deepspeedai / deepspeedai/DeepSpeed

need an an easier way to activate memory usage debug

Open
#1,133 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

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

Description

Currently when there are OOM problems we are being asked to modify deepspeed code to enable see_memory_usage - but it's not trivial - once needs to know where to enable as it's not just one place. And it's an issue in a shared env, unless one creates a separate env to debug.

I suggest adding a new debug feature that can change the behavior of the project from outside w/o needing to modify the source code, that is using env vars to do that. It shouldn't impact performance as it could assign noop functions at startup time.

So I propose to create an env var, e.g.: DS_MEM_SEE_USAGE=stage3,partitioning,engine,etc so really it's about defining the granularity of the areas, remapping those to the corresponding names, and then implementing this in the code.

Thanks.

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

No files or tests are named. Start by locating the existing see_memory_usage calls in the stage3, partitioning, and engine areas, then determine how an environment variable could select their granularity. Done means memory-usage debugging can be enabled externally without source edits and disabled with no performance impact.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning, tooling
Issue type
Feature
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.