Why is Pipeline parallelism not compatible with ZeRO-2 and ZeRO-3?

Open
#1,629 3 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

enhancement
Dominant language
Python
Stars
43.1k
Forks
5k
Avg merge
4d 15h
Merged PRs (30d)
112

Description

Could you explain why Pipeline parallelism is not compatible with ZeRO-2 and ZeRO-3? Are there any design tradeoffs?

Because as far as I know, it is pretty common to train large models with both DataParallel and PipelineParallel together. and with the constraint above, the offload mechanism cannot be enabled due to its dependency on ZeRO-2/3.

Also, Megatron-DeepSpeed:pretrain_gpt.py use GPTModelPipe, which is a subclass of PipelineModule as the model module passed to deepspeed.initialize(), so it's impossible to enable ZeRO-2/3 in the config json, is there any examples to run with ZeRO-2/3?

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the linked Megatron-DeepSpeed pretrain_gpt.py example, especially the GPTModelPipe and PipelineModule usage, and inspect how its configuration is passed to deepspeed.initialize(). Compare that setup with the issue's ZeRO-2/3 and offload questions; done means documenting the compatibility tradeoffs and providing a verified example if one exists.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
distributed-systems, documentation, machine-learning
Issue type
Documentation
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.