vllm - 💡(How to fix) Fix Request for triage permission — active ROCm contributor [2 comments, 3 participants]

Hi maintainers,

I'm requesting triage permission on this repository. I'm an active ROCm/AMD contributor with 12+ open PRs covering AITER integration, MLA attention backends, MXFP4 quantization, and GLM-5 model support on MI355X.

Why triage permission?

• Request reviewers on my own PRs — currently blocked because fork-contributors cannot use requested_reviewers API
• Manage labels to keep ROCm-related PRs properly categorized
• Faster review cycles for customer-facing work (e.g. GLM-5 on MI355X)

Triage permission is read-only for code — it only allows managing issues/PRs metadata (labels, reviewers, milestones).

My contribution history

Open PRs (partial list):

• #38665 — Dual-stream shared experts + GLM-5 MXFP4/FP8 support
• #36855 — Sparse MLA head repeat fix for < 16 heads
• #38947 — AITER MLA prefill kernel integration
• #37800 — MXFP4 linear method + shared expert fusion
• #37353 — Skip head repeat for AITER MLA decode
• #36851 — Sequence parallelism for AMD GPUs
• #36425 — AITER fused allreduce + RMSNorm
• #36227 — Cascade attention for AITER FlashAttention
• #36092 — AITER ops fake impl fixes

GitHub: @ChuanLi1101

cc @tjtanaa @gshtras @simon-mo

Thank you!