deepspeedai / deepspeedai/DeepSpeed
why is AutoTP only applies LinearAllreduce (RowParallel) ? What about ColumnParallel and ParallelEmbedding
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
Hi, I noticed that using AutoTP for training, only LinearAllreduce (RowParallel) are used. What about ColumnParallel and ParallelEmbedding ? Is there a reason to not do it ?
I guess I have to do it manually with prepare_tp_model ? cc: @inkcherry
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
The issue names AutoTP, prepare_tp_model, LinearAllreduce (RowParallel), ColumnParallel, and ParallelEmbedding, but no file or test. Start by tracing how AutoTP handles LinearAllreduce and compare that path with prepare_tp_model for the two omitted layer types. Done means establishing whether the omission is intentional and documenting the reason or providing tested support.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- distributed-systems, machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100