Compare commits
No commits in common. "master" and "updata" have entirely different histories.
|
|
@ -1,79 +0,0 @@
|
|||
# =========================
|
||||
# macOS
|
||||
# =========================
|
||||
.DS_Store
|
||||
.AppleDouble
|
||||
.LSOverride
|
||||
Icon?
|
||||
._*
|
||||
.Spotlight-V100
|
||||
.Trashes
|
||||
.fseventsd
|
||||
|
||||
# =========================
|
||||
# IDE / Editor
|
||||
# =========================
|
||||
.vscode/
|
||||
.idea/
|
||||
*.swp
|
||||
*.swo
|
||||
*~
|
||||
|
||||
# =========================
|
||||
# Python
|
||||
# =========================
|
||||
__pycache__/
|
||||
*.py[cod]
|
||||
*.pyo
|
||||
*.pyd
|
||||
.pytest_cache/
|
||||
.mypy_cache/
|
||||
.ruff_cache/
|
||||
.coverage
|
||||
htmlcov/
|
||||
.env
|
||||
.venv/
|
||||
venv/
|
||||
env/
|
||||
|
||||
# =========================
|
||||
# Jupyter
|
||||
# =========================
|
||||
.ipynb_checkpoints/
|
||||
|
||||
# =========================
|
||||
# Build / Packaging
|
||||
# =========================
|
||||
build/
|
||||
dist/
|
||||
*.egg-info/
|
||||
.eggs/
|
||||
pip-wheel-metadata/
|
||||
|
||||
# =========================
|
||||
# C / C++ / CMake
|
||||
# =========================
|
||||
CMakeFiles/
|
||||
CMakeCache.txt
|
||||
cmake-build-*/
|
||||
Makefile
|
||||
*.o
|
||||
*.so
|
||||
*.dylib
|
||||
*.dll
|
||||
*.a
|
||||
*.lib
|
||||
|
||||
# =========================
|
||||
# Logs / Temp
|
||||
# =========================
|
||||
*.log
|
||||
*.tmp
|
||||
*.temp
|
||||
logs/
|
||||
tmp/
|
||||
|
||||
# =========================
|
||||
# OS / Tool caches
|
||||
# =========================
|
||||
.cache/
|
||||
329
FAQ.md
329
FAQ.md
|
|
@ -1,329 +0,0 @@
|
|||
# 常见问题 FAQ
|
||||
|
||||
> 最后整理：2026-07-13
|
||||
|
||||
本文档汇总沐曦“揭榜挂帅”两项赛题的常见问题。使用 `Ctrl+F`（Windows/Linux）或 `⌘F`（macOS）搜索关键词。
|
||||
|
||||
环境版本、评测配置、时间安排可能调整。请以仓库 README、XPU-OJ 公告、比赛群通知和容器中的实际版本为准。发现内容过期或未覆盖的问题，请[提交 Issue](https://gitlink.org.cn/metax-maca/op_optimization/issues)。
|
||||
|
||||
## 快速导航
|
||||
|
||||
- [重要入口与咨询方式](#entry)：XPU-OJ 开放情况、问题反馈渠道和赛事联系人。
|
||||
- [XPU-OJ、提交与排行榜](#xpuoj)：账号申请、提交环境、测试样例、评测硬件和排名指标。
|
||||
- [赛题一：TileLang 与 Fused MoE](#track-one)：提交限制、算子实现、Baseline、性能优化和决赛加分规则。
|
||||
- [赛题二：AI Agent 与推理算子库](#track-two)：任务范围、Agent 参与证明、Baseline、测试参数和上游代码引用。
|
||||
- [评测规则与通用技术问题](#evaluation)：技术资料发布、正确性与稳定性要求，以及 FAQ 内容纠错。
|
||||
- [环境、镜像与算力资源](#environment)：比赛镜像、MACA 与 PyTorch 版本、开发工具、算力券和资源申请。
|
||||
- [报名、组队与资格审核](#registration)：参赛资格、跨校组队、指导教师、材料盖章和审核流程。
|
||||
|
||||
<a id="entry"></a>
|
||||
|
||||
## 重要入口与咨询方式
|
||||
|
||||
<a id="q-xpuoj-open"></a>
|
||||
|
||||
### ❓ 问题 1：XPU-OJ 平台是否已经上线？
|
||||
|
||||
**回答：** XPU-OJ 已开放。账号申领流程和使用指南见[赛事 XPU-OJ 账号申领说明](赛事XPUOJ账号申领说明.md)。
|
||||
|
||||
<a id="q-contact"></a>
|
||||
|
||||
### ❓ 问题 2：两个赛题的联系人不同，遇到问题应该联系谁？
|
||||
|
||||
**回答：** 请先通过 [GitLink Issue](https://gitlink.org.cn/metax-maca/op_optimization/issues) 提问，维护者会把可复用的答案更新到本文档。比赛群用于接收赛事通知和临时信息。需要单独沟通时，请按对应比赛方案联系章老师或杨老师。
|
||||
|
||||
## XPU-OJ、提交与排行榜
|
||||
|
||||
<a id="q-moe-score-decrease"></a>
|
||||
|
||||
### ❓ 问题 1：XPUOJ测评 MoE 耗时减少了但是分数反而降低了
|
||||
|
||||
**回答：**
|
||||
|
||||
针对近期部分同学反馈的“XPU.OJ”第三方评测系统中基线（baseline）不稳定的问题，我们高度重视，并已第一时间组织排查与测试。在此，我们对因此给大家带来的困扰深表歉意，也衷心感谢各位同学提出的宝贵意见。
|
||||
目前，相关问题已修复完毕。为确保评测的公平性与准确性，我们将对现有榜单进行清空处理。历史提交记录仍可查看，但后续排名将统一以基线修复后重新提交的算子成绩为准。
|
||||
|
||||
比赛期间，我们将持续关注系统运行状态，也欢迎大家继续向我们反馈建议。
|
||||
祝大家比赛顺利，取得理想成绩！
|
||||
|
||||
<a id="q-xpuoj-environment"></a>
|
||||
|
||||
### ❓ 问题 2：XPU-OJ 与模力方舟的运行环境一致吗？
|
||||
|
||||
**回答：** 排行榜评测环境与模力方舟开发环境保持一致。版本调整时以 XPU-OJ 公告为准。
|
||||
|
||||
<a id="q-xpuoj-account"></a>
|
||||
|
||||
### ❓ 问题 3：如何申请 XPU-OJ 账号？
|
||||
|
||||
**回答：** 请按照[赛事 XPU-OJ 账号申领说明](赛事XPUOJ账号申领说明.md)提交申请。账号发放进度以赛事通知和回复邮件为准。
|
||||
|
||||
<a id="q-official-ranking"></a>
|
||||
|
||||
### ❓ 问题 4：赛题一 MoE 初赛排名以哪个入口为准？
|
||||
|
||||
**回答：** 正式排名和初筛结果以 XPU-OJ 的评测结果为准。Sample benchmark 用于本地功能验证、调试和性能对比，不作为正式榜单依据。
|
||||
|
||||
<a id="q-test-cases"></a>
|
||||
|
||||
### ❓ 问题 5：MACA C++、Triton 和 TileLang 是否分别设榜？
|
||||
|
||||
**回答：** 不按语言分别设榜。每个任务支持 MACA C++、Triton 和 TileLang，团队可以使用一种或多种语言提交。排行榜采用通过正确性和稳定性测试后的最高成绩。
|
||||
|
||||
<a id="q-ranking-metric"></a>
|
||||
|
||||
### ❓ 问题 6：排行榜使用 latency、speedup 还是综合 score？
|
||||
|
||||
**回答：** 当前 XPU-OJ 以 speedup 作为核心排名指标。赛事方调整计算方式时，以 XPU-OJ 公告为准。
|
||||
|
||||
<a id="track-one"></a>
|
||||
|
||||
## 赛题一：TileLang 与 Fused MoE
|
||||
|
||||
<a id="q-track-one-submission"></a>
|
||||
|
||||
### ❓ 问题 1：赛题一的提交要求有哪些调整？
|
||||
|
||||
**回答：** 正式提交禁止使用 `MACA Maca running` 方式，也不能使用 PyTorch 实现算子。参赛者需使用 TileLang 实现并提交。
|
||||
|
||||
<a id="q-ops-reference"></a>
|
||||
|
||||
### ❓ 问题 2：OPS 目录中的 TileLang、CUDA、CUTLASS 和 MACA 代码有什么用途？
|
||||
|
||||
**回答：** 这些代码用于解释算子的实现原理和设计思路，可作为 TileLang 实现的参考。
|
||||
|
||||
<a id="q-baseline-modification"></a>
|
||||
|
||||
### ❓ 问题 3：官方 Baseline 可以修改到什么范围？
|
||||
|
||||
**回答：** 参赛者可以重新设计和优化算子实现，但需保持与统一 Workload 测试框架的接口兼容。
|
||||
|
||||
<a id="q-gemm-optimization"></a>
|
||||
|
||||
### ❓ 问题 4：GEMM 计算中可以引入其他优化策略吗？
|
||||
|
||||
**回答：** 可以，前提是实现符合赛题规则和评测要求。
|
||||
|
||||
<a id="q-benchmark-modification"></a>
|
||||
|
||||
### ❓ 问题 5：可以优化 `fusedmoe_benchmark` 吗？
|
||||
|
||||
**回答：** 本地修改 benchmark 不会提高正式成绩，XPU-OJ 使用赛事方的评测框架。参赛者应把优化工作放在规定的算子实现和允许修改的接口上。
|
||||
|
||||
<a id="q-forward-modification"></a>
|
||||
|
||||
### ❓ 问题 6：可以修改 `fusedmoe_benchmark.py` 中的 MoE forward 吗？
|
||||
|
||||
**回答：** 正式成绩以 XPU-OJ 的独立评测为准。本地修改 forward 不能替代对提交算子的优化，也不会改变赛事方的评测代码。
|
||||
|
||||
<a id="q-async-copy"></a>
|
||||
|
||||
### ❓ 问题 7：赛题一允许使用异步拷贝吗？
|
||||
|
||||
**回答：** 不允许。当前比赛规则禁用异步拷贝。
|
||||
|
||||
<a id="q-baseline-performance"></a>
|
||||
|
||||
### ❓ 问题 8：Fused MoE 初赛成绩如何影响决赛？
|
||||
|
||||
**回答：** 当前规则只给初赛前 10 名决赛加分，第 11 名及之后不获得额外初赛加分。
|
||||
|
||||
<a id="track-two"></a>
|
||||
|
||||
## 赛题二：AI Agent 与推理算子库
|
||||
|
||||
<a id="q-track-two-update"></a>
|
||||
|
||||
### ❓ 问题 1：赛题二的内容有什么调整？
|
||||
|
||||
**回答：** “Agent 推理算子库优化 - FlashAttention KV Cache Decode”新增 `mctlass/cute` 要求。参赛者需基于 `mctlass/cute` 实现或优化对应任务。
|
||||
|
||||
<a id="q-track-two-selection"></a>
|
||||
|
||||
### ❓ 问题 2：赛题二可以选择几个任务？
|
||||
|
||||
**回答：** 参赛团队可以从 FlashInfer、FlashAttention、MCTLASS/Fused MoE 等方向选择一项或多项。提交多个任务时，每个任务按赛事规则取有效最高成绩。支持语言包括 Triton、MXMACA C++ 和 TileLang。
|
||||
|
||||
<a id="q-agent-proof"></a>
|
||||
|
||||
### ❓ 问题 3：如何证明 AI Agent 参与了优化过程？
|
||||
|
||||
**回答：** 参赛团队应保留 Agent 配置、Skill 文件、关键提示词、操作日志、代码变更记录、测试结果和复现实验步骤，等待赛事方发布核验细则。
|
||||
|
||||
<a id="q-agent-baseline"></a>
|
||||
|
||||
### ❓ 问题 4：Agent 赛题的性能 baseline 使用哪个版本？
|
||||
|
||||
**回答：** 当前评测使用赛事方提供的 baseline，标准环境为 `PyTorch-Agent / 2.8.0 / Python 3.12 / MACA 3.7.1.5`。赛事方变更 baseline 或评测方式时会发布通知。
|
||||
|
||||
<a id="q-mla-dimensions"></a>
|
||||
|
||||
### ❓ 问题 5：MLA 的 `QK dim = 576, VO dim = 512` 与 `race_tests` 参数冲突吗？
|
||||
|
||||
**回答：** 不冲突。`race_tests` 中的 `dim=512, pe_dim=64` 对应 `QK dim = 576, V dim = 512`。
|
||||
|
||||
<a id="q-nsa-ranking"></a>
|
||||
|
||||
### ❓ 问题 6：NSA 的 109 个测试 case 如何计算榜单成绩？
|
||||
|
||||
**回答：** XPU-OJ 通过统一接口统计测试总运行时间，再根据 baseline 计算整体 speedup。
|
||||
|
||||
<a id="q-upstream-code"></a>
|
||||
|
||||
### ❓ 问题 7：可以引用或修改 FlashInfer、FlashAttention 等上游代码吗？
|
||||
|
||||
**回答：** 可以。参赛团队需遵守上游项目许可证，保留版权和许可证声明，并在提交材料中说明引用范围、迁移工作和自主优化内容。
|
||||
|
||||
<a id="evaluation"></a>
|
||||
|
||||
## 环境、镜像与算力资源
|
||||
|
||||
<a id="q-environment-image"></a>
|
||||
|
||||
### ❓ 问题 1：比赛使用哪个在线算力环境？
|
||||
|
||||
**回答：** 比赛使用模力方舟沐曦算力专区。仓库 README 当前标注的统一镜像为：
|
||||
|
||||
```text
|
||||
PyTorch-Agent / 2.8.0 / Python 3.12 / MACA 3.7.1.5
|
||||
```
|
||||
|
||||
创建实例和连接环境的步骤见[模力方舟快速使用 SOP](模力方舟快速使用SOP.md)。
|
||||
|
||||
<a id="q-download-maca"></a>
|
||||
|
||||
### ❓ 问题 2：如何获取 MACA 镜像或安装包？
|
||||
|
||||
**回答：** 参赛者可以在[模力方舟沐曦算力专区](https://ai.gitee.com/compute/metax)选择 `PyTorch-Agent` 镜像。需要单独获取软件包时，请前往[沐曦开发者社区软件中心](https://developer.metax-tech.com/softnova/docker?chip_name=%E6%9B%A6%E4%BA%91C500%E7%B3%BB%E5%88%97&package_kind=AI&dimension=docker)，并选择与比赛标准环境一致的 MACA `3.7.1.5` 版本。
|
||||
|
||||
<a id="q-download-pytorch"></a>
|
||||
|
||||
### ❓ 问题 3：比赛使用的 PyTorch 镜像可以下载吗？
|
||||
|
||||
**回答：** 可以。请在[沐曦开发者社区](https://developer.metax-tech.com/)或其 [PyTorch 镜像列表](https://developer.metax-tech.com/softnova/docker?chip_name=%E6%9B%A6%E4%BA%91C500%E7%B3%BB%E5%88%97&package_kind=AI&dimension=docker&deliver_type=%E5%88%86%E5%B1%82%E5%8C%85&ai_frame=pytorch)中查询。下载前请核对比赛镜像的 Python、PyTorch 和 MACA 版本。
|
||||
|
||||
<a id="q-version-mismatch"></a>
|
||||
|
||||
### ❓ 问题 4：页面标注的 MACA 版本与容器内版本不一致怎么办？
|
||||
|
||||
**回答：** 比赛标准版本为 MACA `3.7.1.5`。
|
||||
|
||||
<a id="q-pytorch-source"></a>
|
||||
|
||||
### ❓ 问题 5：Linux 版本的 mcprofiler 是否可用？
|
||||
|
||||
**回答：** mcprofiler 的 Linux 版本已经打包进模力方舟上的 pytorch-agent 比赛镜像。
|
||||
|
||||
<a id="q-compute-coupons"></a>
|
||||
|
||||
### ❓ 问题 6：算力券按团队还是按个人领取？
|
||||
|
||||
**回答：** 新人礼和启悟社区算力券按学生个人发放，符合条件的团队成员均可领取。团队主申请人的额度用完后，其他成员可以继续申请资源。
|
||||
|
||||
- [沐曦开发者社区新人礼](https://developer.metax-tech.com/activities/6)
|
||||
- [启悟社区学生算力券](https://developer.metax-tech.com/activities/11)
|
||||
- [赛事算力券活动](https://developer.metax-tech.com/activities/17)
|
||||
|
||||
<a id="q-more-compute"></a>
|
||||
|
||||
### ❓ 问题 7：算力额度不足时可以追加申请吗？
|
||||
|
||||
**回答：** 可以先领取上述活动中的算力券。仍需额外资源时，请发送需求邮件至 `opensource@metax-tech.com`。
|
||||
|
||||
<a id="q-commercial-agent-cost"></a>
|
||||
|
||||
<a id="registration"></a>
|
||||
|
||||
## 报名、组队与资格审核
|
||||
|
||||
<a id="q-register-both"></a>
|
||||
|
||||
### ❓ 问题 1：同一团队或个人可以同时参加两个赛题吗？
|
||||
|
||||
**回答：** 可以。同一团队或个人可以同时报名两个赛题。
|
||||
|
||||
<a id="q-register-multiple-tracks"></a>
|
||||
|
||||
### ❓ 问题 2：同一名学生可以报名不同赛道的不同赛题吗？
|
||||
|
||||
**回答：** 可以。赛事不统一限制学生报名不同赛道或不同赛题，但同一作品不得用相同核心技术内容重复申报不同赛题。
|
||||
|
||||
<a id="q-new-graduate"></a>
|
||||
|
||||
### ❓ 问题 3：本科应届毕业、尚未正式入学的研一新生可以报名吗？
|
||||
|
||||
**回答：** 可以。参赛者可联系原本科学校完成认证手续，并以本科生身份报名。
|
||||
|
||||
<a id="q-advisor-required"></a>
|
||||
|
||||
### ❓ 问题 4：参赛必须配备指导教师吗？
|
||||
|
||||
**回答：** 不强制。填写指导教师时，每支队伍可以设置 1 至 3 名指导教师。
|
||||
|
||||
<a id="q-advisor-team-limit"></a>
|
||||
|
||||
### ❓ 问题 5：一名指导教师最多可以指导几支队伍？
|
||||
|
||||
**回答：** 赛事暂未设置统一的硬性上限。指导教师应根据可投入的时间控制队伍数量。
|
||||
|
||||
<a id="q-cross-school-stamp"></a>
|
||||
|
||||
### ❓ 问题 6：跨校组队时，报名表应该由哪所学校盖章？
|
||||
|
||||
**回答：** 资格审批阶段，每名参赛者需到本人学校的校团委或院团委完成盖章确认。后续材料由团队牵头学生统一整理和提交。
|
||||
|
||||
<a id="q-upload-stamped-form"></a>
|
||||
|
||||
### ❓ 问题 7：提交报名后还可以补充已盖章的报名表扫描件吗？
|
||||
|
||||
**回答：** 审核人员发现材料缺少盖章时，会退回申请。团队补齐材料后可以重新提交。尚未完成盖章的团队应先与学院、校团委或学校相关部门确认办理方式。
|
||||
|
||||
<a id="q-stamp-department"></a>
|
||||
|
||||
### ❓ 问题 8：资格审查材料应该加盖哪个部门的公章？
|
||||
|
||||
**回答：** 各高校的管理口径不同。教务处、学生处等学籍或学生管理部门通常可以办理，参赛团队应以本校校团委或相关管理部门的要求为准。
|
||||
|
||||
<a id="q-no-youth-league"></a>
|
||||
|
||||
### ❓ 问题 9：学校未设校团委，可以用院系公章替代吗？
|
||||
|
||||
**回答：** 赛事原则上要求校级部门公章。学校未设校团委时，可以联系校级学工、双创或教务部门盖章，并提交情况说明。院系公章不能直接替代校级部门公章。
|
||||
|
||||
<a id="q-student-status-proof"></a>
|
||||
|
||||
### ❓ 问题 10：学校无法配合盖章，可以用学籍证明替代吗？
|
||||
|
||||
**回答：** 不可以。参赛团队应使用报名系统导出的报名表，并按要求完成学校盖章。
|
||||
|
||||
<a id="q-public-notice"></a>
|
||||
|
||||
### ❓ 问题 11：公示材料需要包含哪些内容？
|
||||
|
||||
**回答：** 请参考赛事工作群发布的参考文本，并按学校要求调整。跨校团队涉及的学校应分别在学校官网公示，公示渠道原则上使用学校官网。
|
||||
|
||||
<a id="q-review-deadline"></a>
|
||||
|
||||
### ❓ 问题 12：报名审核需要在报名截止日前完成吗？
|
||||
|
||||
**回答：** 原则上需要。往届出现过系统延后关闭的情况，但本届参赛团队不应据此推迟材料提交或审核。
|
||||
|
||||
<a id="q-review-flow"></a>
|
||||
|
||||
### ❓ 问题 13：报名材料的审核顺序是什么？
|
||||
|
||||
**回答：** 学生提交材料后，学校校团委先审核；学校审核通过后，企业再审核。企业审核通过即视为报名成功。
|
||||
|
||||
<a id="q-school-review-account"></a>
|
||||
|
||||
### ❓ 问题 14：后台显示“校团委审核”，但学校不了解审核事项，怎么办？
|
||||
|
||||
**回答：** 校团委需在报名系统内完成审核。省级团委通常会向各高校团委发放账号和密码。学校未收到或不了解安排时，请学校联系省级团委确认。
|
||||
|
||||
## FAQ 维护约定
|
||||
|
||||
1. 参赛者通过 Issue 提交问题。
|
||||
2. 维护者确认答案后更新本文档。
|
||||
3. 每个答案保留稳定锚点；需要时附上来源 Issue 或公告。
|
||||
4. 维护者在原 Issue 中回复 FAQ 锚点链接，并关闭已经解决的问题。
|
||||
5. 涉及版本、日期、评测参数的答案应标注确认日期。
|
||||
22
README.md
22
README.md
|
|
@ -1,22 +1,5 @@
|
|||
# 降低Token 成本，攻坚国产推理生态｜沐曦两大赛题登陆 2026 揭榜挂帅擂台赛，邀青年共破局！
|
||||
|
||||
## 常用入口
|
||||
|
||||
- [常见问题 FAQ](FAQ.md)
|
||||
- [沐曦通用GPU MXMACA编译器内建函数编程指南](https://developer.metax-tech.com/api/client/document/preview/1395/index.html)
|
||||
|
||||
## XPU-OJ 基线修复及榜单调整通知
|
||||
|
||||
针对近期部分同学反馈的“XPU.OJ”第三方评测系统中基线（baseline）不稳定的问题，我们高度重视，并已第一时间组织排查与测试。在此，我们对因此给大家带来的困扰深表歉意，也衷心感谢各位同学提出的宝贵意见。
|
||||
目前，相关问题已修复完毕。为确保评测的公平性与准确性，我们将对现有榜单进行清空处理。历史提交记录仍可查看，但后续排名将统一以基线修复后重新提交的算子成绩为准。
|
||||
|
||||
比赛期间，我们将持续关注系统运行状态，也欢迎大家继续向我们反馈建议。
|
||||
祝大家比赛顺利，取得理想成绩！
|
||||
|
||||
- XPU-OJ 地址：[https://xpuoj.com/](https://xpuoj.com/)
|
||||
- 账号申领说明：[赛事 XPU-OJ 账号申领说明](赛事XPUOJ账号申领说明.md)
|
||||
- 账号申领邮箱：`opensource@metax-tech.com`
|
||||
|
||||
2026 年度中国青年科技创新「揭榜挂帅」擂台赛正式启幕。沐曦股份重磅发布两大 AI 算力硬核榜题，聚焦国产 GPU 大模型推理算子优化，以硬核赛事搭建科研攻关平台，邀全国青年学子、科研人才揭榜攻坚，用技术重构推理效率，用创新拉低每 Token 算力成本！
|
||||
|
||||
## 两大重磅赛题 直击推理成本核心痛点
|
||||
|
|
@ -46,7 +29,7 @@
|
|||
|
||||
### 赛题二：基于 AI Agent 开发范式的国产 GPU 大模型推理算子库优化
|
||||
|
||||
大模型推理具有高并发、长序列、高<EFBFBD><EFBFBD><EFBFBD>用频次等特点，FlashInfer、FlashAttention、Fused MoE 等核心算子直接决定模型服务的吞吐、延迟与显存开销，影响单 Token 综合推理成本。
|
||||
大模型推理具有高并发、长序列、高调用频次等特点，FlashInfer、FlashAttention、Fused MoE 等核心算子直接决定模型服务的吞吐、延迟与显存开销，影响单 Token 综合推理成本。
|
||||
|
||||
本赛题面向沐曦国产 GPU 及 MXMACA 软件栈，鼓励参赛团队构建或使用 AI Agent / Skill 工作流，围绕推理算子库开展代码理解、算子迁移、性能分析、Kernel 优化、自动调优、Benchmark 验证和多轮迭代，探索“Agent 驱动算子优化”的新型开发范式。
|
||||
|
||||
|
|
@ -63,11 +46,10 @@
|
|||
**赛题二相关资料**
|
||||
|
||||
- [赛题二方案：基于 AI Agent 开发范式的国产 GPU 大模型推理算子库优化方案](基于AI%20Agent开发范式的国产GPU大模型推理算子库优化/基于AI%20Agent开发范式的国产GPU大模型算子推理库优化方案.md)
|
||||
- [赛题二选手入口](基于AI%20Agent开发范式的国产GPU大模型推理算子库优化/选手入口.md)
|
||||
- [模力方舟 Agent 部署准备教程](基于AI%20Agent开发范式的国产GPU大模型推理算子库优化/模力方舟Agent部署准备教程.md)
|
||||
- [赛题二说明及资料参考](基于AI%20Agent开发范式的国产GPU大模型推理算子库优化/赛题说明.md)
|
||||
|
||||
### **两个赛题统一使用模力方舟上的镜像PyTorch-Agent / 2.8.0 / Python 3.12 / maca 3.7.1.5**
|
||||
###**两个赛题统一使用镜像 MACA-Pytorch3.7.1.5-torch2.8-py312-ubuntu24.04**
|
||||
|
||||
## 参赛对象
|
||||
|
||||
|
|
|
|||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
|
|
@ -0,0 +1,583 @@
|
|||
# FlashInfer 迁移 Baseline 实战
|
||||
|
||||
## 一、教程定位
|
||||
|
||||
本教程是参赛训练课程的 **FlashInfer Baseline 入门**模块，主要帮助用户快速跑通 FlashInfer 的最小可运行流程。完成本教程后，用户应能够完成源码编译、API 调用、正确性测试和 Benchmark 测试，并记录一份 baseline 性能结果，为后续算子优化提供对比基准。
|
||||
|
||||
## 二、学习目标
|
||||
|
||||
1. 理解 FlashInfer Attention Kernel 的基本作用与适用场景；
|
||||
2. 完成 FlashInfer 环境、工具链的准备与源码编译；
|
||||
3. 跑通 BatchDecode、BatchPrefill、MLA 等典型算子的 API 调用示例；
|
||||
4. 完成各算子在不同参数配置下的 Benchmark 测试；
|
||||
5. 输出各算子的 Baseline 性能结果记录表，为后续算子优化提供对比基准。
|
||||
6. 理解 XPU-OJ 评测`run_kernel`接口与精度要求。
|
||||
7. 理解 Baseline 与 XPU-OJ 评测题包之间的关系，能够根据题包接口实现一个最小正确版 run_kernel。
|
||||
|
||||
|
||||
## 三、适用对象
|
||||
|
||||
适合人群
|
||||
|
||||
* 参赛选手：需要完成 Baseline 入门模块，为后续算子优化做准备
|
||||
* 软件开发者和Vibe Coding开发：希望从事AI相关行业开发，以及用智能体方式来做开发工作
|
||||
* LLM 推理开发者：希望了解 FlashInfer Attention Kernel 的性能表现
|
||||
* 算子优化工程师：希望基于MXMACA软件栈在沐曦国产 GPU 上做算子迁移和优化
|
||||
|
||||
前置基础
|
||||
|
||||
* Python 基础：能够运行和修改 Python 脚本
|
||||
* PyTorch 基础：了解MXMACA软化栈的使用
|
||||
* Linux 命令行：能够使用终端执行命令
|
||||
* 了解 Attention 机制：理解 Q/K/V、KV Cache 等基本概念
|
||||
|
||||
|
||||
## 四、前置准备
|
||||
|
||||
开始实战前，请确认你已经完成以下准备：
|
||||
|
||||
### GPU准备
|
||||
|
||||
* 步骤1：获取算力券[https://developer.metax-tech.com/activities/6](https://developer.metax-tech.com/activities/6)
|
||||
|
||||
* 首次登录需要先进行注册（使用邮箱或者手机号进行注册）
|
||||
|
||||
* 登录成功后进行第二步-邮箱验证，填入自己的邮箱。
|
||||
|
||||
* 第三步，提交申请。
|
||||
|
||||
* 获得兑换码
|
||||
|
||||
|
||||
* 步骤2：兑换算力和登陆平台
|
||||
|
||||
* 访问模力方舟官网：[https://ai.gitee.com/](https://ai.gitee.com/)
|
||||
|
||||
* 进入费用中心 - 算力券 ，点击右上角 “兑换”。
|
||||
|
||||
|
||||
* 步骤3：租用算力
|
||||
|
||||
- 进入算力容器，选择沐曦，租用算力，建议优先选16G显存/32G显存，如下图：
|
||||
* 步骤4：创建实例
|
||||
|
||||
- 基础镜像：maca-pytorch:3.7.1.5-torch2.8-py312-ubuntu24.04-amd64
|
||||
* 步骤5：选择工具-lab进入实例环境
|
||||
|
||||
|
||||
* 步骤6：在JupyterLab Terminal中检查运行环境的配置，确认沐曦 GPU 可见--可以使用`mx-smi`命令查看
|
||||
|
||||
|
||||
### 环境依赖准备
|
||||
|
||||
```bash
|
||||
pip install flashinfer torch pandas numpy
|
||||
```
|
||||
|
||||
### opencode Agent安装
|
||||
|
||||
```shell
|
||||
curl -fsSL https://opencode.ai/install | bash
|
||||
```
|
||||
|
||||
说明：
|
||||
|
||||
自动下载最新版本 自动配置环境 适用于 macOS / Linux / WSL 安装完成后，还会提示你怎么用，OpenCode 包含免费模式，使用方式：
|
||||
|
||||
```shell
|
||||
cd # 进入项目目录
|
||||
opencode
|
||||
```
|
||||
|
||||
### 代码准备
|
||||
|
||||
* 已进入指定项目目录；
|
||||
|
||||
* 准备测试脚本和 Benchmark 脚本。
|
||||
|
||||
|
||||
| 命令 | 说明 |
|
||||
| --- | --- |
|
||||
| `python bench_batch_decode.py` | 运行 Batch Decode 基准测试 |
|
||||
| `python bench_batch_prefill_paged.py` | 运行 Batch Prefill (Paged KV Cache) 基准测试 |
|
||||
| `python bench_batch_prefill_ragged.py` | 运行 Batch Prefill (Ragged KV Cache) 基准测试 |
|
||||
| `python bench_batch_mla.py` | 运行 MLA (Multi-head Latent Attention) 基准测试 |
|
||||
|
||||
## 五、知识预备
|
||||
|
||||
#### LLM推理阶段重要概念：
|
||||
|
||||
* **Prefill阶段**：prefill 阶段是指处理输入 prompt 的阶段
|
||||
|
||||
* 输入：用户一次性给出的完整 prompt，长度为 seq\_len
|
||||
|
||||
* 计算：对 prompt 中的每个 token 并行计算注意力，生成第一个输出 token 及 KV cache
|
||||
|
||||
* 特点：这是**计算密集型（compute-bound）**阶段，因为需要做完整的 seq\_len x seq\_len 注意力矩阵乘法
|
||||
|
||||
* **decode阶段**：
|
||||
|
||||
* 每次只生成 1 个 token，利用 prefill 阶段填充好的 KV cache 做自回归生成
|
||||
|
||||
* **显存带宽密集型（memory-bound**），瓶颈在从显存读取 KV cache 而非计算
|
||||
|
||||
|
||||
prefill = 并行处理用户输入，decode = 逐个生成回答 token
|
||||
|
||||
## 六、项目实践--FlashInfer-Baseline
|
||||
|
||||
## 本地项目运行
|
||||
|
||||
#### Step 1：检查运行环境
|
||||
|
||||
**目标：** 进入terminal确认当前环境满足本模块运行要求。
|
||||
|
||||
**操作：** 检查 GPU、Python、编译工具和依赖版本。
|
||||
|
||||
|
||||
|
||||
**常见问题：**
|
||||
|
||||
| 问题 | 解决方法 |
|
||||
| --- | --- |
|
||||
| `mx-smi: command not found` | 使用 `python -c "import torch; print(torch.cuda.get_device_name(0))"` |
|
||||
| `No GPUs are available` | 检查 MXMACA 驱动是否正确安装 |
|
||||
| `ModuleNotFoundError: No module named 'flashinfer'` | 执行 `pip install flashinfer` 安装 |
|
||||
|
||||
#### Step 2：进入项目目录
|
||||
|
||||
**目标：** 进入本模块所需的源码目录[flashinfer_baseline](https://gitlink.org.cn/metax-maca/op_optimization/tree/master/%E5%9F%BA%E4%BA%8EAI%20Agent%E5%BC%80%E5%8F%91%E8%8C%83%E5%BC%8F%E7%9A%84%E5%9B%BD%E4%BA%A7GPU%E5%A4%A7%E6%A8%A1%E5%9E%8B%E6%8E%A8%E7%90%86%E7%AE%97%E5%AD%90%E5%BA%93%E4%BC%98%E5%8C%96%2Fbaselines%2Fflashinfer_baseline)。
|
||||
|
||||
1. 克隆代码仓库
|
||||
|
||||
```Bash
|
||||
git clone https://gitlink.org.cn/metax-maca/op_optimization.git
|
||||
```
|
||||
|
||||
2. 准备flashinfer_baseline
|
||||
|
||||
在仓库目录 `op_optimization/基于AI Agent开发范式的国产GPU大模型推理算子库优化` 下，找到 `flashinfer_baseline` 文件夹。可以将 `flashinfer_baseline` 整个目录复制到工作目录 `data/` 下。
|
||||
|
||||
**下一步操作：** 切换到 FlashInfer_Baseline 项目目录。
|
||||
|
||||
**命令示例：**
|
||||
|
||||
```bash
|
||||
cd baselines\flashinfer_baseline\FlashInfer_Baseline
|
||||
ls -la
|
||||
```
|
||||
|
||||
**预期结果：**
|
||||
|
||||
```plaintext
|
||||
bench_common.py
|
||||
bench_batch_decode.py
|
||||
bench_batch_prefill_paged.py
|
||||
bench_batch_prefill_ragged.py
|
||||
bench_batch_mla.py
|
||||
README.md
|
||||
...
|
||||
|
||||
```
|
||||
|
||||
#### Step 3：验证项目脚本
|
||||
|
||||
**目标：** 确认所有基准测试脚本可正常执行。
|
||||
|
||||
**操作：** 检查脚本文件是否存在且可读。
|
||||
|
||||
**命令示例：**
|
||||
|
||||
```bash
|
||||
# 检查脚本文件
|
||||
python -c "import os; scripts = ['bench_common.py', 'bench_batch_decode.py', 'bench_batch_prefill_paged.py', 'bench_batch_prefill_ragged.py', 'bench_batch_mla.py']; [print(f'✓ {s}') if os.path.exists(s) else print(f'✗ {s} missing') for s in scripts]"
|
||||
|
||||
# 测试脚本导入
|
||||
python -c "from bench_common import setup_workspace, get_csv_path; print('脚本导入正常')"
|
||||
|
||||
```
|
||||
|
||||
**预期结果：**
|
||||
|
||||
```plaintext
|
||||
✓ bench_common.py
|
||||
✓ bench_batch_decode.py
|
||||
✓ bench_batch_prefill_paged.py
|
||||
✓ bench_batch_prefill_ragged.py
|
||||
✓ bench_batch_mla.py
|
||||
脚本导入正常
|
||||
|
||||
```
|
||||
|
||||
#### Step 4：运行单算子 Benchmark
|
||||
|
||||
**目标：** 执行基准测试，获取 Baseline 性能数据。
|
||||
|
||||
**操作：** 运行 BatchDecode 基准测试脚本。
|
||||
|
||||
**命令示例：**
|
||||
|
||||
```bash
|
||||
python bench_batch_decode.py
|
||||
|
||||
```
|
||||
|
||||
**预期结果：**
|
||||
|
||||
```plaintext
|
||||
[BatchDecodeWithPagedKVCacheWrapper] Starting benchmark, total cases: 144
|
||||
[1/144] bs=1, kv_len=512, hd=64: 0.020ms, 51.53 GB/s, 0.82 TFLOPs
|
||||
[2/144] bs=1, kv_len=512, hd=128: 0.023ms, 45.28 GB/s, 0.72 TFLOPs
|
||||
...
|
||||
|
||||
Results saved to BatchDecodeWithPagedKVCacheWrapper_20260525_xxxxxx.csv
|
||||
|
||||
```
|
||||
|
||||
**常见问题：**
|
||||
|
||||
| 问题 | 解决方法 |
|
||||
| --- | --- |
|
||||
| `out of memory` | 减小 batch\_size 或 seq\_len 参数 |
|
||||
| 运行时间过长 | 脚本会自动调整重复次数，耐心等待 |
|
||||
|
||||
#### Step 5：查看测试结果
|
||||
|
||||
**目标：** 查看并分析 Benchmark 输出结果。
|
||||
|
||||
**操作：** 读取生成的 CSV 结果文件。
|
||||
|
||||
**命令示例：**
|
||||
|
||||
```bash
|
||||
# 查看最新生成的 CSV 文件
|
||||
ls -lt *.csv | head -1
|
||||
|
||||
# 使用 Python 查看结果
|
||||
python -c "
|
||||
import pandas as pd
|
||||
import glob
|
||||
csv_files = glob.glob('BatchDecodeWithPagedKVCacheWrapper_*.csv')
|
||||
if csv_files:
|
||||
df = pd.read_csv(max(csv_files))
|
||||
print(df[['batch_size', 'seq_len_kv', 'head_dim', 'time_ms', 'bandwidth_GB_s', 'tflops']].head(10))
|
||||
"
|
||||
|
||||
```
|
||||
|
||||
**预期结果：**
|
||||
|
||||
```plaintext
|
||||
batch_size seq_len_kv head_dim time_ms bandwidth_GB_s tflops
|
||||
0 1 512 64 0.020429 51.528822 0.821253
|
||||
1 1 512 128 0.023340 45.278052 0.718833
|
||||
2 1 512 256 0.031949 66.153846 1.050256
|
||||
...
