deepspeedai / deepspeedai/DeepSpeed

[BUG] MegatronMoELayerPolicy not mapped

Open
#5,285 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

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

Description

Describe the bug
Not able to start inference for MoE based DS Megatron GPT model because the MegatronMoELayerPolicy (https://github.com/microsoft/DeepSpeed/blob/master/deepspeed/module_inject/containers/megatron_gpt_moe.py) that supports this feature is not being used/referenced anywhere.

To Reproduce
Steps to reproduce the behavior:

  1. Load DS Megatron GPT MoE model
  2. Pass it to deepspeed.init_inference
  3. inference engine (using the BaseContainer instead of BaseMoeContainer) tries to transpose weights of MoE layer with logic written for a single MLP layer

Expected behavior
Inference engine should load without an issue

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 MegatronMoELayerPolicy implementation in deepspeed/module_inject/containers/megatron_gpt_moe.py and trace how deepspeed.init_inference selects between BaseContainer and BaseMoeContainer. Reproduce the Megatron GPT MoE inference setup and verify that the MoE policy is selected and inference loads without the single-MLP weight-transpose failure.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
3/5
Estimated time
1-2 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
45/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.