Lightning-AI / Lightning-AI/lightning-thunder

Fix test_parse_resnet18 test for train=True, dtype=float32, executor=nvfuser

Open
#2,497 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

ci / tests numerical accuracy
Dominant language
Python
Stars
1.5k
Forks
121
PR merge metrics
No merged PRs in 30d

Description

With #2494, this case for this test in test_update_aliases.py is skipped. However, the test consistently fails for this case with an AssertionError
```
E AssertionError: Tensor-likes are not close!
E
E Mismatched elements: 9402 / 9408 (99.9%)
E Greatest absolute difference: 0.09202009439468384 at index (4, 1, 5, 0) (up to 1e-05 allowed)
E Greatest relative difference: 10.512157440185547 at index (39, 1, 3, 0) (up to 1.3e-06 allowed)
E
E The failure occurred for item [0]
```
I verified that this test case in test_inplace_functionalization.py similarly fails in commit 19e6fa4 (with the modification to not skip float32) suggesting the issue is common to both the aliasing and functionalization approaches to in-place support.

cc @borda

Contributor guide

No contributing guide indexed for this repository

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 train=True, dtype=float32, executor=nvfuser case in test_update_aliases.py and reproduce the AssertionError. Compare it with the corresponding case in test_inplace_functionalization.py at commit 19e6fa4, considering the skip introduced by #2494. Done means both in-place support tests pass for this configuration without skipping float32.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
compilers, testing
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
38/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.