Lightning-AI / Lightning-AI/pytorch-lightning

Decouple benchmarks from tests

Open
#13,684 0 comments 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

pl tests
Dominant language
Python
Stars
31.4k
Forks
3.8k
Avg merge
6d 7h
Merged PRs (30d)
6

Description

I've removed the convergence check. It was failing under batch mode. There might be RNG interactions between tests

        results = trainer.test(datamodule=dm)
>       assert results[0]["test_acc"] > 0.7
E       assert 0.6937500238418579 > 0.7

It's really arbitrary and PyTorch doesn't even guarantee performance between versions

These kinds of tests should be benchmark tests. And ideally something like the same model trained with DeepSpeed vs normally

Originally posted by @carmocca in https://github.com/Lightning-AI/lightning/pull/13673#discussion_r922250654


This will contribute to less flakiness in our CI 🚀


cc @borda @akihironitta

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 by locating the test containing trainer.test(datamodule=dm) and the convergence check described in the issue. Review the existing benchmark-test structure and how these tests run in CI. Done means performance assertions no longer make regular tests flaky, with benchmark coverage kept separate.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
machine-learning, testing-qa
Issue type
Refactor
Difficulty
3/5
Estimated time
1-2 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.