deepspeedai / deepspeedai/DeepSpeed

Training accuracy heavily impacted when using activation checkpointing

Open
#1,116 1 comment 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

I just use deepspeed without any additional params and let's get consistently 0.250 accuracy on 200'th batch. When I turn on the activation checkpointing, I get 0.025 accuracy with the same parameters at the same batch number.
My model is like an unet and has conv2d, leakyrelu and batchnorm layers and skip connections. Can this be a problem?
For skip connections I have modified a bit example from you and saved/consumed them when needed in a separate list

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 source file, test, or minimal reproduction is provided. First reproduce the accuracy gap with the stated UNet-like model and activation-checkpointing settings, then compare checkpointed and baseline runs; done means a confirmed cause with a regression test or documented limitation.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.