Skip to content

reorg of test scripts and minor changes to tests#602

Merged
pacman100 merged 2 commits intohuggingface:mainfrom
pacman100:smangrul/fix-deepspeed-fsdp-slow-tests
Aug 3, 2022
Merged

reorg of test scripts and minor changes to tests#602
pacman100 merged 2 commits intohuggingface:mainfrom
pacman100:smangrul/fix-deepspeed-fsdp-slow-tests

Conversation

@pacman100
Copy link
Contributor

@pacman100 pacman100 commented Aug 3, 2022

What does this PR do?

  1. Reorg test scripts with external dependencies such as datasets, transformers ...
  2. lower performance threshold from 0.83 to 0.82 as it leads to assertion error on GCP instance with 2 T4 GPUs which run in CI slow test.
  3. Adding fix to deepspeed as per [BUG] Error when enabling fp16 config with tasks involving FloatTensor Inputs such as Image Classification, Speech Recognition ... deepspeedai/DeepSpeed#2019

All slow tests are now passing on GCP instance with 2 T4 GPUs
report_fsdp.txt
report_deepspeed_mem_test.txt
report_deepspeed_checkpoint_test.txt
report_deepspeed_performance_test.txt

@pacman100 pacman100 requested a review from sgugger August 3, 2022 07:22
@HuggingFaceDocBuilderDev
Copy link

HuggingFaceDocBuilderDev commented Aug 3, 2022

The documentation is not available anymore as the PR was closed or merged.

@pacman100 pacman100 requested a review from muellerzr August 3, 2022 08:55
Copy link
Collaborator

@sgugger sgugger left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks for working on this!
(Let's wait for @muellerzr before merging though)

Copy link
Contributor

@muellerzr muellerzr left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for this!

As discussed offline this likely won't affect our slow multi-gpu runner that is failing, as that seems to be a different issue entirely.

@pacman100 pacman100 merged commit b10fd81 into huggingface:main Aug 3, 2022
@pacman100 pacman100 deleted the smangrul/fix-deepspeed-fsdp-slow-tests branch August 4, 2022 05:44
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants