RFC-0057: Semantic Identity for Inductor Distributed Collectives - #103
RFC-0057: Semantic Identity for Inductor Distributed Collectives#103NmomoN wants to merge 1 commit into
Conversation
|
Hi @NmomoN! Thank you for your pull request and welcome to our community. Action RequiredIn order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you. ProcessIn order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA. Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks! |
|
Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks! |
Summary
This RFC proposes a small, private semantic identity contract for functional distributed collectives in Inductor.
Inductor can identify collective transport types such as all-gather and reduce-scatter, but currently relies on graph structure to infer why those collectives were emitted. This becomes ambiguous when FSDP, tensor parallelism, context parallelism, and user-authored distributed code coexist in the same graph.
Phase 1 starts with FSDP. It introduces purpose-oriented collective roles, distinguishes producer-provided identity from heuristic inference, and defines how that identity should be preserved or invalidated by Inductor graph transformations.
This PR contains a design proposal only and does not change PyTorch behavior.
Scope
Phase 1 covers:
It intentionally does not introduce scheduling priority, static urgency, topology-aware cost models, or a public annotation API.
Feedback requested
Feedback would be particularly helpful on:
Related PyTorch issue: pytorch/pytorch#190792