Question: Does Chakra support alltoallv collective communication?
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 196
- Forks
- 84
- PR merge metrics
- No merged PRs in 30d
Description
I'm working on MoE model simulation using astra-sim 2.0 and noticed that Chakra supports alltoall, but I can't find support for alltoallv (variable-length all-to-all).
Unlike regular alltoall with fixed send/recv sizes, alltoallv uses variable counts and offsets per rank, which is widely used in modern MoE models for dynamic token routing.
- ✅ Supported: alltoall
- ❌ Unclear: alltoallv
Could you clarify whether Chakra already supports alltoallv, or if it's planned for future support?
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 by checking Chakra's documented and implemented collective operations, especially the existing alltoall support. Determine whether variable counts and offsets per rank are supported; done means establishing whether alltoallv already exists or whether future support is planned.
Written by the indexing model from the issue text.
Assessment
- Domain
- distributed-systems
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Needs clarification
- Newbie friendliness
- 30/100