|
||||
|
||||
```
|
||||
|
||||
### **XPU-OJ 评测与提交**
|
||||
|
||||
#### Step 1：访问XPUOJ平台(https://xpuoj.com/)进行登录
|
||||
|
||||
- 完成 baseline benchmark 后，要到统一评测工具XPU-OJ完成提交，进行评测。
|
||||
|
||||
- 等待组委会统一发放 XPU-OJ 账号
|
||||
- 使用分配的用户名和初始密码登录平台
|
||||
|
||||
|
||||
|
||||
#### Step 2：进入比赛页面
|
||||
|
||||
- 登录后进入对应赛题页面
|
||||
- 选择对应的赛题
|
||||
|
||||
|
||||
|
||||
#### Step 3：选择对应的算子题目进行评测
|
||||
|
||||
|
||||
|
||||
#### Step 4：阅读题目要求，接口约定，编写符合`run_kernel`接口规范的CUDA算子
|
||||
|
||||
以下代码仅用于说明接口结构，不代表最优实现，也不作为评分参考：
|
||||
|
||||
```python
|
||||
FlashInfer Ragged Prefill示例代码
|
||||
#include <stdint.h>
|
||||
#include <cuda_bf16.h>
|
||||
#include <cuda_runtime.h>
|
||||
#include <math.h>
|
||||
|
||||
#ifndef max
|
||||
#define max(a,b) ((a) > (b) ? (a) : (b))
|
||||
#endif
|
||||
|
||||
#ifndef min
|
||||
#define min(a,b) ((a) < (b) ? (a) : (b))
|
||||
#endif
|
||||
|
||||
template <int D_QK, int D_VO, int Br, int Bc>
|
||||
__global__ void ragged_prefill_kernel(
|
||||
const __nv_bfloat16* __restrict__ q,
|
||||
const __nv_bfloat16* __restrict__ k,
|
||||
const __nv_bfloat16* __restrict__ v,
|
||||
__nv_bfloat16* __restrict__ output,
|
||||
const int32_t* __restrict__ qo_indptr,
|
||||
const int32_t* __restrict__ kv_indptr,
|
||||
int64_t seq_len,
|
||||
int64_t num_qo_heads,
|
||||
int64_t num_kv_heads,
|
||||
float scale,
|
||||
int64_t causal
|
||||
) {
|
||||
int b = blockIdx.x;
|
||||
int hq = blockIdx.y;
|
||||
int hkv = hq / (num_qo_heads / num_kv_heads);
|
||||
|
||||
int ty = threadIdx.y; // 0~31, Block 内的 query 索引 (对应 Br)
|
||||
int tx = threadIdx.x; // 0~31, Warp 内的线程 (lane)
|
||||
|
||||
// 使用 int64_t 防止长序列+大Batch组合下的整数溢出
|
||||
int64_t q_idx = (int64_t)blockIdx.z * Br + ty;
|
||||
bool is_valid = q_idx < seq_len;
|
||||
|
||||
int64_t cur_qo_start = qo_indptr[b];
|
||||
int64_t cur_kv_start = kv_indptr[b];
|
||||
|
||||
float r_Q[D_QK / 32];
|
||||
if (is_valid) {
|
||||
// 显式使用 int64_t 计算全局偏移量，防止超过 2GB 显存边界
|
||||
int64_t q_offset_base = (cur_qo_start + q_idx) * num_qo_heads * D_QK + hq * D_QK;
|
||||
#pragma unroll
|
||||
for (int d = 0; d < D_QK / 32; ++d) {
|
||||
r_Q[d] = __bfloat162float(q[q_offset_base + tx + d * 32]);
|
||||
}
|
||||
}
|
||||
|
||||
float r_O[D_VO / 32];
|
||||
#pragma unroll
|
||||
for (int d = 0; d < D_VO / 32; ++d) r_O[d] = 0.0f;
|
||||
|
||||
float m = -1e20f;
|
||||
float l = 0.0f;
|
||||
|
||||
extern __shared__ __nv_bfloat16 smem[];
|
||||
__nv_bfloat16* shared_K = smem;
|
||||
__nv_bfloat16* shared_V = smem + Bc * D_QK;
|
||||
|
||||
int tid = ty * 32 + tx;
|
||||
int total_threads = Br * 32;
|
||||
|
||||
int64_t block_max_q = min((int64_t)seq_len - 1, (int64_t)(blockIdx.z * Br + Br - 1));
|
||||
|
||||
int total_k_ldg = (Bc * D_QK) / 8;
|
||||
int total_v_ldg = (Bc * D_VO) / 8;
|
||||
|
||||
for (int64_t kv_start = 0; kv_start <= block_max_q; kv_start += Bc) {
|
||||
|
||||
// 协同加载 K
|
||||
for (int i = tid; i < total_k_ldg; i += total_threads) {
|
||||
int token_idx = i / (D_QK / 8);
|
||||
int d_idx = i % (D_QK / 8);
|
||||
int64_t global_kv_idx = kv_start + token_idx;
|
||||
|
||||
if (global_kv_idx < seq_len) {
|
||||
// 同样强制为 int64_t
|
||||
int64_t k_offset = (cur_kv_start + global_kv_idx) * num_kv_heads * D_QK + hkv * D_QK + d_idx * 8;
|
||||
((float4*)shared_K)[i] = *(reinterpret_cast<const float4*>(&k[k_offset]));
|
||||
} else {
|
||||
((float4*)shared_K)[i] = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
|
||||
}
|
||||
}
|
||||
|
||||
// 协同加载 V
|
||||
for (int i = tid; i < total_v_ldg; i += total_threads) {
|
||||
int token_idx = i / (D_VO / 8);
|
||||
int d_idx = i % (D_VO / 8);
|
||||
int64_t global_kv_idx = kv_start + token_idx;
|
||||
|
||||
if (global_kv_idx < seq_len) {
|
||||
int64_t v_offset = (cur_kv_start + global_kv_idx) * num_kv_heads * D_VO + hkv * D_VO + d_idx * 8;
|
||||
((float4*)shared_V)[i] = *(reinterpret_cast<const float4*>(&v[v_offset]));
|
||||
} else {
|
||||
((float4*)shared_V)[i] = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
|
||||
}
|
||||
}
|
||||
__syncthreads();
|
||||
|
||||
if (is_valid) {
|
||||
int64_t kv_end = causal ? min(kv_start + Bc, q_idx + 1) : min(kv_start + Bc, seq_len);
|
||||
|
||||
for (int j = 0; j < kv_end - kv_start; ++j) {
|
||||
float score = 0.0f;
|
||||
#pragma unroll
|
||||
for (int d = 0; d < D_QK / 32; ++d) {
|
||||
score += r_Q[d] * __bfloat162float(shared_K[j * D_QK + tx + d * 32]);
|
||||
}
|
||||
|
||||
#pragma unroll
|
||||
for (int offset = 16; offset > 0; offset /= 2) {
|
||||
score += __shfl_down_sync(0xffffffff, score, offset);
|
||||
}
|
||||
score = __shfl_sync(0xffffffff, score, 0);
|
||||
score *= scale;
|
||||
|
||||
float m_prev = m;
|
||||
m = max(m, score);
|
||||
float P = __expf(score - m);
|
||||
l = l * __expf(m_prev - m) + P;
|
||||
|
||||
#pragma unroll
|
||||
for (int d = 0; d < D_VO / 32; ++d) {
|
||||
r_O[d] = r_O[d] * __expf(m_prev - m) + P * __bfloat162float(shared_V[j * D_VO + tx + d * 32]);
|
||||
}
|
||||
}
|
||||
}
|
||||
__syncthreads();
|
||||
}
|
||||
|
||||
if (is_valid) {
|
||||
int64_t out_offset_base = (cur_qo_start + q_idx) * num_qo_heads * D_VO + hq * D_VO;
|
||||
#pragma unroll
|
||||
for (int d = 0; d < D_VO / 32; ++d) {
|
||||
float out_val = r_O[d] / l;
|
||||
output[out_offset_base + tx + d * 32] = __float2bfloat16(out_val);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
extern "C" void run_kernel(
|
||||
const __nv_bfloat16* q,
|
||||
const __nv_bfloat16* k,
|
||||
const __nv_bfloat16* v,
|
||||
__nv_bfloat16* output,
|
||||
const int32_t* qo_indptr,
|
||||
const int32_t* kv_indptr,
|
||||
int64_t batch_size,
|
||||
int64_t seq_len,
|
||||
int64_t num_qo_heads,
|
||||
int64_t num_kv_heads,
|
||||
int64_t head_dim_qk,
|
||||
int64_t head_dim_vo,
|
||||
int64_t causal
|
||||
) {
|
||||
float scale = 1.0f / sqrtf((float)head_dim_qk);
|
||||
const int Br = 32;
|
||||
|
||||
dim3 grid(batch_size, num_qo_heads, (seq_len + Br - 1) / Br);
|
||||
dim3 block(32, Br);
|
||||
|
||||
// 根据维度配置绝对安全的共享内存策略
|
||||
if (head_dim_qk == 128 && head_dim_vo == 128) {
|
||||
const int Bc = 64;
|
||||
int smem = Bc * 128 * 2 + Bc * 128 * 2; // 32 KB
|
||||
cudaFuncSetAttribute(ragged_prefill_kernel<128, 128, Br, Bc>, cudaFuncAttributeMaxDynamicSharedMemorySize, smem);
|
||||
ragged_prefill_kernel<128, 128, Br, Bc><<<grid, block, smem>>>(
|
||||
q, k, v, output, qo_indptr, kv_indptr, seq_len, num_qo_heads, num_kv_heads, scale, causal);
|
||||
} else if (head_dim_qk == 192 && head_dim_vo == 128) {
|
||||
const int Bc = 64;
|
||||
int smem = Bc * 192 * 2 + Bc * 128 * 2; // 40 KB
|
||||
cudaFuncSetAttribute(ragged_prefill_kernel<192, 128, Br, Bc>, cudaFuncAttributeMaxDynamicSharedMemorySize, smem);
|
||||
ragged_prefill_kernel<192, 128, Br, Bc><<<grid, block, smem>>>(
|
||||
q, k, v, output, qo_indptr, kv_indptr, seq_len, num_qo_heads, num_kv_heads, scale, causal);
|
||||
} else if (head_dim_qk == 256 && head_dim_vo == 256) {
|
||||
// 关键降级：256维度下，将 Bc 从 64 降为 32。
|
||||
// 将所需的共享内存从 64KB 压回到绝对安全的 32KB，防止在部分环境触发 Launch Failure (OOM/Crash)。
|
||||
const int Bc = 32;
|
||||
int smem = Bc * 256 * 2 + Bc * 256 * 2; // 32 KB
|
||||
cudaFuncSetAttribute(ragged_prefill_kernel<256, 256, Br, Bc>, cudaFuncAttributeMaxDynamicSharedMemorySize, smem);
|
||||
ragged_prefill_kernel<256, 256, Br, Bc><<<grid, block, smem>>>(
|
||||
q, k, v, output, qo_indptr, kv_indptr, seq_len, num_qo_heads, num_kv_heads, scale, causal);
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
#### Step 5：提交代码开始评测
|
||||
|
||||
|
||||
|
||||
## 七、常见问题
|
||||
|
||||
### 环境相关问题
|
||||
|
||||
| 问题 | 原因 | 解决办法 |
|
||||
| --- | --- | --- |
|
||||
| `No GPUs are available` | MXMACA 驱动未安装或 GPU 不可见 | 检查驱动安装，运行 `python -c "import torch; print(torch.cuda.device_count())"` 验证 |
|
||||
| `ModuleNotFoundError: No module named 'flashinfer'` | flashinfer 未安装 | 执行 `pip install flashinfer` |
|
||||
| `out of memory` | GPU 显存不足 | 减小 `batch_size` 或 `seq_len` 参数 |
|
||||
|
||||
### 运行相关问题
|
||||
|
||||
| 问题 | 原因 | 解决办法 |
|
||||
| --- | --- | --- |
|
||||
| Benchmark 运行时间过长 | 参数组合过多， workload 较大 | 耐心等待，脚本会自动调整重复次数 |
|
||||
| `KeyError: 'BatchPrefillWithPagedKVCacheKernel'` | profiler 未捕获目标 kernel | 检查 `target_kernels` 配置是否正确 |
|
||||
| CSV 文件为空 | 测试未正常完成 | 检查 GPU 显存是否充足，重新运行 |
|
||||
|
||||
### 代码相关问题
|
||||
|
||||
| 问题 | 原因 | 解决办法 |
|
||||
| --- | --- | --- |
|
||||
| `ImportError: cannot import name 'xxx' from 'bench_common'` | 函数名拼写错误 | 检查 `bench_common.py` 中的函数名 |
|
||||
| `RuntimeError: error: device-side assert triggered` | 输入参数超出范围 | 检查 `num_qo_heads`、`num_kv_heads`、`head_dim` 配置 |
|
||||
|
||||
### 性能相关问题
|
||||
|
||||
| 问题 | 原因 | 解决办法 |
|
||||
| --- | --- | --- |
|
||||
| TFLOPs 数值异常低 | 工作负载过小，kernel 启动开销占比大 | 增大 `batch_size` 或 `seq_len` |
|
||||
| 带宽数值异常低 | 数据未正确加载到 GPU | 检查 Tensor 是否在 CUDA 设备上 |
|
||||
|
||||
## 八、下一步学习建议
|
||||
|
||||
### 1. 保存你的 Baseline 结果
|
||||
|
||||
将本次运行生成的 CSV 文件妥善保存，后续优化时需要以此作为对比基准。
|
||||
|
||||
```bash
|
||||
# 建议创建 results 目录保存
|
||||
mkdir -p results
|
||||
mv *.csv results/
|
||||
|
||||
```
|
||||
|
||||
### 2. 深入理解 FlashInfer 核心概念
|
||||
|
||||
* 阅读 FlashInfer 官方文档，理解 Paged KV Cache、Ragged KV Cache 的设计理念
|
||||
|
||||
* 学习 MLA (Multi-head Latent Attention) 的原理，了解 DeepSeek 的注意力优化方案
|
||||
|
||||
* 理解 `plan()` 和 `run()` 两阶段设计的作用
|
||||
|
||||
|
||||
### 3. 进入算子优化模块
|
||||
|
||||
参考后续优化模块，学习以下优化技术：
|
||||
|
||||
* **Kernel Tuning**：调整 Block Size、Thread Count 等参数
|
||||
|
||||
* **Memory Optimization**：减少显存占用、优化数据搬运
|
||||
|
||||
* **Compute Optimization**：提升计算效率
|
||||
|
||||
|
||||
### 4. 参考资源
|
||||
|
||||
* FlashInfer 官方仓库：https://github.com/flashinfer-ai/flashinfer
|
||||
|
||||
* FlashInfer 文档：https://flashinfer.ai
|
||||
|
||||
|
||||
### 5. 记录优化过程
|
||||
|
||||
建议维护一份优化日志，记录每次优化的改动和性能变化：
|
||||
|
||||
| 优化项 | 改动内容 | Baseline | 优化后 | 提升比例 |
|
||||
| --- | --- | --- | --- | --- |
|
||||
| 例：调整 block\_size | 16 → 32 | xx ms | xx ms | xx% |
|
||||
| 完成优化后，再次运行本模块的 Benchmark 脚本，对比前后性能变化。 | | | | |
|
||||
|
||||
### 6. 使用多语言完成算子优化加速
|
||||
|
||||
可以使用Triton和TileLang语言实现`run_kernel`接口完成算子优化，对比不同的语言对于性能加速的影响
|
||||
File diff suppressed because it is too large
Load Diff
|
|
@ -0,0 +1,614 @@
|
|||
# Fused MoE Baseline 入门：快速跑通最小闭环教程
|
||||
|
||||
## 一、教程定位
|
||||
|
||||
本教程是参赛训练课程的 baseline 入门模块，主要帮助用户快速跑通 Fused MoE 的最小可运行流程。
|
||||
|
||||
完成本教程后，用户应能够完成源码编译、正确性测试和 benchmark 测试，并记录一份 baseline 性能结果，为后续算子优化提供对比基准。
|
||||
|
||||
## 二、学习目标
|
||||
|
||||
完成本模块后，你将能够：
|
||||
|
||||
1. 理解 Fused MoE 推理算子及 int8 量化在大模型推理加速中的核心作用与优化意义；
|
||||
|
||||
2. 完成环境配置和 Fused MoE baseline 源码准备；
|
||||
|
||||
3. 跑通 Fused MoE 的最小闭环示例；
|
||||
|
||||
4. 理解基于 Trition 与 MXMACA C++ 写Fused MOE 算子；
|
||||
|
||||
5. 完成数值正确性测试，即验证 reference 计算、pybind 计算、Triton 计算这三种方式计算结果是否数值完全一致。
|
||||
|
||||
* reference：基于 PyTorch 架构在 CPU 上运行的**数值基准**实现；
|
||||
|
||||
* pybind：将 MXMACA C++ 算子编译并封装为 Python 可调用的动态库，**实现复杂且迁移成本高**；
|
||||
|
||||
* Triton：基于 Python 编写的高效 GPU Kernel，可利用 Agent 自动调优，**开发效率高、易于迁移**；
|
||||
|
||||
* 要求 pybind 和 Triton 结果均与 reference 一致，鼓励参赛选手持续调优 Triton ，使其性能逼近甚至超越 pybind 性能。
|
||||
|
||||
6. 输出 benchmark 结果对比表。
|
||||
|
||||
|
||||
## 三、适用对象
|
||||
|
||||
**本模块适合以下人员：**
|
||||
|
||||
* 参与基于 AI Agent 开发范式的国产 GPU 大模型推理算子库优化比赛的学生；
|
||||
|
||||
* 对 GPU 推理算子性能优化感兴趣的开发者；
|
||||
|
||||
* 需要了解 Fused MoE 推理性能的研究人员。
|
||||
|
||||
|
||||
**学习本模块前，需掌握以下基础知识：**
|
||||
|
||||
* Python、C++ 编程基础；
|
||||
|
||||
* PyTorch 基础；
|
||||
|
||||
* GPU 推理基本概念。
|
||||
|
||||
|
||||
## 四、前置准备
|
||||
|
||||
**开始实战前，请确认你已经完成以下准备：**
|
||||
|
||||
**环境准备：**
|
||||
|
||||
* 已进入赛事专属镜像环境。
|
||||
|
||||
**工具准备：**
|
||||
|
||||
* 已准备 Agent 工具；
|
||||
* 已配置 Token / API Key；
|
||||
* 已确认 Agent 可以正常调用模型。
|
||||
|
||||
**代码准备：**
|
||||
|
||||
* 已获取 Fused MoE Baseline 源码。
|
||||
|
||||
|
||||
## 五、项目实践1-算力平台 baseline 源码部署
|
||||
|
||||
**项目目标：**在算力平台上拉取 Fused MoE 的 baseline 源码，快速跑通最小闭环，并记录一份 baseline 性能结果，为后续算子优化提供对比基准。
|
||||
|
||||
### 步骤 1：检查运行环境
|
||||
|
||||
**目标：**确认当前环境满足本模块运行要求，包括编译器、MXMACA 工具链及 Python 依赖库。
|
||||
|
||||
**操作：**检查 Python、编译工具、MXMACA 编译器及关键 Python 包（numpy、torch、triton）是否存在。
|
||||
|
||||
**命令示例：**
|
||||
|
||||
```apl
|
||||
python --version # 检查Python版本，Python ≥ 3.8
|
||||
g++ --version # 确认 C++ 编译器存在
|
||||
which mxcc # 确认 MACA 编译器存在
|
||||
|
||||
|
||||
# 检查 Python 依赖
|
||||
python - << 'EOF'
|
||||
import sys
|
||||
deps = ["numpy", "torch", "triton"]
|
||||
missing = []
|
||||
for d in deps:
|
||||
try:
|
||||
__import__(d)
|
||||
except ImportError:
|
||||
missing.append(d)
|
||||
if missing:
|
||||
print(f"[ERROR] Missing packages: {missing}")
|
||||
sys.exit(1)
|
||||
else:
|
||||
print("[OK] numpy, torch, triton are installed.")
|
||||
EOF
|
||||
```
|
||||
|
||||
|
||||
**预期结果：**
|
||||
|
||||
* Python 3.12.11
|
||||
|
||||
* g++ (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0
|
||||
|
||||
* /opt/maca/mxgpu\_llvm/bin/mxcc
|
||||
|
||||
* \[OK\] numpy, torch, triton are installed.
