incorrect outputs are observed for fp16 for bert-base model with the vectorized implementation of the layernorm
Open
@turneram is already working on this.
Since May 23, 2022.
- Dominant language
- C++
- Stars
- 333
- Forks
- 150
- Avg merge
- 4d 19h
- Merged PRs (30d)
- 54
Description
If we run the bert-base-cased_1 model for fp16 and set the input shape as [1, 128], the fp16 outputs is different from fp32, but if we use the non-vectorized implementation, the fp16 outputs are very close to that of the fp32.
Need further investigation on this, will continue investigating this issue, but still want to create an issue to track the problem.
Contributor guide
No contributing guide indexed for this repository
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.