Skip to content

GitHub 历史论文笔记归档 ​

论文总览

来源:KuangjuX/Papers-and-Learning-Notes 的历史 issues,共 34 个论文 issue。

包括 12 篇成段笔记、2 篇未完成摘记和 1 个外部笔记入口。论文阅读状态以主题清单为准。

历史笔记的性能结论和“当前系统”均以发表时的版本与实验背景为限。

补充:Unikraft 的外部文章现已完整本地化,原 issue 来源仍保留。

归档清单 ​

Issue论文主题材料与归档入口
#1Nephele: Extending Virtualization Environments for Cloning Unikernel-based VMs计算基础已有本地笔记;issue 本身仅有来源链接
#2Theseus: an Experiment in Operating System Structure and State Management计算基础论文链接;无笔记正文
#3XPC: Architectural Support for Secure and Efficient Cross Process Call计算基础论文链接;无笔记正文
#4Fast Distributed Deep Learning over RDMA分布式训练与 MoE论文链接;无笔记正文
#5RedLeaf: Isolation and Communication in a Safe Operating System计算基础已有本地笔记;issue 本身仅有来源链接
#6The Demikernel Datapath OS Architecture for Microsecond-scale Datacenter Systems计算基础已有本地笔记;issue 本身仅有来源链接
#7Every Walk’s a Hit: Making Page Walks Single-Access Cache Hits计算基础论文链接;无笔记正文
#8The benefits and costs of writing a POSIX kernel in a high-level language计算基础论文链接;无笔记正文
#9Unikraft: Fast, Specialized Unikernels the Easy Way计算基础本地笔记与来源
#10Alpa: Automating Inter- and Intra-Operator Parallelism for Distributed Deep Learning分布式训练与 MoE论文链接;无笔记正文
#11Ray: A Distributed Framework for Emerging AI Applications分布式训练与 MoE论文链接;无笔记正文
#12SkyBridge: Fast and Secure Inter-Process Communication for Microkernels计算基础论文链接;无笔记正文
#13Firecracker: Lightweight Virtualization for Serverless Applications计算基础论文链接;无笔记正文
#14Keystone: an open framework for architecting trusted execution environments计算基础论文链接;无笔记正文
#15The Deep Learning Compiler: A Comprehensive SurveyGPU 编译与性能已有本地笔记;issue 本身仅有来源链接
#16MLIR: Scaling Compiler Infrastructure for Domain Specific ComputationGPU 编译与性能已有本地笔记;issue 本身仅有来源链接
#17Microsecond-scale Preemption for Concurrent GPU-accelerated DNN Inferences推理与 KV Cache摘要摘记,未完成
#18Fast Segment Anything模型架构与推理能力论文链接;无笔记正文
#19System Virtualization for Neural Processing Units计算基础论文链接;无笔记正文
#20Pushing the Limits of Machine Design: Automated CPU Design with AI计算基础论文链接;无笔记正文
#21Cocktailer: Analyzing and Optimizing Dynamic Control Flow in Deep LearningGPU 编译与性能历史笔记
#22Rammer: Enabling Holistic Deep Learning Compiler Optimizations with rTasksGPU 编译与性能历史笔记
#23The Tensor Algebra CompilerGPU 编译与性能无笔记正文;唯一评论来自他人,仅询问 data layout 论文推荐,不作为个人阅读笔记导入
#24Roller: Fast and Efficient Tensor Compilation for Deep LearningGPU 编译与性能历史笔记
#25Welder: Scheduling Deep Learning Memory Access via Tile-graphGPU 编译与性能历史笔记
#26AStitch: Enabling a New Multi-dimensional Optimization Space for Memory-Intensive ML Training and Inference on Modern SIMT ArchitecturesGPU 编译与性能历史笔记
#27Graphene: An IR for Optimized Tensor Computations on GPUsGPU 编译与性能历史笔记
#28BOLT: BRIDGING THE GAP BETWEEN AUTO-TUNERS AND HARDWARE-NATIVE PERFORMANCEGPU 编译与性能历史笔记
#29PockEngine: Sparse and Efficient Fine-tuning in a Pocket分布式训练与 MoE历史笔记
#30Chimera: An Analytical Optimizing Framework for Effective Compute-intensive Operators FusionGPU 编译与性能历史笔记
#31AMOS: Enabling Automatic Mapping for Tensor Computations On Spatial Accelerators with Hardware AbstractionGPU 编译与性能历史笔记
#32Attention is all you needAttention历史笔记
#33FLASHATTENTION: Fast and Memory-Efficient Exact Attention with IO-AwarenessAttention历史笔记
#34A Loop Transformation Theory and an Algorithm to Maximize Parallelism计算基础开篇摘记,未完成

原 issue 中的推荐与外链 ​

以下短评属于原 issue 作者的记录;保留来源,不将推荐语改写为论文结论。

  • #9:Unikraft 知乎笔记。首次读取返回 403,后续已通过浏览器补入本地全文。
  • #15:system-notes 仓库中的原笔记;当前已有本地版本。
  • #18:原记录为“导师推荐的文章”,仅附 Fast Segment Anything 论文链接。
  • #19:原记录为“导师推荐的关于 NPU 虚拟化的文章”,仅附论文链接。
  • #20:原记录为“一篇比较有意思的使用 AI 自动设计芯片的文章,计算所和寒武纪一起做的”,仅附论文链接。

记录原理、连接知识、积累实践。