|
||||
|
||||
|
||||
**常见问题：**
|
||||
|
||||
| 报错 | 原因 | 解决办法 |
|
||||
| --- | --- | --- |
|
||||
| `g++：command not found` | 未安装 C++ 编译工具 | `apt update && apt install -y build-essential` |
|
||||
| `Python 3.6.x/ Python 3.7.x` | Python 版本过低 | `conda install python=3.12` （推荐3.10+) |
|
||||
| `ModuleNotFoundError: numpy` | 当前 Python 缺少依赖 | `pip install numpy torch triton` |
|
||||
|
||||
|
||||
### 步骤 2：进入项目目录
|
||||
|
||||
**目标：**进入本模块所需的源码目录：https://gitlink.org.cn/metax-maca/op_optimization/tree/master/%E5%9F%BA%E4%BA%8EAI%20Agent%E5%BC%80%E5%8F%91%E8%8C%83%E5%BC%8F%E7%9A%84%E5%9B%BD%E4%BA%A7GPU%E5%A4%A7%E6%A8%A1%E5%9E%8B%E6%8E%A8%E7%90%86%E7%AE%97%E5%AD%90%E5%BA%93%E4%BC%98%E5%8C%96%2Fbaselines%2Ffused_moe
|
||||
|
||||
1. 克隆代码仓库
|
||||
|
||||
```Bash
|
||||
git clone https://gitlink.org.cn/metax-maca/op_optimization.git
|
||||
```
|
||||
|
||||
2. 准备fused_moe
|
||||
|
||||
在仓库目录 `op_optimization/基于AI Agent开发范式的国产GPU大模型推理算子库优化` 下，找到 `fused_moe` 文件夹。可以将 `fused_moe` 整个目录复制到工作目录 `data/` 下。
|
||||
|
||||
**下一步操作：** 切换到 FlashInfer Baseline 项目目录。
|
||||
|
||||
**操作：**切换到指定项目路径。
|
||||
|
||||
**命令示例：**
|
||||
|
||||
```apl
|
||||
cd data/fused_moe # 这里需要替换成自己的项目路径
|
||||
```
|
||||
|
||||
**预期结果：**
|
||||
|
||||
终端提示符路径显示为 fused\_moe 项目路径：
|
||||
|
||||
* (base) root@0a5a9d0c0f06:/data/fused_moe#
|
||||
|
||||
|
||||
### 步骤 3：pybind 编译
|
||||
|
||||
**目标：**将用 C++ 编写的 fused\_moe 算子编译为 Python 可调用的 pybind 模块。
|
||||
|
||||
**操作：**运行 `fused_moe/scripts/build_fused_moe_i8_tn_pybind.sh` 脚本
|
||||
|
||||
**命令示例：**
|
||||
|
||||
```apl
|
||||
bash scripts/build_fused_moe_i8_tn_pybind.sh
|
||||
```
|
||||
|
||||
切换 Python 环境命令示例：
|
||||
|
||||
```apl
|
||||
[SUCCESS] /data/fused_moe/standalone/fused_moe_i8_tn/build/fused_moe_i8_tn_pybind.sh
|
||||
```
|
||||
|
||||
**预期结果：**
|
||||
|
||||
编译成功无报错，终端显示：
|
||||
|
||||
* \[SUCCESS] /data/fused_moe/standalone/fused_moe_i8_tn/build/fused_moe_i8_tn_pybind.so
|
||||
|
||||
且成功生成 `fused_moe/standalone/fused_moe_i8_tn/build/fused_moe_i8_tn_pybind.cpython-310-x86_64-linux-gnu.so` 文件
|
||||
|
||||
|
||||
**常见问题：**
|
||||
|
||||
|
||||
| 报错 | 原因 | 解决办法 |
|
||||
| --- | --- | --- |
|
||||
| `Python.h: No such file or directory` | Python 头文件路径未找到 | 确认 `PYTHON_BIN` 路径正确，脚本自动探测 `sysconfig.get_path('include')` |
|
||||
| `libpython3.x.so: cannot find` | 链接时找不到 Python 库 | 1、执行 `find $CONDA_PREFIX -name "libpython3*.so*"`查找绝对路径<br>2、将该路径赋值给 `LIBPYTHON_PATH` |
|
||||
| `recompile with -fPIC` | 编译未开启位置无关代码 | 确保 `mxcc`/ `g++`编译参数中有 `-fPIC` |
|
||||
| `permission denied` | 无脚本执行权限 | `chmod +x scripts/*.sh` |
|
||||
| `undefined reference to Py_...` | Python 版本不匹配 | 确认编译脚本中`PYTHON_BIN`路径与当前运行的 Python 环境完全一致 |
|
||||
|
||||
### 步骤 4：正确性测试
|
||||
|
||||
**目标：**验证 reference 计算、pybind 计算、Triton 计算这三种方式计算结果的数值是否完全一致。
|
||||
|
||||
**操作：**运行 `fused_moe/scripts/run_fused_moe_i8_tn_pybind_test.sh` 脚本
|
||||
|
||||
**命令示例：**
|
||||
|
||||
```apl
|
||||
bash scripts/run_fused_moe_i8_tn_pybind_test.sh --backend all # 运行全部计算方式
|
||||
|
||||
# --backend：选择计算方式
|
||||
# 只测 pybind：
|
||||
bash scripts/run_fused_moe_i8_tn_pybind_test.sh --backend pybind
|
||||
# 只测 triton：
|
||||
bash scripts/run_fused_moe_i8_tn_pybind_test.sh --backend triton
|
||||
# 只测 reference:
|
||||
bash scripts/run_fused_moe_i8_tn_pybind_test.sh --backend reference
|
||||
```
|
||||
|
||||
|
||||
**预期结果：**
|
||||
|
||||
编译成功无报错，输出示例如下：
|
||||
|
||||
> pybind:fused\_moe\_i8\_tn\_topk1 passed: rows=256, cols=128, sample C\[0\]=0.69531, C\[last\]=-0.44531
|
||||
> pybind:fused\_moe\_i8\_tn\_topk2 passed: rows=512, cols=128, sample C\[0\]=-0.57813, C\[last\]=-0.49805
|
||||
> pybind:fused\_moe\_i8\_tn\_topk3 passed: rows=384, cols=128, sample C\[0\]=-1.08594, C\[last\]=-0.33594
|
||||
|
||||
> reference:fused\_moe\_i8\_tn\_topk1 passed: rows=256, cols=128, sample C\[0\]=0.6934, C\[last\]=-0.4451
|
||||
> reference:fused\_moe\_i8\_tn\_topk2 passed: rows=512, cols=128, sample C\[0\]=-0.5768, C\[last\]=-0.4975
|
||||
> reference:fused\_moe\_i8\_tn\_topk3 passed: rows=384, cols=128, sample C\[0\]=-1.0875, C\[last\]=-0.3362
|
||||
|
||||
> triton:fused\_moe\_i8\_tn\_topk1 passed: rows=256, cols=128, sample C\[0\]=0.69337, C\[last\]=-0.44513
|
||||
> triton:fused\_moe\_i8\_tn\_topk2 passed: rows=512, cols=128, sample C\[0\]=-0.57678, C\[last\]=-0.49749
|
||||
> triton:fused\_moe\_i8\_tn\_topk3 passed: rows=384, cols=128, sample C\[0\]=-1.08748, C\[last\]=-0.33618
|
||||
|
||||
|
||||
**结果解释：**
|
||||
|
||||
* “pybind/reference/Triton”：三种计算方式；
|
||||
|
||||
* “fused\_moe\_i8\_tn\_topk1/2/3 passed”：测试算子通过数值校验，数值误差在允许范围内且无明显异常，否则会报错 FAILED；
|
||||
|
||||
* ”rows=... , cols=...“：输出 Tensor 的行和列的大小；
|
||||
|
||||
* ”sample C\[0\]=... , C\[last\]=...“：首尾采样值，用于辅助定位数值偏差，不作为精度判定依据。
|
||||
|
||||
|
||||
**常见问题：**
|
||||
|
||||
|
||||
| 报错 | 原因 | 解决办法 |
|
||||
| :--- | :--- | :--- |
|
||||
| `ModuleNotFoundError: fused_moe_i8_tn_pybind` | pybind 模块未编译或未加入 `PYTHONPATH` | 回到步骤 3，确认 `.so` 已生成；执行 `export PYTHONPATH=/root/Project/fused_moe:$PYTHONPATH` |
|
||||
| `FAILED: max abs diff too large` | 数值误差超过阈值 | 检查 scale 是否应用位置错误；确认 TopK 索引与权重是否一致 |
|
||||
| `FAILED: shape mismatch` | 输出张量形状不一致 | 检查 Token Permute / Unpermute 逻辑；确认 expert 维度对齐 |
|
||||
| `FAILED: NaN or Inf detected` | 溢出或未初始化内存 | 检查 INT8 乘加是否溢出；确认 GEMM 输出是否反量化 |
|
||||
| 终端长时间无输出 | Kernel 死锁或 Launch 失败 | 减小测试 shape；检查是否触发 MACA 硬件限制 |
|
||||
|
||||
### 步骤5：性能测试
|
||||
|
||||
**目标：**输出 benchmark 结果对比表
|
||||
|
||||
**操作：**运行 `fused_moe/scripts/run_fused_moe_i8_tn_benchmark.sh` 脚本
|
||||
|
||||
**命令示例：**
|
||||
|
||||
```apl
|
||||
bash scripts/run_fused_moe_i8_tn_benchmark.sh --backend all --warmup 5 --iters 20
|
||||
# --backend：选择计算方式
|
||||
# --warmup：设置预热次数
|
||||
# --iters：设置迭代次数
|
||||
```
|
||||
|
||||
|
||||
**预期结果：**
|
||||
|
||||
编译成功无报错，输出示例如下：
|
||||
|
||||
> pybind:fused\_moe\_i8\_tn\_topk1 benchmark: avg\_ms=0.308978, TOPS=0.027149, warmup=5, iters=20
|
||||
> pybind:fused\_moe\_i8\_tn\_topk2 benchmark: avg\_ms=0.304500, TOPS=0.055098, warmup=5, iters=20
|
||||
> pybind:fused\_moe\_i8\_tn\_topk3 benchmark: avg\_ms=0.297775, TOPS=0.042256, warmup=5, iters=20
|
||||
|
||||
> reference:fused\_moe\_i8\_tn\_topk1 benchmark: avg\_ms=1685.43, TOPS=0.000005, warmup=5, iters=20
|
||||
> reference:fused\_moe\_i8\_tn\_topk2 benchmark: avg\_ms=3384.52, TOPS=0.000005, warmup=5, iters=20
|
||||
> reference:fused\_moe\_i8\_tn\_topk3 benchmark: avg\_ms=2532.14, TOPS=0.000005, warmup=5, iters=20
|
||||
|
||||
> triton:fused\_moe\_i8\_tn\_topk1 benchmark: avg\_ms=19.013421, TOPS=0.000441, warmup=5, iters=20
|
||||
> triton:fused\_moe\_i8\_tn\_topk2 benchmark: avg\_ms=16.745914, TOPS=0.001002, warmup=5, iters=20
|
||||
> triton:fused\_moe\_i8\_tn\_topk3 benchmark: avg\_ms=19.630328, TOPS=0.000641, warmup=5, iters=20
|
||||
|
||||
|
||||
**结果解释：**
|
||||
|
||||
* “pybind/reference/Triton”：三种计算方式；
|
||||
|
||||
* “fused\_moe\_i8\_tn\_topk1/2/3”：分别对应选择前 1 / 2 / 3 个专家场景下的 MoE 算子；
|
||||
|
||||
* “avg\_ms”：平均算子执行耗时（毫秒），这里不计算预热时间，只计算正式迭代的<E4BBA3><E79A84><EFBFBD>间<EFBFBD><E997B4><EFBFBD>
|
||||
|
||||
* “TOPS”：Tera Operations Per Second，本次 MoE 算子的总运算量 / 实际耗<E99985><E88097><EFBFBD>；
|
||||
|
||||
* “warmup=5, iters=20”：预热轮数和正式迭代数。
|
||||
|
||||
|
||||
**常见问题：**
|
||||
|
||||
|
||||
| 报错 | 原因 | 解决办法 |
|
||||
| --- | --- | --- |
|
||||
| `ModuleNotFoundError: fused_moe_i8_tn_pybind` | pybind 模块未编译或未加入 `PYTHONPATH` | 回到步骤 3，确认 `.so`已生成；执行 `export PYTHONPATH=/root/Project/fused_moe:$PYTHONPATH` |
|
||||
| 终端长时间无输出 | Kernel 死锁或 MACA 驱动异常 | 减小测试 shape；重启容器或设备 |
|
||||
| avg\_ms 异常抖动（±50%） | 其他进程占用 GPU | 关闭其他占用显存的进程，单机单任务运行 |
|
||||
|
||||
## 六、项目实践2-Kernel Swift 智能算子迁移系统自动调优
|
||||
|
||||
系统链接：[https://deeplink.org.cn/kernelswift/task](https://deeplink.org.cn/kernelswift/task)
|
||||
|
||||
**项目目标：**基于 KernelSwift 智能算子迁移系统，对 Fused MoE 算子进行在线自动调优。通过输入算子代码，一键生成适配沐曦硬件的高性能实现，高效完成算子优化与全流程追踪。
|
||||
|
||||
### 步骤1：复用算子广场的 fused\_moe 算子进行二次优化
|
||||
|
||||
**目标：**通过提交算子广场的 fused\_moe 算子代码发起自动优化流程，实现二次优化
|
||||
|
||||
**操作：**
|
||||
|
||||
1. 进入算子广场：点击左侧导航栏 【算子广场】，进入算子列表页
|
||||
|
||||
搜索 fused\_moe 算子，复制 `input_code.py` 代码，也可直接复制以下代码：
|
||||
|
||||
```python
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
import torch.nn.functional as F
|
||||
|
||||
|
||||
class Model(nn.Module):
|
||||
"""
|
||||
Reference PyTorch MoE forward (no fused kernels).
|
||||
Expects inputs:
|
||||
hidden_states: (M, in_size)
|
||||
w1: (E, hidden_size, in_size) where hidden_size = 2 * up_dim
|
||||
w2: (E, out_size, up_dim)
|
||||
topk_weights: (M, top_k)
|
||||
topk_idx: (M, top_k)
|
||||
top_k: int
|
||||
renormalize: bool
|
||||
"""
|
||||
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
|
||||
def forward(
|
||||
self,
|
||||
hidden_states: torch.Tensor,
|
||||
w1: torch.Tensor,
|
||||
w2: torch.Tensor,
|
||||
topk_weights: torch.Tensor,
|
||||
topk_idx: torch.Tensor,
|
||||
top_k: int,
|
||||
renormalize: bool = True,
|
||||
) -> torch.Tensor:
|
||||
if renormalize:
|
||||
topk_weights = topk_weights / topk_weights.sum(dim=-1, keepdim=True)
|
||||
|
||||
seq_len = hidden_states.size(0)
|
||||
out_size = w2.size(1)
|
||||
output = hidden_states.new_zeros(seq_len, out_size)
|
||||
num_experts = w1.size(0)
|
||||
|
||||
# Accumulate expert contributions
|
||||
for eid in range(num_experts):
|
||||
token_idx, k_idx = torch.where(topk_idx == eid)
|
||||
if token_idx.numel() == 0:
|
||||
continue
|
||||
gate_proj, up_proj = w1[eid].chunk(2, dim=0)
|
||||
down_proj = w2[eid]
|
||||
tmp = F.linear(hidden_states[token_idx], gate_proj)
|
||||
tmp = F.silu(tmp) * F.linear(hidden_states[token_idx], up_proj)
|
||||
tmp = F.linear(tmp, down_proj)
|
||||
tmp = tmp * topk_weights[token_idx, k_idx, None]
|
||||
output.index_add_(0, token_idx, tmp.to(output.dtype))
|
||||
return output
|
||||
|
||||
|
||||
# Hyperparameters
|
||||
seq_len = 128
|
||||
in_size = 128
|
||||
hidden_size = 256 # 2 * up_dim
|
||||
out_size = 128
|
||||
num_experts = 32
|
||||
top_k = 4
|
||||
|
||||
dtype = torch.float16
|
||||
|
||||
def get_inputs():
|
||||
hidden_states = (torch.rand(seq_len, in_size, dtype=dtype) - 0.5) / 2
|
||||
w1 = (torch.rand(num_experts, hidden_size, in_size, dtype=dtype) - 0.5) / 2
|
||||
w2 = (torch.rand(num_experts, out_size, hidden_size//2, dtype=dtype) - 0.5) / 2
|
||||
routing_logits = (torch.rand(seq_len, num_experts, dtype=dtype) - 0.5) / 2
|
||||
routing_weights = torch.softmax(routing_logits, dim=-1, dtype=torch.float32)
|
||||
topk_weights, topk_idx = torch.topk(routing_weights, top_k, dim=-1)
|
||||
return [hidden_states, w1, w2, topk_weights, topk_idx, top_k, True]
|
||||
|
||||
def get_init_inputs():
|
||||
return []
|
||||
```
|
||||
|
||||
2. 进入新建任务页：点击左侧导航栏【新建任务】 ，进入算子提交页面。
|
||||
|
||||
3. 编写算子代码：在 `model.py` 编辑器中输入刚刚复制的 fused\_moe 算子代码。
|
||||
|
||||
如果想自行编写算子代码，需严格遵循标准格式规范：输入代码必须包含 `class Model` 定义算子实现，`get_init_inputs` 和 `get_inputs` 定义测试用例，确保优化过程可验证算子正确性。
|
||||
|
||||
4. 配置优化参数
|
||||
|
||||
* 指定任务名称：支持字母、下划线、数字组合，示例：fused\_moe\_01
|
||||
|
||||
* 选择适配硬件：算子需要适配的目标硬件厂商及型号，建议：沐曦
|
||||
|
||||
* 最大演化轮次：优化算法迭代次数，取值范围40-400，建议默认40，复杂算法可提高至100+
|
||||
|
||||
5. 提交优化任务：点击右下角 \[优化\] 按钮，系统将提交任务并进入 \[生成中\] 状态
|
||||
|
||||
|
||||
|
||||
|
||||
完成上述步骤将看到如下界面：
|
||||
|
||||
|
||||
|
||||
### 步骤2：任务查看与结果管理
|
||||
|
||||
**目标：** 在新建优化任务后可追踪任务进度，获取优化结果
|
||||
|
||||
**操作：**
|
||||
|
||||
1. 查看任务列表：点击左侧【任务查看】，可看到所有提交的优化任务
|
||||
|
||||
* 任务状态：排队中、环境初始化、算子预编译、精度验证、性能调优、已完成、失败
|
||||
|
||||
* 任务信息：任务名称、进度、创建时间、适配硬件
|
||||
|
||||
* 操作按钮：查看详情、删除任务
|
||||
|
||||
|
||||
|
||||
|
||||
2. 追踪任务进度：当前任务状态为【运行中】时，点击任务列表中的【查看详情】按钮，追踪任务进度：
|
||||
|
||||
* 左侧：原始算子代码（输入的 `input_code.py`）
|
||||
|
||||
* 右侧：任务进度条，包含以下阶段：
|
||||
|
||||
1. 环境初始化：准备目标硬件编译环境
|
||||
|
||||
2. 算子预编译：验证算子代码是可正常编译
|
||||
|
||||
3. 精度验证：验证优化后算子输出与原始算子误差的可接受范围
|
||||
|
||||
4. 性能调优：按设定的演化轮次迭代优化算子性能
|
||||
|
||||
|
||||
* 顶部：任务名称、创建/更新时间、适配硬件、当前轮次进度
|
||||
|
||||
|
||||
|
||||
|
||||
3. 获取优化结果：当前任务状态为【已完成】时，可在详情页查看优化结果：
|
||||
|
||||
* 优化后算子代码支持一键复制
|
||||
|
||||
* 算子加速比（基准耗时 / 优化后耗时）、性能数据（如延迟、吞吐量）
|
||||
|
||||
* 可点击【Diff 对比】查看优化前后代码差异，理解性能提升逻辑
|
||||
|
||||
|
||||
|
||||
|
||||
4. 任务异常处理
|
||||
|
||||
* 任务失败：查看错误日志，常见原因包括代码不符合规范、测试用例错误、硬件适配问题，修改后重新提交任务；
|
||||
|
||||
* 排队时间长：可调整提交时间，或联系平台管理员确认资源状态。
|
||||
|
||||
|
||||
## 七、Agent使用说明
|
||||
|
||||
在本模块中，Agent可以帮助你完成以下任务：
|
||||
|
||||
1. **环境检查**
|
||||
|
||||
```plaintext
|
||||
我正在算力平台部署 fused_moe_baseline 源码。
|
||||
需要的环境信息如下：
|
||||
- Python 3.12
|
||||
- g++ 13.3.0
|
||||
- mxcc 已安装
|
||||
- numpy / torch / triton 已安装
|
||||
|
||||
请帮我确认：
|
||||
1. 当前环境是否满足编译与运行要求？
|
||||
2. 是否有潜在的不兼容风险（如 Python 与 libpython 版本）？
|
||||
```
|
||||
|
||||
2. **运行测试**
|
||||
|
||||
```plaintext
|
||||
请帮我运行 scripts/run_fused_moe_i8_tn_pybind_test.sh 脚本
|
||||
```
|
||||
|
||||
3. **分析结果**
|
||||
|
||||
```plaintext
|
||||
这是性能测试结果：
|
||||
pybind: avg_ms=0.30, TOPS=0.027
|
||||
triton: avg_ms=19.01, TOPS=0.0004
|
||||
reference: avg_ms=1685, TOPS=0.000005
|
||||
|
||||
请分析：
|
||||
1. 为什么 pybind 比 Triton 快这么多？
|
||||
2. TOPS 指标是否可信？
|
||||
3. 当前结果是否已经具备提交价值？
|
||||
```
|
||||
|
||||
4. **报错检查**
|
||||
|
||||
```plaintext
|
||||
编译 pybind 时出现以下错误：
|
||||
/usr/bin/ld: cannot find -lpython3.10
|
||||
|
||||
已知：
|
||||
- 使用的是 Conda Python 3.10
|
||||
- mxcc 编译正常
|
||||
|
||||
请一步一步告诉我：
|
||||
1. 错误原因是什么？
|
||||
2. 如何用 find 命令定位 libpython3.10.so？
|
||||
3. 如何在 build_fused_moe_i8_tn_pybind.sh 中正确指定路径？
|
||||
```
|
||||
|
||||
5. **代码理解**
|
||||
|
||||
```plaintext
|
||||
请帮我梳理释 benchmark_fused_moe_i8_tn.py 代码整体框架
|
||||
```
|
||||
|
||||
6. **KernelSwift 系统搜索算子**
|
||||
|
||||
|
||||
```plaintext
|
||||
请帮我在算子广场检索 fused_moe 算子
|
||||
```
|
||||
|
||||
## 八、常见问题与注意事项
|
||||
|
||||
### 算力平台 basline 源码部署项目：
|
||||
|
||||
1. 环境准备与依赖问题
|
||||
|
||||
* 确保算力平台已正确安装 Python 和 C++、MACA 编译器及相关运行时库，避免因环境缺失导致编译失败；
|
||||
|
||||
* 镜像环境使用 Conda Python 作为默认运行环境，避免系统 Python 与 Conda Python 混用，防止 `Python.h`或 `libpython`路径错误。
|
||||
|
||||
2. pybind 编译与链接
|
||||
|
||||
* 若`Python.h not found`，请检查脚本中`PYTHON_INCLUDE`是否指向当前 Python 的 `include`目录；
|
||||
|
||||
* 若`libpython not found`，请直接指定 Conda 下的`**libpython3.x.so**`绝对路径，避免链接系统静态库；
|
||||
|
||||
* 编译 `pybind`模块时，务必开启 `-fPIC`，否则会出现 `recompile with -fPIC`错误。
|
||||
|
||||
3. 性能测试建议
|
||||
|
||||
* benchmark 应在关闭其他占用 GPU 的任务后执行，避免干扰性能数据；
|
||||
|
||||
* 多次运行取平均值，避免单次抖动影响结果；
|
||||
|
||||
* 性能对比应基于相同随机种子、相同 shape、相同 TopK、相同 batch size的条件下进行，降低误差。
|
||||
|
||||
|
||||
### Kernel Swift 智能算子迁移系统自动调优项目：
|
||||
|
||||
1. 代码规范问题
|
||||
|
||||
输入代码需符合以下标准格式：
|
||||
|
||||
* `class Model`，表示待优化的算子实现；
|
||||
|
||||
* `def get_init_inputs`，表示 module init 的输入测试样例；
|
||||
|
||||
* `def get_inputs`，表示 module forward 的输入测试样例。
|
||||
|
||||
2. 性能优化建议
|
||||
|
||||
* 对于复杂算子，可适当提高最大演化轮次（如 100-200），获得更高加速比；
|
||||
|
||||
* 优先选择算子广场中已有优化案例的算子类型，降低适配失败概率。
|
||||
|
||||
3. 硬件适配问题
|
||||
|
||||
* 提交任务前确认目标硬件支持的算子类型；
|
||||
|
||||
* 优化失败时，可尝试更换适配硬件，或调整算子实现逻辑。
|
||||
|
||||
|
||||
## 九、下一步学习建议
|
||||
|
||||
完成本模块后，建议继续学习以下内容：
|
||||
|
||||
1. **研读 fused\_moe源码：**理解代码的底层逻辑，可尝试修改 `build_fused_moe_i8_tn_pybind.sh`中的编译参数，观察其对 `avg_ms` 的影响；
|
||||
|
||||
2. **算子优化基础：**了解如何分析 Kernel 性能瓶颈；
|
||||
|
||||
3. **性能对比分析：**将 baseline 结果与优化后的结果进行对比分析，明确后续优化方向。
|
||||
Binary file not shown.
