Skip to content

docs: UltraEP on Intel XPU implementation task list - #98

Draft
mingfeima wants to merge 2 commits into
mainfrom
cursor/ultraep-xpu-task-list-e411
Draft

docs: UltraEP on Intel XPU implementation task list#98
mingfeima wants to merge 2 commits into
mainfrom
cursor/ultraep-xpu-task-list-e411

Conversation

@mingfeima

@mingfeima mingfeima commented Jul 23, 2026

Copy link
Copy Markdown
Owner

Summary

Adds an English, phased implementation task list for bringing UltraEP-style real-time MoE expert load balancing to Intel XPU in SGLang.

Contents

  • Phase 0: Hardware gates, scope, success metrics
  • Phase 1: XPU MoE A2A + expert GEMM foundation (blocker)
  • Phase 2: Portable quota-driven planner + remapping
  • Phase 3: XeLink / Intel SHMEM data plane (tile streaming, relay fan-out)
  • Phase 4: SGLang integration, flags, docs, CI
  • Phase 5: Evaluation and hardening
  • Phase 6: Stretch (training path, multi-node, quantized replication)

Also registers the page in Mintlify nav and links it from the existing XPU hardware page.

Notes

This is a planning checklist only — no runtime UltraEP/XPU code changes.

Open in Web Open in Cursor 

CI States

Latest PR Test (Base): ✅ Run #29984462210
Latest PR Test (Extra): ❌ Run #29984462087

cursoragent and others added 2 commits July 23, 2026 06:14
Phased engineering checklist covering XPU MoE A2A prerequisites,
quota planner, XeLink/Intel SHMEM data plane, SGLang integration,
and evaluation for UltraEP-style real-time expert load balancing.

Co-authored-by: Ma Mingfei <mingfei.ma@intel.com>
Register the new page in docs.json and add a Related link on the
Intel XPU hardware page.

Co-authored-by: Ma Mingfei <mingfei.ma@intel.com>
@github-actions github-actions Bot added the documentation Improvements or additions to documentation label Jul 23, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

documentation Improvements or additions to documentation

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants