Repository navigation
Conversation
Enable Denglin/QY native and torch paths, linked flash-attn varlen/kvcache providers via *_no_sinks, and reshape_and_cache_flash for paged caching.
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.
Co-authored-by: Cursor <cursoragent@cursor.com>
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>
Format failing clang-format-21 paths and wrap Device::Type::kQy behind WITH_QY so unit jobs building against InfiniRT master still compile.
WITHOUT this, WITH_QY torch ops fail to instantiate StreamGuard. Co-authored-by: Cursor <cursoragent@cursor.com>
QY required smoke (PR head)
|
QY required smoke (PR head) — detailed
Also on this tip: clang-format-21 CI green; |
Co-authored-by: Cursor <cursoragent@cursor.com>
CI note (infra re-run follow-up)Tip: What we did
Current classification (not a
|
| Check | Runner (latest tip) | Result |
|---|---|---|
| clang-format / ruff / docs | — | green |
ci / unit / metax |
Metax-Server-006 | green (others still running) |
ci / unit / nvidia |
Nvidia-Server-006 | red — actions/checkout → proxy :11312 Connection refused (exit 128); never reaches build |
ci / unit / ascend |
Ascend-910C-Server-000 | red — same |
| shadow nvidia / ascend | Nvidia-Server-001 / Ascend-910C-Server-002 | red — same |
ci / unit / iluvatar (+ shadow) |
Iluvatar-Server-006 / 000 | still in_progress at comment time (past checkout) |
Required format is green. Remaining nvidia/ascend unit reds are runner proxy checkout infra (127.0.0.1:11312), not a QY/kQy/Device::Type failure. No further Ops code commits planned for this flake; needs org runner/proxy attention.
| if(WITH_QY) | ||
| add_compile_definitions(WITH_QY=1) | ||
|
|
||
| set(QY_ROOT "") |
There was a problem hiding this comment.
CMake 清空用户指定的 QY_ROOT。 -DQY_ROOT=... 被覆盖,非默认 SDK 路径配置失败
|
|
||
| template <> | ||
| __device__ __forceinline__ half mulScalarValue<half>(half value, double alpha) { | ||
| return __hmul(value, __float2half(static_cast<float>(alpha))); |
There was a problem hiding this comment.
标量乘法提前把 alpha 转为 FP16。 输入约 0.01、alpha=100000 时,结果应为 1000,当前计算会得到 inf
| for (; d + 4 <= dim; d += 4) { | ||
| *reinterpret_cast<uint64_t *>(out + d) = | ||
| *reinterpret_cast<const uint64_t *>(row + d); | ||
| } |
There was a problem hiding this comment.
代码里的“前进 4 个元素”在 FP32 下等于前进 16 字节,而每次实际只复制了 8 字节。
会导致FP32 pack/unpack 漏拷一半数据。
|
cuda/qy下面不放具体算子的文件,请全部迁移到ops文件夹下面对应的算子文件夹下面。 |
|
我发现include/下面的launcher都没有被调用,可以等用到的时候再加;如果加的话,也是加在调用算子的文件夹下面 |
Summary
master.instantiate.h/stream_guard.hwireWITH_QYC10 specialization.Device::Type::kQybehindWITH_QYso non-QY unit jobs still compile against InfiniRT master (nokQyenum).Motivation / related
clang-formatchecking GitHub Action #54, CCL feat: addallbase #90, Core #1583, LM feat: add public C++ operator API #618.Type of change
kQyCI gate; QYstream_guardinclude)Platforms affected
WITH_QYgate (no functional change when QY off)Validation
DENGLIN_DEVICES=3.clang-formatchecking GitHub Action #54 head047340f45efa;pytest tests/test_add.py tests/test_gemm.py --devices qy→ 1633 passed, 6 skipped.nov_milestone/logs/prhead-verify-20261113/ops-pytest-add-gemm.logChecklist
feat/qy-backend-20261112)127.0.0.1:11312connection refused (infra, notkQycompile)Stack PR index (merge order)
71e85162ab96047340f45efa691c3092c4c90b69ee9002bd36beae5376f4→ baserefactor/adopt-modern-infini-stackITW pin: https://github.com/Ceng23333/InfiniTensorWorktree/tree/v2026.11.12-qy-moe-bf16-warpnk