Binary file not shown.
Binary file not shown.
|
|
|
@ -0,0 +1,49 @@
|
|||
batch_size,seq_len_kv,heads,headdim,time_ms,bandwidth_GB_s
|
||||
1,512,8,128,0.0322,65.27
|
||||
2,512,8,128,0.0324,129.61
|
||||
4,512,8,128,0.0332,253.01
|
||||
8,512,8,128,0.0355,472.66
|
||||
16,512,8,128,0.0546,614.98
|
||||
32,512,8,128,0.0817,822.60
|
||||
64,512,8,128,0.1297,1035.62
|
||||
128,512,8,128,0.2453,1095.50
|
||||
1,1024,8,128,0.0578,72.55
|
||||
2,1024,8,128,0.0586,143.14
|
||||
4,1024,8,128,0.0597,281.24
|
||||
8,1024,8,128,0.0625,536.80
|
||||
16,1024,8,128,0.0982,683.59
|
||||
32,1024,8,128,0.1493,899.64
|
||||
64,1024,8,128,0.2403,1117.62
|
||||
128,1024,8,128,0.4594,1169.27
|
||||
1,2048,8,128,0.1101,76.24
|
||||
2,2048,8,128,0.1107,151.64
|
||||
4,2048,8,128,0.1119,299.88
|
||||
8,2048,8,128,0.1159,578.98
|
||||
16,2048,8,128,0.1849,726.23
|
||||
32,2048,8,128,0.2843,944.47
|
||||
64,2048,8,128,0.4607,1165.56
|
||||
128,2048,8,128,0.8868,1211.07
|
||||
1,4096,8,128,0.2139,78.46
|
||||
2,4096,8,128,0.2151,156.01
|
||||
4,4096,8,128,0.2163,310.36
|
||||
8,4096,8,128,0.2227,602.81
|
||||
16,4096,8,128,0.3574,751.13
|
||||
32,4096,8,128,0.5540,969.23
|
||||
64,4096,8,128,0.9016,1191.07
|
||||
128,4096,8,128,1.7414,1233.34
|
||||
1,8192,8,128,0.4215,79.61
|
||||
2,8192,8,128,0.4226,158.81
|
||||
4,8192,8,128,0.4242,316.39
|
||||
8,8192,8,128,0.4362,615.46
|
||||
16,8192,8,128,0.7035,763.14
|
||||
32,8192,8,128,1.0934,982.11
|
||||
64,8192,8,128,1.7814,1205.57
|
||||
128,8192,8,128,3.4505,1244.82
|
||||
1,16384,8,128,0.8356,80.32
|
||||
2,16384,8,128,0.8377,160.23
|
||||
4,16384,8,128,0.8407,319.30
|
||||
8,16384,8,128,0.8625,622.51
|
||||
16,16384,8,128,1.3934,770.60
|
||||
32,16384,8,128,2.1695,989.88
|
||||
64,16384,8,128,3.5397,1213.41
|
||||
128,16384,8,128,6.8668,1250.98
|
||||
|
|
|
@ -0,0 +1,145 @@
|
|||
api,batch_size,seq_len_q,seq_len_kv,num_qo_heads,num_kv_heads,head_dim,time_ms,bandwidth_GB_s,tflops
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,512,32,8,64,0.02042879999999998,51.528822055137894,0.8212531328320811
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,512,32,4,128,0.02333952000000001,45.27805199078642,0.718832949435121
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,512,32,4,256,0.0319488,66.15384615384615,1.0502564102564103
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,1024,32,8,64,0.023262719999999973,90.32684054143292,1.4424122372620245
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,1024,32,4,128,0.025041919999999992,84.07278675117566,1.3399304845634845
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,1024,32,4,256,0.033387520000000004,126.11562643766291,2.0099984664928687
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,2048,32,8,64,0.028298240000000037,148.36258368011562,2.371485435136599
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,2048,32,4,128,0.027745280000000008,151.4670603432367,2.418748846650673
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,2048,32,4,256,0.03723775999999999,225.7115358174069,3.604344837068611
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,4096,32,8,64,0.03886591999999997,215.93992886312756,3.4533526544592306
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,4096,32,4,128,0.03426815999999998,245.03212311370103,3.916689078141344
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,4096,32,4,256,0.066048,254.26356589147287,4.064248062015504
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,8192,32,8,64,0.052495359999999984,319.6722910367698,5.1135082414902975
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,8192,32,4,128,0.04628480000000001,362.6548672566371,5.799646017699114
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,8192,32,4,256,0.08975359999999999,374.0330861380491,5.981608670849972
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,16384,32,8,64,0.08625152000000001,389.0775258221536,6.224480588863825
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,16384,32,4,128,0.0638464,525.6776263031276,8.408789093825181
|
||||
BatchDecodeWithPagedKVCacheWrapper,1,1,16384,32,4,256,0.13059071999999994,514.0123892417473,8.222190857053247
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,512,32,8,64,0.02342912,89.86013986013987,1.4321678321678322
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,512,32,4,128,0.02486784,84.99073502161829,1.3493102738315832
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,512,32,4,256,0.03340287999999998,126.54812998160644,2.009074187614961
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,1024,32,8,64,0.02839040000000001,148.02524797114512,2.3637871956717755
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,1024,32,4,128,0.028165120000000012,149.5000908925649,2.382694055626249
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,1024,32,4,256,0.03740160000000001,225.16084873374396,3.5885557837097872
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,2048,32,8,64,0.03881984000000001,216.30176734370872,3.457451859667633
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,2048,32,4,128,0.03601408000000001,233.38072220642587,3.7268126243957904
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,2048,32,4,256,0.06728704000000002,249.82498858621207,3.9894080048698815
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,4096,32,8,64,0.052490240000000014,319.7815060476004,5.114007023019897
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,4096,32,4,128,0.04626431999999999,362.9924745462595,5.802213368747235
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,4096,32,4,256,0.08993791999999999,373.44870773084375,5.969349880450872
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,8192,32,8,64,0.08536063999999999,393.18618042226495,6.289443378119003
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,8192,32,4,128,0.0630784,532.2077922077922,8.51116883116883
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,8192,32,4,256,0.12952576,518.3650881492608,8.289793659577834
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,16384,32,8,64,0.15207424000000003,441.34401723789637,7.0606423809844445
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,16384,32,4,128,0.10330112,649.8017446471055,10.394290245836638
|
||||
BatchDecodeWithPagedKVCacheWrapper,2,1,16384,32,4,256,0.2281984,588.3060354498541,9.410599057662106
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,512,32,8,64,0.0283904,148.3137962128043,2.3637871956717764
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,512,32,4,128,0.028078080000000036,150.5470459518598,2.3900802334062696
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,512,32,4,256,0.03707903999999999,228.00331400165706,3.619773543220106
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,1024,32,8,64,0.03844096000000004,218.64677677144357,3.4915290356952546
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,1024,32,4,128,0.03641856000000004,231.23857725291697,3.6854210600309254
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,1024,32,4,256,0.06640640000000002,253.63145720894363,4.04231303006939
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,2048,32,8,64,0.059007999999999984,284.5986984815619,4.5491366594360105
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,2048,32,4,128,0.04641792000000003,362.1442753143611,5.783013456871825
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,2048,32,4,256,0.08961023999999998,375.1799794309223,5.991178151068451
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,4096,32,8,64,0.09185279999999997,365.484949832776,5.84490523968785
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,4096,32,4,128,0.06349823999999998,528.9469440412838,8.454894371875506
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,4096,32,4,256,0.1303347200000001,515.3991200502825,8.238340666247638
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,8192,32,8,64,0.16568319999999992,405.1421508034613,6.480692212608161
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,8192,32,4,128,0.10290176,652.4828341128471,10.43463031147378
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,8192,32,4,256,0.22947840000000008,585.1673360107093,9.358107987505575
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,16384,32,8,64,0.30601215999999987,438.65613706331163,7.017641547316293
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,16384,32,4,128,0.18384895999999992,730.2216776205863,11.680695109724857
|
||||
BatchDecodeWithPagedKVCacheWrapper,4,1,16384,32,4,256,0.4362026666666668,615.5418398787107,9.846265564630505
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,512,32,8,64,0.038655999999999975,217.85430463576174,3.472105960264903
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,512,32,4,128,0.03645951999999999,231.87754528858312,3.6812807190001418
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,512,32,4,256,0.06676480000000001,253.25153374233125,4.020613496932515
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,1024,32,8,64,0.05858303999999996,286.9428421604617,4.582135990211505
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,1024,32,4,128,0.04676608000000001,360.14889424129615,5.73996058681848
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,1024,32,4,256,0.08992768000000002,374.58437713504884,5.970029606012297
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,2048,32,8,64,0.092416,363.43490304709144,5.8092853185595565
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,2048,32,4,128,0.07130112000000002,471.5208961654457,7.5296280338934345
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,2048,32,4,256,0.14862335999999993,452.41835469202175,7.224583161085851
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,4096,32,8,64,0.16396288000000003,409.4928803397451,6.548688483637271
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,4096,32,4,128,0.11201536000000002,599.6891854831337,9.585665965810401
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,4096,32,4,256,0.24935424000000006,538.7869081351894,8.61218019793848
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,8192,32,8,64,0.3056947200000001,439.16524302415155,7.024928817874248
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,8192,32,4,128,0.20128768000000002,667.1211273337741,10.668728697156228
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,8192,32,4,256,0.46690133333333317,575.2104541716168,9.198875628255509
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,16384,32,8,64,0.5866495999999998,457.6296037702917,7.321179961598886
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,16384,32,4,128,0.37337600000000015,719.1169009256082,11.503062050051419
|
||||
BatchDecodeWithPagedKVCacheWrapper,8,1,16384,32,4,256,0.8934826666666666,601.0211546726517,9.613991308915525
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,512,32,8,64,0.0698112,241.26145947928126,3.845163182984965
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,512,32,4,128,0.04724735999999999,357.8673602080625,5.681491114000869
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,512,32,4,256,0.08954879999999998,377.6329331046313,5.995288736420813
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,1024,32,8,64,0.12070911999999998,278.52052935188334,4.447641669494402
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,1024,32,4,128,0.07076864000000004,475.9947909130369,7.586282737664589
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,1024,32,4,256,0.14710784000000002,457.9702074342197,7.2990115550605585
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,2048,32,8,64,0.22239232000000014,302.05359609540454,4.8281425545630325
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,2048,32,4,128,0.11209728000000002,599.8355713894217,9.578660820316067
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,2048,32,4,256,0.2504192,537.0190145164587,8.575555101206296
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,4096,32,8,64,0.42098688000000006,318.97256275539985,5.101070247129791
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,4096,32,4,128,0.2027008,662.7936347562515,10.594352109118466
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,4096,32,4,256,0.46432,578.6905582356995,9.250015713301172
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,8192,32,8,64,0.8234496000000004,326.06851955480926,5.215822918609709
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,8192,32,4,128,0.3726506666666667,720.6924662239522,11.525451797572705
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,8192,32,4,256,0.8939733333333334,600.8378952392316,9.608714568667223
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,16384,32,8,64,1.6324906666666663,328.9062896122735,5.261858317107276
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,16384,32,4,128,0.7114879999999999,754.7590177206082,12.073196725735361
|
||||
BatchDecodeWithPagedKVCacheWrapper,16,1,16384,32,4,256,1.742272,616.4387466480549,9.860612570253094
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,512,32,8,64,0.08406016,400.730905104154,6.386746254111341
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,512,32,4,128,0.08498175999999999,397.92746113989637,6.317484034220991
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,512,32,4,256,0.1808896,373.8918765921313,5.935895839230116
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,1024,32,8,64,0.14712832,457.0155902004454,7.297995545657015
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,1024,32,4,128,0.14887935999999996,452.5208061077104,7.212160396175805
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,1024,32,4,256,0.3279462400000001,410.8661712358707,6.548279522887651
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,2048,32,8,64,0.27223039999999993,493.51137859695325,7.888478465299983
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,2048,32,4,128,0.27833343999999993,483.1610316029581,7.715507155733786
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,2048,32,4,256,0.6300373333333331,426.89493109403054,6.817004435715981
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,4096,32,8,64,0.52494336,511.6104868913858,8.181772784019977
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,4096,32,4,128,0.5080533333333332,528.8767583455806,8.453772496325847
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,4096,32,4,256,1.2449493333333332,431.66029782202656,6.899826653186422
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,8192,32,8,64,1.0273706666666667,522.6954607749491,8.361086091615102
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,8192,32,4,128,1.0078719999999999,532.9377698755399,8.522842773685548
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,8192,32,4,256,2.446784,439.05228741073995,7.021408176610604
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,16384,32,8,64,2.0322986666666663,528.4030903594223,8.453417534430637
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,16384,32,4,128,2.018026666666667,532.2050425498176,8.513202262276018
|
||||
BatchDecodeWithPagedKVCacheWrapper,32,1,16384,32,4,256,4.847957333333333,443.0748433429558,7.087467154826447
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,512,32,8,64,0.13077504,515.1671756322919,8.210602145485865
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,512,32,4,128,0.14377984000000002,470.39384659212305,7.4679581226408365
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,512,32,4,256,0.27039743999999993,500.2499431947286,7.9419525865333656
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,1024,32,8,64,0.231424,581.0973451327434,9.279433628318584
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,1024,32,4,128,0.25729023999999995,523.6965692907746,8.34654143118682
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,1024,32,4,256,0.502016,536.8036715961244,8.555439061703213
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,2048,32,8,64,0.4335923199999999,619.7010131544766,9.905542828802874
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,2048,32,4,128,0.47517866666666664,566.018137739068,9.038636616683128
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,2048,32,4,256,0.9693866666666666,554.9070422535212,8.861205633802816
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,4096,32,8,64,0.8388479999999999,640.3222705424583,10.240156252384224
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,4096,32,4,128,0.9261013333333336,580.2768883462716,9.275372232844207
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,4096,32,4,256,1.906474666666667,563.7580287805205,9.011328335161021
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,8192,32,8,64,1.6543999999999999,649.1803481624759,10.384350328820116
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,8192,32,4,128,1.8147413333333327,591.9665201137942,9.466841840453299
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,8192,32,4,256,3.774634666666667,569.2026947596871,9.10279839037844
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,16384,32,8,64,3.2680746666666667,657.1899393567508,10.513755612274872
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,16384,32,4,128,3.5912106666666666,598.129192458031,9.567731207451676
|
||||
BatchDecodeWithPagedKVCacheWrapper,64,1,16384,32,4,256,7.526272,570.802632697835,9.130612969608327
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,512,32,8,64,0.2176000000000001,619.2188235294115,9.86895058823529
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,512,32,4,128,0.21536768,628.0715100798782,9.971243818942565
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,512,32,4,256,0.45757866666666663,591.2264441232692,9.386292694298103
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,1024,32,8,64,0.39856127999999985,674.826576229382,10.776177997019683
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,1024,32,4,128,0.39381333333333335,684.2938244853738,10.906099241603465
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,1024,32,4,256,0.8577493333333336,628.3514810853829,10.014504539010616
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,2048,32,8,64,0.7606186666666664,706.5238121949853,11.293352330734283
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,2048,32,4,128,0.7354026666666665,731.4625203063357,11.680586679043863
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,2048,32,4,256,1.6673066666666665,645.2556074467406,10.30396478792144
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,4096,32,8,64,1.4816639999999999,725.0403006349618,11.594983197270098
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,4096,32,4,128,1.4354773333333333,748.7338009749138,11.968053263583403
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,4096,32,4,256,3.2697173333333325,657.4209880731792,10.508473627893627
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,8192,32,8,64,2.9226666666666676,734.9479708029195,11.75629734306569
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,8192,32,4,128,2.825301333333333,760.4612643087983,12.161442024827087
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,8192,32,4,256,6.484309333333334,662.6865294520187,10.5978097594357
