fp16 via pytorch
Open
Nobody has claimed this yet.
- Dominant language
- Cuda
- Stars
- 50
- Forks
- 9
- PR merge metrics
- No merged PRs in 30d
Description
According to the readme, this library supports fp16 embedding tables, however the pytorch bindings are hardcoded to fp32. Trying to flip the hardcoding to at::Half (for testing) doesnt seem to do the trick. Is there something else you might recommend?
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.
Research direction
Start with the README's claim that fp16 embedding tables are supported and trace the PyTorch bindings where the data type is hardcoded to fp32. Reproduce the issue with the attempted at::Half change and compare the binding behavior with the documented fp16 path. Done means fp16 embedding tables work through the PyTorch bindings without relying on fp32 hardcoding.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- pytorch
- Domain
- machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 30/100