Skip to content

feat(qy): QY ops backend with FA and longctx MoE/RoPE kernels (2026.11.12) - #1038

Open
Ceng23333 wants to merge 4 commits into
InfiniTensor:masterfrom
Ceng23333:feat/qy-backend-20261112
Open

feat(qy): QY ops backend with FA and longctx MoE/RoPE kernels (2026.11.12)#1038
Ceng23333 wants to merge 4 commits into
InfiniTensor:masterfrom
Ceng23333:feat/qy-backend-20261112

Conversation

@Ceng23333

@Ceng23333 Ceng23333 commented Nov 12, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Summary

  • Rebase QY (Denglin) native CUDA ops backend with linked FlashAttention (varlen + KV-cache) onto current master.
  • Add longctx MoE/RoPE QY kernels for MiniCPM5 FA/longctx/cudagraph stack.
  • Adapt torch instantiate.h for WITH_QY after upstream device-instantiation refactor.

Validation

  • QY smoke/pytest -m smoke: not run on PR host (no QY SDK).
  • Integration-validated via ITW pin v2026.11.12-qy-moe-bf16-warpnk.

Stack PR index (merge order)

  1. InfiniOps: feat(qy): QY ops backend with FA and longctx MoE/RoPE kernels (2026.11.12) #1038
  2. InfiniRT: feat(qy): enable QY CUDA runtime device backend (2026.11.12) InfiniRT#54
  3. InfiniCCL: feat(qy): enable InfiniCCL NCCL backend for Denglin QY (2026.11.12) InfiniCCL#90
  4. InfiniCore: chore(qy): pin InfiniRT/Ops/CCL for QY MoE bf16-warpnk (2026.11.12) InfiniCore#1583
  5. InfiniLM: feat(qy): MiniCPM5 FA/longctx/cudagraph + MoE bf16/warpnk surpass (2026.11.12) InfiniLM#618 → base refactor/adopt-modern-infini-stack

ITW pin: https://github.com/Ceng23333/InfiniTensorWorktree/tree/v2026.11.12-qy-moe-bf16-warpnk

Supersedes working tip of #1002 (left open; do not close automatically).

Ceng23333 and others added 4 commits November 12, 2026 17:26
@Ceng23333 bench
feat(qy): add QY ops backend with linked FA and paged KV flash
772130d 
Enable Denglin/QY native and torch paths, linked flash-attn varlen/kvcache
providers via *_no_sinks, and reshape_and_cache_flash for paged caching.
@Ceng23333 bench
feat(qy): add longctx MoE/RoPE QY kernels for MiniCPM5
97adba6 
Promote mul_scalar, silu_and_mul, moe_sum/align/topk_sigmoid, and rope/rope_pack kernels from the longctx adapting lane onto feat/qy-backend.
@Ceng23333 @cursoragent bench
style(qy): clang-format QY backend changes
7ee8485 
Co-authored-by: Cursor <cursoragent@cursor.com>
bench @cursoragent
fix(qy): instantiate torch Operator templates for WITH_QY
27b3487 
Upstream moved device instantiation to instantiate.h macros; add kQy
alongside the rebased QY backend so torch providers link under WITH_QY.

Co-authored-by: Cursor <cursoragent@cursor.com>
@Ceng23333
Ceng23333 requested a review from a team November 12, 2026 09:32
This was referenced Nov 12, 2026
Open
Open
Open
Open
Open

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@Ceng23333