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,16384,32,8,64,5.794901333333332,741.2536188134122,11.858610316747416
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,16384,32,4,128,5.61536,765.0472760428539,12.237768680191476
|
||||
BatchDecodeWithPagedKVCacheWrapper,128,1,16384,32,4,256,12.908458666666668,665.6125232199165,10.647200957222468
|
||||
|
|
|
@ -0,0 +1,33 @@
|
|||
api,batch_size,seq_len,num_heads,head_dim_ckv,head_dim_kpe,time_ms,bandwidth_GB_s,tflops
|
||||
BatchMLAPagedAttentionWrapper,1,1024,64,512,64,0.035975679999999996,34.83953604212624,3.963964989681919
|
||||
BatchMLAPagedAttentionWrapper,1,4096,64,512,64,0.05349631999999998,89.58223668469162,10.662889409963158
|
||||
BatchMLAPagedAttentionWrapper,1,8192,64,512,64,0.06174719999999999,154.0298507462687,18.47615257048093
|
||||
BatchMLAPagedAttentionWrapper,1,16384,64,512,64,0.08995584000000004,210.63775292410136,25.3646831156265
|
||||
BatchMLAPagedAttentionWrapper,4,1024,64,512,64,0.05086207999999998,98.5705657338434,11.215139923495071
|
||||
BatchMLAPagedAttentionWrapper,4,4096,64,512,64,0.08034559999999999,238.58531145451653,28.39858531145452
|
||||
BatchMLAPagedAttentionWrapper,4,8192,64,512,64,0.10866687999999997,350.0942329438373,41.99442140972485
|
||||
BatchMLAPagedAttentionWrapper,4,16384,64,512,64,0.16821760000000002,450.56155836250184,54.2559488662304
|
||||
BatchMLAPagedAttentionWrapper,16,1024,64,512,64,0.06735359999999997,297.7423033067276,33.87645762067656
|
||||
BatchMLAPagedAttentionWrapper,16,4096,64,512,64,0.14288383999999996,536.6395528003728,63.87570143691549
|
||||
BatchMLAPagedAttentionWrapper,16,8192,64,512,64,0.21618431999999987,703.9113289992544,84.43540682321462
|
||||
BatchMLAPagedAttentionWrapper,16,16384,64,512,64,0.39363328000000025,770.1826837405613,92.74424666532254
|
||||
BatchMLAPagedAttentionWrapper,64,1024,64,512,64,0.15278592,525.0226198853926,59.73590697362689
|
||||
BatchMLAPagedAttentionWrapper,64,4096,64,512,64,0.4850483199999999,632.3256206721838,75.26512413443676
|
||||
BatchMLAPagedAttentionWrapper,64,8192,64,512,64,0.9133465600000001,666.4484158127227,79.94166423750474
|
||||
BatchMLAPagedAttentionWrapper,64,16384,64,512,64,1.7720038399999998,684.3541287134007,82.40890045926764
|
||||
BatchMLAPagedAttentionWrapper,1,1024,128,512,64,0.04499968000000001,29.491409716691315,6.338104448742746
|
||||
BatchMLAPagedAttentionWrapper,1,4096,128,512,64,0.05375743999999999,90.51859612362495,21.222191532930147
|
||||
BatchMLAPagedAttentionWrapper,1,8192,128,512,64,0.08302080000000002,115.44865864939868,27.48349059512796
|
||||
BatchMLAPagedAttentionWrapper,1,16384,128,512,64,0.11321343999999998,168.01736613603475,40.30795947901592
|
||||
BatchMLAPagedAttentionWrapper,4,1024,128,512,64,0.05178880000000003,102.50123578843295,22.028907563025196
|
||||
BatchMLAPagedAttentionWrapper,4,4096,128,512,64,0.11032576,176.4247261926861,41.36298496380175
|
||||
BatchMLAPagedAttentionWrapper,4,8192,128,512,64,0.1688268800000001,227.08800873415404,54.06014435615937
|
||||
BatchMLAPagedAttentionWrapper,4,16384,128,512,64,0.30781695999999986,247.18357299091002,59.30021207408457
|
||||
BatchMLAPagedAttentionWrapper,16,1024,128,512,64,0.10527487999999995,201.69734698344004,43.34749896651511
|
||||
BatchMLAPagedAttentionWrapper,16,4096,128,512,64,0.2629478400000002,296.0920614521874,69.41913273750409
|
||||
BatchMLAPagedAttentionWrapper,16,8192,128,512,64,0.3962367999999998,387.02674764181444,92.13485980100793
|
||||
BatchMLAPagedAttentionWrapper,16,16384,128,512,64,0.7528985599999998,404.23663979381246,96.97779742333418
|
||||
BatchMLAPagedAttentionWrapper,64,1024,128,512,64,0.3242547199999998,261.9380714026308,56.29404872811108
|
||||
BatchMLAPagedAttentionWrapper,64,4096,128,512,64,1.1793126399999994,264.07507342582215,61.91271216426548
|
||||
BatchMLAPagedAttentionWrapper,64,8192,128,512,64,2.3186406399999986,264.55887532446616,62.98038839860932
|
||||
BatchMLAPagedAttentionWrapper,64,16384,128,512,64,4.6020608,264.53295358462015,63.462389746784744
|
||||
|
|
|
@ -0,0 +1,33 @@
|
|||
api,batch_size,seq_len,num_qo_heads,num_kv_heads,head_dim,time_ms,bandwidth_GB_s,tflops
|
||||
BatchPrefillWithPagedKVCacheWrapper,1,1024,32,4,128,0.3529011200000001,29.71302556364796,24.34091054174041
|
||||
BatchPrefillWithPagedKVCacheWrapper,1,4096,32,4,128,4.62532608,9.068126068205768,29.714435500296666
|
||||
BatchPrefillWithPagedKVCacheWrapper,1,8192,32,4,128,18.113853439999996,4.631045529757804,30.350019983820744
|
||||
BatchPrefillWithPagedKVCacheWrapper,1,16384,32,4,128,71.05519616000001,2.36115258372119,30.948099145350383
|
||||
BatchPrefillWithPagedKVCacheWrapper,4,1024,32,4,128,1.2374374399999997,33.89507917264893,27.766848858234006
|
||||
BatchPrefillWithPagedKVCacheWrapper,4,4096,32,4,128,17.896878079999997,9.374381344614939,30.71797279003423
|
||||
BatchPrefillWithPagedKVCacheWrapper,4,8192,32,4,128,71.25501952,4.709062214288198,30.861310127559136
|
||||
BatchPrefillWithPagedKVCacheWrapper,4,16384,32,4,128,283.27072767999994,2.3690716139159393,31.051895457919
|
||||
BatchPrefillWithPagedKVCacheWrapper,16,1024,32,4,128,4.752537600000002,35.301595509733566,28.919067041573737
|
||||
BatchPrefillWithPagedKVCacheWrapper,16,4096,32,4,128,70.51405312000001,9.517090711803915,31.185602844439067
|
||||
BatchPrefillWithPagedKVCacheWrapper,16,8192,32,4,128,284.16772266666663,4.723186952426669,30.953878011423416
|
||||
BatchPrefillWithPagedKVCacheWrapper,16,16384,32,4,128,1129.139136,2.377346134250013,31.160351250841774
|
||||
BatchPrefillWithPagedKVCacheWrapper,64,1024,32,4,128,18.757478399999997,35.77712449878125,29.3086203894016
|
||||
BatchPrefillWithPagedKVCacheWrapper,64,4096,32,4,128,281.4907093333333,9.536210151864244,31.248253425628754
|
||||
BatchPrefillWithPagedKVCacheWrapper,64,8192,32,4,128,1134.7048106666668,4.731370722616177,31.007511167737377
|
||||
BatchPrefillWithPagedKVCacheWrapper,64,16384,32,4,128,4514.139178666666,2.378619226173592,31.177037921302507
|
||||
BatchPrefillWithPagedKVCacheWrapper,1,1024,32,4,256,0.7928422399999997,26.4510629504301,21.668710768992337
|
||||
BatchPrefillWithPagedKVCacheWrapper,1,4096,32,4,256,12.533002240000002,6.69321511267838,21.932327281224513
|
||||
BatchPrefillWithPagedKVCacheWrapper,1,8192,32,4,256,49.81321727999999,3.368024977325858,22.072688491402744
|
||||
BatchPrefillWithPagedKVCacheWrapper,1,16384,32,4,256,190.01136128,1.765917141688929,23.14622915954513
|
||||
BatchPrefillWithPagedKVCacheWrapper,4,1024,32,4,256,3.111116800000001,26.963333552761494,22.088362846422218
|
||||
BatchPrefillWithPagedKVCacheWrapper,4,4096,32,4,256,47.738091520000026,7.02885912101079,23.032165567728153
|
||||
BatchPrefillWithPagedKVCacheWrapper,4,8192,32,4,256,190.14286336,3.529391680241077,23.130221315627924
|
||||
BatchPrefillWithPagedKVCacheWrapper,4,16384,32,4,256,759.6848640000004,1.76675532658763,23.157215416649382
|
||||
BatchPrefillWithPagedKVCacheWrapper,16,1024,32,4,256,12.28442624,27.31461066593534,22.376129057534232
|
||||
BatchPrefillWithPagedKVCacheWrapper,16,4096,32,4,256,191.34602666666663,7.014398487291994,22.984780963158407
|
||||
BatchPrefillWithPagedKVCacheWrapper,16,8192,32,4,256,759.7649706666668,3.5331380935403933,23.15477380982632
|
||||
BatchPrefillWithPagedKVCacheWrapper,16,16384,32,4,256,3028.668266666667,1.77263029400997,23.234219789647476
|
||||
BatchPrefillWithPagedKVCacheWrapper,64,1024,32,4,256,49.26948266666667,27.241554149868346,22.316281159572153
|
||||
BatchPrefillWithPagedKVCacheWrapper,64,4096,32,4,256,763.6229333333335,7.030576067909256,23.037791659325052
|
||||
BatchPrefillWithPagedKVCacheWrapper,64,8192,32,4,256,3037.7449386666663,3.534667477616765,23.16479678130923
|
||||
BatchPrefillWithPagedKVCacheWrapper,64,16384,32,4,256,12110.653866666667,1.7732185822854112,23.241930601731337
|
||||
|
|
|
@ -0,0 +1,49 @@
|
|||
api,batch_size,seq_len,num_qo_heads,num_kv_heads,head_dim_qk,head_dim_vo,time_ms,bandwidth_GB_s,tflops
|
||||
BatchPrefillWithRaggedKVCacheWrapper,1,1024,32,4,128,128,0.031580159999999996,66.66666666666667,272.00415045395596
|
||||
BatchPrefillWithRaggedKVCacheWrapper,1,4096,32,4,128,128,0.0424448,197.82870928829917,3238.0634016887816
|
||||
BatchPrefillWithRaggedKVCacheWrapper,1,8192,32,4,128,128,0.057313279999999994,292.871180989816,9592.119206717885
|
||||
BatchPrefillWithRaggedKVCacheWrapper,1,16384,32,4,128,128,0.06972416000000001,481.36290204141574,31538.89922161844
|
||||
BatchPrefillWithRaggedKVCacheWrapper,4,1024,32,4,128,128,0.04327423999999998,194.60482725982024,793.9998106956938
|
||||
BatchPrefillWithRaggedKVCacheWrapper,4,4096,32,4,128,128,0.06579199999999998,510.5058365758757,8355.967501945528
|
||||
BatchPrefillWithRaggedKVCacheWrapper,4,8192,32,4,128,128,0.09618432000000002,698.0517406579366,22862.596060896405
|
||||
BatchPrefillWithRaggedKVCacheWrapper,4,16384,32,4,128,128,0.15411199999999997,871.12292358804,57075.97735548174
|
||||
BatchPrefillWithRaggedKVCacheWrapper,16,1024,32,4,128,128,0.07452671999999999,451.99230557845567,1844.1567463588901
|
||||
BatchPrefillWithRaggedKVCacheWrapper,16,4096,32,4,128,128,0.1668906666666667,805.0108653969065,13176.43041083983
|
||||
BatchPrefillWithRaggedKVCacheWrapper,16,8192,32,4,128,128,0.2874026666666667,934.46080760095,30605.46766745843
|
||||
BatchPrefillWithRaggedKVCacheWrapper,16,16384,32,4,128,128,0.5342506666666667,1005.1498622369525,65857.42289917343
|
||||
BatchPrefillWithRaggedKVCacheWrapper,64,1024,32,4,128,128,0.15733333333333333,856.4111186440679,3494.2106814915255
|
||||
BatchPrefillWithRaggedKVCacheWrapper,64,4096,32,4,128,128,0.5614719999999999,957.1184315513509,15666.129428017784
|
||||
BatchPrefillWithRaggedKVCacheWrapper,64,8192,32,4,128,128,1.1031466666666667,973.8198414233224,31894.55505055115
|
||||
BatchPrefillWithRaggedKVCacheWrapper,64,16384,32,4,128,128,2.1813759999999998,984.7032038493136,64517.75776176505
|
||||
BatchPrefillWithRaggedKVCacheWrapper,1,1024,32,4,192,128,0.03564544000000001,73.88681413386956,301.22838264866414
|
||||
BatchPrefillWithRaggedKVCacheWrapper,1,4096,32,4,192,128,0.04922368,213.27231121281466,3490.1635115456625
|
||||
BatchPrefillWithRaggedKVCacheWrapper,1,8192,32,4,192,128,0.061327359999999984,342.16062781766584,11205.353815328106
|
||||
BatchPrefillWithRaggedKVCacheWrapper,1,16384,32,4,192,128,0.08377343999999999,500.8189707859675,32812.059161471705
|
||||
BatchPrefillWithRaggedKVCacheWrapper,4,1024,32,4,192,128,0.049623040000000056,212.29880313660726,865.5187783739157
|
||||
BatchPrefillWithRaggedKVCacheWrapper,4,4096,32,4,192,128,0.08634367999999998,486.3377609108161,7958.831119544594
|
||||
BatchPrefillWithRaggedKVCacheWrapper,4,8192,32,4,192,128,0.13644799999999999,615.1444652908068,20145.249981238278
|
||||
BatchPrefillWithRaggedKVCacheWrapper,4,16384,32,4,192,128,0.2321706666666666,722.8359827253516,47357.904577781876
|
||||
BatchPrefillWithRaggedKVCacheWrapper,16,1024,32,4,192,128,0.09042944,465.99479107688825,1899.8093081191257
|
||||
BatchPrefillWithRaggedKVCacheWrapper,16,4096,32,4,192,128,0.3087573333333334,544.0154770952807,8902.716705589717
|
||||
BatchPrefillWithRaggedKVCacheWrapper,16,8192,32,4,192,128,0.5995946666666665,559.9464882943145,18337.58185156195
|
||||
BatchPrefillWithRaggedKVCacheWrapper,16,16384,32,4,192,128,1.1809706666666668,568.4182232017052,37240.94624227753
|
||||
BatchPrefillWithRaggedKVCacheWrapper,64,1024,32,4,192,128,0.2555306666666667,659.6413424611787,2689.2849156787443
|
||||
BatchPrefillWithRaggedKVCacheWrapper,64,4096,32,4,192,128,0.9085866666666667,739.472740079831,12101.340115520075
|
||||
BatchPrefillWithRaggedKVCacheWrapper,64,8192,32,4,192,128,1.7810773333333334,754.017631276351,24693.1810808739
|
||||
BatchPrefillWithRaggedKVCacheWrapper,64,16384,32,4,192,128,3.5260586666666662,761.5134193267346,49891.92667360423
|
||||
BatchPrefillWithRaggedKVCacheWrapper,1,1024,32,4,256,256,0.044037119999999964,95.61678874549479,390.12245087780525
|
||||
BatchPrefillWithRaggedKVCacheWrapper,1,4096,32,4,256,256,0.08118271999999997,206.86175580222005,3385.916448032292
|
||||
BatchPrefillWithRaggedKVCacheWrapper,1,8192,32,4,256,256,0.11204607999999996,299.6161579235972,9813.030743922503
|
||||
BatchPrefillWithRaggedKVCacheWrapper,1,16384,32,4,256,256,0.14619648000000002,459.1440778875113,30083.12177628353
|
||||
BatchPrefillWithRaggedKVCacheWrapper,4,1024,32,4,256,256,0.07792639999999999,216.1366622864652,881.8510381077531
|
||||
BatchPrefillWithRaggedKVCacheWrapper,4,4096,32,4,256,256,0.13784064000000001,487.3337790654483,7976.686902904687
|
||||
BatchPrefillWithRaggedKVCacheWrapper,4,8192,32,4,256,256,0.22408533333333336,599.2505712109672,19626.65938766184
|
||||
BatchPrefillWithRaggedKVCacheWrapper,4,16384,32,4,256,256,0.3959893333333334,678.0510720827496,44425.908890852275
|
||||
BatchPrefillWithRaggedKVCacheWrapper,16,1024,32,4,256,256,0.15150079999999996,444.6907739101049,1814.366042581954
|
||||
BatchPrefillWithRaggedKVCacheWrapper,16,4096,32,4,256,256,0.4274346666666664,628.6284687562392,10289.400589339195
|
||||
BatchPrefillWithRaggedKVCacheWrapper,16,8192,32,4,256,256,0.7913173333333334,678.7833823093305,22231.518637802277
|
||||
BatchPrefillWithRaggedKVCacheWrapper,16,16384,32,4,256,256,1.5360853333333337,699.1824898616742,45810.43946625186
|
||||
BatchPrefillWithRaggedKVCacheWrapper,64,1024,32,4,256,256,0.43906133333333336,613.773091686507,2504.2324256352945
|
||||
BatchPrefillWithRaggedKVCacheWrapper,64,4096,32,4,256,256,1.6363946666666664,656.8039006075145,10750.576497692491
|
||||
BatchPrefillWithRaggedKVCacheWrapper,64,8192,32,4,256,256,3.234005333333333,664.3564257160574,21759.006842803803
|
||||
BatchPrefillWithRaggedKVCacheWrapper,64,16384,32,4,256,256,6.420821333333334,669.0757535484556,43837.84598543405
|
||||
|
Binary file not shown.
|
Binary file not shown.
|
Binary file not shown.
|
Binary file not shown.
|
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
|
|
@ -0,0 +1,16 @@
|
|||
{
|
||||
"id": 193,
|
||||
"displayId": 20001,
|
||||
"type": "Traditional",
|
||||
"isPublic": false,
|
||||
"locales": [
|
||||
"zh_CN"
|
||||
],
|
||||
"samples": [
|
||||
{
|
||||
"inputData": "1\n",
|
||||
"outputData": ""
|
||||
}
|
||||
],
|
||||
"problemTagIds": []
|
||||
}
|
||||
|
|
@ -0,0 +1,307 @@
|
|||
from __future__ import annotations
|
||||
|
||||
|
||||
HEAD_DIM_CONFIGS = [(128, 128)]
|
||||
BATCH_SIZES = [1, 4, 16]
|
||||
SEQ_LENS = [1024, 4096, 8192, 16384]
|
||||
NUM_QO_HEADS = 32
|
||||
NUM_KV_HEADS = 4
|
||||
CAUSAL = 1
|
||||
|
||||
|
||||
def _build_cases():
|
||||
cases = []
|
||||
for head_dim_qk, head_dim_vo in HEAD_DIM_CONFIGS:
|
||||
for batch_size in BATCH_SIZES:
|
||||
for seq_len in SEQ_LENS:
|
||||
cases.append(
|
||||
(
|
||||
batch_size,
|
||||
seq_len,
|
||||
NUM_QO_HEADS,
|
||||
NUM_KV_HEADS,
|
||||
head_dim_qk,
|
||||
head_dim_vo,
|
||||
CAUSAL,
|
||||
)
|
||||
)
|
||||
return cases
|
||||
|
||||
|
||||
TESTCASES = _build_cases()
|
||||
|
||||
|
||||
def getNumOfTestcases() -> int:
|
||||
return len(TESTCASES)
|
||||
|
||||
|
||||
try:
|
||||
from pathlib import Path
|
||||
from typing import List, Tuple, Union
|
||||
import sys
|
||||
|
||||
import torch
|
||||
|
||||
KernelArg = Union[torch.Tensor, int, float]
|
||||
CURRENT_CASE = None
|
||||
|
||||
def _ensure_flashinfer_importable():
|
||||
try:
|
||||
import flashinfer # noqa: F401
|
||||
|
||||
return
|
||||
except ImportError:
|
||||
pass
|
||||
|
||||
here = Path(__file__).resolve()
|
||||
for parent in here.parents:
|
||||
candidate = parent / "McFlashInfer"
|
||||
if (candidate / "flashinfer").is_dir():
|
||||
sys.path.insert(0, str(candidate))
|
||||
return
|
||||
|
||||
def _get_testcase_index() -> int:
|
||||
try:
|
||||
raw = input().strip()
|
||||
except EOFError:
|
||||
return 0
|
||||
if raw == "":
|
||||
return 0
|
||||
try:
|
||||
testcase_id = int(raw.split()[0])
|
||||
except ValueError:
|
||||
return 0
|
||||
if 1 <= testcase_id <= len(TESTCASES):
|
||||
return testcase_id - 1
|
||||
if 0 <= testcase_id < len(TESTCASES):
|
||||
return testcase_id
|
||||
return 0
|
||||
|
||||
def _compute_reps(batch_size: int, seq_len: int, head_dim: int, base_reps: int = 100) -> int:
|
||||
workload = batch_size * seq_len * head_dim
|
||||
if workload < 1e5:
|
||||
return base_reps
|
||||
if workload < 1e6:
|
||||
return base_reps // 2
|
||||
if workload < 1e7:
|
||||
return base_reps // 4
|
||||
if workload < 1e8:
|
||||
return base_reps // 8
|
||||
if workload < 1e9:
|
||||
return base_reps // 16
|
||||
return base_reps // 32
|
||||
|
||||
def getTestCaseSize() -> Tuple[List[Tuple[int, ...]], Tuple[int, int]]:
|
||||
testcase_id = _get_testcase_index()
|
||||
global CURRENT_CASE
|
||||
batch_size, seq_len, num_qo_heads, num_kv_heads, head_dim_qk, head_dim_vo, causal = TESTCASES[testcase_id]
|
||||
CURRENT_CASE = (
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_qo_heads,
|
||||
num_kv_heads,
|
||||
head_dim_qk,
|
||||
head_dim_vo,
|
||||
causal,
|
||||
20260610 + testcase_id,
|
||||
)
|
||||
qo_len = batch_size * seq_len
|
||||
kv_len = batch_size * seq_len
|
||||
warmup = 3
|
||||
iters = max(1, _compute_reps(batch_size, seq_len, head_dim_qk + head_dim_vo))
|
||||
return [
|
||||
(qo_len, num_qo_heads, head_dim_qk),
|
||||
(kv_len, num_kv_heads, head_dim_qk),
|
||||
(kv_len, num_kv_heads, head_dim_vo),
|
||||
(qo_len, num_qo_heads, head_dim_vo),
|
||||
(batch_size + 1,),
|
||||
(batch_size + 1,),
|
||||
(), (), (), (), (), (), (),
|
||||
], (warmup, iters)
|
||||
|
||||
def genTestCase(testcase_sizes, device: str = "cuda") -> List[KernelArg]:
|
||||
del testcase_sizes
|
||||
batch_size, seq_len, num_qo_heads, num_kv_heads, head_dim_qk, head_dim_vo, causal, seed = CURRENT_CASE
|
||||
gen = torch.Generator(device=device)
|
||||
gen.manual_seed(seed)
|
||||
dtype = torch.bfloat16
|
||||
qo_len = batch_size * seq_len
|
||||
q = torch.rand(
|
||||
qo_len,
|
||||
num_qo_heads,
|
||||
head_dim_qk,
|
||||
dtype=dtype,
|
||||
device=device,
|
||||
generator=gen,
|
||||
).contiguous()
|
||||
kv_len = batch_size * seq_len
|
||||
k = torch.rand(
|
||||
kv_len,
|
||||
num_kv_heads,
|
||||
head_dim_qk,
|
||||
dtype=dtype,
|
||||
device=device,
|
||||
generator=gen,
|
||||
).contiguous()
|
||||
v = torch.rand(
|
||||
kv_len,
|
||||
num_kv_heads,
|
||||
head_dim_vo,
|
||||
dtype=dtype,
|
||||
device=device,
|
||||
generator=gen,
|
||||
).contiguous()
|
||||
output = torch.empty(
|
||||
qo_len,
|
||||
num_qo_heads,
|
||||
head_dim_vo,
|
||||
dtype=dtype,
|
||||
device=device,
|
||||
)
|
||||
qo_indptr = torch.arange(0, batch_size + 1, dtype=torch.int32, device=device) * seq_len
|
||||
kv_indptr = qo_indptr.clone()
|
||||
return [
|
||||
q,
|
||||
k,
|
||||
v,
|
||||
output,
|
||||
qo_indptr,
|
||||
kv_indptr,
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_qo_heads,
|
||||
num_kv_heads,
|
||||
head_dim_qk,
|
||||
head_dim_vo,
|
||||
causal,
|
||||
]
|
||||
|
||||
def baseline(
|
||||
q,
|
||||
k,
|
||||
v,
|
||||
output,
|
||||
qo_indptr,
|
||||
kv_indptr,
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_qo_heads,
|
||||
num_kv_heads,
|
||||
head_dim_qk,
|
||||
head_dim_vo,
|
||||
causal,
|
||||
):
|
||||
_ensure_flashinfer_importable()
|
||||
import flashinfer
|
||||
|
||||
