deepspeedai / deepspeedai/DeepSpeed
[BUG] inference ops unit tests are failing
Open
@loadams is already working on this.
Since Dec 9, 2024.
bug
inference
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
It was seen that tests under unit/ops/transformer/inference are not being run in any CI job.
Some tests are failing in that directory (examples will be provided below), I have talked to @loadams about it and he tried running them on a V100 setup.
The results he got were 440 failed, 2598 passed, 8 skipped for those tests.
Example for the tests we saw them failing were:
- unit/ops/transformer/inference/test_bias_geglu.py::test_gated_silu and the failure was on different results than the reference.
- unit/ops/transformer/inference/test_layer_norm.py::test_layer_norm and the failure was
Feature '.bf16' requires .target sm_80 or higher
A hint that could help, those tests are permutated over the supported dtypes, and the failures are always in dtype2 (I assume it is bf16).
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.