Skip to content

Single-controller Realm backend#1629

Open
elliottslaughter wants to merge 77 commits intoflexflow:masterfrom
elliottslaughter:realm-execution
Open

Single-controller Realm backend#1629
elliottslaughter wants to merge 77 commits intoflexflow:masterfrom
elliottslaughter:realm-execution

Conversation

@elliottslaughter
Copy link
Contributor

@elliottslaughter elliottslaughter commented Feb 12, 2026

This PR implements a single-controller Realm execution backend.

Feature overview:

  • Single-controller (i.e., controller runs on one node)
  • One task per dynamic graph node
    • Tasks are remote-launched if necessary (i.e., distributed execution should work, except for the caveats below)
  • Tensors are allocated as instances in the memory nearest the assigned processor
  • Most of the infrastructure for distributed execution is present, except:
    • Copies are not inserted, so data will not be moved between devices
    • PCG operators representing data movement (split, replicate, etc.) are not implemented

This change is Reviewable

@elliottslaughter elliottslaughter marked this pull request as ready for review February 19, 2026 00:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant

Comments