workspace_buffer = torch.empty(128 * 1024 * 1024, dtype=torch.uint8, device=q.device)
|
||||
wrapper = flashinfer.BatchPrefillWithRaggedKVCacheWrapper(
|
||||
workspace_buffer,
|
||||
kv_layout="NHD",
|
||||
backend="auto",
|
||||
)
|
||||
wrapper.plan(
|
||||
qo_indptr,
|
||||
kv_indptr,
|
||||
int(num_qo_heads),
|
||||
int(num_kv_heads),
|
||||
int(head_dim_qk),
|
||||
int(head_dim_vo),
|
||||
causal=bool(causal),
|
||||
q_data_type=torch.bfloat16,
|
||||
kv_data_type=torch.bfloat16,
|
||||
)
|
||||
wrapper.run(q, k, v, out=output)
|
||||
return [
|
||||
q,
|
||||
k,
|
||||
v,
|
||||
output,
|
||||
qo_indptr,
|
||||
kv_indptr,
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_qo_heads,
|
||||
num_kv_heads,
|
||||
head_dim_qk,
|
||||
head_dim_vo,
|
||||
causal,
|
||||
]
|
||||
|
||||
def check(
|
||||
testcase_sizes,
|
||||
original_input_tensors,
|
||||
target_kernel_input_tensors,
|
||||
baseline_input_tensors,
|
||||
rtol=1e-2,
|
||||
atol=1e-2,
|
||||
) -> bool:
|
||||
del testcase_sizes, original_input_tensors
|
||||
output_t = target_kernel_input_tensors[3]
|
||||
output_ref = baseline_input_tensors[3]
|
||||
if output_t.shape != output_ref.shape:
|
||||
print(f"[FAIL] shape mismatch: target {output_t.shape}, ref {output_ref.shape}", file=sys.stderr)
|
||||
return False
|
||||
if output_t.dtype != output_ref.dtype:
|
||||
print(f"[FAIL] dtype mismatch: target {output_t.dtype}, ref {output_ref.dtype}", file=sys.stderr)
|
||||
return False
|
||||
if not torch.allclose(output_t.float(), output_ref.float(), rtol=rtol, atol=atol):
|
||||
diff = (output_t.float() - output_ref.float()).abs()
|
||||
print(
|
||||
f"[FAIL] allclose failed: max_abs_diff={float(diff.max().item()):.6f}, "
|
||||
f"mean_abs_diff={float(diff.mean().item()):.6f} (rtol={rtol}, atol={atol})",
|
||||
file=sys.stderr,
|
||||
)
|
||||
return False
|
||||
return True
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
|
||||
INPUT_CLASS = [
|
||||
"INPUT",
|
||||
"INPUT",
|
||||
"INPUT",
|
||||
"OUTPUT",
|
||||
"INPUT",
|
||||
"INPUT",
|
||||
"INPUT",
|
||||
"INPUT",
|
||||
"INPUT",
|
||||
"INPUT",
|
||||
"INPUT",
|
||||
"INPUT",
|
||||
"INPUT",
|
||||
]
|
||||
|
||||
|
||||
def getWorkload(testcase_sizes) -> dict:
|
||||
raw_sizes = testcase_sizes[0] if isinstance(testcase_sizes, tuple) and len(testcase_sizes) == 2 else testcase_sizes
|
||||
q_shape, k_shape, v_shape, output_shape, qo_indptr_shape, kv_indptr_shape = raw_sizes[:6]
|
||||
qo_len, num_qo_heads, head_dim_qk = q_shape
|
||||
kv_len, num_kv_heads, k_dim = k_shape
|
||||
v_len, v_heads, head_dim_vo = v_shape
|
||||
assert k_dim == head_dim_qk
|
||||
assert v_len == kv_len
|
||||
assert v_heads == num_kv_heads
|
||||
assert qo_len == kv_len
|
||||
assert output_shape == (qo_len, num_qo_heads, head_dim_vo)
|
||||
assert qo_indptr_shape == kv_indptr_shape
|
||||
batch_size = qo_indptr_shape[0] - 1
|
||||
seq_len = kv_len // batch_size
|
||||
flops = batch_size * seq_len * seq_len * num_qo_heads * (head_dim_qk + head_dim_vo)
|
||||
memory_bytes = (
|
||||
qo_len * num_qo_heads * head_dim_qk * 2
|
||||
+ kv_len * num_kv_heads * head_dim_qk * 2
|
||||
+ kv_len * num_kv_heads * head_dim_vo * 2
|
||||
+ qo_len * num_qo_heads * head_dim_vo * 2
|
||||
+ (batch_size + 1) * 4 * 2
|
||||
)
|
||||
return {
|
||||
"flops": flops,
|
||||
"memory_bytes": memory_bytes,
|
||||
"dtype": "bf16",
|
||||
}
|
||||
|
||||
|
||||
DESIGNED_VRAM_SIZE = 48
|
||||
|
|
@ -0,0 +1,23 @@
|
|||
---
|
||||
sectionTitle: "题目描述"
|
||||
type: "Text"
|
||||
---
|
||||
你需要实现 FlashInfer ragged KV cache prefill 的CUDA C++前向算子。
|
||||
|
||||
本题输入采用 FlashInfer `BatchPrefillWithRaggedKVCacheWrapper` 的 ragged `NHD` 布局。每个 batch 中有 `seq_len` 个 query token，KV cache 中也有 `seq_len` 个 token：
|
||||
|
||||
其中 query heads 采用 GQA 布局：`num_qo_heads` 个 query/output heads 共享 `num_kv_heads` 个 KV heads，`G = num_qo_heads / num_kv_heads`。
|
||||
|
||||
评测程序会调用你提交代码中的 `run_kernel` 函数。你需要根据 `qo_indptr` 和 `kv_indptr` 读取 ragged Q/K/V，并将结果写入 `output`。
|
||||
|
||||
baseline 使用 FlashInfer ragged prefill 的 Python API：
|
||||
|
||||
```python
|
||||
wrapper = flashinfer.BatchPrefillWithRaggedKVCacheWrapper(workspace, kv_layout="NHD", backend="auto")
|
||||
wrapper.plan(qo_indptr, kv_indptr, num_qo_heads, num_kv_heads,
|
||||
head_dim_qk, head_dim_vo, causal=True,
|
||||
q_data_type=torch.bfloat16, kv_data_type=torch.bfloat16)
|
||||
wrapper.run(q, k, v, out=output)
|
||||
```
|
||||
|
||||
如何提交代码详见[评测指南](/d/2)。
|
||||
|
|
@ -0,0 +1,41 @@
|
|||
---
|
||||
sectionTitle: "接口约定"
|
||||
type: "codeSample"
|
||||
lang: "cuda"
|
||||
---
|
||||
你必须在提交的 CUDA 源码中提供如下 **C 符号**，函数名、参数类型、顺序必须完全一致，并使用 `extern "C"` 防止 name mangling：
|
||||
|
||||
```cpp
|
||||
#include <stdint.h>
|
||||
#include <cuda_bf16.h>
|
||||
|
||||
extern "C" void run_kernel(
|
||||
const __nv_bfloat16* q,
|
||||
const __nv_bfloat16* k,
|
||||
const __nv_bfloat16* v,
|
||||
__nv_bfloat16* output,
|
||||
const int32_t* qo_indptr,
|
||||
const int32_t* kv_indptr,
|
||||
int64_t batch_size,
|
||||
int64_t seq_len,
|
||||
int64_t num_qo_heads,
|
||||
int64_t num_kv_heads,
|
||||
int64_t head_dim_qk,
|
||||
int64_t head_dim_vo,
|
||||
int64_t causal
|
||||
);
|
||||
```
|
||||
|
||||
### 参数说明
|
||||
|
||||
* `q`：query tensor，shape `(batch_size * seq_len, num_qo_heads, head_dim_qk)`，连续 `bf16`
|
||||
* `k`：key tensor，shape `(batch_size * seq_len, num_kv_heads, head_dim_qk)`，连续 `bf16`
|
||||
* `v`：value tensor，shape `(batch_size * seq_len, num_kv_heads, head_dim_vo)`，连续 `bf16`
|
||||
* `output`：输出缓冲区，shape `(batch_size * seq_len, num_qo_heads, head_dim_vo)`，连续 `bf16`
|
||||
* `qo_indptr`：query/output ragged indptr，shape `(batch_size + 1)`，连续 `int32`
|
||||
* `kv_indptr`：KV ragged indptr，shape `(batch_size + 1)`，连续 `int32`
|
||||
* `causal`：是否启用 causal mask，评测中固定为 `1`
|
||||
|
||||
本题测试中 `qo_indptr[b + 1] - qo_indptr[b] == seq_len`，`kv_indptr[b + 1] - kv_indptr[b] == seq_len`。
|
||||
|
||||
`run_kernel` 内部需要自行计算合适的 launch 配置并启动 CUDA kernel。为保证计时准确，不建议在 `run_kernel` 内部做 `cudaDeviceSynchronize()` 或显式同步。
|
||||
|
|
@ -0,0 +1,26 @@
|
|||
---
|
||||
sectionTitle: "输入格式"
|
||||
type: "Text"
|
||||
---
|
||||
## 数据范围
|
||||
|
||||
- 数据类型：`q/kv_data/output` 均为 `bfloat16`
|
||||
- KV layout：`NHD`
|
||||
- `num_qo_heads = 32`
|
||||
- `num_kv_heads = 4`
|
||||
- `page_block_size = 16`
|
||||
- `causal = 0`
|
||||
- `head_dim` 固定为 `128`
|
||||
- `batch_size` 取值为 `1, 4, 16`
|
||||
- `seq_len` 取值为 `1024, 4096, 8192, 16384`
|
||||
|
||||
测试点顺序与 `McFlashInfer/benchmarks/bench_batch_prefill_paged.py` 中的 cases 一致，即：
|
||||
|
||||
```python
|
||||
for head_dim in [128]:
|
||||
for batch_size in [1, 4, 16]:
|
||||
for seq_len in [1024, 4096, 8192, 16384]:
|
||||
...
|
||||
```
|
||||
|
||||
输出参与 `torch.allclose` 校验，容差为 `rtol=1e-2, atol=1e-2`。
|
||||
|
|
@ -0,0 +1,28 @@
|
|||
pytorch参考实现：
|
||||
|
||||
```python
|
||||
def baseline(q, kv_data, output, qo_indptr, kv_indptr, kv_indices, last_page_len,
|
||||
batch_size, seq_len, num_qo_heads, num_kv_heads,
|
||||
head_dim, page_block_size, causal):
|
||||
workspace_buffer = torch.empty(128 * 1024 * 1024, dtype=torch.uint8, device=q.device)
|
||||
wrapper = flashinfer.BatchPrefillWithPagedKVCacheWrapper(
|
||||
workspace_buffer,
|
||||
kv_layout="NHD",
|
||||
backend="auto",
|
||||
)
|
||||
wrapper.plan(
|
||||
qo_indptr,
|
||||
kv_indptr,
|
||||
kv_indices,
|
||||
last_page_len,
|
||||
num_qo_heads,
|
||||
num_kv_heads,
|
||||
head_dim,
|
||||
page_block_size,
|
||||
causal=bool(causal),
|
||||
q_data_type=torch.bfloat16,
|
||||
kv_data_type=torch.bfloat16,
|
||||
)
|
||||
wrapper.run(q, kv_data, out=output)
|
||||
```
|
||||
|
||||
|
|
@ -0,0 +1 @@
|
|||
FlashInfer Ragged Prefill
|
||||
|
|
@ -0,0 +1,16 @@
|
|||
{
|
||||
"id": 194,
|
||||
"displayId": 20002,
|
||||
"type": "Traditional",
|
||||
"isPublic": false,
|
||||
"locales": [
|
||||
"zh_CN"
|
||||
],
|
||||
"samples": [
|
||||
{
|
||||
"inputData": "1\n",
|
||||
"outputData": ""
|
||||
}
|
||||
],
|
||||
"problemTagIds": []
|
||||
}
|
||||
|
|
@ -0,0 +1,273 @@
|
|||
from __future__ import annotations
|
||||
|
||||
|
||||
HEAD_DIMS = [128]
|
||||
BATCH_SIZES = [1, 4, 16]
|
||||
SEQ_LENS = [1024, 4096, 8192, 16384]
|
||||
NUM_QO_HEADS = 32
|
||||
PAGE_BLOCK_SIZE = 16
|
||||
CAUSAL = 0
|
||||
|
||||
|
||||
def _build_cases():
|
||||
cases = []
|
||||
for head_dim in HEAD_DIMS:
|
||||
for batch_size in BATCH_SIZES:
|
||||
for seq_len in SEQ_LENS:
|
||||
num_kv_heads = 8 if head_dim == 64 else 4
|
||||
cases.append(
|
||||
(
|
||||
batch_size,
|
||||
seq_len,
|
||||
NUM_QO_HEADS,
|
||||
num_kv_heads,
|
||||
head_dim,
|
||||
PAGE_BLOCK_SIZE,
|
||||
CAUSAL,
|
||||
)
|
||||
)
|
||||
return cases
|
||||
|
||||
|
||||
TESTCASES = _build_cases()
|
||||
|
||||
|
||||
def getNumOfTestcases() -> int:
|
||||
return len(TESTCASES)
|
||||
|
||||
|
||||
try:
|
||||
from pathlib import Path
|
||||
from typing import List, Tuple, Union
|
||||
import sys
|
||||
|
||||
import torch
|
||||
|
||||
KernelArg = Union[torch.Tensor, int, float]
|
||||
CURRENT_CASE = None
|
||||
|
||||
def _ensure_flashinfer_importable():
|
||||
try:
|
||||
import flashinfer # noqa: F401
|
||||
|
||||
return
|
||||
except ImportError:
|
||||
pass
|
||||
|
||||
here = Path(__file__).resolve()
|
||||
for parent in here.parents:
|
||||
candidate = parent / "McFlashInfer"
|
||||
if (candidate / "flashinfer").is_dir():
|
||||
sys.path.insert(0, str(candidate))
|
||||
return
|
||||
|
||||
def _get_testcase_index() -> int:
|
||||
try:
|
||||
raw = input().strip()
|
||||
except EOFError:
|
||||
return 0
|
||||
if raw == "":
|
||||
return 0
|
||||
try:
|
||||
testcase_id = int(raw.split()[0])
|
||||
except ValueError:
|
||||
return 0
|
||||
if 1 <= testcase_id <= len(TESTCASES):
|
||||
return testcase_id - 1
|
||||
if 0 <= testcase_id < len(TESTCASES):
|
||||
return testcase_id
|
||||
return 0
|
||||
|
||||
def _compute_reps(batch_size: int, seq_len: int, head_dim: int, base_reps: int = 100) -> int:
|
||||
workload = batch_size * seq_len * head_dim
|
||||
if workload < 1e5:
|
||||
return base_reps
|
||||
if workload < 1e6:
|
||||
return base_reps // 2
|
||||
if workload < 1e7:
|
||||
return base_reps // 4
|
||||
if workload < 1e8:
|
||||
return base_reps // 8
|
||||
if workload < 1e9:
|
||||
return base_reps // 16
|
||||
return base_reps // 32
|
||||
|
||||
def _setup_paged_kv_indptr(batch_size: int, seq_len: int, page_block_size: int, device: str):
|
||||
seq_lens = torch.full((batch_size,), seq_len, dtype=torch.int32, device=device)
|
||||
seq_lens_blocks = torch.div(seq_lens + page_block_size - 1, page_block_size, rounding_mode="floor")
|
||||
kv_indptr = torch.empty((batch_size + 1,), dtype=torch.int32, device=device)
|
||||
kv_indptr[0] = 0
|
||||
kv_indptr[1:] = torch.cumsum(seq_lens_blocks, dim=0)
|
||||
num_blocks = int(kv_indptr[-1].item())
|
||||
last_page_len = (seq_lens - 1) % page_block_size + 1
|
||||
return kv_indptr, last_page_len, num_blocks
|
||||
|
||||
def getTestCaseSize() -> Tuple[List[Tuple[int, ...]], Tuple[int, int]]:
|
||||
testcase_id = _get_testcase_index()
|
||||
global CURRENT_CASE
|
||||
batch_size, seq_len, num_qo_heads, num_kv_heads, head_dim, page_block_size, causal = TESTCASES[testcase_id]
|
||||
num_blocks = batch_size * ((seq_len + page_block_size - 1) // page_block_size)
|
||||
qo_len = batch_size * seq_len
|
||||
CURRENT_CASE = (
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_qo_heads,
|
||||
num_kv_heads,
|
||||
head_dim,
|
||||
page_block_size,
|
||||
causal,
|
||||
num_blocks,
|
||||
20260620 + testcase_id,
|
||||
)
|
||||
warmup = 3
|
||||
iters = max(1, _compute_reps(batch_size, seq_len, head_dim))
|
||||
return [
|
||||
(qo_len, num_qo_heads, head_dim),
|
||||
(num_blocks, 2, page_block_size, num_kv_heads, head_dim),
|
||||
(qo_len, num_qo_heads, head_dim),
|
||||
(batch_size + 1,),
|
||||
(batch_size + 1,),
|
||||
(num_blocks,),
|
||||
(batch_size,),
|
||||
(), (), (), (), (), (), (),
|
||||
], (warmup, iters)
|
||||
|
||||
def genTestCase(testcase_sizes, device: str = "cuda") -> List[KernelArg]:
|
||||
del testcase_sizes
|
||||
(
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_qo_heads,
|
||||
num_kv_heads,
|
||||
head_dim,
|
||||
page_block_size,
|
||||
causal,
|
||||
num_blocks,
|
||||
seed,
|
||||
) = CURRENT_CASE
|
||||
gen = torch.Generator(device=device)
|
||||
gen.manual_seed(seed)
|
||||
dtype = torch.bfloat16
|
||||
qo_len = batch_size * seq_len
|
||||
q = torch.rand(qo_len, num_qo_heads, head_dim, dtype=dtype, device=device, generator=gen).contiguous()
|
||||
kv_data = torch.randn(
|
||||
num_blocks,
|
||||
2,
|
||||
page_block_size,
|
||||
num_kv_heads,
|
||||
head_dim,
|
||||
dtype=dtype,
|
||||
device=device,
|
||||
generator=gen,
|
||||
).contiguous()
|
||||
output = torch.empty(qo_len, num_qo_heads, head_dim, dtype=dtype, device=device)
|
||||
qo_indptr = torch.arange(0, batch_size + 1, dtype=torch.int32, device=device) * seq_len
|
||||
kv_indptr, last_page_len, num_blocks_check = _setup_paged_kv_indptr(
|
||||
batch_size,
|
||||
seq_len,
|
||||
page_block_size,
|
||||
device,
|
||||
)
|
||||
assert num_blocks_check == num_blocks
|
||||
kv_indices = torch.arange(num_blocks, dtype=torch.int32, device=device)
|
||||
return [
|
||||
q,
|
||||
kv_data,
|
||||
output,
|
||||
qo_indptr,
|
||||
kv_indptr,
|
||||
kv_indices,
|
||||
last_page_len,
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_qo_heads,
|
||||
num_kv_heads,
|
||||
head_dim,
|
||||
page_block_size,
|
||||
causal,
|
||||
]
|
||||
|
||||
def baseline(
|
||||
q,
|
||||
kv_data,
|
||||
output,
|
||||
qo_indptr,
|
||||
kv_indptr,
|
||||
kv_indices,
|
||||
last_page_len,
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_qo_heads,
|
||||
num_kv_heads,
|
||||
head_dim,
|
||||
page_block_size,
|
||||
causal,
|
||||
):
|
||||
_ensure_flashinfer_importable()
|
||||
import flashinfer
|
||||
|
||||
workspace_buffer = torch.empty(128 * 1024 * 1024, dtype=torch.uint8, device=q.device)
|
||||
wrapper = flashinfer.BatchPrefillWithPagedKVCacheWrapper(
|
||||
workspace_buffer,
|
||||
kv_layout="NHD",
|
||||
backend="auto",
|
||||
)
|
||||
wrapper.plan(
|
||||
qo_indptr,
|
||||
kv_indptr,
|
||||
kv_indices,
|
||||
last_page_len,
|
||||
int(num_qo_heads),
|
||||
int(num_kv_heads),
|
||||
int(head_dim),
|
||||
int(page_block_size),
|
||||
causal=bool(causal),
|
||||
q_data_type=torch.bfloat16,
|
||||
kv_data_type=torch.bfloat16,
|
||||
)
|
||||
wrapper.run(q, kv_data, out=output)
|
||||
return [
|
||||
q,
|
||||
kv_data,
|
||||
output,
|
||||
qo_indptr,
|
||||
kv_indptr,
|
||||
kv_indices,
|
||||
last_page_len,
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_qo_heads,
|
||||
num_kv_heads,
|
||||
head_dim,
|
||||
page_block_size,
|
||||
causal,
|
||||
]
|
||||
|
||||
def check(
|
||||
testcase_sizes,
|
||||
original_input_tensors,
|
||||
target_kernel_input_tensors,
|
||||
baseline_input_tensors,
|
||||
rtol=1e-2,
|
||||
atol=1e-2,
|
||||
) -> bool:
|
||||
del testcase_sizes, original_input_tensors
|
||||
output_t = target_kernel_input_tensors[2]
|
||||
output_ref = baseline_input_tensors[2]
|
||||
if output_t.shape != output_ref.shape:
|
||||
print(f"[FAIL] shape mismatch: target {output_t.shape}, ref {output_ref.shape}", file=sys.stderr)
|
||||
return False
|
||||
if output_t.dtype != output_ref.dtype:
|
||||
print(f"[FAIL] dtype mismatch: target {output_t.dtype}, ref {output_ref.dtype}", file=sys.stderr)
|
||||
return False
|
||||
if not torch.allclose(output_t.float(), output_ref.float(), rtol=rtol, atol=atol):
|
||||
diff = (output_t.float() - output_ref.float()).abs()
|
||||
print(
|
||||
f"[FAIL] allclose failed: max_abs_diff={float(diff.max().item()):.6f}, "
|
||||
f"mean_abs_diff={float(diff.mean().item()):.6f} (rtol={rtol}, atol={atol})",
|
||||
file=sys.stderr,
|
||||
)
|
||||
return False
|
||||
return True
|
||||
except Exception:
|
||||
pass
|
||||
|
|
@ -0,0 +1,22 @@
|
|||
---
|
||||
sectionTitle: "题目描述"
|
||||
type: "Text"
|
||||
---
|
||||
你需要实现 FlashInfer paged KV cache prefill 的 CUDA C++ 前向算子。
|
||||
|
||||
本题输入采用 FlashInfer `BatchPrefillWithPagedKVCacheWrapper` 的 paged `NHD` 布局。每个 batch 中有 `seq_len` 个 query token，KV cache 也有 `seq_len` 个 token，并按 page 存储。
|
||||
|
||||
评测程序会调用你提交代码中的 `run_kernel` 函数。你需要根据 `qo_indptr`、`kv_indptr`、`kv_indices` 和 `last_page_len` 读取 paged KV cache，并将结果写入 `output`。
|
||||
|
||||
baseline 使用 FlashInfer paged prefill 的 Python API：
|
||||
|
||||
```python
|
||||
wrapper = flashinfer.BatchPrefillWithPagedKVCacheWrapper(workspace, kv_layout="NHD", backend="auto")
|
||||
wrapper.plan(qo_indptr, kv_indptr, kv_indices, last_page_len,
|
||||
num_qo_heads, num_kv_heads, head_dim, page_block_size,
|
||||
causal=bool(causal),
|
||||
q_data_type=torch.bfloat16, kv_data_type=torch.bfloat16)
|
||||
wrapper.run(q, kv_data, out=output)
|
||||
```
|
||||
|
||||
如何提交代码详见[评测指南](/d/2)。
|
||||
|
|
@ -0,0 +1,42 @@
|
|||
---
|
||||
sectionTitle: "接口约定"
|
||||
type: "codeSample"
|
||||
lang: "cuda"
|
||||
---
|
||||
你必须在提交的 CUDA 源码中提供如下 **C 符号**，函数名、参数类型、顺序必须完全一致，并使用 `extern "C"` 防止 name mangling：
|
||||
|
||||
```cpp
|
||||
#include <stdint.h>
|
||||
#include <cuda_bf16.h>
|
||||
|
||||
extern "C" void run_kernel(
|
||||
const __nv_bfloat16* q,
|
||||
const __nv_bfloat16* kv_data,
|
||||
__nv_bfloat16* output,
|
||||
const int32_t* qo_indptr,
|
||||
const int32_t* kv_indptr,
|
||||
const int32_t* kv_indices,
|
||||
const int32_t* last_page_len,
|
||||
int64_t batch_size,
|
||||
int64_t seq_len,
|
||||
int64_t num_qo_heads,
|
||||
int64_t num_kv_heads,
|
||||
int64_t head_dim,
|
||||
int64_t page_block_size,
|
||||
int64_t causal
|
||||
);
|
||||
```
|
||||
|
||||
### 参数说明
|
||||
|
||||
* `q`：query tensor，shape `(batch_size * seq_len, num_qo_heads, head_dim)`，连续 `bf16`
|
||||
* `kv_data`：paged KV cache，shape `(num_blocks, 2, page_block_size, num_kv_heads, head_dim)`，连续 `bf16`，其中 `kv_data[:, 0]` 为 key，`kv_data[:, 1]` 为 value
|
||||
* `output`：输出缓冲区，shape `(batch_size * seq_len, num_qo_heads, head_dim)`，连续 `bf16`
|
||||
* `qo_indptr`：query/output indptr，shape `(batch_size + 1)`，连续 `int32`
|
||||
* `kv_indptr`：paged KV indptr，shape `(batch_size + 1)`，连续 `int32`
|
||||
* `kv_indices`：page index，shape `(num_blocks)`，连续 `int32`
|
||||
* `last_page_len`：每个 batch 最后一个 page 的有效 token 数，shape `(batch_size)`，连续 `int32`
|
||||
* `page_block_size`：page size，评测中固定为 `16`
|
||||
* `causal`：是否启用 causal mask，本题按 benchmark case 固定为 `0`
|
||||
|
||||
`run_kernel` 内部需要自行计算合适的 launch 配置并启动 CUDA kernel。为保证计时准确，不建议在 `run_kernel` 内部做 `cudaDeviceSynchronize()` 或显式同步。
|
||||
|
|
@ -0,0 +1,26 @@
|
|||
---
|
||||
sectionTitle: "输入格式"
|
||||
type: "Text"
|
||||
---
|
||||
## 数据范围
|
||||
|
||||
- 数据类型：`q/kv_data/output` 均为 `bfloat16`
|
||||
- KV layout：`NHD`
|
||||
- `num_qo_heads = 32`
|
||||
- `num_kv_heads = 4`
|
||||
- `page_block_size = 16`
|
||||
- `causal = 0`
|
||||
- `head_dim` 固定为 `128`
|
||||
- `batch_size` 取值为 `1, 4, 16`
|
||||
- `seq_len` 取值为 `1024, 4096, 8192, 16384`
|
||||
|
||||
测试点顺序与 `McFlashInfer/benchmarks/bench_batch_prefill_paged.py` 中的 cases 一致，即：
|
||||
|
||||
```python
|
||||
for head_dim in [128]:
|
||||
for batch_size in [1, 4, 16]:
|
||||
for seq_len in [1024, 4096, 8192, 16384]:
|
||||
...
