EnzymeAD / EnzymeAD/Enzyme-JAX
hlo opt gives incorrect results for neural gcm
Open
python
- Dominant language
- MLIR
- Stars
- 131
- Forks
- 53
- Avg merge
- 1d 10h
- Merged PRs (30d)
- 193
Description
added tests for this in https://github.com/EnzymeAD/Enzyme-JAX/pull/1790 (asserts are skipped for now). we should investigate the source. one good thing is that it fails even for partopt which is a significantly smaller subset of optimizations
Contributor guide
No contributing guide indexed for this repository
Research direction
Start with the tests added in Enzyme-JAX pull request #1790, where the assertions are currently skipped. Reproduce the incorrect results with HLO optimization enabled, then compare against partopt to narrow the source of the failure. Done means identifying and fixing the source so the neural GCM assertions pass under both optimization modes.
Written by the indexing model from the issue text.
Assessment
- Domain
- compilers, machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100