|
||||
```
|
||||
|
||||
输出参与 `torch.allclose` 校验，容差为 `rtol=1e-2, atol=1e-2`。
|
||||
|
|
@ -0,0 +1,31 @@
|
|||
---
|
||||
sectionTitle: "输出格式"
|
||||
type: "Text"
|
||||
---
|
||||
PyTorch 参考实现
|
||||
|
||||
```python
|
||||
def baseline(q, kv_data, output, qo_indptr, kv_indptr, kv_indices, last_page_len,
|
||||
batch_size, seq_len, num_qo_heads, num_kv_heads,
|
||||
head_dim, page_block_size, causal):
|
||||
workspace_buffer = torch.empty(128 * 1024 * 1024, dtype=torch.uint8, device=q.device)
|
||||
wrapper = flashinfer.BatchPrefillWithPagedKVCacheWrapper(
|
||||
workspace_buffer,
|
||||
kv_layout="NHD",
|
||||
backend="auto",
|
||||
)
|
||||
wrapper.plan(
|
||||
qo_indptr,
|
||||
kv_indptr,
|
||||
kv_indices,
|
||||
last_page_len,
|
||||
num_qo_heads,
|
||||
num_kv_heads,
|
||||
head_dim,
|
||||
page_block_size,
|
||||
causal=bool(causal),
|
||||
q_data_type=torch.bfloat16,
|
||||
kv_data_type=torch.bfloat16,
|
||||
)
|
||||
wrapper.run(q, kv_data, out=output)
|
||||
```
|
||||
|
|
@ -0,0 +1 @@
|
|||
FlashInfer Paged Prefill
|
||||
|
|
@ -0,0 +1,16 @@
|
|||
{
|
||||
"id": 195,
|
||||
"displayId": 20003,
|
||||
"type": "Traditional",
|
||||
"isPublic": false,
|
||||
"locales": [
|
||||
"zh_CN"
|
||||
],
|
||||
"samples": [
|
||||
{
|
||||
"inputData": "1\n",
|
||||
"outputData": ""
|
||||
}
|
||||
],
|
||||
"problemTagIds": []
|
||||
}
|
||||
|
|
@ -0,0 +1,245 @@
|
|||
from __future__ import annotations
|
||||
|
||||
|
||||
HEAD_DIM_CKV = 512
|
||||
HEAD_DIM_KPE = 64
|
||||
NUM_HEADS_LIST = [64, 128]
|
||||
BATCH_SIZES = [1, 4, 16]
|
||||
SEQ_LENS = [1024, 4096, 8192, 16384]
|
||||
PAGE_SIZE = 1
|
||||
CAUSAL = 0
|
||||
|
||||
|
||||
def _build_cases():
|
||||
cases = []
|
||||
for num_heads in NUM_HEADS_LIST:
|
||||
for batch_size in BATCH_SIZES:
|
||||
for seq_len in SEQ_LENS:
|
||||
cases.append(
|
||||
(
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_heads,
|
||||
HEAD_DIM_CKV,
|
||||
HEAD_DIM_KPE,
|
||||
PAGE_SIZE,
|
||||
CAUSAL,
|
||||
)
|
||||
)
|
||||
return cases
|
||||
|
||||
|
||||
TESTCASES = _build_cases()
|
||||
|
||||
|
||||
def getNumOfTestcases() -> int:
|
||||
return len(TESTCASES)
|
||||
|
||||
|
||||
try:
|
||||
from pathlib import Path
|
||||
from typing import List, Tuple, Union
|
||||
import sys
|
||||
|
||||
import torch
|
||||
|
||||
KernelArg = Union[torch.Tensor, int, float]
|
||||
CURRENT_CASE = None
|
||||
|
||||
def _ensure_flashinfer_importable():
|
||||
try:
|
||||
import flashinfer # noqa: F401
|
||||
|
||||
return
|
||||
except ImportError:
|
||||
pass
|
||||
|
||||
here = Path(__file__).resolve()
|
||||
for parent in here.parents:
|
||||
candidate = parent / "McFlashInfer"
|
||||
if (candidate / "flashinfer").is_dir():
|
||||
sys.path.insert(0, str(candidate))
|
||||
return
|
||||
|
||||
def _get_testcase_index() -> int:
|
||||
try:
|
||||
raw = input().strip()
|
||||
except EOFError:
|
||||
return 0
|
||||
if raw == "":
|
||||
return 0
|
||||
try:
|
||||
testcase_id = int(raw.split()[0])
|
||||
except ValueError:
|
||||
return 0
|
||||
if 1 <= testcase_id <= len(TESTCASES):
|
||||
return testcase_id - 1
|
||||
if 0 <= testcase_id < len(TESTCASES):
|
||||
return testcase_id
|
||||
return 0
|
||||
|
||||
def _compute_reps(batch_size: int, seq_len: int, head_dim: int, base_reps: int = 100) -> int:
|
||||
workload = batch_size * seq_len * head_dim
|
||||
if workload < 1e5:
|
||||
return base_reps
|
||||
if workload < 1e6:
|
||||
return base_reps // 2
|
||||
if workload < 1e7:
|
||||
return base_reps // 4
|
||||
if workload < 1e8:
|
||||
return base_reps // 8
|
||||
if workload < 1e9:
|
||||
return base_reps // 16
|
||||
return base_reps // 32
|
||||
|
||||
def getTestCaseSize() -> Tuple[List[Tuple[int, ...]], Tuple[int, int]]:
|
||||
testcase_id = _get_testcase_index()
|
||||
global CURRENT_CASE
|
||||
batch_size, seq_len, num_heads, head_dim_ckv, head_dim_kpe, page_size, causal = TESTCASES[testcase_id]
|
||||
CURRENT_CASE = (
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_heads,
|
||||
head_dim_ckv,
|
||||
head_dim_kpe,
|
||||
page_size,
|
||||
causal,
|
||||
20260630 + testcase_id,
|
||||
)
|
||||
warmup = 3
|
||||
iters = max(1, _compute_reps(batch_size, seq_len, head_dim_ckv + head_dim_kpe))
|
||||
kv_len = batch_size * seq_len
|
||||
return [
|
||||
(batch_size, num_heads, head_dim_ckv),
|
||||
(batch_size, num_heads, head_dim_kpe),
|
||||
(kv_len, 1, head_dim_ckv),
|
||||
(kv_len, 1, head_dim_kpe),
|
||||
(batch_size, num_heads, head_dim_ckv),
|
||||
(batch_size + 1,),
|
||||
(batch_size + 1,),
|
||||
(kv_len,),
|
||||
(batch_size,),
|
||||
(), (), (), (), (), (), (),
|
||||
], (warmup, iters)
|
||||
|
||||
def genTestCase(testcase_sizes, device: str = "cuda") -> List[KernelArg]:
|
||||
del testcase_sizes
|
||||
batch_size, seq_len, num_heads, head_dim_ckv, head_dim_kpe, page_size, causal, seed = CURRENT_CASE
|
||||
gen = torch.Generator(device=device)
|
||||
gen.manual_seed(seed)
|
||||
dtype = torch.bfloat16
|
||||
kv_len = batch_size * seq_len
|
||||
q_nope = torch.randn(batch_size, num_heads, head_dim_ckv, dtype=dtype, device=device, generator=gen).contiguous()
|
||||
q_pe = torch.zeros(batch_size, num_heads, head_dim_kpe, dtype=dtype, device=device)
|
||||
ckv = torch.randn(kv_len, 1, head_dim_ckv, dtype=dtype, device=device, generator=gen).contiguous()
|
||||
kpe = torch.zeros(kv_len, 1, head_dim_kpe, dtype=dtype, device=device)
|
||||
output = torch.empty(batch_size, num_heads, head_dim_ckv, dtype=dtype, device=device)
|
||||
q_indptr = torch.arange(0, batch_size + 1, dtype=torch.int32, device=device)
|
||||
kv_indptr = q_indptr * seq_len
|
||||
kv_indices = torch.arange(0, kv_len, dtype=torch.int32, device=device)
|
||||
kv_lens = torch.full((batch_size,), seq_len, dtype=torch.int32, device=device)
|
||||
return [
|
||||
q_nope,
|
||||
q_pe,
|
||||
ckv,
|
||||
kpe,
|
||||
output,
|
||||
q_indptr,
|
||||
kv_indptr,
|
||||
kv_indices,
|
||||
kv_lens,
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_heads,
|
||||
head_dim_ckv,
|
||||
head_dim_kpe,
|
||||
page_size,
|
||||
causal,
|
||||
]
|
||||
|
||||
def baseline(
|
||||
q_nope,
|
||||
q_pe,
|
||||
ckv,
|
||||
kpe,
|
||||
output,
|
||||
q_indptr,
|
||||
kv_indptr,
|
||||
kv_indices,
|
||||
kv_lens,
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_heads,
|
||||
head_dim_ckv,
|
||||
head_dim_kpe,
|
||||
page_size,
|
||||
causal,
|
||||
):
|
||||
_ensure_flashinfer_importable()
|
||||
import flashinfer
|
||||
|
||||
sm_scale = 1.0 / ((int(head_dim_ckv) + int(head_dim_kpe)) ** 0.5)
|
||||
workspace_buffer = torch.empty(128 * 1024 * 1024, dtype=torch.uint8, device=q_nope.device)
|
||||
wrapper = flashinfer.mla.BatchMLAPagedAttentionWrapper(workspace_buffer, backend="auto")
|
||||
wrapper.plan(
|
||||
q_indptr,
|
||||
kv_indptr,
|
||||
kv_indices,
|
||||
kv_lens,
|
||||
int(num_heads),
|
||||
int(head_dim_ckv),
|
||||
int(head_dim_kpe),
|
||||
int(page_size),
|
||||
bool(causal),
|
||||
sm_scale,
|
||||
q_nope.dtype,
|
||||
ckv.dtype,
|
||||
)
|
||||
wrapper.run(q_nope, q_pe, ckv, kpe, out=output, return_lse=False)
|
||||
return [
|
||||
q_nope,
|
||||
q_pe,
|
||||
ckv,
|
||||
kpe,
|
||||
output,
|
||||
q_indptr,
|
||||
kv_indptr,
|
||||
kv_indices,
|
||||
kv_lens,
|
||||
batch_size,
|
||||
seq_len,
|
||||
num_heads,
|
||||
head_dim_ckv,
|
||||
head_dim_kpe,
|
||||
page_size,
|
||||
causal,
|
||||
]
|
||||
|
||||
def check(
|
||||
testcase_sizes,
|
||||
original_input_tensors,
|
||||
target_kernel_input_tensors,
|
||||
baseline_input_tensors,
|
||||
rtol=1e-2,
|
||||
atol=1e-2,
|
||||
) -> bool:
|
||||
del testcase_sizes, original_input_tensors
|
||||
output_t = target_kernel_input_tensors[4]
|
||||
output_ref = baseline_input_tensors[4]
|
||||
if output_t.shape != output_ref.shape:
|
||||
print(f"[FAIL] shape mismatch: target {output_t.shape}, ref {output_ref.shape}", file=sys.stderr)
|
||||
return False
|
||||
if output_t.dtype != output_ref.dtype:
|
||||
print(f"[FAIL] dtype mismatch: target {output_t.dtype}, ref {output_ref.dtype}", file=sys.stderr)
|
||||
return False
|
||||
if not torch.allclose(output_t.float(), output_ref.float(), rtol=rtol, atol=atol):
|
||||
diff = (output_t.float() - output_ref.float()).abs()
|
||||
print(
|
||||
f"[FAIL] allclose failed: max_abs_diff={float(diff.max().item()):.6f}, "
|
||||
f"mean_abs_diff={float(diff.mean().item()):.6f} (rtol={rtol}, atol={atol})",
|
||||
file=sys.stderr,
|
||||
)
|
||||
return False
|
||||
return True
|
||||
except Exception:
|
||||
pass
|
||||
|
|
@ -0,0 +1,22 @@
|
|||
---
|
||||
sectionTitle: "题目描述"
|
||||
type: "Text"
|
||||
---
|
||||
你需要实现 FlashInfer MLA paged attention 的 CUDA C++ 前向算子。
|
||||
|
||||
本题使用 `BatchMLAPagedAttentionWrapper` 的 DeepSeek MLA decode 配置：`q_nope` 表示不含 RoPE 的 query，`q_pe` 表示 RoPE 部分的 query，`ckv` 与 `kpe` 分别表示 compressed KV cache 与 RoPE KV cache。MLA page size 固定为 `1`。
|
||||
|
||||
评测程序会调用你提交代码中的 `run_kernel` 函数。你需要根据 `q_indptr`、`kv_indptr`、`kv_indices` 和 `kv_lens` 读取 cache，并将结果写入 `output`。
|
||||
|
||||
baseline 使用 FlashInfer MLA 的 Python API：
|
||||
|
||||
```python
|
||||
wrapper = flashinfer.mla.BatchMLAPagedAttentionWrapper(workspace, backend="auto")
|
||||
wrapper.plan(q_indptr, kv_indptr, kv_indices, kv_lens,
|
||||
num_heads, head_dim_ckv, head_dim_kpe,
|
||||
page_size, False, sm_scale,
|
||||
q_nope.dtype, ckv.dtype)
|
||||
wrapper.run(q_nope, q_pe, ckv, kpe, out=output, return_lse=False)
|
||||
```
|
||||
|
||||
如何提交代码详见[评测指南](/d/2)。
|
||||
|
|
@ -0,0 +1,46 @@
|
|||
---
|
||||
sectionTitle: "接口约定"
|
||||
type: "codeSample"
|
||||
lang: "cuda"
|
||||
---
|
||||
你必须在提交的 CUDA 源码中提供如下 **C 符号**，函数名、参数类型、顺序必须完全一致，并使用 `extern "C"` 防止 name mangling：
|
||||
|
||||
```cpp
|
||||
#include <stdint.h>
|
||||
#include <cuda_bf16.h>
|
||||
|
||||
extern "C" void run_kernel(
|
||||
const __nv_bfloat16* q_nope,
|
||||
const __nv_bfloat16* q_pe,
|
||||
const __nv_bfloat16* ckv,
|
||||
const __nv_bfloat16* kpe,
|
||||
__nv_bfloat16* output,
|
||||
const int32_t* q_indptr,
|
||||
const int32_t* kv_indptr,
|
||||
const int32_t* kv_indices,
|
||||
const int32_t* kv_lens,
|
||||
int64_t batch_size,
|
||||
int64_t seq_len,
|
||||
int64_t num_heads,
|
||||
int64_t head_dim_ckv,
|
||||
int64_t head_dim_kpe,
|
||||
int64_t page_size,
|
||||
int64_t causal
|
||||
);
|
||||
```
|
||||
|
||||
### 参数说明
|
||||
|
||||
* `q_nope`：query 的 compressed/nope 部分，shape `(batch_size, num_heads, head_dim_ckv)`，连续 `bf16`
|
||||
* `q_pe`：query 的 RoPE 部分，shape `(batch_size, num_heads, head_dim_kpe)`，连续 `bf16`
|
||||
* `ckv`：compressed KV cache，shape `(batch_size * seq_len, 1, head_dim_ckv)`，连续 `bf16`
|
||||
* `kpe`：RoPE KV cache，shape `(batch_size * seq_len, 1, head_dim_kpe)`，连续 `bf16`
|
||||
* `output`：输出缓冲区，shape `(batch_size, num_heads, head_dim_ckv)`，连续 `bf16`
|
||||
* `q_indptr`：decode query indptr，shape `(batch_size + 1)`，内容为 `[0, 1, ..., batch_size]`
|
||||
* `kv_indptr`：KV indptr，shape `(batch_size + 1)`，每段长度为 `seq_len`
|
||||
* `kv_indices`：page index，shape `(batch_size * seq_len)`，连续 `int32`
|
||||
* `kv_lens`：每个 batch 的 KV 长度，shape `(batch_size)`，连续 `int32`
|
||||
* `page_size`：评测中固定为 `1`
|
||||
* `causal`：评测中固定为 `0`
|
||||
|
||||
`run_kernel` 内部需要自行计算合适的 launch 配置并启动 CUDA kernel。为保证计时准确，不建议在 `run_kernel` 内部做 `cudaDeviceSynchronize()` 或显式同步。
|
||||
Some files were not shown because too many files have changed in this diff Show More
Loading…
Reference in New Issue